(navigation image)
Home American Libraries | Canadian Libraries | Universal Library | Community Texts | Project Gutenberg | Children's Library | Biodiversity Heritage Library | Additional Collections
Search: Advanced Search
Anonymous User (login or join us)
See other formats

Full text of "Memoirs of the California Academy of Sciences"

Nina G. Jablonski, Series Editor 

The Edited Proceedings of a Paul L. and Phyllis Wattis Foundation 
Memoirs of the California Academy of Sciences Number 24 



Edited by Nina G. Jablonski and Leslie C. Aiello 

Wattis Symposium Series in Anthropology 


The Origin and Diversification 
of Language 


Nina G. Jablonski 

California Academy of Sciences, San Francisco, California 

Leslie C. Aiello 
University College London, London, England 

with the editorial assistance of Nancy Gee 

Wattis Symposium Series in Anthropology 


Number 24 

San Francisco, California 
July 17, 1998 


Alan E. Leviton, Editor 

Katie Martin, Managing Editor 

©1998 by the California Academy of Sciences 
Golden Gate Park 
San Francisco, California 94 1 1 8 

All rights reserved. No part of this publication may be reproduced or transmitted in any 
form orby any means, electronic or mechanical, including photocopying, recording, or any in- 
formation storage or retrieval system, without permission in writing from the publisher. 

Library of Congress Catalog Card Number: 98-71454 

ISBN 0-940228-44-0 (cloth) 
ISBN 0-040228-46-7 (paper) 

Printed in the United States of America by the Allen Press 
Distributed by the University of California Press 

The Origin and Diversification 
of Language 


Preface ix 

Animal Communication and Human Language, 

by Peter Maiier 1 

The Foundations of Human Language, by Le^/ze C ^/e//o .... 21 

Comparative Aspects of Human Brain Evolution: 
Scaling, Energy Costs and Confounding Variables, 
by Robert D. Martin 35 

Organization of Semantic Knowledge and the Origin 

of Words in thQ Brain, by Alex Martin 69 

Neanderthals, Modem Humans and the Archaeological 

Evidence for Language, by Paw/ .4. Me//ar5 89 

The Evolution of the Human Language Faculty, 

by Steven Pinker 117 

The Origin and Dispersal of Languages: Linguistic 

Evidence, by Johanna Nichols 127 

The Origins of World Linguistic Diversity: 

An Archaeological Perspective, by Co/m /^e/T/rew 171 

Index 193 


Since the beginnings of humankind, there has been no greater innovation than lan- 
guage. The signal importance of language in human evolution has long been appreci- 
ated, but until recently the tools available for study of the origin and diversification of 
language were limited to relatively simple comparative anatomical and linguistic 
studies, and to studies of language acquisition in apes. The study of the origin and di- 
versification of language today benefits from advances in those areas plus many oth- 
ers, from neurobiology and molecular biology, to cognitive and developmental 
psychology, to animal behavior, archaeology and historical linguistics. 

This short volume represents the edited and peer-reviewed proceedings of the 
Third Wattis Symposium in Anthropology, on the Origin and Diversification of Lan- 
guage, held at the California Academy of Sciences on 12 April 1997. Like the previ- 
ous two Wattis Symposia, the aim of the Third was to address a topic of broad interest 
in anthropology, the illumination of which would benefit from the contributions of 
workers fVom many disciplines. While many approaches could have been taken in the 
organization of the symposium and this volume, coverage of some of the startling ad- 
vances made in recent years in the study of the biological underpinnings of language 
was deemed of high priority. The first half of this book is, thus, devoted to explora- 
fions of the similarities and differences between animal communicafion and human 
language (Marler), the anatomical foundations of languages (Aiello and R.D. Mar- 
tin), and the functional neuroanatomy of the organization of semantic knowledge in 
the brain (A. Martin). In the latter half of the book the archaeological evidence for the 
origin of language is explored (Mellars), followed by an examination of language as 
an adaptation in the evolutionary sense (Pinker). The book is rounded out by two 
chapters on the diversification of languages, one from the point of view of the ar- 
chaeological record (Renfrew), one from that of historical linguistics (Nichols). This 
is intended as a provocative, not a comprehensive, collection of studies of language 
origin and diversification, and one ideally suited for advanced undergraduate or 
graduate students eager to learn of the present and emerging controversies in this im- 
portant field. 

The Third Wattis Symposium in Anthropology and this volume would not have 
been possible without the support of many devoted individuals. Mrs. Phyllis Wattis 
herself deserves greatest thanks for her continued generous financial support and her 
inspiring enthusiasm, which seems only to increase from one symposium to another. 
Within the California Academy of Sciences itself, Nancy Gee of the Department of 
Anthropology is thanked for the work she contributed toward the organization of the 
symposium, but most especially for her painstaking work in typesetting this volume. 

Nina G. Jablonski 

Irvine Chair and Curator of Anthropology 

California Academy of Sciences 

23 March 1998 

Animal Communication and 
Human Language 

Peter Marler 

Center for Animal Behavior 
University of California 
Davis, CA 95616 

A reductionistic approach is taken to some similarities and contrasts between 
language and the natural communicative behavior of animals, especially vocaliza- 
tions of birds and nonhuman primates. Three basic questions are addressed: 1) 
What do animal sounds mean; 2) Do animals "intend" to communicate; and 3) Do 
animals speak in sentences? Using as a point of departure the widespread view 
that animal sounds are all nonsymbolic displays of emotion, a case is made that the 
alarm signals and food calls of some animals are indeed symbolic. They function 
referentially in the sense that they "stand for" something specific in the environ- 
ment such as food or a particular predator. The presence or absence of a potential 
"audience" for their calls affects the readiness of animals to give alarm signals 
when they sight a predator, or food calls when they discover something edible, as 
would be expected if they "intend" to communicate. Some animals, like humans, 
have vocalizations that are learned, and passed by tradition from generation to 
generation. But only humans have the capacity to recombine them into sentences. 
It is true that certain animals, especially songbirds, can recombine learned sounds 
into many different sequences, thus creating large signal repertoires. The se- 
quences are not sentences, however, but affective, nonsymbolic songs, more like 
music than language. 

Anyone who ventures to speculate about the relationship between animal commu- 
nication and language is confronted by obstacles on all sides. How should the com- 
parison be conducted? We are a single species. Animals are legion, and each has its 
own way of communicating, some completely alien to us. What should we compare? 
Only by the most indirect means can we gain access to the mysteries of the electrical 
senses offish, the substrate vibrations that frogs and many insects use to talk to one 
another, the ultrasonic signals of bats and rodents, or the infrasound of elephants. The 
most ubiquitous medium of all for biological communication, the sense of smell, is 
one that many of us choose to mask or ignore. Whenever living things of any species 
gather together, more often than not they eventually communicate, the leading sub- 
ject is usually sex, and the primary vehicles will be, not sounds, but pheromones, per- 
haps sometimes more important in our own lives than we acknowledge. 

The Origin an J Diversification of Language Memoirs of the Calit'omia Academy of Sciences 

Editors. N.G. Jabionski & L.C. Aiello Number 24, Copyright ©1998 


Comparisons of the intricacies of visual communication, both paralinguistic and lin- 
guistic, including signing and facial expressions, in animals and in humans, would 
surely be illuminating {cf. Marler & Evans 1997). But if the subject is commonalties 
and contrasts with language, communication by ear has to be the primary focus. 

Then there are other problems. We are bom and bred as users of language. From 
birth, and perhaps even in the womb, we bring to bear every physiological, behavioral 
and cognitive specialization we possess, as we develop our uniquely human system of 
language and thought. As insiders we have a privileged, intimate view of the almost 
unlimited potential of language, and our insights are authentic to a degree that we can 
never hope to attain with the communication systems of other species. By compari- 
son with language studies, in all of their many guises, investigations of animal com- 
munication are still in their infancy. If we are ever to make any scientific progress, 
our present state of relative ignorance almost forces us to simplify, and focus, not on 
the highest achievements, but on the fundamental principles that underlie communi- 
cation in animals. It is in this reductionistic spirit that I pose three basic questions, 
drawing illustrations from the animal vocal communication systems of the animals 
that I know best, namely birds and monkeys. The first question is what do animal 
sounds mean, if anything at all? Are they just displays of emotion, or is there reason to 
believe that some of their calls serve as symbols? Secondly, do we have any indica- 
tions that animals intend to communicate rather than simply signaling reflexively? 
Finally and most importantly, I will grapple with one aspect of the great linguistic 
theme that we identify especially with the work of Noam Chomsky, the question of 
syntax. Adopting once more a reductionistic approach, I will pose the question: Do 
animals speak in sentences? In what follows I shall discuss only the natural commu- 
nicative behavior of animals. I shall not consider any of the studies in which humans 
have taught animals to perform somewhat human-like behavior, using signing, 
sounds or other tokens as vehicles for communicating with their trainer. In such stud- 
ies it is not easy to determine how much of the behavior comes naturally, and how 
much is inculcated by the experimenter. 

What Do Animal Sounds Mean? 

The thinking of zoologists about the semantics of natural animal calls, especially 
the calls of nonhuman primates, has undergone something of a revolution in the past 
couple of decades. Not so long ago, speculations about how best to interpret the calls 
of monkeys were all based on what Donald Griffin (1992) apdy described as the 
"groans of pain" (GOP) concept of animal communication. The universal assump- 
tion underlying the GOP model is that vocalizations of monkeys and other animals 
are displays of emotion or affect, much like our own facial expressions. Only humans 
are thought to have progressed beyond this condition, and to have achieved symbolic 
signaling. Premack (1975:593) stated the prevailing view clearly and succinctly: 
"Man has both affective and symbolic communication. All other species, except 
when tutored by man, have only the affective form." According to this view, sym- 
bolic signals are presumably those that have identifiable referents in the organism's 
environment to which the signal can be said to refer, in an abstract, noniconic fashion. 
For an animal communication system to qualify as symbolic, information about one 
or more external referents has to be both encoded by signalers and decoded by receiv- 


Premack's is not an idiosyncratic view. Affective or emotional signaling is pre- 
sumed to be a more basic and primitive communicative mode, and as such is a critical 
antithesis to the kind of symbolic functioning that epitomizes language. This view re- 
curs repeatedly in the literamre of many disciplines (Table 1 ). The underlying logic is 
often not made fully explicit, however, complicating the task of generalizing from hu- 
mans to animals, and making it more difficult to decide whether the affective/sym- 
bolic dichotomy does in fact provide a useftil basis for comparing animal and human 
communication. Because it is so widespread, I have taken it as one point of departure 
in analyzing the communicative behavior of animals. There are others, such as semi- 
otic theory, that may prove to be more illuminating in the long run (e.g., Marler 1961, 
1992), making use of basic distinctions between signals as icons, indexical signs and 
symbols (Sebeok 1976). For present purposes, borrowing and adapting from Morris 
(1946), Cherry (1957), and Zivin (1985), I shall use the term ''symbol" to refer to 
noniconic signals that stand for or represent a referent that is external to the organism, 
either currently or from memory, and can be interpreted as such by another organism. 
This usage is somewhat different from that of Smith ( 198 1 ) who defines a referent as 
"anything that becomes knowable or predictable through the performance of a sig- 
nal," presumably whether external or internal. However I shall take the position that a 
symbol cannot refer to itself Thus, a sign or manifestation of an internal state, such as 
a motivational or emotional condition, cannot symbolize itself It is rather an index of 
that state. An internal state can, however, mediate in the production of a symbol with 
a set of external objects and events as its referents if experience of those referents or 
"designate" reliably engenders that state. Signals that lack the property of symboliza- 
tion of external referent I shall call simply nonsymbolic signals. Expressions of emo- 
tion are viewed by many as falling into this category (Table 1), although, as I will 
show, it can be argued that this is not always true. 

If we accept that, by definition, affective signals have no clear or specific sym- 
bolic referents, emotion-based calls are surely widespread in animals, and perhaps 

Table l. interpretations of animal signals as manifestations'of affect, emotion, or motiva tion: State- 
ments from a variety of sources about the meaning of animal signals, especially monkey calls. 

A . "The noises made by monkeys express their mood, and are effective in communicating it to others." 

(Rowell & Hinde 1962:279, zoologists). 

B. "Nonhuman primates can send complex messages about their motivational states; they communicate 

almost nothing about the state of their physical environments." (Lancaster 1965:64, anthropolo- 

C. "All signals (of animals) appear to be clearly related to the immediate emotional states of the signal- 

ing individuals and their levels of arousal." (Bastian 1965:598, linguist). 

D. "The use of both the face and the voice by rhesus monkeys in their natural habitat seems to b e restricted 

to circumstances that connote emotion." (Myers 1976:747-748, neurobiologist). 

E. "The nonhuman primate does not use the auditoi^ medium to communicate whatever concep tual 
knowledge it possesses. The vocal repertoire appears to relate to affective rather than cognitive 
dimensions, the nature of the signal reflecting the emotional disposition of the cal ler." (Marin etal. 
1979: 184, psychiatrists). 

F. "The signal emitted by an animal is an expression of its affective condition, and the reception of the 

signal indicates the infection of others by the same condition — nothing more." (Luria 1982:29, 

G. "Man has both affective and symbolic communication. All other species, except when tu tored by 
man, have only the affective form." (Premack 1975:593, psychologist). 


even the rule. But there are others that do not fit neatly into the GOP mold. A few 
years ago the revisionist process began in earnest with descriptive studies of the re- 
markably rich repertoire of alarm calls of the vervet monkey in Africa by Struhsaker 
(1967). Seyfarth, Cheney and Marler (1980a, b) took the further step of playing tape- 
recordings of alarm calls to free-ranging monkeys. Vervet monkeys live, not like 
most of its dozen or so congeners, in the depths of the African rain forest, but on the 
forest edge. They often venture out on to the savannah, where they are exposed to 
many predators, hence perhaps the enrichment of their alarm call repertoire. Differ- 
ent predators call for different escape strategies and distinct alarm calls aid them in 
deciding which strategy to adopt. Some vervet alarm calls are quite general, simply 
leading companions to become more vigilant, but other calls are so specific that it is 
not unreasonable to think of them as labels or names, such as the leopard call, the 
snake call, or the eagle call. This viewpoint seemed all the more reasonable after a 
long series of playbacks of tape recordings had been conducted in the absence of any 
predators in their natural habitat, in the African bush, at Amboseli, in Kenya. The 
calls elicited those natural reactions that were already known to be specific and appro- 
priate to the particular predator. They differed in ways that made good ecological 
sense, given the different hunting strategies of these predators. The monkeys 
searched the sky and ran into bushes in response to eagle calls, leaped up into the can- 
opy of fever trees in response to leopard calls, and reared up on their hind legs and 
scanned the underbrush when a snake call was played. In other words, there was 
every indication that, in a formalistic sense, the calls served as symbols for the preda- 

Inspired by this new point of view on what animal calls mean, especially as it was 
developed at length by Cheney and Seyfarth (1981, 1985, 1988, 1990), there have 
been many other demonstrations of animal alarm calls that display what came to be 
defined as "functional reference" (Marler, Evans & Hauser 1992; Evans & Marler 
1 995). The calls function as though they "stand for" the object or "referent that they 
represent. In other words, they fiinction as abstract "symbols." They are also "non- 
iconic" in the sense that their acoustic structure has an arbitrary relationship with the 
physical features of the referent. To model what seems to be taking place, we have to 
postulate a formal series of translations or transformations from between (a) the 
predator, (b) the caller's perception and identification of the predator, (c) production 
of the appropriate call, and (d) the response that another monkey gives when the call is 
heard. The question remains, what is really going on in the monkeys' heads? Are 
they cogitating, or acting reflexively and unconsciously? From one point of view, a 
cognitive interpretation of this behavioral sequence seems natural. But in the absence 
of detailed experiments, and without the benefits of introspection, we have no way of 
knowing whether the transformations that go on in a vervet's head when it hears and 
responds to another's alarm call involve cognition, conscious or otherwise, or 
whether they are reflexive, and perhaps even innate, and thus quite un-language like. 
Yet the alarm calls clearly fiinction referential ly, as symbols for both callers and lis- 
teners. It is to capture this concept that the term "functional reference" was coined, 
permitting the issue of reference to be discussed while remaining agnostic on the na- 
ture of the underlying mental and neural processes. 

In addition to alarm calls, some animals also give food calls (Table 2). In both pri- 
mates and birds there are now enough well-documented accounts that we can assume 
that functionally referential food calls are not uncommon in higher vertebrates, con- 
veying as they seem to, not just that food has been found, but with some inkling, un- 
derstood by others, as to the quality and quantity of the food as well (e.g., Marler etal. 


Table 2. Bird and mammal calls with 
tion symbolically. 

'meaning": A sampling ofsome bird and mammal calls that func- 

red jungle fowl 

alarm calls 

food calls 


alaiTTi calls 

food calls 

lapwings (3 spp) 

alarm calls 


alarm calls 


food calls 

rhesus macaques 

food calls 

toque macaques 

food calls 

vervet monkeys 

alarm calls 

diana monkeys 

alarm calls 


food calls 

ring-tailed lemurs 

alarm calls 

Malaysian tree squirrels (3 spp) 

alarm calls 

alpine marmots 

alarm calls 

Collias 1987 

Gyger etal. 1987 
Evans et al. 1 993 
Evans & Marler 1 994 

Walters 1990 

Marler 1959b 

Hauser^ra/. 1993 

Hauser & Marler 1993a, b 

Dittus 1 984 

Seyfarth ei al. 1 980 

Zuberbuhler £>? a/. 1997 

Benz 1993 
Benz et al. 1 992 

Macedonia 1990 

Tamura&Yong 1993 

Boero 1992 

1986a; Elowson et al. 1 99 1 ; review in Hauser 1 996). Food calling is especially wide- 
spread in gallinaceous birds such as pheasants, quail and chickens {e.g., Collias 1 960, 
1 987), studies of which have provided some evidence of deceptive use of food calls to 
attract others when in fact no food is present (Gyger & Marler 1988). Whether the de- 
ception is intentional or the manifestation of a learned habit is moot, and since the 
subjects in these studies were not monkeys but chickens, most observers display a 
chauvinistic reluctance to entertain cognitive interpretations of the behavior of a crea- 
ture as lowly as domestic fowl. 

Aside from the slippery issue of deception, studies of these cases of referential 
alarm and food calls seem to indicate that the GOP point of view is not tenable, assum- 
ing that is, that we accept the widespread view that displays of emotion do not have 
symbolic meanings in the usual sense. This is, in fact, a matter for debate. The key 
point may be, not that emotional displays are completely lacking in semantic mean- 
ing, but rather that there is a difference in the kinds of meanings they encode, perhaps 
more generalized and unspecified than those we tend to view as diagnostic of lan- 
guage. When a person responds to situations they encounter with either a grunt of 
rage or a cry of fear we can be fairly sure that the external situations they have encoun- 
tered are different and we might hazard a guess as to their nature, but at best only in 
very general terms. A wide range of situations can result in production of the very 
same angry grunt. Nevertheless the entire set of situations that merit an angry grunt, if 
we could even characterize it, would be different from the equivalent sets for cries of 
fear or chuckles of pleasure. So there may be some degree of referential ity less spe- 
cific than in emotional displays, but this will typically be much less specific than is 


achievable in language. This relative lack of specificity, compared with what we can 
accomplish with language, may be one factor that gives rise to the judgment shared by 
many (Table 1) that there is little common ground between emotional and symbolic 
behavior. On this basis we may conclude that these cases of highly specific functional 
reference in animal communication provide some indication of a language-like at- 
tribute, but several issues remain equivocal. 

We have only limited information on the role of experience in the development of 
functionally referential communication, an issue that is critical if we are to under- 
stand the role of cognition. Learning does seem to play a role in both linking calling to 
appropriate external stimuli and in developing responses to different alarm call types. 
Seyfarth and Cheney ( 1980) gathered data in the field on what elicits the alarm calls of 
young vervets as they grow up (Figure 1). Eagle calls given by adults are quite spe- 
cific but, in contrast, infants give eagle calls to almost anything moving above in 
free-space, even a falling leaf, but never to a snake or a ground predator, such as a 
leopard. So the relationship between referents and call types sharpens with experi- 
ence, as though the monkeys are developing predator-related concepts, perhaps hint- 
ing at a role for cognition in development. There are also indications of innate 
underpinnings to this behavior. In call production the monkeys behave as though they 
are innately able as youngsters, provoked to alarm calling, to divide up the world of 
predators into several, broad ill-defined classes leaving it to individual and social ex- 
perience to bring each referent class to a focus on a particular call type as they mature. 
The role of innate predispositions raises some uncertainty about whether call produc- 
tion is a truly mindful process based on acquired internal representations of predators. 
On the other hand, there is a growing conviction that a great deal of human cognitive 
processing also has innate underpinnings (Hirschfeld & Gelman 1994; Hauser 1996; 
Pinker 1994). In any case, it seems probable that experience-related mentation plays 
a substantial role in development of how a nonhuman primate receiver responds on 
hearing the call (Cheney & Seyfarth 1990, 1992), even though actual call structure is 
minimally dependent on experience, as with all monkey and ape vocalizations. The 
innateness of the call morphology of monkeys is, of course, a radical contrast with 
speech, but the probable importance of individual experience in the emergence of call 
meaning, is a hint of language-like potential in monkey vocal behavior. 

We have to equivocate on another aspect of the meanings of the naturally occur- 
ring vocal signals of animals. Alarm and food calls have been described as though it is 
reasonable to think of them as labels for things, like predators and food. In fact, we do 
not know what kind of label these calls represent. Our window on what they mean is 
provided by the responses that they elicit in others. As a result, we cannot distinguish 
between the labeling of an object, whether a predator or food, and a prescription for 
the actions relating to that object {cf. Marler 1961, 1992), a distinction that looms 
large in the semiotic theories of Morris ( 1 946). One way to verbal ize the message of a 
leopard alarm call would be "run rapidly away from bushes and up into the nearest fe- 
ver tree," a good move if a leopard is around, because they hunt by stealth and ambush 
and cannot run or climb very rapidly. Alternatively, a more "language-like" interpre- 
tation would be that a leopard call elicits in the mind of the listener an internal repre- 
sentation of a leopard. Then the monkey would decide on the basis of past experience 
what is its best escape strategy, varying with the circumstances in which the individ- 
ual finds itself There are many important issues like this that we cannot yet address, 
with different implications for the kind of cognitive and brain processing that is im- 
plied. Nevertheless linkages between call and referent are more specific that we usu- 
ally associate with emotional displays. 
















■ — i^ 

•O 1) 

u « 

— c 

c ^ 

■ - o. 

T3 1> 

QJ J= 

1> tn 


« uT 


60 -S 


5> > 

/ . 



5: > 


ra oxj 

/ /I 

^ o 

/ / ^ 




/ ID 

2 ^ 




> J= 

s s 





> -J 






c "ah 

\ c 

'^ « 

•o — 

1 \\ -fc 

/ X 

_ij g 

/ \^ 


E li 

\ ^= 



Before leaving the issue of call meaning it is important to acknowledge that non- 
symbolic, affective signals can make a significant contribution to communication in 
some circumstances, as is indicated by this quotation from Norbert Wiener's 1948 
book on cybernetics. 

Suppose 1 find myself in the woods with an intelligent savage, who cannot speak my 
language, and whose language I cannot speak. Even without any code of sign language 
common to the two of us, 1 can learn a great deal from him. All 1 need to do is to be alert 
to those movements when he is showing the signs of emotion or interest. I then cast my 
eyes around, perhaps paying attention to the direction of his glance, and fix in memory 
what I see or hear. It will not be long before I discover the things which seem important 
to him, not because he has communicated them to me by language, but because 1 myself 
have observed them (p. 1 57). 

Thus with no other signaling elements than signs of arousal and the indexical 
property of gaze direction, such behavior has rich communicative potential. The 
theme that we are inclined to underestimate the potential of affective signaling, is ech- 
oed by Premack ( 1975). 

Consider two ways in which you could benefit from my knowledge of the conditions 
next door. I could return and tell you, 'The apples next door are ripe.' Alternatively, I 
could come back from next door chipper and smiling. On still another occasion I could 
return and tell you, 'A tiger is next door." Alternatively, I could return mute with fright, 
disclosing an ashen face and quaking limbs. The same dichotomy could be arranged on 
numerous occasions. I could say, 'The peaches next door are ripe' or say nothing and 
manifest an intemiediate amount of positive affect since I am only moderately fond of 
peaches. Likewise, I might report, 'A snake is next door,' also an intermediate amount 
of affect since I am less shaken by snakes than by tigers (p. 591 ). 

Premack develops fiirther the differences between two kinds of signaling, referen- 
tial (= symbolic) and affective (= excited or aroused), suggesting that information of 
the first kind consists of explicit properties of the world next door while information 
of the second kind consists of affective states, that he assumes to be positive or nega- 
tive and varying in degree. He goes on. 

Since changes in the affective states are caused by changes in the conditions next 
door, the two kinds of information are obviously related. In the simplest case we could 
arrange that exactly the condition referred to in the symbolic communication be the 
cause of the affective state (p. 591 ). 

As Premack indicates, as long as there are perceptible signs of the signaler's state 
of arousal, and some concordance between the preferences and aversions of commu- 
nicants then a significant amount of information can be transmitted by an affective 
system. While he explicitly restricts himself to "whaf rather than "where" one may 
note, harking back to the Wiener quotation, that adding an indexical component to an 
affective signal — pointing or looking — not only indicates where, but also adds a 
highly specific connotation — not apple trees in general, but one in particular. 

While Wiener refers to the potential of an arousal system with a single dimension, 
Premack implicates at least two, a dimension of positive affect concerned with attrac- 
tion, and a dimension of negative affect concerned with apprehension and avoidance. 
While Premack seems to have had human affective signaling in mind, parallels may 
exist in animals. There are indications that a wide range of species, including non- 
human primates, tend to use high-pitched sounds in nonaggressive and fearful situa- 
tions and low-pitched, harsh sounds in aggressive, and attractive situations, in accor- 


dance with what has been proscribed as a set of basic "motivational-structural rules 
(Morton 1977. 1982; Hauser 1993, 1996;Collias 1960), adding further to the commu- 
nicative potential of affective, nonsymbolic signals. However, even with these added 
dimensions, emotion-based models are inadequate to explain the details of the cases 
of alarm calling and food calling behavior we have considered, signals satisfy the cri- 
teria for ftinctional reference, and thus qualify as instances of symbolic communica- 

Do Animals "Intend" to Communicate? 

Whenever professional students oflanguage behavior discuss the subject of lin- 
guistic reference, there is recurrent concern with the issue of intentionality. This is a 
philosophically complex matter that others can deal with better than I {e.g., Cheney & 
Seyfarth 1992, 1996). I will not grapple with it here, except for one simple question. 
When an animal calls, does it care whether or not there is someone in range to hear the 
call? There are in fact some intriguing indications that presence of an "audience" can 
have strong effects on a signaler's behavior (Figure 2). For example, a bird that spots 
food, or catches sight of a predator, may or may not call, depending on whether there 
is a companion nearby to get the message. A male food-calling chicken will call more 
with an audience than without (Table 3), especially if the companion is female, and 
especially if she is a newcomer and thus a potential addition to his harem (Mar\er etal. 
1986a, b; Evans & Marler 1994). 

The modulation of a bird's calling behavior by an audience perhaps hints at an in- 
tent to communicate. Moreover, "audience effects" are not simple. They vary from 
one call to another, in ways that seem to be functionally adaptive, again perhaps sug- 
gesting cognitive complexities (Table 3). For example, there are strong audience ef- 
fects with aerial alarm call signals that are obviously directed at companions, but none 
with the class of ground predator alarm calls, including mobbing calls, aimed at the 
predator as well as at companions (Table 3) (Marler & Evans 1996). 

One possible "cognitive" interpretation of these effects of a signal audience is that 
animals possess an intention to communicate, and a determination to change what the 
other is doing. But alternative interpretations are possible. The behavior could be 
simply reflexive, with several reflexive circuits modulating each calling system in 
different ways. It would be illuminating to know whether, in cases where alarm call- 
ing behavior is audience-modulated, a calling animal is more likely to sound the 
alarm with an audience that is relaxed and apparently unaware of danger, and less 
likely to do so if it is obvious that the audience has itself already detected the danger it 
is in. Even then it is not clear that we could confidently infer intentionality or whether 
we should invoke yet another reflexive circuit. As so often happens with animal stud- 
ies, it is a major and perhaps, as some believe, even an insurmountable challenge, to 
determine for sure what is going on in an animal's head, as indeed would be the case in 
ourselves, if it were not for the benefit of introspection. 

Do Animals Speak in Sentences? 

A primary source of the power of speech is its two-level temporal structure, what 
Hockett (1960) calls, the duality of patterning. The most basic requirement for 
speech-like behavior is a large lexicon of words that can be arranged into many differ- 







Chicken Empty Cage Bobwhite 






C 60 



"I r 

Food item 
Non-food item 

Male Familiar Unfamiliar Empty 

female female Cage 

Figure 2. Histograms of alarm calls (left) and food calls (right) given by cockerels in response to a hawk 
image overhead, and to food. An adjacent cage was either empty or contained a female chicke n or a bob- 
white quail. Most aerial alarm calls are given with a chicken as audience. More food calls are given with a 
female as an audience than with a male, or with no audience. Note that deceptive food calling with a non - 
food item (peanut shell), occurs especially with an unfamiliar hen as an audience (after Karakashian elal. 
l988;Marler<?/a/. 1986b). 

ent sentences, what Pinker (this volume) defines as a "discrete combinatorial sys- 
tem." It follows that there is also a need for an efficient way to produce all of these 
words. The most economical method is to have a small repertoire of distinct articula- 
tor/ gestures or phonemes, averaging up to 40 or so in speech, with no inherent mean- 
ing in themselves, but with the potential to be sequenced in as many different ways as 
you choose. When meanings are attached to these sequences they become words, and 
when words are properly sequenced, you have a sentence, the essence of spoken lan- 
guage. It is useful to have different terms for these two levels of syntactical organiza- 
tion. The higher level, at which words and sentences are meaningful, is appropriately 


Table 3 . A chart of audience effects on the alarm and food cal Is of male chickens. The criteria for an effec - 
tive audience vary from call to call, in haimony with different functional requirements. No te there is no 
audience effect with the ground alarm call, which is addressed to the predator as well as to co mpanions. 

Is there an audience effect? 
Is audience gender relevant? 
Is audience familiarity relevant? 

Alarm Calls 

Food Calls 

Aerial alarm 

Ground alarm 











termed "lexical syntax," involving what I call "lexicodes." The lower level, at which 
meaningless sounds are combined into sequences, may be termed ''phonological syn- 
tax," involving what I call "phonocodes" for short. The distinction between "lexicod- 
ing" and "phonocoding" is useful in making comparisons between animal 
communication and language. Is there any evidence that any of these steps towards 
language has been taken by animals'? 

We can begin with lexicoding at the level of the sentence and work down in reduc- 
tionistic fashion. Evidence has already been presented to show that some animal 
sounds possess meanings in the conventional sense. However, although there are 
cases of animals stringing calls together, the strings all seem to mean the same thing 
(e.g.,Hailman&Ficken 1987; Hailmaner«/. 1985, 1987; Robinson 1984;Mitani& 
Marler 1989). There does not seem to be any recorded natural example of an animal 
unambiguously sequencing calls to make a sentence, where the sequence has a new 
meaning compiled from the meanings of its parts. Natural lexicoding appears to be a 
purely human phenomenon. The only animals that do anything remotely similar have 
been tutored by humans. 

If there are no animal sentences, how about words and phonemes, or their equiva- 
lent? Have animals discovered the trick of phonocoding? The meaningful animal 
signals that I have discussed — alarm calls and food calls of monkeys and birds — all 
seem to come as an indivisible package. They can be repeated, and given quickly or 
slowly, loudly or softly, but there is no obvious analogue to phonological syntax. But 
if we widen the search to include animal vocalizations, not of a referential nature, but 
of a more classical, affective kind, impoverished in referential content, but rich in 
emotional content, here we can find many cases of phonological syntax. Learned 
birdsongs in particular provide us with an abundance of illustrations. A perusal of the 
literature on the structure of learned birdsongs {e.g., Kroodsma & Miller 1982), re- 
veals case after case where unusual ly large repertoires of songs are created by recom- 
bining in different sequences the same basic set of notes, syllables and phrases, 
minimal acoustic units that serve as the bird's equivalent of phonemes and syllables. 
Each species has its own set of phonocoding rules for creating many different songs 
out of these sound units. Two illustrations will serve, one simple, one more complex. 

The male swamp sparrow has two or three simple two-second songs, each consist- 
ing of a string of identical repeated syllables. Unlike the alarm and food calls we have 
discussed, bird songs typically have no referential meaning. As often as not, the onset 
of singing is endogenously motivated, rather than being triggered by something in the 
environment, although certain birds, such as wood warblers, may be an exception. 
Songs ser\'e as self-advertising signals, broadcasting the singer's species, sex and 


population membership, and the fact that he is in an emotional state of sexual and ag- 
gressive arousal. Each swamp sparrow song syllable is made up of two to six differ- 
ent notes, themselves meaningless, arranged in a distinctive cluster (Figure 3). The 
constituent notes are all drawn from a simple species-wide repertoire of six note types 
with a set of "phonocoding" rules for assembling them into a song (Marler & Pickert 
1984). With the right combination you can describe any natural swamp sparrow 
song, just as you can describe the speech patterns of any language with the right com- 
bination of phonemes. Swamp sparrow song thus provides us with a clear case of 
phonological syntax. 

The song of the winter wren is more complicated. A male has a learned repertoire 
of up to 5- 10 song types, each up to 10 seconds in duration. Each song in a male's rep- 
ertoire is different but close inspection reveals phrases that recur again and again, but 
in different sequences (Figure 4). Evidently when a young male learns a set of songs 
from adults, he breaks them down into segments, and then creates variety and en- 
larges the song repertoire by rearranging them in different patterns (Kroodsma 1980; 
Kroodsma & Momose 1991). The mistle thrush engages in very similar behavior 
(Marler 1959a). As in the swamp sparrow, the songs function as affective rather than 
symbolic signals, and the variety is generated, not to diversify meaning, but rather to 
maintain the interest of anyone who is listening, and to alleviate habituation. 

Other songbirds take this process to extreme, creating hundreds of sequences and 
song types, especially, it seems, in species in which incessant male singing has as- 
sumed an important role in attracting and stimulating females. Some, such as the 
mockingbird and its relatives, even incorporate sounds of other species as a device for 
enlarging and diversifying the repertoire (Kroodsma & Miller 1982). The record is 
held by a male brown thrasher, with an individual song repertoire numbered in the 
many hundreds that he uses to bombard the female incessantly (Kroodsma & Parker 

These learned birdsongs present us with one very clear parallel with speech be- 
havior. They make extensive use of the process of syntactical recombination, but of 
course there is a crucial contrast with language. Generally speaking, the song se- 
quences are not meaningfully distinct, in the referential sense. Semantically, each 
winter wren song means the same thing. The variety is introduced, not to enrich 
meaning, but to create diversity for its own sake, to alleviate boredom in singer and 
listener, perhaps with individual differences serving to impress the listener with the 
singer's virtuosity, but not to convey knowledge. 

This is not, of course, to say that the great signal diversification that results is bio- 
logically insignificant. On the contrary, learned birdsongs provide an immensely rich 
source of cues for individual, sexual and species recognition, and for assessment of 
the status of the singer as a potential rival, mate, and provider of resources, as well as 
proffering iconic signs of a bird's current motivational state. But these are separate is- 
sues, distinct from the question of symbolic reference, which is the present concern. 
Also there are cases of songbirds that have functionally distinct types of singing. 
North American wood warblers (Lein 1972; Kroodsma 1981, 1988; Kroodsma €'/«/. 
1989; Stacier 1991; Spector 1992; Byers 1995). Often one song type is used more in 
conflicts between males and another is used more when males are courting females as 
though the two song types express different states of the singing male, one more ag- 
gressive and the other more sexual (Catchpole & Slater 1995). But I do not know of 
any case where different learned song types display contrasting modes of functional 
reference in the sense that production of one type or the other is strictly triggered by 





■vvvvvvv'V wiwNm 




I II Ul IV V \1 







Figure 3. The "phonocoding" iTjles for male swamp sparrow song. Songs are about two seconds long, 
with up to 1 repeated "syllables." A syllable consists of 2 -6 notes, drawn from a species-universal set of 
six note types. The rules for sequencing note types vary with the local dialect. Thus in the New York birds, 
syllables tend to begin with a type one note and end with a type six note. There is an opposi te rule in Minne- 
sota (after Marler & Pickeit 1 984). 

encounters with a male or female. Nor is there any evidence that rearrangements of 
the same set of song elements have distinct referential meanings. 

Learned birdsongs provide many cases of phonocoding, but none of lexicoding, 
an obvious and crucial contrast with language. The use of phonological syntax for 
generating the wonderfully diverse sound patterns of birdsong depends on the 
leamability of the songs. The only other comer of the animal kingdom where we find 
anything equivalent, apart from speech, is in the learned songs of cetaceans. Songs of 
the humpback whale display clear evidence of phonocoding (Payne et al. 1983). I 
know of nothing equivalent in the vast majority of animals whose vocal repertoires 
are innate. It is true that the pant-hooting of chimpanzees, and the songs of gibbons 
are made up of repeated units, but each individual has but one basic pattern. They 
have no song repertoires (Marler & Tenaza 1977). The only parallels I know of occur 
in the innate songs of some New World monkeys, which seem to indulge in phono- 
coding to some degree (Robinson 1979, 1984). 

We can draw the inference that learned vocal behavior, evident in no primate other 
than humans, is an early prerequisite for the emergence of speech behavior. The prob- 
able sequence of evolutionary events thus becomes clearer. As a first step the special 
brain mechanisms required for vocal learning must have evolved in the immediately 
prehuman lineage. Then selective pressures for an enlarged vocal repertoire would 
have led immediately to exploitation of the potential for phonological syntax and an 




explosive increase in the size of the lexicon. But what gave the human brain its unique 
power was not just the capacity, shared with the brains of songbirds and whales, to 
learn and produce new sounds in an infinite number of combinations. Much more re- 
markable was the emergence for the first time ever of the ability to attach new mean- 
ings to newly learned sounds, and, above all, perhaps in response to the growing need 
for communication about tool construction and use, aided and abetted by the demands 
of social intelligence, the brain mechanisms necessary to retain the newly-acquired 
meanings as the sounds were combined into sentences, something that no other or- 
ganism has ever achieved. 


So what emerges from these reductionistically-minded speculations about the 
relationship between animal communication and language? We have seen that al- 
though some animal sounds do have symbolic meanings, these particular signals 
come as an indivisible package, with no underlying combinatorial phonocode. Pho- 
nological syntax can be found in animal signals, however, operating on basically 
similar principles to those underlying speech. It is largely restricted to those few ani- 
mal groups — cetaceans and certain birds — in which part of the natural vocal reper- 
toire is learned, and transmitted by cultural tradition from generation to generation. 
However, these naturally learned animal sounds all appear to function non- 
symbolically, as affective displays. So far as we know, the increased signal diversity 
that results in no way enriches semantic meaning, despite the great enlargement of the 
vocal repertoire that the capacity for phonological syntax makes possible. There is no 
evidence that any creature other than humans has ever taken that further step, and spo- 
ken naturally in sentences. The neuroanatomical and neurophysiological require- 
ments for taking that further step remain an unresolved mystery. Until cognitive 
neuroscience provides us with some notion of the kind of circuitry that is required to 
learn to speak in sentences before we can understand the crucial evolutionary steps in 
the structure and fianctioning of the human brain that made language possible. 


Thanks are due to Marc Hauser for thoughtful and constructive comments on the 
manuscript, to Donald Kroodsma for permission to use his wonderful data on winter 
wren song, and to three anonymous reviewers. 

Literature Cited 

Bastian, J. 1965. Primate signaling systems and human languages. Pages 585-606 in I. De- 
Vore, ed.. Primate Behavior: Field Studies of Monkeys and Apes. Holt, Rinehart and Win- 
ston, New York. 

Benz,J.J. 1993. Food-elicited vocalizations in golden lion tamarins: Design features for repre- 
sentational communication. Anim. Behav. 45:443-455. 

Benz, J. J., D.W. Leger, & J. A. French. 1992. The relation between food preference and food- 
elicited vocalizations in golden lion tanparins {Leontopilhecus rosalia). Jour. Coinp. Psy- 
chol. 106:142-149. 


Boero, D.L. 1992. Alarm calling in Alpine marmot (Mannota mannota L.): Evidence for se- 
mantic communication. EthoL, Ecol. Evol. 4:125-138. 

Byers, B.E. 1995. Song types. Repertoires and song variability in a population of chestnut- 
sided warblers. Condor 97:390-401. 

Catchpole, C.K. & P.J.B. Slater. 1995. Bird Song: Biological Themes and Variations . Cam- 
bridge University Press, Cambridge, U.K. 

Cheney, D.L. & R.M. Seyfarth. 1 98 1 . Selective forces affecting the predator alarm calls ofver- 
vet monkeys. Behaviour 76:25-61. 

. 1985. Vervetmonkey alarm calls: Manipulation through shared information? Behav- 
iour 94:\50-\66. 

— . 1988. Assessment of meaning and the detection of unreliable signals by vervet mon- 

keys. Ani/n. Behav. 36:477^86. 
. 1990. How Monkeys See the World: Inside the Mind of Another Species. Chicago 

University Press, Chicago. 
. 1 992. Meaning, reference, and intentionality in the natural vocalizations of monkeys. 

Pages 315-330 in T. Nishida, W.C. McGrew, P. Marler, M. Pickford, & F. de Waal, eds.. 
Topics in Primatology, Vol. 1, Human Origins. Tokyo University Press, Tokyo. 

Cherry, C. 1957. On Human Communication. John Wiley & Sons Inc., New York. 

Collias, N.E. 1960. An ecological and functional classification of animal sounds. Pages 
368-391 in W.E. Lanyon & W.N. Tavolga, eds.. Animal Sounds and Communication. 
American Institute of Biological Science Pub. No. 7, Washington, D.C. 

. 1987. The vocal repertoire of the redjunglefowl: A spectrograph ic classification and 

the code of communication. Condor 89:5 1 0-524. 

Dittus, W. 1984. Toque macaque food calls: Semantic communication concerning food distri- 
bution in the environment. Anim. Behav. 32:470-477. 

Elowson, A.M., P.L. Tannenbaum, & C.T. Snowdon. 1991. Food-associated calls coirelate 
with food preferences in cotton-top tamarins. Anim. Behav. 42:931-937. 

Evans, C.S., L. Evans, & P. Marler. 1 993. On the meaning of alarm calls: Functional reference 
in an avian vocal system. Anim. Behav. 46:23-38. 

Evans, C.S., & P. Marler. 1994. Food-calling and audience effects in male chickens. Callus 
gallus: Their relationships to food availability, courtship and social facilitation. Anim. Be- 
hav. 47:1159-1170. 

. 1995. Language and animal communication: Parallels and contrasts. Pages 341-382 

in H. Roitblatt, eds.. Comparative Approaches to Cognitive Science. MIT Press, Cam- 

Griffin, D.R. 1992. Animal Minds. Chicago University Press, Chicago. 

Gyger,M.& P. Marler 1988. Food calling in thedomesticfowl(G«//i/.9grt//i«): The roleof ex- 
ternal referents and deception. Anim. Behav. 36:358-365. 

Hailman, J. P. & M.S. Ficken. 1987. Combinatorial animal communication with computable 
syntax: Chick-a-dee calling qualifies as "language" by structural linguistics. Anim. Behav. 

Hailman, J. P., M.S. Ficken, & R.W. Ficken, 1 985. The "chick-a-dee" calls ofParusatricapil- 
his: A recombinant system of animal communication compared with written English. 
Semiotica 56:\9\-224. 

. 1987. Constraints on the structure of combinatorial "chick-a-dee" calls. Ethologv 


Hauser, M.D. 1993. The evolution of nonhuman primate vocalizations: Effects of phylogeny, 
body weight and motivational state. Amer. Nat. 142:528-542. 

. 1996. The Evolution of Communication. MIT Press, Cambridge. 


Hauser, M.D. & P. Marler. 1993a. Food-associated calls to rhesus macaques (Macaco mu- 
latto). 1. Sociological factors. Behav. Ecol. 4:194—205. 
. 1993b. Food-associated calls in rhesus macaques {Macaco mulatto). 11. Costs and 

benefits of call production and suppression. Behav. Ecol. 4:206-212. 
Hauser, M.D., P. Teixidor, L. Field, & R. Flaherty, 1993. Food-elicited calls in chimpanzees: 

Effects of food quantity and divisibility? Anim. Behav. 45:81 7-819. 
Hirschfeid, L.A. & S.A. Gelman 1994. Mapping the Mind: Domain Specificity in Cognition 

and Culture. Cambridge University Press, Cambridge, U.K. 
Hockett, C.F. 1 960. Logical considerations in the study of animal communication. Pages 

392^30 in W.E. Lanyon & W.N. Tavolga, eds.. Animal Sounds and Communication. 

American Institute of Biological Sciences, Washington, D.C. 
Karakashian, S.J.,M.Gyger, & P. Marler. 1988. Audience effects on alarm calling in chickens 

(Callus gallus). Jour. Comp. Psychol. 102:129-135. 
Kroodsma, D.E. 1980. Winter Wren singing behavior: A pinnacle of song complexity. Condor 

. 1981. Geographical variation and functions ofsong types in warblers (Parulidae). Auk 

. 1988. Song types and their use: developmental flexibility of the male Blue-winged 

Warbler. Ethology 79:235-247. 
Kroodsma, D.E., R.C. Bereson, B.E. Byers, & E. Minear. 1989. Use ofsong types by the 

Chestnut-sided Warbler: Evidence for both intra- and inter-sexual functions. Conad. Jour. 

Zool. 67:447-^56. 
Kroodsma, D.E. & E.H. Miller, 1982. Acoustic Communication in Birds. Academic Press, 

New York. 
Kroodsma, D.E. & H. Momose 1991. Songs of the Japanese population of the Winter Wren 

(Troglodytes troglodytes). Co/zt/o/- 93:424-432. 
Kroodsma, D.E. & L.D. Parker. 1977. Vocal virtuosity in the Brown Thrasher. Auk 94:783- 

Lancaster. J. 1965. Primate Behavior and the Emergence of Human Culture. Holt, Rinehart 

Winston, New York. 
Luria, A. 1982. Language and Cognition. MIT Press, Cambridge. 
Macedonia, J. M. 1991. What is communicated in the antipredator calls of lemurs? Evidence 

from playback experiments with ring-tailed and ruffed lemurs. Ethology 86: 1 77-190. 
Marin, O., Schwarz, M.F. & E.M. Saffran. 1979. Origins and distribution of language. Pages 

1 79-2 1 3 in M.S. Gazzaniga, ed.. Handbook of Behavioral Biology II. Plenum Press, New 

Marler, P. 1955. Characteristics of some animal calls. Nature 176:6-7. 
. 1959a. Developments in the study of animal communication. Pages 150-206 and 

329-335 in P.R. Bell, ed., Doi-win 's Biological Work. Cambridge University Press, Cam- 
bridge, U.K. 

. 1959b. The voice of the chaffinch and its function as a language. Ibis 98:231-261. 

. 1961. The logical analysis of animal communication. Jour. Theoret. Biol. 1:295-317. 

. 1992. Functions of arousal and emotion in primate communication: A semiotic ap- 

proach. Pages 235-248 inJ. Nishida, W.C. McGrew, P. Marier, M. Pickford,& F.B.M. de- 
Waal, eds.. Topics in Primatology, Vol. I, Human Origins. University of Tokyo Press, 
Marier, P. & C.S. Evans. 1997. Animal sounds and human faces: Do they have anything in 
common? Pages 133-157 in J. A. Russell & J.M. Fernandez-Dols, eds.. The Psychology of 
Facial E.xpression. Cambridge University Press, Cambridge, U.K. 


Marler, P., A. Dufty, & R. Pickert. 1986a. Vocal communication in the domestic chicken: I. 
Does a sender communicate information about the quality of a food referent to a receiver? 
Anim. Behav. 34:188-193. 

. 1 986b. Vocal communication in the domestic chicken: II. Is a sender sensitive to the 

presence and nature of a receiver? Anim. Behav. 34: 194—198. 

Marler, P., C.S. Evans, & M.D. Hauser. 1992. Animal signals? Reference, motivation or both? 
Pages 66-86 in H. Papoucek, U. Jurgens, & M. Papoucek, eds.. Nonverbal Vocal Commu- 
nication: Comparative and Developmental Approaches. Cambridge University Press, 
Cambridge, U.K. 

Marler, P. & C.S. Evans. 1996. Bird calls: Just emotional displays or something more? Ibis 

Marler, P. & R. Pickert. 1984. Species-universal microstructure in the learned song of the 
swamp sparrow (A/e/o5/7/za georg/aAifl). Anim. Behav. 32:673-689. 

Marler P. & R. Tenaza. 1977. Signaling behavior of apes, with special reference to vocaliza- 
tion. Pages 965-1033 /// T. Sebeok, ed.. How Animals Communicate. Indiana University 
Press, Bloomington. 

Mitani, J.C. & P. Marler. 1989. A phonological analysis of male gibbon singing behavior, fie- 
/zflv/our 109:20-45. 

Morris, C. 1946. Signs, Language and Behavior. Prentice Hall, New York. 

Morton, E.S. 1 977. On the occun^ence and significance of motivation-structural rules in some 
birds and mammal sounds. Amer. Nat. 1 1 1 : 855-869. 

. 1982. Grading, discreteness, redundancy, and motivational-structural rules. Pages 

183—212 in D. Kroodsma& E.H. Miller, eds.. Acoustic Communication in Birds, Vol. I. 
Academic Press, New York. 

Myers, R.E. 1976. Comparative neurology of vocalization and speech: Proof of a dichotomy. 
Pages 745-757 in S.R. Hamad, H.D. Steklis, & L. Lancaster, eds.. Origins and Evolution of 
Language and Speech. New York Academy of Sciences, New York. 

Payne, K., P. Tyack, & R. Payne. 1 983. Progressive changes in the songs of humpback whales 
(Megaptera novaeangliae): A detailed analysis of two seasons in Hawaii. Pages 9-57 in R. 
Payne, ed.. Communication and Behavior of Whales. AAAS Selected Symposia Series. 
Westview Press, Boulder. 

Pinker, S. 1994. The Language Instinct. William Morrow and Company Inc., New York. 

Premack, D. 1975. On the origins of language. Pages 591-605 in M.S. Gazzaniga & C.B. 
Blakemore, eds.. Handbook of Psychobiology. Academic Press, New York. 

Robinson, J. G. 1979. An analysis of the organization of vocal communication in the titi mon- 
key Callicebus moloch. Zeits. Tierpsychol. A9.}i%\^05. 

. 1984. Syntactic structures in the vocalizations of wedge-capped capuchin monkeys. 

Cebiis nigrivittatus. Behaviour 90:46-79. 

Rowell, T.E. & Hinde, R.A. 1 962. Vocal communication in rhesus monkey {Macaca mulatta). 
Proc. Zool. Soc. Lond 138:279-294. 

Sebeok, T. A. 1976. Contributions to the Doctrine of Signs. Indiana University, Bloomington. 

Seyfarth, R.M. & D.L. Cheney. 1 980. The ontogeny of vervet monkey alarm-calling behavior: 
A preliminary report. Zeits. Tierpsychol. 54:37-56. 

Seyfarth, R.M., D.L. Cheney, & P. Marler. 1980a. Monkey responses to three different alarm 
calls: Evidence of predator classification and semantic communication. Science 

Seyfarth, R.M., D.L. Cheney, & P. Marler. 1980b. Vervetmonkey alarm calls: Semantic com- 
munication in a free-ranging primate. Anim. Behav. 28:1070-1094. 

Smith, W.J. 1981. Referents of animal communication. Anim. Behav. 29:]273-\215. 


Spector, D.A. 1992. Wood-warbler song systems: A review of paruline singing behaviors. 

Cun: Orintfiol. 9:]99-23S. 
Stacier, C.A. 1 989. Characteristics, use and significance of two singing behaviors in Grace's 

warbler {Dendroica graciae). Auk 1 06:49-63. 
Struhsaker, T.T. 1967. Auditory communication among vervet monkeys (Cercopitheciis 

aethiops). Pages 281-324 //; S.A. Altmann, ed.. Social Communication Among Primates. 

Chicago University Press, Chicago. 
Tamura, N. & U.S. Young. 1993. Vocalizations in response to predators in three species of 

Malaysian Ca//o5c/»n/5 (Sciuridae). Jour. Mammal. 74:703-714. 
Walters, J. R. 1990. Anti-predator behavior of lapwings: Field evidence of discriminative abili- 
ties. Wilson Bull. 102:49-70. 
Wiener, N. 1948. Cybernetics. Wiley and Sons, New York. 

Zivin,G.,ed. 1985. The Development of E.xpressive Behavior. Academic Press, New York. 
Zuberbuhler, K., R. Noe, & R.M. Seyfarth. 1 997. Diana monkey long-distance calls: Messages 

for conspecifics and predators. Anim. Behav. 53:589-604. 


The Foundations of Human 

Leslie C. Aiello 

Department of Anthropology 
University College London 
Gower Street 
London WC1E6BT, UK 

The foundations of modern human language are often sought in the communi - 
cation capabilities of our closest living relatives, the chimpanzees. Many lines of 
modern research suggest, however, that fully developed human language did not 
appear until after about 500,000 years ago, while at least 5 million years separates 
us from the last common ancestor we share with the chimpanzees. It would be na- 
ive to think that modern human language developed directly from a chimpanzee 
type of communication system without significant intermediary stages. 

It is argued that the most important intermediary stage in the development of 
modern human language had little to do initially with vocal communication. 
Rather it was related to a major transition in early human lifestyle that took place 
about two million years ago in Africa. At this time our ancestors moved from a for- 
ested, or woodland, environment where they were partially tree-living to a more 
open-country lifestyle. This new environment required new biological and behav- 
ioral adaptations, many of which are directly related to vocal communication. 
These included living in larger groups, having larger home ranges, and adopting a 
different diet and locomotor pattern. During this period there was also a major 
increase in brain size. When these adaptations are viewed in the context of the 
archaeological and paleontological records and of other fields such as developmen - 
tal psychology and neuroanatomy, a picture emerges of a human ancestor with 
preadaptations what would better suit it to the later development of modern lan- 
guage than any living primate. 

In the search for the foundations of modem human language, one must clearly dis- 
tinguish between human speech and human language. Human speech is defined as the 
sounds that are characteristic of human language. The ability to produce human 
speech is dependent on the unique structure of the human vocal tract (Figure 1, see 
also Aiello & Dean 1990). Humans have a muscular tongue, a distinctively shaped 
jaw and a larynx which is located low in the throat (Lieberman & Crelin 197 1 ; Lieber- 
manetal. 1992;Duchin 1990). Humans also have the neurological coordination that 
permits the necessary complex articulatory movements of the jaw. lips and tongue in 
respect to the teeth, palate and pharynx. This allows us to produce the wide range of 

The Origin and Divcrsi/icutioii of Language Memoirs of the California Academy of Sciences 

Editors. N.G. Jablonski & L.C. Aielio Number 24, Copyright ©1998 



•s ^ ^ 

C ly— • 

« c S 

-C u 1-4 












































































o O X 

CO .3 CO 




— ^ 













vowels and consonants characteristic of human speech. The production of human 
speech is also not innate. It requires voluntary learning to control and to coordinate 
the vocal apparatus. 

In contrast to speech, human language is our ability to communicate vocally. It is 
characterized by three distinctive features. Firstly, human language uses a finite 
number of sounds to generate an infinite number of meanings (Deacon 1992). 
Sounds, or phonemes, are combined into words and words are combined into sen- 
tences. The order of the sound, or syntax, is fijndamentally important. Second, human 
language is symbolic in nature. The individual phonemes, words or sentences are ar- 
bitrary in relation to the meaning that is being conveyed. And third, the sounds are not 
tied to immediate events. This involves off-line thinking where real or imaginaiy 
events can be discussed with reference to both the past and the future (Bickerton 
1990, 1995). 

In this paper, I will argue that these two aspects of human vocal communication, 
the ability to produce human speech and the ability to produce human language, did 
not appear in our evolutionary history at the same time or for the same reasons. Many 
of the unique anatomical features involved in the ability to produce human speech, as 
well as some of the cognitive precursors of human language, significantly preceded 
the appearance of fully developed modem human language involving syntax, sym- 
bolic reference and off-line thinking. Modem human language is built on foundations 
that extend back over two million years into our evolutionary history. As we will find, 
many of these foundations are the result of adaptations that initially had little, if any- 
thing, to do with vocal communication. 

The Search for the Foundations of Language 

In the search for the foundations of human language, a logical place to begin is 
with the first definite evidence for its appearance. Many authors have argued that the 
only direct evidence for symbolic human language is the presence in the archaeologi- 
cal record of evidence for symbolic thinking. The focus has been on apparent differ- 
ences between the material remains left by the Neanderthals and anatomically 
modem humans in Europe (White 1982; Chase & Dibble 1987;Mellars 1991; Noble 
& Davidson 1991; Davidson & Noble 1993; Milo & Quiatt 1993). In this volume 
Mellars provides a particularly clear discussion of the virtual lack of convincing evi- 
dence for explicit symbolic thinking in Neanderthals. He also notes that the Neander- 
thals apparently lacked any interest in the visual appearance of the tools they made. 
Different types of tools, which were apparently used for different purposes, grade into 
one another without clear-cut divisions. Mellars suggests that this might indicate that 
Neanderthals lacked a highly structured vocabulary which would provide names for 
distinct artifact types. The absence of complexity in many other aspects of Neander- 
thal culture also suggests to him that Neanderthals may have been deficient in the ca- 
pacity to organize and structure their activities. The greater degree of both 
chronological and spatial variation in technology associated with anatomically mod- 
em humans, in contrast to the Neanderthals, suggests to him that early modem hu- 
mans had more sharply defined ethnic divisions. These last two points are significant 
because language is used to transmit rules of complex behavior and specifically de- 
fined cultural traditions. 

The apparent difference between Neanderthal and anatomically modem human 
archaeological remains and inferred linguistic and cognitive abilities are particularly 

24 AlELLO 

relevant in the context of newly published genetic evidence. Krings and his coworkers 
(Krings et al. 1 997) suggest that all living modem humans are more closely related to 
each other than any of us are to the Neanderthals. Their evidence also suggests that the 
split leading to the Neanderthals on the one hand and to modem humans on the other 
occurred about 600,000 years ago. The line leading to modem humans, therefore, 
would have been separate from the line leading to Neanderthals for a significant 
length of time. Different cognitive processes, possibly including language, could 
have developed in the two lineages. This possibility is interesting in the context of 
new analyses of Neanderthal brain size. Whereas previous work emphasized the un- 
usually large brain sizes of the Neanderthals, new estimates of their body masses tell a 
slightly different story (Ruffe/ al. 1997). The large Neanderthal brains are smaller 
than those of modem humans in relation to their considerable body masses. 

It is possible that the Neanderthals may not have had a system of vocal communi- 
cation characterized by any or all of the factors that define modem human language. 
This is particularly interesting because Neanderthals did apparently have a vocal ap- 
paratus that would have been compatible with the production of the full range of 
sounds necessary for human speech. The best evidence for this is the existence of a 
Neanderthal hyoid bone from the site of Kebara in Israel (Arensburg et al. 1989, 
1990). This hyoid is fundamentally modem in form. It implies that the Neanderthal 
larynx was located low in the throat, in a position that would be consistent with the 
production of the full range of sounds characteristic of human speech (Figure 1). 

Analysis of the Neanderthal jaw also shows that it was shaped in such a way as to 
give the muscles that move the tongue proper leverage. Duchin ( 1 990) has argued that 
unless the jaw is short and broad the tongue simply cannot be positioned in the oral 
cavity in the variety of ways necessary to produce the full range of sounds characteris- 
tic of human speech. In her analyses Neanderthal jaws fall clearly within the human 
range of variation. Assuming that Neanderthals had the necessary cortical control of 
their tongues, there is no anatomical reason why they could not have produced human 
speech. Because of the unusually large Neanderthal nasal passages however, their 
speech would most probably have had a considerable nasal quality. 

This evidence for Neanderthal speech, and particularly for the inferred low posi- 
tion of the Neanderthal larynx, directly contradicts the work of Lieberman and his col- 
leagues (see particularly Liebermane/ a/. 1992; Lieberman &Crelin 1971). For more 
than twenty years they have argued that the Neanderthals were different from modem 
humans and similar to all other primates in having a very high larynx and a vocal tract 
incapable of producing the ftill range of human speech. Not only does the previously 
cited evidence argue against this interpretation, but also the conclusions of Lieberman 
and his coworkers have been repeatedly criticized on the basis of a variety of other 
anatomical criteria (see particularly Schepartz 1993; Houghton 1993; McCarthy & D. 
Lieberman, 1997; but see P. Lieberman et al.\992, 1994). 

The evidence presented here is consistent with other arguments that have been 
made for the appearance of the modem human vocal tract even earlier in our evolu- 
tionary history. The descent of the larynx and the change of the shape of the jaw, as 
well as other necessary anatomical and behavioral foundations for human language, 
may have been simple consequences of fundamental morphological and lifestyle 
changes that accompanied the evolution of Homo ergaster (early African //omo erec- 
tiis) almost two million years ago (Aiello 1996a, 1996b). 


Anatomical Foundations For Human Speech 
in the Plio-PIeistocene 

Homo ergaster appears in the fossil record at about 1.8 million years ago and was 
the first known hominin to have resembled modem humans in its overall body pro- 
portions, and particularly in the length of its legs in relation to its inferred body mass 
(Walker & Leakey 1993). It is also the first known hominm to be a dedicated biped, 
lacking all of the features in its postcranial skeleton that indicated arboreal locomo- 
tion in the earlier australopithecines. 

Homo ergaster occupied substantially different habitats than those occupied by 
earlier hominins (Cachel & Harris 1995; Reid 1997). In particular. Homo ergaster 
lived in drier, more open (savanna) parts of the environment (Reid 1997; Stanley 
1992). Earlier australopithecine species are found in fairly wooded, well-watered re- 
gions, while Paranthropus (P. aethiopicus, P. robustus, P. boisei), who first appeared 
about 2.6 million years ago and survived until about 1 .4 million years ago, occupied 
both wooded and open environments, but always near wetlands (Reid 1997). 

Homo ergaster was also the first hominin to have relatively small teeth (McHenry 
1988) and jaws (Wood & Aiello, in press), reflecting a fundamental change in diet. 
This new diet most probably involved significantly greater amounts of animal-based 
products. Such a change in diet is a necessary consequence of the increase in brain 
size observed in early Homo (Aiello & Wheeler 1995; Aiello 1997; Aiello in press). 
The brain is very expensive in metabolic terms, requiring per unit mass over 22 times 
the amount of energy required by an equivalent amount of muscle tissue. In order to 
grow and maintain a large brain, its energy requirements have to be met. Humans do 
not have an unusually high basal metabolic rate to provide energy for our large brains. 
Rather we have enough energy for our relatively large brains because one of our other 
energetically expensive organs, the gut, is relatively small. (Aiello & Wheeler 1995; 
Aiello 1997). A reduction in gut size relative to body size is only possible by moving 
to higher-quality and more easy to digest food. In the case of our early ancestors, high 
quality food that would have been readily available would have been animal-based 
products such as meat, fat and bone marrow. At this stage in human evolution there is 
no evidence of any type of food preparation, such as cooking, that would enhance the 
digestibility of more difficult to digest foods. 

Direct archaeological evidence for a major shift in diet comes from a number of 
different sources. The microwear on the stone tools indicates that they were used at 
least in part for the acquisition and preparation of meat, fat and maiTow (Keeley & 
Toth 1981; Bunn 1981; Potts & Shipman 1981). Analysis of the animal bones left in 
the archaeological sites also shows that the hominins were better able to control parts 
of the landscape and to protect carcasses from scavengers (Monahan 1996). A greater 
variety of game animals are represented in the sites and the animal bone is more frag- 
mentary (Isaac 1975; Cachel & Harris 1995). Cachel and Harris (1995) note that the 
large body size in Homo ergaster, which is 50% greater than the average body mass of 
the australopithecines, also may reflect increasing dietary protein from animal-based 

These major morphological and lifestyle changes characteristic of Homo ergaster 
are fundamental to both human speech and human language. The change in shape and 
robusticity of the jaw resulting from dietary change is particularly important (Aiello 
1 996a, 1 996b). In Duchin's ( 1 990) analysis of jaw shape in relation to speech produc- 
tion, she concluded that Middle Pleistocene Homo erectus, as well as the Neander- 


thals, could have produced human speech. More recently, Oniko (n.d.) has 
demonstrated that other early members of the genus Homo would also have had the 
necessary jaw proportions. In contrast, none of the australopithecines or paran- 
thropines would have had jaw proportions compatible with the production of the full 
range of human speech sounds. 

The small jaw and less projecting face oi Homo ergaster may also be associated 
directly with the descent of the larynx from its high ape-like position to a position 
lower in the throat. In bipedal hominins, the head is balanced on the vertebral column 
which lies under the skull in a more anterior position than it occupies in the apes. In 
Homo ergaster the anterior position of the vertebral column (and foramen magnum) 
together with the reduction of the face and jaws may simply have squeezed the larynx 
into its lower position in the throat (Aiello 1996a, 1996b). Alternatively, reduction in 
the size of the face and the jaw may actually require the descent of the larynx. 

Acoustic features of animal calls communicate cues of individual- and kinship- 
related identity (Owren 1996). These identity features do not involve complex articu- 
lations of the vocal apparatus, but rather result from minor individual variations in the 
actual vocal tract anatomy that produce variations in resonance. A short face reduces 
the length of the vocal tract. The shorter the vocal tract, the higher and more widely 
spaced are the resonance frequencies, and the less effective are the resonance fre- 
quencies in providing cues to individual identity (Owren 1996). A lower larynx may 
simply result from the need to maintain a minimum overall vocal tract length to pre- 
serve the identity-signaling system (Owren 1996). This, together with the advanta- 
geous shape of the jaw resulting from dietary change, would have provided one of the 
first building blocks in the anatomical foundations for the evolution of human speech 
Neither of these features would have been selected for in the context of human speech 
but would have been consequences of the dietary changes that were taking place at 
this time in human evolution. 

By the Early Pleistocene Homo ergaster would, therefore, have had at least two 
anatomical prerequisites for human speech, a low larynx and an advantageously 
shaped jaw. But what was the linguistic ability of Homo ergaster? If these hominins 
did have the morphological adaptations outlined above, were they communicating 
vocally in any way that modem humans might recognize as language? If not, when 
did fully developed modem human language appearand what factors underlie this re- 
markable form of vocal communication? 

Tool Making and the Cognitive Foundations 
for Human Language 

One of the problems in reconstructing the linguistic capabilities of our early an- 
cestors is that we have no living analogue. Homo ergaster was unlike living apes in 
anatomy and habitat preference. Likewise these hominins were unlike modem hu- 

One of the fundamental differences between living apes and modem humans is 
that apes live in the present time. They are unreflective, concrete in their actions and 
situation bound. Their cognitive capabilities are episodic in nature (Donald 1991). 
Any linguistic abilities that they might acquire through laboratory teaching by hu- 
mans, are immediate, short-term responses to the environment. This contrasts with 
human abstract symbolic memory. Our consciousness allows us to think about and 


act on events in the past and plan for the future. It also allows us to engage in collective 
plans and representations with others. 

One of the first steps in the evolution of language is to break the episodic nature of 
cognition and to acquire the ability to produce conscious, self-initiated, representa- 
tional acts. The first tangible evidence of this cognitive breakthrough may well be the 
Acheulian tool tradition (Wynn 1991; Donald 1991). The Acheulian tool tradition 
first appears in Africa at about 1.4 million years ago (Asfaw et al. 1992). It persists 
with an impressive degree of stylistic uniformity in Africa and the western half of 
Eurasia until about 150,000 years ago when it is replaced by more elaborate Middle 
Paleolithic stone tool industries (Gowlett 1992). It is characterized by well formed, 
symmetrical handaxes that have been interpreted to reflect the technical demands of a 
diet based on a higher proportion of animal-based foodstuffs (Shipman & Walker 
1989). In a sense, the Acheulian would have provided the artificial claws and teeth al- 
lowing the early hominins to butcher the animals that they either hunted or scavenged. 

The Acheulian represents a level of cognitive ability over and above the earlier 
Olduwan tool tradition (Wynn 199 1 ). Acheulian hand axes could not have been pro- 
duced by trial and error as could Olduwan chopping tools. Rather they would require 
a clear conscious idea of the shape of the tool and the sequence of tool-making action 
to achieve that shape. This would have involved a level of intentionality and autocue- 
ing that is not apparent in the production of the earlier Olduwan tools and is unknown 
in modem primates. The distribution and character of the slightly earlier stone tools 
from the Okote Member of the Koobi Fora Formation, Kenya {ca. 1.6 million years 
ago) also show evidence of anticipation of a fiiture need for stone and flexible manu- 
facturing strategies based on the size of the raw material available (Rogers 1997). 

In Donald's (1991, 1993) view the cognitive ability reflected in the Acheulian is 
part of a larger complex which he terms mimetic culture. He sees mimesis as a revolu- 
tion in motor skill that "rests on the ability to produce conscious, self-initiated, repre- 
sentational acts that are intentional but not linguistic" (Donald 1991:168). The main 
element in mimesis is the ability to intentionally re-represent a situation and reflect 
on it. This can be done individually, such as rehearsing an event to one's self as in the 
manufacture of an Acheulian tool. Importantly, it also can be done in the context of 
social communication. 

Group Size, Social Intelligence and the Cognitive 
Foundations of Human Language 

In nonhuman primates vocalization is primarily controlled by the limbic system 
and the cingulate gyrus. One of the first steps in the evolution of human language 
would be to give voluntary control to the modulation of vocalizations that are already 
so important in primate sociality (Orwen 1996). The ability to regulate the volume, 
pitch, and tone of the voice for emphasis would have been an important initial step in 
the ability to produce the rapid, consciously controlled, vocalizations necessary for 
human language (Donald 1991). 

Possible selection pressures for increased vocal mimesis may lie with the unique 
open habitat and dietary transition characteristic of Homo ergaster. Those non- 
human primates that today live in open environments tend to have larger group sizes 
than those living in forested environments (Foley 1987). Reasons for this might in- 
clude the increased predator pressure in such open environments or the competition 
with mammalian carnivores for meat (Cachel & Harris 1 995). Protection from both of 


these dangers would be gained through larger group sizes (Aiello & Dunbar 1993). 
Large group sizes, however, pose problems of social cohesion and intragroup compe- 
tition. Nonhuman primates reinforce their social networks through mutual grooming 
and there is a strong correlation between group size and time spent in grooming be- 
havior. Primates cannot spend more than about 20% of their daily time budgets in 
grooming and still have time for other necessary activities such as feeding, resting or 
traveling (Dunbar 1992, 1993, 1994). As group sizes increase and more grooming 
time is required, other activities suffer and this ultimately affects individual survival 
and fitness. Because some primate calls already have the strong social function of sig- 
naling individual identity (Orwen 1996), the conscious exaggeration of calls for 
more emphasis in the context of vocal grooming (Aiello & Dunbar 1993) is a logical 
first step in vocal mimesis (Donald 1 99 1 ). Furthermore, larger group sizes have costs 
in terms of intragroup competition for resources. Brain size (neocortex ratio), group 
size and use of deception all intercorrelate strongly in higher primates (Byrne & 
Whitten 1992; Byrne 1996). Vocal mimesis may have evolved to allow the unemo- 
tional use of calls to manipulate the behavior of others. 

One other aspect of Homo ergaster morphology which may be important in the 
context of an early generalized mimetic adaptation is the locomotor pattern of these 
hominins. Dedicated bipedal locomotion with a total absence of any specific adapta- 
tions for tree living, is first apparent in Homo ergaster. This type of locomotion would 
have freed not only the hands but also the entire upper body from locomotor function. 
Furthermore, the change in body proportions and reduction of the gut would have pro- 
vided these early hominins with a waist and rendered the upper body much more mo- 
bile than is inferred for the earlier australopithecines (Schmid 1991: Aiello & 
Wheeler 1995). The motor-vocal cross modalities that are postulated for mimesis at 
this stage in human evolution with re-representation and autocueing, which would es- 
sentially be replaying events in the mind, may well have also heralded the first ap- 
pearance of rhythm and use of the entire body to reenact, or rehearse, events (Donald 
1991, 1993). In the first instance this may simply have taken the form of rehearsing 
the movements used in stone tool making or playing with vocalizations in the context 
of vocal mimesis. It could have rapidly come to involve the use of the whole body. 
The ubiquitous occurrence of rhythm, dance and song in all human cultures today 
may attest to the deep evolutionary roots of this uniquely human ability in the dietary 
and habitat transition of our early Homo ergaster ancestors. The cross modal mimetic 
ability may also underlie the later evolving human ability to communicate symboli- 
cally using a variety of modalities (voice, gesture and also the written word). 

Primate Social Intelligence and the Evolution of Language 

An important question is why other primates living in large group sizes in open 
habitats have not also developed vocal mimesis? The answer may simply be that they 
did not also undergo a major shift in their diet of the type posUilated for Homo ergas- 
ter. They would have not undergone the anatomical changes connected with adoption 
of a higher quality diet nor would they have been subjected to the complexities of so- 
cial organization presupposed by this dietary transition. These social complexities are 
not simply the elaborate mental maps or more sophisticated organization necessary 
for successful hunting and scavenging postulated by those who normally argue a con- 
nection between feeding behavior and cognitive evolution (Parker & Gibson 1979; 
Clutton-Brock & Harvey 1980; Gibson 1986). There are fijndamental changes in in- 


terpersonal relationships connected with the dietary transition (Hawkes et al. 1997, 
1997a. 1997b; Aiello in press). These fundamental changes are required by the basic 
fact that a high quality diet composed at least in part of animal-based products would 
not be directly accessible to weanlings (Hawkes et al. 1997, 1997a, 1997b). 

In nonhuman primates, weanlings and juveniles tend to forage for themselves. 
Both they and their mothers are limited to resources which can be managed without 
undue skill or learning. A high quality diet which included significant amounts of 
meat, fat or manow. or for that matter plant food requiring extensive extraction or 
preparation, would presuppose regular provisioning of the weanling. This implies a 
significant amount of mother-infant food sharing as well as increased maternal in- 
vestment to train the infant to efficiently obtain the food resource. The period of ma- 
ternal investment in the offspring would be extended significantly past the weaning 
period resulting in increased energetic stress on the female as she came into her next 
fertile period. This has at least two important implications for human evolution. 

The first implication is that it opens up the opportunity for kin other than the 
mother (particularly grandmothers) to enhance their own reproductive fitness by pro- 
visioning the weanling. Hawkes and her coworkers argue that this may well be the 
driving factor behind the evolution of the complex of unique human life-history fea- 
tures including the menopause and extended longevity as well as a relatively late age 
at maturity and the relatively high level of human fertility in relation to the apes 
(Hawkes et al. 1997b; submitted). 

The second important implication revolves around the fact that there was un- 
doubtedly a higher level of adult mortality in human prehistory than in the present 
day. Many (most) females would not have had surviving senior female kin to aid in 
provisioning. Under these circumstances, hominin mothers would have had a strong 
incentive to encourage provisioning from other members of the group and particu- 
larly from males (Aiello in press; Key & Aiello in press). Aspects of female reproduc- 
tive physiology such as concealed ovulation, continuous sexual receptivity and 
reproductive seasonality would be expected to lead to improved levels of attentive- 
ness and investment by the males in the females (Dunbar 1988; Ridley 1989; Powered 
al. 1997). This would also be expected to increase the levels of deceptive behavior be- 
tween the sexes. This is because males and females have fundamentally different and 
potentially conflicting reproductive strategies (Trivers 1972). Whereas females are 
limited in the number of children they can conceive, bear and raise to maturity, males 
are only limited by the number of females that they can inseminate. There would be a 
strong incentive for females to use deceptive tactics to encourage provisioning from 
the male (in possible return for sexual access) while at the same time there would be 
an equally strong incentive for the males to use deceptive behavior to gain sexual ac- 
cess to the female without engaging in the levels of provisioning that might be to her 
best benefit. A transition to a high quality diet would therefore be expected to increase 
the level of interpersonal interaction between members of the group and at the same 
time provide a strong selective pressure for enhanced levels of social intelligence 
above the level observed in living nonhuman primates. 

Social intelligence is a fundamental prerequisite for the origin of fully developed 
modem human language. There are similarities in reasoning processes or procedures 
between primate social intelligence and the computational basis of language process- 
ing including both the semantic aspects of language and syntax (Worden in press). 
Social intelligence has all the fundamental features of language meanings. It is stmc- 
tured, complex and open-ended, discrete valued, extended in space and time, and de- 
pendent on sensory data of all modalities. 


In Worden's view, primates store social events as scripts (or procedures). The 
learning of general rules about events results from the comparison of these scripts. 
The level of social intelligence characterizing a particular species is determined by 
the complexity of the scripts or procedures that can be stored in the brain. This com- 
plexity is determined by both the size of the brain and its design features. For exam- 
ple, the procedures that allow primates with basic social intelligence (such as the 
vei"vet monkeys) to recognize other individuals would also enable them to recognize 
and assign meaning to alarm calls. The more complex social intelligence of the chim- 
panzees, involving the controversial presence of Theory of Mind (the ability to under- 
stand what another individual is thinking) (Povinelli & Preuss 1995), provides the 
procedural basis for the linguistic capabilities evident in laboratory trained animals. 
The highly developed social intelligence of humans, involving higher order Theory 
of Mind, provides the procedural basis for syntax. The procedures that primates use 
for complex social planning and anticipation are, therefore, one and the same as those 
operations that provide the cognitive foundations for human language learning and 

Linguistic Capabilities of Homo ergaster and Beyond 

The transition to a higher quality diet would have had fiindamental morphological 
and cognitive implications for the later development of modem human language. The 
morphological adaptations are driven by the reduction of the face and jaw that corre- 
lates with the reduced chewing requirements of a higher quality animal-based diet. 
Not only can this be directly related to the lowering of the larynx and the potential 
ability to articulate a wider range of vowel sounds, but it can also be related to a 
greater potential for facial expression (Lieberman 1984). The different geometry of 
the Homo ergaster lower face would be expected to have altered the insertion points 
of the facial muscles resulting in a greater range of facial expression. At the same 
time, the tools required by the new diet imply a level of cognitive ability that exceeds 
that seen in living apes or inferred for the earlier australopithecines. As reflected in 
the Acheulian tool tradition, this cognitive ability would have begun to break the epi- 
sodic nature of culture and allow our evolutionary ancestors to re-represent and re- 
flect on past events. If Donald is correct in the cross-modal nature of mimesis, we 
would also expect at this time to have the beginnings of conscious control of vocaliza- 
tions. Initially this would have only involved the ability to modulate and extend exist- 
ing vocalizations for purposes of social control or manipulation. 

On the basis of this reasoning the linguistic capability of Homo ergaster may have 
included more structure than is apparent in the symbolic communication of labora- 
tory trained chimpanzees. This conclusion is based on the inferred more complex 
level of social intelligence required as a consequence of the dietary-based fundamen- 
tal changes in social organization. It is also fair to assume that there would have been a 
wider range of vocalizations available to the early hominins and that these would 
have been under more cortical control than in modem nonhuman primates. There is 
no clear evidence as to the nature of the symbolic content of these vocalizations. Evi- 
dence from laboratoiy trained chimpanzees suggests that they have the capability of 
learning a respectable lexicon and it would be reasonable to assume that the early 
hominins would have had at least this ability. 

The best evidence we have for the evolution of fully developed modem human 
language from this basis is the rapid increase of brain size that gets underway about 


500,000 years ago, over one million years after the first appearance of Homo ergaster 
(Deacon 1992; Aiello 1996b; Ruff era/. 1997). The enlarging human brain would not 
only provide increased memory capacity for an expanded lexicon but also for the re- 
quired procedural templates underlying social intelligence and syntactically based 
language. The prefrontal cortex of the brain, which is particularly large in humans as a 
result of overall brain expansion (Semendeferi et al. 1997), is specifically responsible 
for many features of language production and comprehension as well as the unique 
human ability to reflect on our own mental states and those of others (Theory of Mind) 
(Povinelli & Preuss 1995). 

At present there is no clear reason for this runaway evolution of brain size with its 
implications for language evolution and social intelligence. There is also no clear idea 
of how the Neanderthals fit into this picture. Aiello and Dunbar (1993) have specu- 
lated that the expansion of the brain may well have to do with increased group size and 
intergroup competition, a view which is shared by many others in the field {e.g., Don- 
ald 1991; Pinker 1994). However, one certain thing is that the evolution of increased 
social intelligence would be closely linked with the evolution of language. The reason 
for this is simply that an increased ability to communicate symbolically would be tied 
with the increased ability to cheat. Higher level social intelligence (Theory of Mind) 
is without a doubt an ability to read and manipulate others in order to protect one's 
self against manipulators. The procedural templates involved in this increased social 
intelligence would be the same procedural templates involved in the ability to gener- 
ate more complex linguistic structures. The one would go hand-in-hand with the other 
and both abilities would be expected to evolve rapidly in evolutionary time. 


I would like to thank the Paul L. and Phyllis Wattis Foundation Endowment and 
the California Academy of Sciences for making the 1997 Wattis Symposium on The 
Origin and Diversification of Language possible. I would also like to thank Dr. Nina 
Jablonski for inviting me to participate in this symposium. I would also like to thank 
the following people for discussion and criticism of the arguments presented here: 
Robin Dunbar, Kathleen Gibson, Catherine Key, Bob Martin. Paul Mellars, James 
O'Connell, Michael Orwen, Camilla Power, Todd Preuss, Peter Wheeler, Elizabeth 
Whitcomb, and Robert Worden. 

Literature Cited 

Aiello, L.C. 1996a. Hominine preadaptations for language and cognition. Pages 269-288 in P. 
Mellars & K. Gibson, eds.. Modelling the early human mind, McDonald Institute 
Monograph Series, Cambridge, U.K. 

. 1996b. Terrestriality, bipedalism and the origin of language. Pages 269-289 in W.G. 

Runciman, J. Maynard Smith & R.l.M. Dunbar, eds., Evolution of social behaviour pat- 
terns in primates and man. (Proceedings of the British Academy, vol. 88), The British 
Academy, London, U.K. 

. 1997. Brains and guts in human evolution: the expensive tissue hypothesis. Brazil. 

Jour Genet. 20:141-148 

in press. The expensive tissue hypothesis and the evolution of the human adaptive 

niche: A study in comparative anatomy in J. Bayley, ed.. Proceedings of the Science in Ar- 


chaeologv Conference. HBMC Archaeology Reports Series, English Heritage, London, 

Aiello, L.C. & C. Dean. 1990. An Introduction to Human Evolutionaiy Anatomy. Academic 
Press, London, U.K. 

Aiello, L.C. & R.l.M. Dunbar. 1993. Neocortex size, group size and the evolution of language 
in the hominins. Curr. Anthropol. 34:184-193. 

Aiello, L.C. & P. Wheeler. 1 995. The expensive tissue hypothesis: The brain and the digestive 
system in human and primate evolution. Curr. Anthropol. 36:199-221. 

Arensburg, B., A.M. Tillier, B. Vandermeersch, H. Duday, L.A. Schepartz & Y. Rak. 1 989. A 
Middle Paleolithic human hyoid bone. Nature 338:758-760. 

Arensburg B., L.A. Schepartz, A.M. Tillier, B. Vandermeersch, & Y. Rak. 1 990. A reappraisal 
of the anatomical basis for speech in Middle Paleolithic hominins. Amer. Jour. Phys. An- 
thropol. 83:137-146. 

Asfaw, B., Y. Beyene, G. Suwa, R.C. Walter, T.D. White, G. WoldeGabriel & T. Yemane. 
1992. The earliest Acheulean from Konso-Gardula. Nature 360:732-735. 

Bickerton, D. 1990. Language and Species. University of Chicago Press, Chicago. 

. 1995. Language and Human Behaviour. UCL Press, London, U.K. 

Bunn, H. 1 98 1 . Archaeological evidence for meat eating by Plio-Pleistocene hominins from 
Koobi Fora and Olduvai Gorge. Nature 291:574-577. 

Byrne, R.W. 1996. Relating brain size to intelligence in primates. Pages 1-8 m P.A. Mellars& 
K.R. Gibson, eds.. Modelling the Early Human Mind. Macdonald Institute for Archaeologi- 
cal Research, Cambridge, U.K. 

Byrne, R.W. & A. Whiten. 1992. Cognitive evolution in primates: Evidence from tactical de- 
ception. Man 27:609-627. 

Cachel, S. & J. W.K. Harris ( 1 995) Ranging patterns, land-use and subsistence in Homo erectus 
from the perspective of evolutionary ecology. Pages 51-66 in J.R.F. Bower & S. Sartono, 
eds.. Evolution and Ecology ofWomo erectus (volume I of the Proceedings of the Pithecan- 
thropus Centennial 1893-1993 Congress) Pithecanthropus Centennial Foundation, Leiden 
University, The Netherlands. 

Chase, P.G. & H.L. Dibble. 1 987. Paleolithic symbolism: a review of current evidence and in- 
terpretations. Jour. Anthropol. Archaeol. 6:2632-96. 

Clutton-Brock, T.H. & P. Harvey. 1980. Primates, brains, and ecology. Jour. Zool. London 

Davidson, 1. & W. Noble. 1993. Tools and language in human evolution. Pages 363-398 in 
K.R. Gibson, & T. Ingold, eds.. Tools and Language in Human Evolution . Cambridge Uni- 
versity Press, Cambridge, U.K. 

Deacon, T. 1992. The neural circuitry underlying primate calls and human language. Pages 
121-162 in J. Wind, B. Chiarelli, B. Bichakjian & A. Nocentini, eds. Language Origin: A 
Multidisciplinary Approach. NATO ASl Series - Series D: Behavioural and Social Sci- 
ences, vol. 61. Kluwer Academic Publishers, Dordrecht, The Netherlands. 

Donald, M. 1991. Origins of the Modern Mind. Harvard University Press, Cambridge. 

. 1993. Precis of origins of the modem mind: Three stages in the evolution of culture 

and cognition. Behav. Brain Sci. 16:737-791. 

Duchin, L.E. 1990. The evolution of articulate speech: Comparative anatomy of the oral cavity 
in Pan and Homo. Jour. Human Evol. 19:687-698. 

Dunbar, R.l.M. 1988. Primate Social Systems , Croom Helm, London, U.K. 

. 1992. Neocortex size as a constraint on group size in primates. Jour. Hum. Evol. 


. 1993. Co-evolution of neocortex size, group size and language in humans. Behav. 

Brain Sci. 16:681-735. 


. 1 994. Neocortex size and group size in primates: A test of the hypothesis. Jour. Hum. 

Evol. 28:287-296. 

Foley, R. 1987. Another Unique Species. Longman Scientific and Technical, London, U.K. 

Gibson, K.R. 1986. Cognition, brain size, and the extraction of embedded food resources. 
Pages 93-105 in J.G. Else & P.C. Lee, eds., Primate Ontogeny, Cognition, and Social Be- 
havior. Cambridge University Press, Cambridge, U.K. 

Gowlett, J.A.J. 1 992. Tools — the Palaeolithic record. Pages 350-364 in S. Jones, R. Martin & 
D. Pilbeani, eds. Cambridge Encyclopaedia of Human Evolution. Cambridge University 
Press, Cambridge, U.K. 

Hawkes, K., O'Connell, J.F., & Rogers, L. 1997. The behavioral ecology of modem hunter- 
gatherers, and human evolution. Trends Ecol. Evol. 12:29-32. 

Hawkes, K., J.F. O'Connell & N.G. Blurton Jones. 1997a. Menopause: Evolutionary causes, 
fossil and archaeological consequences. Jour. Hum. Evol. 32:A8-A9 (abstract). 

Hawkes, K., J.F. O'Connell, & N.G. Blurton Jones. 1997b. Hadza women's time allocation, 
offspring provisioning, and the evolution of long post-menopausal lifespans. Curr. Anthro- 
pol. 38:551-577. 

Hawkes, K., J.F. O'Connell, N.G. Blurton Jones, H. Alvarez, & E.L. Chamov, submitted. 
Grandmothering and the evolution of human life histories. 

Houghton, P. 1993. Neandertal supralaryngeal vocal ti'act. Amer. Jour. Phys. Anthropol. 

Isaac, G.Ll. 1975. Stratigraphy and cultural patterns in East Africa during the middle ranges of 
Pleistocene time. Pages 495-542 //; K.W. Butzer & G.Ll. Isaac, eds.. After the Australo- 
pithecines. Mouton Press, The Hague, The Netherlands. 

Keeley, L. & N. Toth. 1981. Microwear polishes on early stone tools from Koobi Fora, Kenya. 
Nature 295:464. 

Key, C. & Aiello, L.C., in press. A prisoner's dilemma model of the evolution of paternal care, 
Folia Primatol. 

Krings, M., A. Stone, R. W. Schmitz, H. Krainitzki, M. Stoneking & S. Paabo. 1 997. Neander- 
thal DNA sequences and the origin of modem humans. Cell 90: 19-30. 

Lieberman, P. 1 984. The Biology and Evolution of Language. Harvard University Press, Cam - 

Liebemian, P. 1 994. Functional tongues and Neanderthal vocal-tract reconstruction — a reply. 
Amer. Jour. Phys. Anthropol. 95:443-450. 

Lieberman, P. & E.S. Crelin. 1971. On the speech of Neanderthal man. Linguistic Inquiry 

Lieberman, P., J.T. Laitman, J.S. Reidenberg & P.J. Gannon. 1992. The anatomy, physiology, 
acoustics and perception of speech: essential elements in analysis of the evolution of human 
speech. Jo»r Hum. Evol. 23:447-468. 

McCarthy, R.C. & D.E. Lieberman. 1 997. Reconstructing vocal tract dimensions from cranial 
base flexion: an ontogenetic comparison of cranial base angulation in humans and chim- 
panzees. Amer. Jour. Phys. Anthropol. Supplement 24:163-164 (abstract). 

McHenry, H.M. 1 988. New estimates of body weight in early hominids and their significance 
to encephalization and megadontia in "robust' australopithecines. Pages 133-148 in F.E. 
Grine, ed. Evolutionary History of the 'Robust' Australopithecines. Aldine de Gruyter, 
New York. 

Mellars, P. 1991. Cognitive changes in the emergence of modem humans. Camb. Archaeol. 
Jour 1:63-76. 

Milo, R.G. & D. Quiatt. 1993. Glottogenesis and anatomically modem Homo sapiens: the evi- 
dence for and implications of a late origin of vocal language. Curr Anthropol. 34:569-598. 


Monahan, CM. 1 996. New zooarchaeological date from Bed II, Olduvai Gorge, Tanzania: im- 
plications for hominid behavior in the Early Pleistocene. Jour. Hum. Evol. 31 :93-128. 

Nobel, W. & 1. Davidson. 1991. The evolutionary emergence of modem human behaviour: 
Language and its archaeology. Man 26:222-253. 

Oniko, A. n.d. The mandible and the evolution of human language. Unpublished Masters The- 
sis. Department of Anthropology, University College London, U.K. 

Orwen, Michael J. n.d. An "acoustic-signature" model of early speech evolution. A paper pre- 
sented at the 1996 Annual Meeting of the Language Origins Society, July 1 1-15, Univer- 
sity of Maryland, Baltimore County. 

Parker, S.T. & K.R. Gibson, 1979. A developmental model for the evolution of language and 
intelligence in early hominids. Behav. Brain Sci. 2:367-407. 

Pinker, S. 1994. The Language Instinct. Penguin Books, London, U.K. 

Potts, R., & P. Shipman. 1981. Cutmarks made by stone tools on bones from Olduvai Gorge, 
Tanzania. Nature 291 : 577-580. 

Power, C, C. Arthur, & L.C. Aiello. 1997. On seasonal reproductive synchrony as an evolu- 
tionary stable strategy in human evolution. Curr. Anthropol. 38:88-91. 

Povinelli, D.J. & T.M. Preuss. 1995. Theory of mind: evolutionary history of a cognitive spe- 
cialization. Trends Neurosci. 18:418-424. 

Reid, K.E. 1997. Early hominid evolution and ecological change through the African Plio- 
Pleistocene. Jour. Hum. Evol. 32:289-322. 

Ridley, M. 1989. The number of males in a primate troop. /l/7//». Behav. 34:1848-1858. 

Rogers, M.J. 1997. Early Homo erectus land use and future planning: Inferences from lithic 
discard patterns across a 1 .6 million-year-old paleolandscape at East Turkana, Kenya. Jour 
Hum. Evol. 32:A17-A18. 

Ruff, C.B., E. Trinkaus, & T.W. Holliday. 1 997. Body mass and encephalization in Pleistocene 
Homo. Nature 387:173-176. 

Schepartz, L.A. 1993. Language and modern human origins. Yearb.Phys. Anthropol. 

Schmid, P. 1 99 1 . The trunk of the australopithecines. Pages 225-234 in Y. Coppens and B. Se- 
nut, eds., Origine(s) de la bipedie chez les Hominides. Editions du CNRS, Paris, France. 

Semendeferi, K. H. Damasio, R. Frank, & G.W. Van Hoesen. 1997. The evolution of the fron- 
tal lobes: A volumetric analysis based on three-dimensional reconstructions of magnetic 
resonance scans of human and ape brains. Jour. Hum. Evol. 32:375-388. 

Shipman, P. & A. Walker. 1 989. The costs of being a predator. Jour. Hum. Evol. 1 8:373-392. 

Stanley, S.M. 1992. An ecological theory forthe origin of //omo.Pa/eo^/o/ogv' 18: 237-257. 

Trivers, R.L. 1972. Parental investment and sexual selection. Pages 136-179 /// B.J. Cambell, 
ed.. Sexual Selection and the Descent of Man, Aldine, Chicago. 

White, R. 1982. Rethinking the Middle/Upper Paleolithic transition. Curr. Anthropol. 

Walker, A. & R.E. Leakey. 1993. The Nariokotome Homo erectus Skeleton. Harvard Univer- 
sity Press, Cambridge. 

Wood, B. A. & L.C. Aiello. in press. Taxonomic and functional implications of mandible-based 
early hominin body mass predictions. Amer Jour Phys. Anthropol. 

Worden, R. in press. The evolution of language from social intelligence. In]. Aitchison, J. Hur- 
ford & C. Knight, eds., Proceedings of the Edinburgh Conference on the Evolution of Hu- 
man Language. Cambridge University Press, Cambridge, U.K. 
Wynn, T. 1991. Tools, grammar and the archaeology of cognition. Camb. Archaeol. Jour. 


Comparative Aspects of Human 
Brain Evolution: Scaling, Energy 
Costs and Confounding Variables 

Robert D. Martin 

Anthropologisches Institut & Museum 
Universitat Zurich-Irchel 
Winterthurerstrasse 190 
CH-8057 Zurich, Switzerland 

Broad comparisons including at least nonhuman primates and perhaps other 
mammals yield vital background information for understanding human evolution, 
not least with respect to the brain. Analyses of many species and characters can re- 
veal general principles applicable to the special case of human evolution and are, 
indeed, essential for testing of explanations based on simple correlations. Biological 
systems are complex and a simple association or correlation between any two fea- 
tures will often not reveal a direct causal connection. Interactions with potentially 
confounding variables must be exhaustively explored in order to establish the crite- 
rion of isolation. Multiple allometric scaling analyses provide a useful method for 
comparative studies, although there are three major practical problems (choice of 
line- fitting procedure, recognition of grade shifts, appropriate control for degree of 
phylogenetic relatedness) in addition to the universal problem of confounding vari - 

One special example of simple association is the apparent temporal coincidence 
of features in the fossil record (e.g., between the earliest finds of stone tools and the 
first evidence of increased brain size) and here the estimation of times of origin is of 
particular importance. Broad comparative evidence indicates that there has been 
systematic underestimation of dates of origin in the primate tree, such that w idely 
cited times of divergence among hominids are probably too recent. Combined with 
mounting evidence of a relatively marked divergence between Neanderthals and 
modern humans, this has direct relevance for interpretations of the latter stages of 
hominid evolution. Because these hominids had slightly bigger brains than us, it is 
automatically assumed that they had comparable capacities. We should, however, 
consider the implications of the fact that brain expansion in Neanderthals was 
probably to some extent a parallel development. As part of a general recalibration 
of times of origin in the primate tree, the inferred time of origin of modern humans 
must also be pushed back in time, and this has implications for the spread of lan- 

Certain human features, including spoken language, are unique and there is no 
direct comparative evidence. Even here, however, comparisons can provide useful 
supporting information. Numerous explanations have been proposed to explain 
the three-fold expansion of human brain size in just a few million years, several in- 

Tlic Origin and Diversification of Language Memoirs oftlie Califomia Academy of Sciences 

Editors. N.G.Jabionski & L.C. Aieilo Number 24. Copyright ©19^8 


volving the development of language, certain aspects of behavior linked to lan- 
guage or a combination thereof. In fact, appropriate analysis of brain size reveals 
that brain expansion (relative to body size) was already under way in australo- 
pithecines, so there is no temporal association with the first appearance of stone 
tools. Further, there is no convincing evidence for a direct link between overall 
brain size and any specific behavioral feature in primates generally, although indi- 
vidual parts of the brain may be linked to specific features. Here, brain evolution is 
explained in terms of a two-phase model in which the overall size of the brain is 
constrained by the resources provided by the mother (maternal energy hypothe- 
sis), while the sizes of individual components may be subjected to selection. Possi- 
ble evidence of specific selection is provided by a strong correlation between the 
size of the neocortex and group size, although the potential influence of confound- 
ing variables has yet to be fully explored. Hence, concatenation of this correlation 
with others to suggest that language evolved as an "inexpensive" substitute for so- 
cial grooming because of the emergence of large group sizes in hominids must be 
treated with great caution. 

This contribution reflects the strongly held conviction that a broad-based com- 
parative framework is essential for well-founded interpretations of the biological ori- 
gins of human language. There are at least two reasons for this. Firstly, comparisons 
should be widely spread across species to generate a more reliable basis for interpre- 
tation of human evolution. In the absence of comparative evidence, it is all too easy to 
fall into the trap of special pleading, basing arguments on their individual plausibility 
rather than on general principles. Even in cases where we are confronted with features 
that are in many ways unique to humans, as is the case with spoken language, interpre- 
tations will be more convincing if we can identify underlying biological principles 
that have general validity. For instance, if we can succeed in establishing a general re- 
lationship across primates (or across mammals generally) between the size of the 
brain (or of some specific component of the brain) and a particular form of behavior 
relevant to the origin of language, we can be far more confident about any conclusions 
that we may draw. Secondly, any analyses should be widely spread across a spectrum 
of characters for the species compared in order to counter the problem of indirect as- 
sociation between features. In analyses restricted to just two features, it is common to 
find a correlation between them, but reliable inference of a causal link requires far 
more detailed investigation. If the two features examined initially are actually de- 
pendent on a third feature that has not been considered, the secondary association be- 
tween them may be misinterpreted if considered in isolation. 

A graphic recent example of a confounding variable underlying a simply correla- 
tion is provided by an analysis that seemed to indicate a significant correlation be- 
tween the initial letter of an author's surname and probability of citation in the 
literature. Names with initial letters earlier in the alphabet showed a higher citation 
rate even after excluding cases in which authors' names were in obvious alphabetical 
order (Tregenza 1997). This was interpreted as evidence for a causal link of some 
kind, and it was even suggested that a correction factor should be applied when con- 
sidering citation rates of individual authors. At least one subsequent attempt was 
made to provide a causal explanation for the observed correlation (Alexander & An- 
drews 1997). However, reanalysis indicated that surnames beginning with initials 
earlier in the alphabet are simply more common. When this confounding factor was 
taken into account, the association between alphabetic order and citation rate was no 
longer significant (Shevlin & Davies 1997). In other words, the reanalysis indicated 
that authors with initial letters earlier in the alphabet are cited more often primarily 
because there are more of them! To quote the authors of that revised analysis: "In or- 


der to assume any causal relation, the condition of isolation must be established, that 
is, the relationship between the two variables exists when isolated from all other con- 
founding variables." Of course, hypotheses concerning causal links should be tested 
with experiments wherever possible. As direct experiments are ruled out when con- 
sidering hypotheses concerning evolutionary history, we should at least test our infer- 
ences through repeated analyses and confrontation with alternative hypotheses, 
including as many species and as many features as we can. 

Morphologically, humans are quite clearly distinguished fi-om other primates in 
three major contexts: dentition, locomotor pattern and brain size. These features can 
be traced through the fossil record. One striking feature is that human brain size is 
both absolutely and relatively the largest among primates, being some three times 
bigger (on average) than in our closest zoological relatives the great apes. Of course, 
humans are equally strikingly distinguished from other mammals by behavioral char- 
acteristics, some of which are connected with biological adaptations and may leave 
traces in the fossil record. For instance, tools manufactured with durable materials 
can be traced comparatively easily through the fossil record. Other human behavioral 
peculiarities such as burial of the dead and artistic creations can similarly leave pre- 
served traces, although the first recorded cases occur relatively late in the record and 
have no obvious link to biological features. Spoken language, a universal and unique 
feature of modern humans, is both fascinating and tantalizing. In the first place, it 
clearly combines both biological and cultural components that are universal features 
of modem humans. There can be little doubt that — as a species-typical feature — hu- 
mans possess a special adaptation of the central nervous system, which is equipped 
with a "universal grammar" to develop a specific language prescribed by the cultural 
environment. This special adaptation is referred to as the "language organ" or "lan- 
guage acquisition device" by Noam Chomsky and as the "language instinct" by Ste- 
ven Pinker (see his chapter in this volume). There are also specific regions of the 
brain, notably Broca's area and Wernicke's area, that have long been recognized as 
having some direct association with language functions. Lateralization, with a con- 
centration of language functions on the left side, is ajso a typical feature of the human 
brain. On the other hand, it is relatively difficult to trace any biological feature of lan- 
guage back through time and we are generally obliged to fall back on indirect infer- 
ence. This being the case, any hypotheses that are advanced must be exhaustively 
tested before we can be confident that any correlations we may find truly indicate a 
causal connection. 

A key feature linking the morphological and behavioral peculiarities of modem 
humans is the brain. It is generally believed that the impressive overall size of the hu- 
man brain must be connected in some way with our wide-ranging behavioral skills, 
notably including spoken language. Indeed, the connection between our large brain 
and language may seem to be almost self-evident. In addition, it seems reasonable to 
expect that certain aspects of the brain that are clearly linked to language, such as lat- 
eralization of brain function and enlargement of specific brain areas, may be traceable 
back through the fossil record. A similar expectation applies to the vocal tract, which 
is directly involved in the production of human speech. 

In this contribution, I shall consider in turn a number of general issues connected 
with a broad comparative approach to the origins of human language. It is important 
first to consider the pattern and timing of primate evolution, particularly with respect 
to the radiation of apes and humans (hominoids). The overall evolutionary tree of the 
primates provides an essential backdrop for broad consideration of, for example, the 
brain and its relationship to behavior. Indeed, knowledge of primate phylogeny is a 


prerequisite for reliable interpretation of comparative evidence. It is also necessary to 
examine the question of times of origin, as reference to specific "established" dates is 
becoming increasingly common in the literature. Certain inferences concerning hu- 
man evolution are based on apparent temporal associations, so errors in inference of 
dates may lead directly to errors in interpretation of such associations. A close focus 
on the radiation of the hominoids is then required to establish a reliable starting-point 
for discussion of early hominid ancestry. There are also certain fundamental issues 
within the hominid branch of the primate phylogenetic tree. One major point con- 
cerns the biological characteristics and relationships of the Neanderthals, considered 
by some to be no more than a subpopulation of the human species but interpreted by 
others as a separate species with distinctive adaptations. Having dealt with these gen- 
eral aspects of the phylogenetic tree of primates, attention will be directed to the issue 
of relative brain size in modern primates in comparison to other mammals. The gen- 
eral relationships that emerge will then be used as a basis for discussing relative brain 
size in hominid evolution. On this foundation, it will be possible to examine various 
explanations that have been proposed to account for differences in relative brain size 
among primate species, notably with respect to evolution of the human brain. Finally, 
evolution of individual brain parts, particularly the neocortex, and a postulated con- 
nection with the development of language will be considered. 

The Pattern and Timing of Primate Evolution 

Broad comparative treatment of primates of course requires an understanding of 
their phylogenetic relationships. There is now a general consensus regarding the gen- 
eral pattern of evolutionary relationships among primates, supported for modem spe- 
cies by a substantial body of molecular data (Martin 1 990; Purvis 1 995 ). Knowledge 
of primate phylogeny is needed not only for general background information but also 
because comparative studies should take into account potentially confounding effects 
of differential degrees of phylogenetic relatedness between the species compared 
(Felsenstein 1985; Harvey & Pagel 1991; see later). 

For discussion of the origins of language in the context of hominid evolution, the 
evolutionary divergence between apes and humans is, of course, of special relevance. 
For the finer relationships among living hominoids there is also a general consensus. 
It is widely accepted that the gibbons branched away first, followed at a later stage by 
the orangutans, and that our closest zoological relatives are the African great apes 
(gorillas and chimpanzees). There is a common tendency to assume that it has now 
been established beyond reasonable doubt that chimpanzees (bonobos and common 
chimpanzees) are closer to humans than gorillas. Because of this, many authors sim- 
ply take chimpanzees as models for discussion of the earliest ancestry of hominids. 
This, however, is unjustifiable on two counts. Firstly, no modem species should ever 
be taken as a direct model for an ancestral condition. Although chimpanzees are un- 
doubtedly more primitive than humans in many respects, they are by no means primi- 
tive in all respects and inference of the earliest ancestral condition for hominids 
requires a carefijl process of character-by-character reconstmction, using a wide 
range of comparative data. Secondly, despite the fact that most molecular studies in- 
dicate a closer relationship between chimpanzees and humans, to the exclusion of go- 
rillas, the evidence is by no means conclusive. The morphological evidence in fact 
indicates that chimpanzees and gorillas are closer to each other than either is to hu- 
mans. For the time being, therefore, it is safer to adopt the more cautious view that the 


relationship between gorillas, chimpanzees and humans is an unresolved trichotomy. 
Arguments depending on a specific relationship between chimpanzees (or bonobos) 
and humans must be treated with caution. 

An overall knowledge of phylogenetic relationships among primates is also im- 
portant in another crucial respect, namely for inference of times of origin of individ- 
ual groups and lineages. Because of the burgeoning number of references to specific 
times of origin in the literature, notably for dates of divergence between African apes 
and humans, it is easily forgotten that such dates are very provisional. In the case of 
molecular data, it is vital to bear in mind a major distinction between the pattern of 
branching relationships within an inferred tree (its topology) and the assignment of 
dates to individual branching-points (nodes) in the tree. Whereas the topology of the 
tree may be based exclusively on analysis of a given set of molecular data and can be 
determined and tested with increasingly sophisticated techniques, dating of nodes al- 
most always depends upon calibration of the tree using at least one date derived from 
the fossil record. Only in the case of limited trees for modern human populations has it 
so far been possible to use certain indications of mutation rates as a basis for inferring 
dates without reference to external paleontological data. Thus, the reliability of the di- 
vergence dates used for almost all molecular trees depends directly on the reliability 
of calibration dates derived from paleontological evidence (often and regrettably just 
one calibration date per tree). 

As noted previously (Martin 1986, 1 990), there is a potentially serious inaccuracy 
in dates of divergence derived from the fossil record in that it is common practice to 
equate the geological age of the first known representative of any group or lineage 
with the actual time of origin. While this may seem to be the only objectively safe ap- 
proach to the problem, it must be recognized that such dates can only indicate mini- 
mum times of origin. If the fossil record is very incomplete and patchy, the actual date 
of origin may be considerably earlier than the first known fossil representative of a 
group or lineage. Using a very simple calculation, it was estimated that less than 3% 
of species in the primate phylogenetic tree are currently known from the fossil record 
and that this low sampling level is likely to lead to a substantial underestimation of di- 
vergence times (Martin 1 993). It was suggested on this basis that the primates of mod- 
ern aspect may have originated some 80 million years ago (Mya), rather than about 55 
Mya, as is still widely assumed. In the absence of a proper model based on realistic as- 
sumptions, the general problem was illustrated with a grossly simplified calculation 
(uniform rate of increase in number of species overtime; uniform survival time of 1 
My per species) using a scaled-down tree. Subsequently, Gingerich and Uhen (1994) 
challenged the interpretation that primates could have originated at such an early 
date. Using a direct probability calculation based on the simple illustration originally 
provided (Martin 1993). they concluded that the ancestor of modem primates could 
not have originated prior to 63 Mya. (Note that this, in itself, is a tacit admission that 
the date of 55 Mya for the earliest known fossil primate of modem aspect could be in- 
creased at least by up to 14.5% in order to obtain the actual time of origin.) Gingerich 
and Uhen went on to infer that the probability of primates originating 80 Mya was less 
than five in a billion. 

Proper assessment of the problems involved in inferring divergence times from a 
very patchy primate fossil record necessitates calculations based on a more compre- 
hensive model than that used by Martin (1993) or Gingerich and Uhen (1994). Sev- 
eral obvious complicating factors should be taken into account. Known fossil species 
tend to be clumped in the phylogenetic tree and therefore cannot be regarded as statis- 
tically independent data points for probability calculations. Further, the average body 


size of primates has tended to increase over time. Because the remains of larger- 
bodied species are generally more likely to be preserved as fossils and are then more 
easily detectable, the probability of fossilization and discovery has presumably also 
increased over time. The shape of the phylogenetic tree will also influence calcula- 
tions. Rather than increasing uniformly over time [(as assumed in the very simple il- 
lustration provided by Martin (1993)], species numbers might conceivably increase 
rapidly at first and then reach an early maximum or, alternatively, increase very 
slowly for some considerable time before reaching a maximum. It should also be 
noted that, regardless of the model tree that is used, underestimation of the time of ori- 
gin will necessarily lead to underestimation of the expected number of fossil species 
to be considered in any calculation and hence to underestimation of the date of origin. 
Finally, the probability of discovery of mammalian fossils is also related to latitude, 
with a bias towards temperate regions and against tropical regions, so it is necessary 
to allow for a special development in the course of primate evolution. Modem pri- 
mates are typically inhabitants of tropical and subtropical forests, now predominantly 
restricted to the southern continents. At the beginning of the Eocene, primates of 
modem aspect appeared relatively abmptly in the fossil record of the northern conti- 
nents and then disappeared equally abmptly at the end of the Eocene. This transitional 
northward expansion of primates undoubtedly reflects an increase in world tempera- 
tures during the Eocene, accompanied by an expansion of tropical and subtropical 
forests. A striking proportion (approximately 12%) of known fossil primate species 
are Eocene forms from the northern hemisphere and the sampling density of the pri- 
mate phylogenetic tree should more properly be calculated separately for fossils oc- 
curring in the present geographical area in which living primates occur. In 
combination, these factors greatly reinforce the likelihood that the origin of primates 
was markedly earlier than has generally been proposed to date. 

Four recent phylogenetic reconstructions using DNA sequence data, all of which 
took the laudable approach of calibrating the ancestral primate node using external 
dates, have provided strong and consistent empirical support for an early origin of pri- 
mates. At the same time, they have provided backing for the proposal that the initial 
diversification of primates was influenced by continental drift (Martin 1990). In the 
first study (Janke et al. 1994), the complete coding region of the mitochondrial ge- 
nome of a marsupial (opossum) was compared with that of 6 placental mammals 
(cow, whale, rat, mouse, seal, human). A phylogenetic tree was inferred from the se- 
quences of eight genes showing approximate constancy of rates of change and cali- 
brated with the widely accepted date of 130 Mya for the divergence between 
marsupials and placentals (undoubtedly a minimum figure). This calibration indi- 
cated the following dates of divergence among placentals: rodents (rat + mousej ver- 
sus cow, whale, seal and human = 1 14 ±15 Mya; cow + whale + seal versus human 
{i.e., primates) = 93 ±12 Mya. 

In the second study. Hedges, Parker, Sibley, and Kumar ( 1 996) used sequence in- 
formation for 48 genes of humans, mice and cattle and for a smaller sample of genes 
for four bird taxa to generate a tree showing divergences between the corresponding 
orders of mammals and birds. They calibrated their tree by taking a date of 3 10 Mya 
for the divergence between birdlike reptiles (diapsids) and mammal-like reptiles 
(synapsids). This age must itself be a minimum date, but it is likely to be reasonably 
reliable because the reptiles concerned were relatively large-bodied {i.e., more likely 
to be preserved and discovered as fossils) and are documented from rich fossil sites. 
On this basis, it was inferred that primates diverged from the other orders of mammals 
examined at least 90 Mya. The resulting calibration of the tree indicates that the initial 


radiations of both birds and mammals began during the mid-Cretaceous, coinciding 
with a period of maximum subdivision of land masses through a combination of con- 
tinental drift and extensive formation of epicontinental seas. 

In a third approach, Amason, Gullberg, Janke, and Xu (1996) used data for com- 
plete mitochondrial DNA sequences to generate a tree showing divergences between 
various mammal species, including a number of primates. They calibrated their tree 
by taking a date of 55 Mya for the first known whale (cetacean). This age must simi- 
larly be a minimum date, but is also likely to be reasonably reliable because cetaceans 
are relatively large-bodied mammals. The calibration applied by Amason, Gullberg, 
Janke, and Xu indicated that primates diverged from other orders of mammals at 
about 90 Mya, thus confirming the estimate provided by Hedges, Parker, Sibley, and 
Kumar (1996). 

Finally, combined analysis of DNA sequences from three mitochondrial genes 
and two nuclear genes indicates that adaptive radiation from a specific common an- 
cestor gave rise to a group of African mammals containing golden moles, hyraxes, 
manatees, elephants, elephant shrews and aardvarks (Springer e/ al. 1997). The re- 
sults "suggest that the base of this radiation occurred during Africa's window of isola- 
tion in the Cretaceous period before land connections were developed with Europe in 
the early Cenozoic era." The mean divergence time between this African group of 
mammals and other orders of mammals (including primates) is estimated at about 90 

Recalibration of the time of origin of primates is of crucial significance because it 
has spin-off effects for dating of nodes throughout the primate tree, including the date 
of the split between hominids and African great apes and the estimated time of origin 
of Homo sapiens. It now seems highly likely that hominids originated at an earlier 
date than has commonly been assumed in recent discussions, and this must be borne 
in mind when discussing inferred dates of origin for human languages in relation to 
the fossil record. For instance, if the inferred age of the common ancestor of all mod- 
em humans is increased by 50% or more through recalibration, this has radical impli- 
cations for comparison with known fossil specimens, the archaeological record and 
inference of temporal relationships between language families. 

Some Fundamental Issues in Human Evolution 

Much has been written about attempts to define what is meant by the term "hu- 
man" in an evolutionary context and to pinpoint the emergence of humankind. This is, 
among other things, reflected in discussions of the emergence of the genus Homo in 
the fossil record. In biological terms, however, it must be accepted that clear defini- 
tion of the emergence of "humankind" is a futile exercise. Any phylogenetic tree is 
characterized not only by discontinuities between species, which arise through the 
process of speciation, but also by continuity along lineages, reflecting direct descent 
from parent to offspring. Whatever process of speciation one may propose, the fact 
remains that in any arbitrary subdivision of a continuous line of descent the parents of 
the first humans must be the last apes in that lineage. Although we can of course list 
several unique biological features distinguishing modem humans from modem Afri- 
can great apes, those features must have appeared progressively and without interrup- 
tion over the course of evolutionary time. Further, the well-known phenomenon of 
mosaic evolution applies to hominid phylogeny just as it does to the evolution of any 
other group of organisms. The features that now distinguish humans from great apes 


did not emerge in concert at one particular point in time. Changes in the dentition, lo- 
comotor apparatus and relative brain size began early and progressed at different 
rates, while special features such as the manufacture of durable tools and spoken lan- 
guage are relatively late de\ elopments. 

Against this background, attempts to define the emergence of human uniqueness 
at a particular point in the hominid evolutionar\' tree are doomed to failure. One good 
example is provided by attempts to recognize a particular brain size (commonly as- 
sessed through cranial capacity) as defining the threshold to humanity, the "cerebral 
rubicon" (Vallois 1954). This threshold was set by Vallois at 800 cc. but was subse- 
quently lowered to 600 cc in the classical paper defining //omo habilis (Leakey etal. 
1964). Such arbitrary definitions, which reflect a belief in a firm connection between 
brain size and behavioral capacity, are of little value. In the first place, there is an al- 
most twofold variation in cranial capacity within hominoid species generally. In 
modem humans, for example, cranial capacitv' can vary between about 1000 cc and 
2000 cc. Further, there is a difference betv^een the sexes, reflecting sexual dimor- 
phism in body size. In modem humans, although sexual dimorphism in body size is 
relatively limited, average cranial capacity is about 10% greater in males than in fe- 
males. Presumably, some degree of sexual dimorphism in body and brain size was 
present in early hominids. Thus, if the "cerebral mbicon" is defined on the basis of av- 
erage brain size, rather than on the relative size of any individual's brain, the axerage 
female must have attained human status some tens of thousands of years later than the 
a\erage male! 

Of course, cranial capacity is but one example of the many attempts that have been 
made to define human uniqueness according to criteria that may or may not be recog- 
nizable in the fossil record. Language is clearly a strong potential candidate here. Al- 
though there has been a great deal of work on the training of great apes to use sign 
"language" (see Savage-Rumbaugh 1986 for a review), it is clear that there is a major 
gulf between the best performance of great apes and human language (Terrace 1979; 
Sebeok & Umiker-Sebeok 1980: Seidenberg & Pettito 1987). It would therefore be of 
great benefit if morphological features associated with spoken language could be 
traced back through the fossil record. There is. for example, the possibility of detect- 
ing signs of brain lateralization in fossil hominids. Greater development of the cere- 
bral cortex on one side of the brain (a petalia) can be detected on brain casts and three 
studies have reported asymmetry in endocasts of fossil hominids (LeMay 1976; Hol- 
low ay & de Lacoste-Lareymondie 1982: Holloway 1988). In fact, however, some de- 
gree of cerebral asymmetrv' has been reported for great apes (LeMay et al. 1982). 
Further, asymmetry in the lengths of certain sulci indicates that in some Old World 
monkeys the prefrontal and parietal cortical areas are significantly larger on the left 
side of the brain (Talk 1978). It has. for example, been demonstrated (contrary to ear- 
lier reports) that there is a characteristic difference in size of the Sylvian sulcus be- 
tween left and right hemispheres in macaques (Falk et al. 1986). This external 
evidence of brain lateralization in macaques links up with demonstration of domi- 
nance of the left hemisphere in processing of vocal communication (Heffner & Heff- 
ner 1984; see also Falk 1987). Hence, some form of lateralization seems to have been 
already present in Old World monkeys and apes long before the origin of hominids. 
Indeed, consistent lateralization in sulcal dimensions has also been found in a New 
World monkey, the capuchin, although it was not detectable in another neotropical 
species, the spider monkey (Gilissen 1 992). Thus, some kind of externally obser\ able 
difference between left and right hemispheres is found in several nonhuman primates 
and certain pattems seem to have arisen independently in at least two separate linea- 


ges. Nevertheless, it should be noted that Holloway and de Lacoste-Lareymondie 
(1982) went beyond the mere recognition of cerebral asymmetry in fossil hominids. 
They made the point that modem humans and fossil hominids are distinguished from 
other primates by a combination of petalia of the left occipital cortex with petalia of 
the right frontal cortex. Although individual petalias may occur in nonhuman pri- 
mates, this particular combination is apparently restricted to hominids. 

In any case, caution is needed with respect to interpretations of brain lateraliza- 
tion. It has recently been suggested that this phenomenon may be a secondary conse- 
quence of large brain size, with hemispheric specialization arising as a consequence 
of delay in conduction between the two halves of the brain (Ringo etal. 1994). Should 
this interpretation be borne out, the causal link between brain lateralization and lan- 
guage might not be so direct as has commonly been assumed and the association 
could (at least in part) be a secondary consequence of the overall size of the brain. The 
apparent causal link between hemispheric specialization and language may conceiva- 
bly reflect yet another indirect correlation. 

An alternative possibility is that features associated with the vocal tract and 
speech may be detectable in the fossil record. Humans are once again unique in com- 
parison to all other primates in that there is postnatal descent of the larynx producing a 
unique configuration of the oral region that is related to the production of various 
sound components in human speech. There have been suggestions that recognizable 
features of the cranial base indicate whether descent of the larynx has occurred and 
that Neanderthal skulls lack these features (Lieberman et al. 1972; Laitman et al. 
1979; Lieberman 1984; Crelin 1987). However, this interpretation has been ques- 
tioned (Falk 1975; Holloway 1983; Houghton 1993) and a fossilized hyoid bone from 
a Neanderthal failed to reveal any significant difference from the modem human con- 
dition (Arensburg et al. 1 988). In a modified interpretation, Laitman ( 1 985) linked re- 
organization of the human lar>'nx to the development of basicranial flexion and traced 
the beginnings of such flexion back to Homo erectiis, although it is by no means as 
pronounced as in Homo sapiens. On a different tack altogether, MacLamon (1993) 
compared dimensions of the vertebral canal between modem humans and the 
Nariokotome //omo erectus specimen. A unique feature of modem humans in com- 
parison to nonhuman primates is that the thoracic part of the canal is enlarged. The 
specific thoracic enlargement is not seen in the Homo erectus specimen. This differ- 
ence between Homo erectus and Homo sapiens could simply be due to adaptation for 
increased muscular movement or control of the trunk. An alternative possibility, 
however, is that it is associated with increased muscular control of breathing and 
hence with the development of speech. This evidence, therefore, might indicate that 
Homo erectus was not capable of speech. In stark contrast, Tobias (1987) has argued 
that several features of brain casts in Homo hahilis (marked increase in overall size, 
pronounced transverse expansion of the cerebmm, inferred presence of Broca's and 
Wernicke's areas) indicate a "new level of organization in cerebral evolution." On 
this basis, he specifically suggests that the development of human language began 
with Homo habilis. The search for fossilizable indicators of the emergence of human 
language continues, but it seems very likely that — as with so many other features of 
human evolution — the process was a gradual one and that no clear threshold to hu- 
manity will be recognizable. 

One particular aspect of hominid phylogeny that is of special importance with re- 
spect to the origin of language and its potential relationship to the evolution of the 
brain is the status of the Neanderthals. Although many authors have interpreted the 
Neanderthals as a subpopulation of the species Homo sapiens, accumulating evi- 


dence has indicated that they were quite distinctive in many respects and should be re- 
garded as a separate species. Numerous morphological differences separating adult 
Neanderthals from adult modem humans have been reported {e.g., Rak 1986, 1990; 
Stringer & Gamble 1993; HubVm etal. 1996: Schwartz & Tattersall 1996; Rak etal. 
1 996). Even more significantly, morphological differences have been found to distin- 
guish very young Neanderthals from modem humans of comparable age {e.g., Zol- 
likofer et al. 1 995, in press), further increasing the likelihood of a clear separation. In 
the light of this accumulated morphological evidence, recognition of a separate spe- 
cies Homo neanderthalensis has increasingly seemed to be justified. This conclusion 
has now received strong support from an analysis of mitochondrial DNA extracted 
from the Neanderthal type specimen (Krings et al. 1997). There is a marked differ- 
ence between the Neanderthal DNA and that of modem humans and the time of sepa- 
ration between the two lineages has been estimated at a minimum of 600,000 years 
ago. It should be emphasized that this is most certainly a minimum figure. In the light 
of the comments made above about the need for recalibration of the entire phyloge- 
netic tree of primates, it seems quite likely that the time of separation between Homo 
sapiens and Homo neanderthalensis may be pushed back to one million years ago or 
even earlier. 

Such a clear separation between Homo neanderthalensis and Homo sapiens has a 
number of interesting implications that have been obscured by the long-standing in- 
terpretation that Neanderthals are no more than an extreme variant of the modem hu- 
man condition. Once we have accepted that Neanderthals were a separate species, it is 
a logical next step to ask how deep the separation goes. Because of a general tendency 
to equate the human condition with possession of a large brain, a clear distinction be- 
tween modem humans and Neanderthals (whose average brain size was in fact some- 
what greater than in present-day humans) seemed inherently unlikely. However, we 
should now consider the possibility that expansion of the brain took place in parallel, 
at least to some extent, during the evolution of Neanderthals and that the shared fea- 
ture of a large brain size may in fact make Neanderthals seem closer to us than they 
really are. Indeed, if a major part of the expansion of the Neanderthal brain took place 
as a parallel phenomenon, we should contemplate the possibility that the intemal or- 
ganization of that brain may have differed in important respects, perhaps including 
the degree of adaptation for spoken language. As a starting point in examining this in- 
triguing possibility, it would be worthwhile to see whether there are any consistent 
differences in extemal morphology of endocranial casts between Neanderthals and 
modem humans. 

Relative Brain Size in Mammals 

The brain has always occupied a central place in discussions of human evolution, 
not least because it is obviously connected in some way with behavior. Assessment of 
the evolution of the brain is, however, a complex matter. In comparisons between spe- 
cies, it is clear that absolute brain size is not a usefiil indicator of behavioral capacity. 
Brain size generally increases with body size such that an elephant, for example, has a 
brain approximately four times bigger than the human brain. A simple ratio between 
brain size and body size is also uninformative. The relationship between brain size 
and body size is curvilinear. As a result, although brain size generally increases with 
body size, the ratio between brain size and body size progressively declines. In the 
lesser mouse lemur (M/croceZ)W5), one of the most primitive living primates, the brain 
represents 3% of body mass, whereas in humans it represents only 2%. 



A practical solution to this problem is provided by allometric analysis, which is 
based on a simple general scaling model for comparative quantitative studies. This 
approach is. in essence, relatively straightforward. The relationship between body 
size (X) and any feature of interest (Y). such as brain size, is taken to conform ap- 
proximately to a power function of the form: Y = k • X^ (where a is the scaling expo- 
nent and k the scaling coefficient). This equation can be converted to a linear 
relationship through the simple expedient of converting the X and Y values to loga- 
rithmic form, yielding the following formula: log Y =a- logX + log k. A best-fit line 
detemiined for a given data set can be taken as describing the general scaling trend, 
while deviations of individual species above or below the line (reflected by positive 
or negative residual values) indicate special adaptations of individual species. The re- 
sidual values can therefore be taken as an indicator of relative brain size. (It should be 
noted that, for technical reasons, residual values should be kept in logarithmic form as 
far as possible. One graphic illustration of the problems that can otherwise arise is 
provided later.) While it should be noted that estimates of relative brain size differ ac- 
cording to the data base used and the taxonomic level of analysis (Holloway & Post 
1 982). use of this concept undoubtedly represents a major advance over mere reliance 
on absolute brain size or on simple ratios to body size. 

Although the basic concept underlying allometric analysis is theoretically quite 
straightfoi-ward, there are in practice at least three major problems that largely con- 
cern statistical aspects (Figure I). These issues have been extensively discussed else- 
where and will only be briefly summarized here. Firstly, there is continuing 

6 8 10 

X variable 


FIGURE 1. Illustration of three tundamenlal problems in allometric analyses: Problem 1 (above graph): The 
choice of best-tit line can affect the results of allometric analysis, particularly if the re is considerable scat- 
ter in the data. In the least-squares regression (continuous line), the sum of the vertical deviations of indi- 
vidual points {e.g., V) is minimized, in this inherently asymmetrical approach, it is assumed that the X 
variable is independent and free of eiTor. In the reduced major axis (hatched line), the sum of areas of the 
triangles subtended by individual points {e.g., T) is minimized. In this approach, no a priori distinction is 
made between the variables. 



40 - 


30 - 

^^^^'^ GRADE 2 

o .-* 



20 - 
10 - 

9 **** GRADE 1 




T 1 1 1 1 1 

8 10 12 14 

X variable 




.2 ^° 








, 1-^^>T^ 

1 1 1 1 1 


2 4 6 8 10 12 14 

X variable 

Figure l {continued). Problem 2 (top graph): It is possible that a given data set includes distinct subsets 
(grades). Commonly, the subsets show similar scaling patterns, reflected by close similarit y in the slopes of 
the lines (scaling exponent a), but they are separated by a vertical displacement (grade sh itt). Fitting of a 
single line to a date set containing grades is likely to generate misleading results 

Problem 3 (bottom graph): Statistical testing of scaling differences between species is subject to potential 
bias due to the differential degi"ees of phylogenetic relatedness between them. 


discussion about the most suitable procedure for determining a best-fit line to de- 
scribe a scaling relationship. Discussions have mainly revolved around choice be- 
tv^een the least-squares regression, the major axis and the reduced major axis (Harvey 
& Mace 1982; Martin & Barbour 1989). In the analyses reported here, the reduced 
major axis has generally been taken as the line of best fit, but the use of an alternative 
procedure would not have affected the results presented. Secondly, it is possible that a 
given data set may contain subgroups of species, separated by grade differences 
(Martin 1989a, 1996). If grade differences are present, fitting of a single best-fit line 
to the data set concerned may yield misleading results. Examples of grade differences 
are provided in the following text. Finally, it has been suggested that the results of al- 
lometric analysis may be biased by differential degrees of phylogenetic relationship 
between the species included in the sample (Felsenstein, 1985). Felsenstein sug- 
gested using "independent contrasts" (differences between successive pairs of taxa in 
the tree), rather than the raw values, as an appropriate procedure to eliminate the ef- 
fects of differential degrees of phylogenetic relationship. The basis for this method 
was subsequently examined in more detail by Harvey and Pagel (Pagel & Harvey 
1989; Hai-vey & Pagel 1991 ) and the now widely used computer program CAIC was 
then designed for its application in comparative analyses (Purvis & Rambaut 1995). 
Contrast analysis using the CAIC program has been applied in some of the analyses 
reported here as one way of testing for potential biasing effects of phylogenetic relat- 

At this point, it is essential to note that there is a fourth potential problem that is 
commonly overlooked: The danger of indirect correlation universally applies to bi- 
variate allometric analyses. A correlation between any two variables, as illustrated in 
a bivariate plot (e.g., Figure 1), does not necessarily indicate a causal connection be- 
tween them even if the correlation seems to be very strong. Here, as elsewhere, the 
criterion of isolation must be respected. It is common to find indirect conelations be- 
tween variables in bivariate allometric analyses because both are related to one or 
more additional variables that have not been included in the analysis. For this reason, 
it is important to conduct extensive analyses in order to test for the presence of indi- 
rect correlations. One useful approach here is partial correlation. Using this tech- 
nique, we can determine whether a conelation between two variables remains after 
the influence of other potentially confounding variables has been excluded. Success- 
ful use of this approach, however, relies upon effective identification of the key inter- 
acting variables in any particular analysis. 

Having briefly reviewed the major problems involved in scaling analyses, it is 
now possible to turn to a number of practical applications. When brain size is scaled to 
body size in a large sample of placental mammals, it emerges that modem humans do, 
in fact, have the largest relative brain sizes. Examination of residual values for differ- 
ent mammalian orders (Figure 2) reveals that primates are a special case. On average, 
primates have larger relative brain sizes than members of other mammalian orders. It 
is, however, important to place this finding in proper perspective. Following the ex- 
ample of many other authors, Dunbar (1992) states that "primates have larger brains 
. . . than other animals." Such blanket statements imply that all primates have bigger 
brains than all other mammals. This is certainly not true when absolute brain size is 
considered. As already noted above, the brain of an elephant is some four times bigger 
than the human brain (the biggest among primates) and several other examples could 
be cited. It is, however, not even true when the size of the brain relative to body size is 
considered. After humans, the next highest relative brain sizes among placental mam- 
mals are found not in other primates but in dolphins (see Figure 2). Further, there is 




8 - 

4 - 






1 1 1 I 1 1 





1 1 1 1 1 1 1 1 

- 1 1 , , n , , 


1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 

-1 — r- 'III 

20 - 


1 , , r-nhn-rri 




— — — 1 — 1 — 1 — (— -1 — 1 — 1 — 1 — 1 — 1 — 1 — 1 — 

i — I — I — r 

8 - 




-1 — I — I — I — I — \ — I — I — I — I — r 

1 1 1 r -I — 1 — 1 — 1 — 1 — 1 — 1 — F- 1 — !■ 






— 1 — 1 — r I ^ 

8 - 

4 - 

n n 




-1 — I — I — I — I — I — I — I — I — r 


4 : 

2 - 

n rrr 






T — I — I — I — I — I — I — I — I — 1 — r 
0.05 0.25 0.45 0.65 0.85 

Residual Values for Brain Weight [logarithmic] 

Figure 2. Logarithmic residual values for brain size in placental mammals, indicating the size of the brain 
relative to body size. Primates are clearly a special case in that the values are generally shifted to the right, 
but there is considerable overlap with other mammals. Humans (indicated in black) have the largest rela- 
tive brain sizes. The next largest relative brain sizes are found not in other primates but in dolphins, mem - 
bers of the order Cetacea. 


considerable overlap betu'een primates and nonprimates in residual values for brain 

The fact that the average value for relative brain size in primates is higher than the 
average for other mammalian orders (illustrated by the rightward shift of the residual 
values for primates in Figure 2) can be traced to a difference in neonatal brain size be- 
tween primates and nonprimates (Figure 3). As a crude approximation, it can be 
stated that a primate neonate has a brain roughly twice as big as any other neonatal 
mammal of the same body size. This reflects the fact that, at any given body size, pri- 
mate fetuses uniformly have larger brains than other mammals throughout develop- 
ment, doubtless because of an early emphasis on brain development in primate 
embryos (Sacher 1982; Martin 1983). However, because of differences among mam- 
mal orders in postnatal development, the marked distinction between primates and 
nonprimates that is so clearly identifiable at birth becomes partially obscured by the 
time that adulthood is reached (Figure 2). On the other hand, the fact that the higher 
average value for relative brain size in primates is connected with a difference in de- 
velopment clearly emphasizes the importance of ontogenetic processes in establish- 
ing difference in brain size between species. 

So far, this discussion has been confined to the overall size of the brain. It is, in 
principle, possible that overall brain size reveals little about individual parts of the 
brain. It turns out, however, that the scaling of individual brain parts in the mammal- 
ian brain follows a reasonably consistent pattern. In a wide-ranging study of scaling 
of individual brain parts in insectivores, primates and bats, Finiay and Darlington 
(1995) showed that there is a relatively tight association with overall brain size. In 
other words, it is possible to predict the sizes of individual brain parts fi-om overall 
brain size to an approximate accuracy of ± 20%. At the same time, it is important to 
recognize that there is some degree of latitude in the relationships concerned (re- 
flected by scatter of points around the best-fit lines), such that there are differences of 
detail between species. This indicates that some adjustment of the relationships be- 
tween individual brain parts does take place, although a basic pattern of brain organi- 
zation is still recognizable among mammals generally. The study conducted by 
Finiay and Darlington ( 1 995) was also significant in that it further emphasized the im- 
portance of brain development in establishing this basic pattern of mammalian brain 

Relative Brain Size in Hominid Evolution 

Overall, the fossil record of hominids reveals a progressive increase in absolute 
brain size over time, although the rate of increase has not been uniform (see Aiello, 
this volume). Taking four fossil hominid species as generally representative of suc- 
cessive stages, the following sequence in absolute brain size can be documented: ^W5- 
tralopithecus africanus = 440 cc (n ^ 7); Homo hahilis = 630 cc (n = 7); Homo erectus 
= 990 cc (n = 28); modem Homo sapiens = 1345 cc (Figure 4). Interestingly, Homo 
neanderthalensis had a larger average brain size than modern humans, with a mean 
value of 14 12 cc (n = 22). Two comments can be made on these numbers. Firstly, it is 
clear that absolute brain size in hominids has generally increased over a period of at 
least three million years. Secondly, the average value iov Australopithecus africanus 
(440 cc) still lies within the range of average values for great apes, (385 cc for Pan 
troglodytes, 405 cc for Pongopygmaeus, and 496 cc for Gorilla gorilla), whereas the 







10 - 

— 8 - 





loQe Neonatal Body Weight [g] 

Figure 3. Logarithmic plot of brain size against body size for neonates of placental mammals (after Mar - 
tin 1 996). Primate neonates (dark symbols; n = 43) clearly tend to have larger brains at any giv en body size 
than other placental mammal neonates (pale symbols; n = 74) This is a prime example of a distinct grade 
shift (see Figure lb). (Figure reprinted with kind peimission from News in Physiological Sciences.) 

values for all three Homo species lie above that level. Because of this, many authors 
have concluded that expansion of the hominid brain first began with the gcnusHomo. 
As has already been seen, absolute brain size is not very informative across mam- 
mals generally and it is therefore necessary to examine relative brain size in fossil 
hominids. Here, however, we encounter a problem with respect to the estimation of 
body size. Initially, it was believed that all gracile australopithecines, such as Austra- 
lopithecus africamis, were relatively small-bodied with an average body mass of 25- 
30 kg. Using this value, Jerison ( 1973) concluded that relative brain size was, in fact, 
greater xn Australopithecus africamis than in modem great apes. In recent years, how- 
ever, the consensus view has been that gracile australopithecines showed quite 
marked sexual dimorphism and that their average body mass was well above 30 kg. 
Jungers (1988), for example, estimated an average body mass of 46 kg for Australo- 
pithecus africamis using a formula based on apes and humans and an average body 
mass of 53 kg using a formula based only apes. Similarly, McHenry (1988) estimated 
an average body mass of 46 kg iox Australopithecus africamis (although he later re- 
ported that a value of 36 kg is obtained using a human formula, compared with 46 kg 
using an ape formula — see McHenry 1992). If the average cranial capacity of440 cc 
^ov Australopithecus africamis is examined in relation to an average body mass of 46 
kg for males and females, the relative brain size of this hominid species is found to fall 
within the range of average values for modem great apes. For this reason, it has been 
widely concluded that expansion of the hominid brain first began with the genus 
Homo even if cranial capacity is considered in relation to body size. Because the earli- 
est reliable evidence for manufacture of stone tools appears in the fossil record at 



2000 n 





O 1200 

CO 1000- 


"(5 800 




600 - 

400 - 

200 - 


















1 125 



Figure 4. Cranial capacities for five fossil hominid species and modem humans. Note the wide range of 
variation within Honw species (mean: value in heavy type opposite black spot; maximum and minimum: 
values in italics opposite upper and lower ends of white boxes). The hatched zone contains the values for 
modem gieat apes (pale hatching: overall range; dark hatching: Range of means for chimpanzee, gorilla 
and orangutan). The "Cerebral Rubicon" was originally set at 800 cc, above the maximum recorded for 
gi-eat apes, but was later reduced to 600 cc when Homo habilis was recognized as a distinct species. 


about the same time as the earliest specimens attributable to the genus Homo, some 
authors have concluded that there is a causal connection between tools and the initial 
increase in relative brain size above the level found in modern great apes. Because it 
has also been suggested that there is a connection between the mental processes re- 
quired for manufacture of tools to fit certain design requirements and the mental pro- 
cesses required for language, it might therefore seem that there is some connection 
between relative brain size in human evolution and language. 

There is a major flaw in recent arguments based on revised body mass figures for 
australopithecines in that all of the values for cranial capacity oi Australopithecus af- 
ricanus are derived from skulls that would be interpreted as females by those who be- 
lieve in the existence of strong sexual dimorphism in this species. The vexed problem 
of inferring sexual dimorphism in the fossil record can therefore be circumvented by 
limiting analysis to these "female" gracile australopithecines and female great apes 
(Martin 1989b). When this is done, it emerges that the cranial capacity of "female" 
Australopithecus africanus relative to body size was in fact some 50% greater than in 
modern female great apes. This conclusion has recently been broadly confirmed by a 
study in which body weight was estimated for individual hominid specimens rather 
than for entire species (Kappelmann 1996). 

It is also important here to consider the implications of another widespread fal- 
lacy, arising from the implicit notion of "frozen ancestors." If a modem species is 
taken as a model for an ancestral condition, this implies that the model species itself 
has remained unchanged over an extensive period of evolutionary time. By taking 
any modem great ape species as a baseline for comparison y^'ith Australopithecus, it is 
inherently implied that the ape's brain size (both absolutely and relatively) has re- 
mained unchanged since hominids diverged from the common ancestor. There has, 
however, been a general trend among mammals for relative brain size to increase in 
all lineages, albeit at different rates (Martin 1990), so it is likely that brain size has 
also increased at least to some extent in the lineages leading to modem great apes 
since they diverged from hominids. Thus, the relative brain size of Australopithecus 
africanus was, if anything, even larger in relation to the common ancestor than the 
above comparison of female brain sizes has indicated. Overall, therefore, there is a 
clear implication that the progressive increase in brain size relative to body size dur- 
ing hominid evolution was initiated prior to the first record of Homo and prior to the 
earliest date for recognizable stone tools. 

Explanations for Differences in Relative Brain Size 

Numerous explanations have been proposed to account for differences in relative 
brain size among primates. By extension, these may be applied to interpretations of 
hominid evolution, although a number of additional factors have been invoked for 
this specific case {e.g., hunting behavior, use and manufacture of tools, language, 
general "intelligence"). As noted by Dunbar (1992), two basic types of explanation 
have been invoked to account for differences in relative brain size among primates 
generally: Ecological or social explanations. At least three versions of the ecological 
type of explanation can be identified: ( 1 ) frugi vores have relatively bigger brains than 
folivores; (2) relative brain size is associated with the size of the home range; (3) "ex- 
tractive foraging" favors the evolution of a relatively larger brain. Similarly, at least 
three versions of the type of explanation invoking social knowledge can be identified, 
linking relative brain size to: ( I ) the number of familiar individuals with which social 


interaction occurs; (2) tracking of social interactions between other individuals; (3) 
the nature of relationships between individuals. It should be noted that these alterna- 
tive versions are not necessarily mutually exclusive and that it may be difficult to 
tease out the main factors involved. For example, the ecological type of explanation is 
confounded by the fact that both diet and range size are closely associated with group 

The question of confounding variables is of central importance in the examination 
of potential links between behavior and relative brain size. In fact, it has been sug- 
gested by a number of authors that there may be some connection between an addi- 
tional factor, metabolism, and brain size (Armstrong 1982, 1983, 1985; Hofman 
1983a, 1983b; Martin 1981, 1983). One preliminary indication that there might be 
such a connection derives from the observation that the value of the scaling exponent 
(a) is approximately the same (close to 0.75) in the allometric equations relating brain 
mass and basal metabolic rate to body mass for large samples of mammalian species. 
Taken alone, this correspondence in the value of the scaling exponent values for brain 
size and basal metabolic rate does not provide very convincing evidence of a link, al- 
though it is suggestive. 

Some authors have proposed that the size of the adult brain is directly linked to that 
individual adult's metabolic capacity. Armstrong (1982) postulated that large brains 
have high energy demands that "are met by an increase in the energy supply," while 
Hofman ( 1 983a) suggested that "the energy demands of the brain must be compatible 
with the oxygen production and transport by the body as a whole." It is, however, 
readily apparent that the connection cannot be such a direct one. For mammals gener- 
ally, the range of variation in relative brain size is markedly greater than the range of 
variation in basal metabolic rate relative to body size. Whereas relative brain size 
shows a fivefold range of variation around the best- fit line (species at the extremes 
have brains five times as big and a fifth as big as a species lying on the best-fit line), 
relative metabolic rate shows only a twofold range (species at the extremes have 
brains twice as big and half as big as a species lying on the best-fit line). Thus, there is 
a great deal of variation in adult brain size that cannot be explained by a direct rela- 
tionship with basal metabolic rate of the adults possessing those brains. This is also 
obvious from examination of the data for primates. As has been shown above (Figure 
2), the average value for relative brain size in primates is higher than the average 
value for any other order of placental mammals . Yet primates do not differ from other 
orders of placental mammals with respect to the average value for basal metabolic 
rate relative to body size (Figure 5). Further, the residual value basal metabolic rate in 
humans does not differ from the average for primates despite the fact that the relative 
size of the human brain is so outstanding. Accordingly, the higher average value for 
relative brain size in primates must be attributable to some factor other than a direct 
connection with basal metabolic rate. It should be noted here that the "expensive tis- 
sue hypothesis" proposed by Aiello and Wheeler (1995) is a variant on the theme that 
the connection between metabolism and brain size is applicable at the level of the in- 
dividual adult. Their approach differs significantly, however, in invoking a tradeoff 
between the size of the gut and the size of the brain, as both have relatively high en- 
ergy requirements. Hence, the size of the gut is a potentially confounding factor that 
could influence the relationship between brain size and metabolism. 

The search for some indirect link between basal metabolic rate and brain size that 
would also explain a number of other findings (e.g., the difference between primate 
and nonprimates at birth illustrated in Figure 3) led to formulation of the maternal en- 
ergy hypothesis (Martin 1981, 1983. 1996). In this, it is proposed that the metabolic 


turnover of the mother determines the resources available for prenatal development 
of the brain in the fetus (provided via the placenta) and for postnatal brain develop- 
ment up to the time of weaning (provided via lactation). The brain is relatively un- 
usual compared to other organs of the body in that it reaches its definitive size 
relatively early in ontogeny, such that a major proportion of brain development has 
been completed by the time of weaning. In other words, development of the brain is 
heavily dependent on resources provided by the mother (Figure 6). Thus, the size of 
the brain in an adult individual may be linked not to that individual's own basal meta- 
bolic rate but to the metabolic turnover of its mother. On the one hand, this would ac- 
count for the general similarity in scaling patterns between brain size and basal 
metabolic rate relative to body mass (as the values for relative metabolic rate of an 
adult individual and its mother would be very similar). On the other hand, the greater 
scatter in residual values for brain size could be attributable to the effects of interven- 
ing variables, such as gestation period, lactation period and efficiency of transfer 
across the placenta and mammary glands. One specific testable prediction that arises 
from this is that there should be a correlation between gestation period and brain size 
in addition to any correlation between basal metabolic rate and brain size. The mater- 
nal energy hypothesis also differs from other hypotheses in that it could account for 
the scaling of brain size in the absence of specific selection pressures favoring an in- 
crease in the processing capacity of the central nervous system. The potential influ- 
ence of the maternal contribution to brain development should therefore be 
considered as one of the factors involved in brain evolution, in addition to any specific 
selection pressures that might be invoked (Figure 6). 

At this point, it is necessary to acknowledge certain challenges that have been 
made to the maternal energy hypothesis. The first of these concerns a test of the pro- 
posed connection between brain size and basal metabolic rate that was conducted by 
McNab and Eisenberg ( 1 989), with seemingly negative results. This finding has been 
specifically cited by subsequent authors (e.g., Aiello & Wheeler 1 995; Aboitiz 1996) 
in discussions that omitted any further consideration of the maternal energy hypothe- 
sis. For instance, in presenting their "expensive tissue hypothesis" Aiello and 
Wheeler (1995:211) specifically stated: "These conclusions are derived from the 
general observation that there is no significant correlation between relative basal 
metabolic rate and relative brain size in humans and other encephalized mammals." 

Quite rightly, McNab and Eisenberg argued that effective demonstration of a link 
between basal metabolic rate and brain size requires removal of the effect of body 
size, as this is a potentially confounding variable. It is, in principle, possible that the 
observed correlation between basal metabolic rate and brain size is an indirect asso- 
ciation arising from that fact that both of these variables are correlated with body 
mass. They analyzed data for 174 mammal species (including monotremes and mar- 
supials) to examine the relationship between relative basal metabolic rate and relative 
brain size. Although a positive trend was indeed found as predicted, the coirelation 
did not quite reach significance (p = 0.08). It should be noted at once that there is a cru- 
cial difference here between hypotheses that invoke some direct link between basal 
metabolic rate and brain size in the individual adult and the maternal energy hypothe- 
sis, which links a mother's metabolism to the completed brain size of her offspring. 
While it is true that the former predict a very tight association between relative basal 
metabolic rate and relative brain size, the latter necessarily predicts a weaker associa- 
tion because other variables are involved in the causal chain (Figure 6). Nevertheless, 
the correlation reported by McNab and Eisenberg is surprisingly weak, even with re- 
spect to the maternal energy hypothesis. There was, however, a basic statistical error 



3 - 


1 1 — 1 — 1 — 1 — 1 — 1 — 1 — r-T 


1 - 

, , n r 

- Ill 



|— 1 


1 1 n , , n 


H — I — p — I — I — 1 — 1 — I — I — I — I — I — I — I — I — I — r— T — I — I — I — I — r 


8 - 
6 - 



4 - 
2 - 

n - 


1 — 1 — l— 1 — 1 — 1 — 1 — 1 — 1 — 

: , m rr 




« - 

6 - 



4 - 
2 - 

, 1 n , 


^ =U 

— 1 — 1 — 1 — 1 — r I I 




-I — I — I — r 



m n m 

-1 — I — I — r 
-1.4 -1.2 -1.0 -0.8 -0.6 -0.4 -0.2 0.0 0.2 0.4 0.6 0.8 1.0 1.2 1.4 

Residual Values for Basal Metabolic Rate [logarithmic] 

Figure 5. Logai'ithmic residual values for basal melabolic rale in placental mammals, indicating the rate of 
metabolism relative to body size. Primates, including humans, do not have relatively higher basal meta - 
bolic rates than other mammals. Funher, humans (mdicated in black) do not have relatively higher basal 
metabolic rates than other placental mammals, despite hav ing the largest relative brain siz e. Accordingly, 
the higher average value for relative brain size in primates (Figure 2) cannot be simply at tributed to a higher 
average value for basal metabolic rate. 



in the analysis conducted. The statistical test used to determine significance was 
based on the assumption of normality of distribution in the values compared, yet it 
was applied after the logarithmic residual values had been converted to quotients (an- 
tilogarithmic form). Because logarithmic residual values are normally distributed, 
derived quotient values necessarily show a skewed distribution favoring higher val- 
ues. (For example, the logarithmic residual values for relative brain size in placental 
mammals are normally distributed around the value of corresponding to species ly- 
ing directly on the best- fit line. By contrast, derived quotient values for species below 
the line take values between and 1, whereas values for species above the line take 
values between 1 and 5.) The appropriate procedure for assessing significance with 
the derived quotient values is a non-parametric test, such as the Spearman rank corre- 
lation, which yields the following result: rs = 0.17; p = 0.025. Alternatively, a para- 
metric test (Pearson correlation) can be applied to the logarithmic residual values, 
yielding the following result: r = 0. 1 58; p = 0.040. In both cases, the con-elation be- 
tween the residual values for basal metabolic rate and brain size is, in fact, significant. 
The correlation is not very strong and this reinforces the case for rejecting any direct 
link in the adult individual between basal metabolic rate and brain size of the kind pro- 
posed by Armstrong (1982) and Hofman (1983a, 1983b). Nevertheless, the fact that 
the correlation is weak but significant (indicating participation of other variables, as 



Maternal Energy Resources 









FIGURE 6. Diagram illustrating the mateiTial energy hypothesis (revised from Martin 1 989b). Be cause the 
brain is unusual in that the major part of its growth occurs early during development, mate mal resources 
provided through the placenta and during lactation are particularly important. By the time of weaning, 
most of the growth in size of the brain has been achieved. Because there is a resulting link be tween a 
mother's metabolism and the ultimate brain size achieved by her offspring, there will be a secondai-y cor - 
relation between that offspring's brain size and its own metabolism once adulthood is reache d. Scaling of 
maternal resources provides a passive contribution to overall scaling of brain size, whil e individual selec- 
tion pressures favor the enlargement of particular pails of the brain. 


is specifically predicted by the maternal energy hypothesis) should be considered in 
any overall discussion of the evolution of relative brain size. 

Incidentally, given that there is evidence of a connection between basal metabolic 
rate and total metabolic turnover, it is incorrect to claim that the Aiello/Wheeler hy- 
pothesis escapes the requirement for a correlation between residual values for brain 
size and basal metabolic rate. Consider two species of the same body size, one with a 
relatively high metabolic rate and one with a relatively low metabolic rate. The spe- 
cies with a relatively higher metabolic rate will be expected to have a larger total en- 
ergy budget, such that (other things being equal) any "tradeoff with the digestive 
tract should leave more energy available to support a large brain. Extrapolation of this 
simple illustration to a larger sample covering a range of body sizes should surely re- 
quire a positive correlation between residual values for brain size and basal metabolic 

The maternal energy hypothesis has been challenged on two quite different 
grounds by Pagel and Harvey ( 1988, 1990). The first relates to the prediction that pre- 
cocial mammals (those such as primates that give birth to small litters of well- 
developed offspring) should give birth to neonates with larger brains than altricial 
mammals (those that give birth to large litters of poorly developed offspring). Pagel 
and Harvey (1988) stated: "But if species that give birth to precocial offspring also 
have higher metabolic rates and longer gestations for their size, then Martin's and 
Hofman's predictions would be supported." This is misleading, as it implies that de- 
velopment of larger brains requires both higher metabolic rates and longer gestations. 
It is obvious, however, from the model illustrated in Figure 6 that these two factors 
can operate independently. Although brain size in the neonate will depend on a com- 
bination of maternal basal metabolic rate and gestation period, it is not necessary for 
both to be elevated in order to produce a large-brained offspring. Indeed, there is evi- 
dence that an increase in gestation length may occur in compensation for a relatively 
low basal metabolic rate (Martin 1996). 

The second challenge mounted by Pagel and Harvey is more substantial, as it is ar- 
gued that some of the findings reported (Martin 198.1. 1983) were invalid because of 
the failure to take litter size into account. This highlights an important issue that re- 
quires clarification. Even if the maternal hypothesis is correct in a general sense, there 
are in fact two different ways in which a mother's energy resources might constrain 
the development of her offspring's brain. It was initially argued (Martin 1 983 ) that "it 
is the mother's metabolic turnover which, both in direct terms (through the physiol- 
ogy of gestation) and in indirect terms (through the partitioning of resources between 
maintenance and reproduction) determines the size of the neonate's brain." At the 
time, it was unclear to what extent these two aspects of maternal resource allocation 
explained the findings, but the additional factor of litter size is obviously of great rele- 
vance here. If the postulated maternal constraint operates predominantly at the level 
of partitioning of resources between maintenance and reproduction {i.e., within a 
life-history framework), then litter size must clearly be taken into account, as an in- 
crease in litter size (other things being equal) must increase the total cost to the 
mother. In that case, however, various other potentially variable factors must also be 
taken into account. For instance, the interval between births will, of course, have a di- 
rect effect on the total cost to the mother, so an increase in litter size could be offset by 
an increase in interbirth interval. On the other hand, if the maternal constraint is pre- 
dominantly physiological in kind, the development of the individual fetus may be of 
prime importance. It can be argued that, regardless of litter size, the development of a 
fetus depends mainly on maternal metabolic turnover and the length of gestation. Al- 


though the presence of competitors within the uterus obviously has some effect, be- 
cause maternal resources are finite, data for variation in fetal growth according to 
litter size within a species show that this has a relatively limited effect. Of course, it is 
impossible to achieve a complete separation between long-term evolutionary effects 
{e.g., life-history adaptation) and direct physiological effects, as both basal metabolic 
rate and gestation period have evolved to exhibit particular values for any species. 
However, given that inclusion of litter size in the analyses yields results conflicting 
with predictions based on the overall allocation of maternal resources (Pagel & Har- 
vey 1988, 1990), we must consider the possibility that it is the direct physiological 
constraint that predominates in the relationship between maternal resources and the 
development of her offspring's brain. 

The possibility of direct physiological constraints in the model presented in Figure 
6 can be tested with a three-step procedure, using a sample of 5 1 placental mammal 
species from 14 different orders for which reliable data are available on body mass, 
brain mass, basal metabolic rate and gestation period. In the first step, residual values 
are calculated for brain mass, basal metabolic rate and gestation period relative to 
body mass. In a second step, residual values for brain size can be examined in relation 
to residual values for basal metabolic rate, following the example set by McNab and 
Eisenberg (1989). In fact, the positive correladon that emerges (r = 0.38: r" = 0. 14; p = 
0.005; see Figure 7a) is highly significant and markedly stronger than that found with 
the data originally reported by McNab and Eisenberg. There could be several reasons 
for this. The data set used by those authors was very different, notably including two 
egg-laying monotreme species and 15 marsupial species that are conspicuously dif- 
ferent from placental mammals in their reproductive biology. Further, their sample 
was heavily biased towards rodents (n = 78, representing 45% of the species exam- 
ined), while only 34 species (20%) were from orders characterized by precocial off- 
spring (artiodactyls, bats, primates). Whatever the explanation may be, the present 
analysis agrees with that of McNab and Eisenberg (following correction of their 
analysis as explained above) in showing a significant correlation between relative 
brain size and relative basal metabolic rate. 

Support for the model presented in Figure 6 is, however, even more convincing 
when the effect of gestation period is included. In the first place, there is a positive and 
significant correlation between the residuals for brain size and the residuals for gesta- 
fion period (r = 0.377; r" = 0. 142; p = 0.006). Further, the model requires that the ef- 
fects of basal metabolic rate and gestation period should interact. One way of testing 
this is to examine the relationship between the residuals for brain size and the sum of 
residuals for basal metabolic rate and gestation period. When this is done, there is a 
marked improvement over the correlation based on basal metabolic rate or gestation 
period alone (r = 0.55; r~ = 0.30; p 0.001; see Figure 7b). This indicates that basal 
metabolic rate and gestation period together account for about 30% of the variation in 
relative brain size between species, which is a very satisfactory result given the other 
factors that may be involved in brain development. It should be noted, incidentally, 
that the analyses reported by Pagel and Harvey (1988, 1990) clearly indicated a link 
between gestation period and brain size; it was the link between brain size and basal 
metabolic rate that they questioned. Hence, there is good evidence for a link between 
gestation period and brain size, and the only hypothesis that predicts such a link is the 
maternal energy hypothesis. An alternative approach, incidentally, is to examine par- 
tial correlation coefficients obtained in a four-way analysis of body size, brain size, 
basal metabolic rate and gestation period. When this is done using the same sample of 


5 1 placental mammal species, it emerges that body size, basal metabolic rate and ges- 
tation period influence brain size to approximately equal degrees (Martin, 1996). 

There is a potential problem in all of the statistical analyses discussed up to this 
point, in that the results might be biased by differential degrees of phylogenetic rela- 
tionship between the species compared (Figure 1 ). McNab and Eisenberg ( 1 989) took 
no account of phylogenetic relationship between the species they examined, and it 
has already been noted above that their sample was heavily biased towards altricial 
mammals, particularly rodents. Pagel and Harvey (1988, 1990) were well aware of 
the potential problem of differential phylogenetic relatedness and took steps to elimi- 
nate it by conducting their analyses at the level of the family rather than at the level of 
the species. Although this approach should at least reduce any effect of phylogenetic 
relatedness, it has a distinct disadvantage in that it drastically reduces sample sizes 
and may hence dilute or extinguish any signal in the data for this reason alone. In a re- 
cent review of the maternal energy hypothesis (Martin, 1996), the possible existence 
of a bias arising fi-om differential degrees of phylogenetic relatedness was not explic- 
itly examined, for two reasons. Firstly, the sample of 51 mammal species analyzed in- 
cluded representatives from 1 4 different orders of placental mammals and took only a 
single species for each genus (thus at least excluding the problem of species-rich gen- 
era). Secondly, the method of "independent contrasts," particularly advocated by 
Harvey and Pagel (1991) and by Purvis and Rambaut (1995) and widely used in re- 
cent allometric analyses, is unlikely to work well if numerous marked grade shifts are 
present in the data. Despite these reservations, the analysis of data for 51 placental 
mammal species summarized above has been repeated using that method as an addi- 
tional test. In fact, the coirelation between residual values for brain size and residual 
values for basal metabolic rate (both calculated from their contrasts relative to body 
mass contrasts) is improved relative to the analysis of the raw data: r = 0.465; r = 
0.216; p = 0.001. On the other hand, the correlation between residual contrast values 
for brain size and residual contrast values for gestation period is reduced relative to 
the analysis of the raw data and (although remaining positive) becomes non- 
significant: r=0.203; r"' = 0.041;p = 0.1 16. Inthiscase, combining the residual values 
for gestation period with those for brain size does not improve the correlation with 
brain size. One interpretation of these results would be that the application of contrast 
analysis has removed a bias due to phylogenetic relatedness between the species ex- 
amined and that there is, in fact, no association between relative brain size and the 
relative length of gestation period. This seems inherently unlikely in view of the re- 
sults obtained from the analysis conducted at the family level by Pagel and Harvey 
(1988, 1 990), which presumably greatly reduced any bias due to differential phyloge- 
netic relatedness and would at the most be expected to lose coiTelations because of the 
dramatic reduction in sample size. Their findings indicated the opposite conclusion, 
namely that there is a significant correlation between relative brain size and the rela- 
tive length of gestation but not between relative brain size and relative basal metabo- 
lism. An alternative interpretation of the results of the contrast analysis reported here 
is that the scaling of basal metabolic rate is only mildly affected by grade shifts and 
that the results therefore survive application of the contrast method, whereas the scal- 
ing of gestation period is greatly affected by grade shifts that interfere with the con- 
trast calculations. It is well established that there are major grade shifts in the scaling 
of mammalian gestation periods, most notably involving a fourfold difference in rela- 
tive values between altricial and precocial mammals (Martin & MacLamon 1985). 
Whatever the explanation may be, it seems clear that there is a significant relationship 
between relative brain size and relative basal metabolism in the sample examined. 




r = 0.38 ^ 

9 o 9 

5 1 - 


« 9 9.o^^^^ 





:?•>?'"" • 

Brain vers 



5.1 - 

* / ••• • 












-2 - 

1 1 1 


-1 1 

Residuals for BMR versus Body weight 








« -1 




r = 0.55 

a • 




9 ^ -^ 
o* 9 

-'• • • • » • 
•• • 





-2-10 1 2 

Combined Residuals for Gestation and BMR versus Body weight 

Figure 7. Upper graph: Plot of logarithmic residual values for brain size against logaiithmic re sidual val- 
ues for basal metabolic rate for 5 1 placental mammal species. Although the conelation is highly significant 
(r = 0.38; p = 0.005), 86% ofthe variance in brain size remains unexplained. Lower graph: Plotoflogarith - 
mic residual values for brain size against combined residual values for basal metabolic rate and gestation 
period for the same 51 placental mammal species. The correlation is improved by the inclusion ofthe effect 
of gestation period (r = 0.55; p 0.00 1 ) and the unexplained variance is reduced to 70%. 


A number of additional lines of evidence support the inference that maternal re- 
sources are of particular importance for the evolution of the brain. For instance, ex- 
perimental work on genomic imprinting has shown that maternally active genes 
specifically favor brain development (Keveme et al. 1996; Martin 1996). More re- 
cently, reexamination of data on the results of human IQ tests has revealed that there 
is an important contribution of the uterine environment (Devlin et al. 1997). 

The maternal energy hypothesis addresses the issue of maternal resources re- 
quired for brain development and the resulting imposition of constraints on overall 
brain size. It is, however, obvious that selection is likely to influence the size of par- 
ticular components of the brain, thus ultimately affecting overall brain size. Until re- 
cently, it was unclear how the concepts of maternal energy constraints and of 
selection of individual brain functions could be combined in a single model. A possi- 
ble connection has now been provided by the two-phase hypothesis of brain size evo- 
lution proposed by Aboitiz (1996). His proposal is that brain size is influenced by 
both "passive" growth (general adjustment to body size) and "active" growth (adapta- 
tion to particular behavioral requirements). A strong possibility would seem to be that 
the maternal energy hypothesis could account (at least in part) for passive adjustment 
of overall brain size to body size (with the refinement that an increase in maternal 
metabolic turnover or an increase in the length of gestation, or both factors combined, 
would also permit an increase in overall brain size), while selection for particular 
brain functions would eventually translate into an increase in brain size. One corol- 
lary of this is that individual behavioral features are unlikely to be reflected more than 
very indirectly by overall brain size. The search for links between particular behav- 
ioral developments, such as language, and increases in brain size should therefore fo- 
cus particularly on associations between behavior and particular parts of the brain. 

Neocortex Size, Group Size and Language 

An apt illustration of the advantages and limitations of broad-based comparisons 
in seeking to explain the origin of human language is provided by analyses examining 
relationships between individual parts of the brain and a simple measure of social 
complexity (the typical size of a group of interacting individuals). A prime example is 
provided by an analysis of the relationship between group size and the volume of the 
neocortex for 36 primate species conducted by Dunbar (1992). Although there are 
various problems involved in defining typical group size in a consistent manner 
across primates, this limitation is probably not severe enough to invalidate the rela- 
tively strong correlations that have been determined for some variables. Dunbar re- 
ported that, after allowing for the effects of potentially confounding variables such as 
body mass, there is no significant con-elation between the relative size of the neocor- 
tex and feeding behavior or ranging behavior (home range size; daily travel distance), 
but that the relative size of the neocortex does show a significant correlation with 
group size. 

One technical comment is necessary here. Although Dunbar's preferred measure 
was the "neocortex ratio" (the ratio between the volume of the neocortex and the vol- 
ume of the rest of the brain), ratios are difficult to interpret because the value can be 
modified by a change in either the numerator or the denominator, or perhaps both to- 
gether. Thus, if an increased value of the neocortex ratio is found in any given species, 
it is not clear to what extent this is due to an increase in the volume of the neocortex as 
opposed to a decrease in the volume of the rest of the brain. Given that it is the volume 


of the neocortex (rather than its volumetric relationship with the rest of the brain) 
which is likely to reflect processing capacity, this variable should be considered in its 
own right. [See Barton (1993) and Deacon (1993) for comments on additional prob- 
lems arising from the use of the neocortex ratio.)] In fact, the basic results of Dunbar's 
original analysis remain largely intact regardless of whether neocortex volume or his 
neocortex ratio is taken. 

In the original analysis of his data, Dunbar did not explicitly take account of poten- 
tial problems due to differential degrees of phylogenetic relatedness between the spe- 
cies compared, although he did subsequently apply a form of contrast analysis to a 
subset of his sample (Dunbar 1995). In order to conduct a ftirther check on his find- 
ings, his data have been reanalyzed here using the "independent contrasts" method 
(Purvis & Rambaut 1995). A significant correlation between the residual values for 
neocortex volume and for group size {i.e., after excluding the influence of body size 
for both variables) is still found following calculation of contrast values: r= 0.467; r 
= 0.218; p = 0.005. A very similar result was reported by Barton (1993, 1996), al- 
though in his later paper he examined contrast values for the residuals for neocortex 
volume in relation to group size rather than residuals of group size, arguing that body 
size has little influence on group size. However, Barton (1996) showed that a signifi- 
cant relationship can only be demonstrated for diurnal monkeys and apes. The rela- 
tionship for prosimians taken alone is not significant, although this is perhaps 
attributable to the small sample size. 

There seems to be a robust correlation between the size of primate groups and the 
size of the neocortex, after allowing for the effect of body size, although it cannot be 
ruled out that additional confounding variables might be found in the future. Indeed, 
Barton (1996) showed for diurnal monkeys and apes that neocortex size is correlated 
not only with group size but also independently with degree of fioigivory, so social 
group size is clearly not the only factor linked to the relative size of the neocortex. 
Here as elsewhere, the criterion of isoladon must be firmly established and a great 
deal of further testing is required before it can be accepted with any confidence that 
there is a direct causal link between the size of the neocortex and social group size. 

The suspicion that there may be far more to this story than initially meets the eye is 
reinforced by a closer examination of Dunbar's concept of "neocortex ratio." In some 
analyses at least, this ratio seems to yield better correlations with group size than neo- 
cortex volume itself, although it is surely the processing capacity of the neocortex that 
should be at stake, not its size in relation to the rest of the brain. The higher correla- 
tions reported for the neocortex ratio arise because the strong positive correlation be- 
tween relative group size and relative neocortex volume is matched by an equally 
strong negative correlation between relative group size and the relative size of the rest 
of the brain. (This results holds regardless of whether the analysis is based on raw val- 
ues or on contrast values and regardless of whether it is conducted with the entire sam- 
ple or just with monkeys and apes.) In parallel with the argument that a positive 
correlation between group size and neocortex volume reflects a causal relationship, 
one might equally well postulate that increase in brain size is causally related to a re- 
duction in the size of the rest of the brain in primates. One possible explanation might 
be that there is a tradeoff between the size of the neocortex and the size of the rest of 
the brain in adaptation to group size. This, in itself, would reinforce the idea that there 
is some kind of constraint on overall brain size, as suggested by the maternal energy 
hypothesis. Hence, a much more detailed analysis is required to tease out the causal 
factors involved. 



The identified relationships between the volume of the neocortex and group size 
or degree of frugivory in any case provide a good potential illustration of the "active" 
phase of the model suggested by Aboitiz (Figure 8). It can be argued that behavioral 
requirements associated with group size or diet have served as selections pressure fa- 
voring the development of a specific part of the brain, namely the neocortex. By con- 
trast, there is no correlation between group size and other parts of the brain, so 
selection associated with this variable has not favored a direct increase in brain size as 
a whole but an indirect increase through enlargement of the neocortex. 

In subsequent papers, it has been specifically proposed that the relationship be- 
tween group size and neocortex volume is directly connected with the evolution of 
human language (Aiello & Dunbar 1993; Dunbar 1993). Indeed, Aiello and Dunbar 
(1993) suggested that "the need for large groups among our early ancestors was the 
driving force behind not only the evolution of language but also hominid encephaliza- 
tion." The chain of argument sets out from the reported significant correlation be- 
tween group size and neocortex size in nonhuman primates. This empirical bivariate 
relationship is extrapolated to derive an expected "natural" group size of 150 for hu- 
mans from the volume of the human neocortex. This inferred group size is then con- 
sidered in connection with another empirical relationship reported by Dunbar ( 1 99 1 ), 
namely between social grooming time and group size for Old World monkeys and 
apes. This relationship is interpreted on the basis of the assumption that social groom- 
ing is the primary factor in maintaining the cohesion of primate groups and social 
bonding over time. Taking the inferred human group size of 150, it is predicted (by 
extrapolation beyond the range of data for Old World monkeys and apes) that such a 
group size would be expected to spend some 40% time in grooming, whereas 20% 
seems to be the upper limit for nonhuman primates. On this basis, it is argued that lan- 

^ ACTIVE y. 
V/, GROWTH y// 






rearrangements in 
connectivity and 
cell differentiation 


Brain cell 

Minor epigenetic 

Limited genetic 
to maintain 

Figure 8. illustration of the two-phase model of brain evolution proposed by Aboitiz (1996). Passive 
change in brain size is primarily a reflection of body size and it is argued here that mater nal resources play a 
major part in this. Active change in the brain is brought about by selection pressures which lead to enlarge - 
ment of specific parts of the brain. In the absence of compensatory effects, enlargement of part of the brain 
will, of course, lead to an increase in overall brain size. 


guage evolved as an alternative means of maintaining group cohesion, being more ef- 
ficient and "cheaper" than grooming and allowing larger mean group sizes. Using 
estimates of the size of the neocortex from endocranial volumes of fossil hominids, 
Aiello and Dunbar ( 1 993 ) went on to infer that some kind of language emerged "early 
in the evolution of the genus Homo.'"' 

It should be noted, however, that demonstration of apparent links between social 
group size, relative size of the neocortex and grooming in nonhuman primates does 
not necessarily imply a link with language in humans. While it is possible that the de- 
velopment of language is a phenomenon directly associated with the emergence of 
greater social complexity during human evolution, this is an additional hypothesis 
that must be adequately tested in its own right. 


I am very grateful to Nina Jablonski for her invitation to participate in the Third 
Wattis Symposium and would also like to acknowledge the kindness and efficiency 
of all those involved in the organization of this event, notably Kathleen Quinlan. The 
section on calibration of phylogenetic trees profited from valuable discussion with Si- 
mon Tavare, while the sections on allometric scaling greatly benefited from recent 
discussions with Andrew Barbour, Kate Jones, Ann MacLarnon and Caroline Ross. 
Andrew Purvis kindly provided a copy of the computer program CAIC, used in some 
of the analyses reported, and has been very generous with advice on its application. 
Positive and helpfiil critical comments from two anonymous reviewers are also grate- 
fiilly acknowledged. The concept of independent brain size evolution in Neanderthals 
was prompted by collaboration with Marcia Ponce de Leon. Christopher Zolloikofer 
and Peter Stucki in a project on computerized skull reconstruction supported by the 
Swiss National Science Foundation (projects 3 1 00-032360.9 1 and 3 1 00-04249 1 .94). 

Literature Cited 

Aboitiz, F. 1 996. Does bigger mean better? Evolutionary determinants of brain size and struc- 
ture. Brain Behav. Evol. 47:225-245. 

Aiello, L.C. & R.I.M. Dunbar. 1 993. Neocortex size, group size, and the evolution of language. 
Citrr. Anthropol. 34:184-193. 

Aiello, L.C. & P. Wheeler. 1 995. The expensive tissue hypothesis: The brain and the digestive 
system in human and primate evolution. Ciin: Anthropol. 36: 199-221. 

Alexander, J. & J. Andrews. 1997. Alphabets and citations. Nature 386: 112-113. 

Arensburg, B., A.-M. Tillier, B. Vandermeersch, H. Duday, L. Schepartz, & Y. Rak. 1988. A 
Middle Palaeolithic human hyoid bone. Nature 338:758-760. 

Armstrong, E. 1982. A look at relative brain size in mammals. Newosci. Lett. 34: 101-104. 

. 1983. Relative brain size and metabolism in mammals. Science 220: 1 302-1304. 

. 1985. Relative brain size in monkeys and prosimians. Amer. Jour. Phys. Anthropol. 


Amason, U., A. GuUberg, A. Janke, & X.-F. Xu. 1996 Pattern and timing of evolutionary di- 
vergences between hominoids based on analyses of complete mtDNAs. Jour. Mol. Evol. 

Barton, R.A. 1993. Independent contrasts analysis of neocortical size and socioecology in pri- 
mates. 5e/za v. Brain Sci. 16:694-695. 


. 1996. Neocortex size and behavioural ecology in primates. Proc. Roy. Soc. Lond. B 

Crelin, E.S. 1987. The Human Vocal Tract. Vantage, New York. 
Deacon, T.W. 1993. Confounded correlations, again. Behav. Brain Sci. 16:698-699. 
Devlin, B., M. Daniels, & K. Roeder. 1997. The heritability of IQ. Nature 388:468-471. 
Dunbar, R.I.M. 1991 . Functional significance of social grooming in primates. Folia Primatol. 

. 1992. Neocortex size as a constraint on group size in primates. Jour. Hum. Evol. 

. 1993. Coevolution of neocortical size, group size and language in humans. Behav. 

Brain Sci. 16:681-735. 
. 1 995. Neocortex size and group size in primates: A test of the hypothesis. Jour. Hum. 

Evol. 28:287-296. 
Falk, D. 1 975. Comparative anatomy of the larynx in man and the chimpanzee: Implications for 

language in Neanderthal. Amer. Jour. Phys. Anthropol. 43:123-132. 
. 1978. External neuroanatomy of Old World monkeys (Cercopithecoidea). Contrib. 

Primatol. 15:1-95. 
. 1987. Brain lateralization in primates and its evolution. Yearh. Phys. Anthropol. 


Falk, D., J. Cheverud, M.W. Vannier, & G.C. Conroy. 1986. Advanced computer graphics 
technology reveals cortical asymmetry in endocasts of rhesus monkey. Folia Primatol. 

Felsenstein, J. 1985. Phylogenies and the comparative method. Amer. Nat. 125:1-15. 

Finlay, B.L. & R.B. Darlington. 1 995. Linked regularities in the development and evolution of 
mammalian brains. 5c/e/7ce 268:1578-1584. 

Gilissen, E. 1992. Les scissures neocorticales du singe capucin {Cebus)\ mise en evidence 
d'une asymetrie de la scissure sylvienne et comparaison avec d'autres primates. C. R. Acad. 
Sci. Paris, ser III 314:1 65-1 70. 

Gingerich, P.D. & M.D. Uhen. 1 994. Time of origin of primates. Jow: Hum. Evol. 27:443^M5. 

Harvey, P. H.&G.M. Mace. 1982. Comparison between taxa and adaptive trends: Problems of 
methodology. Pages 343-361 in King's College Sociobiology Group, ed.. Current Prob- 
lems in Sociobiology. Cambridge University Press, Cambridge, U.K. 

Harvey, P. H. & M.D. Pagel. 1 99 1 . The Comparative Method in Evolutionaiy Biology. Oxford 
University Press, Oxford, U.K. 

Hedges, S.B., P.H. Parker, C.G. Sibley, &. S. Kumar. 1996. Continental breakup and the ordi- 
nal diversification of birds and mammals. Nature 381:226-229. 

Heffner, H.E. & R.S. Heffner. 1984. Temporal lobe lesions and perception of species-specific 
vocalizations by macaques. Science 226:75-76. 

Hofman, M.A. 1983a. Energy metabolism, brain size and longevity in mammals. Quart. Rev. 
Biol. 58:495-512. 

. 1 983b. Evolution of the brain in neonatal and adult placental mammals: A theoretical 

approach. Joj/r Theor. Biol. 105:317-322. 

Holloway, R.L. 1983. Human palaeontological evidence relevant to language behaviour. Hum. 
Neurobiol. 2:105-114. 

. 1988. Brain. Pages 98-105 in I. Tattersall, E. Delson, & J. Van Couvering, eds.. En- 
cyclopedia of Human Evolution and Prehistoty. Garland, New York. 

Holloway, R.L. & M.C.deLacoste-Lareymondie. 1982. Brain endocast asymmetry in pongids 
and hominids: Some preliminary findings on the palaeontology of cerebral dominance. 
Amer Jour Phys. Anthropol. 58:101-1 10. 


Holloway, R.L. & D.G. Post. 1 982. The relativity of relative brain measures and hominid mo- 
saic evolution. Pages 57-76 in E. Armstrong, & D. Falk, eds.. Primate Brain Evolution. 
Plenum Press, New York. 

Houghton, P. 1993. Neandertal supralaryngeal vocal tract. Amer. Jour. Phys. Anthropol. 

Hublin, J.-J.,F.Spoor, M. Braun,F. Zonneveld, & S. Condemi. 1996. A late Neanderthal asso- 
ciated with Upper Palaeolithic artefacts. Nature 381:224-226. 

Janke, A., G. Feldmaier-Fuchs,W. K. Thomas, A. von Haeseler, & S. Paabo. 1 994. The marsu- 
pial mitochondrial genome and the evolution of placental mammals. Genetics 

Jerison, H.J. 1973. Evolution of the Brain and Intelligence. Academic Press, New York. 

Jungers, W.L. 1988. New estimates of body size in australopithecines. Pages 1 15-125 in F.E. 
Grine, ed.. Evolutionary Histoty of the "Robust " Australopithecines. Aldine de Gruyter, 
New York. 

Kappelmann, J. 1996. The evolution of body mass and relative brain size in fossil hominids. 
Jour. Hum. Evol. 30:243-276. 

Keveme, E.B., F.L. Martel, & CM. Nevison. 1 996. Primate brain evolution: genetic and func- 
tional considerations. Proc. Roy. Soc. Land. B 262:689-696. 

Krings, M., A. Stone, R.W. Schmitz, H. Krainitzki, M. Stoneking, & S. Paabo. 1997. Neander- 
tal DNA sequences and the origin of modem humans. Cell 90: 19-30. 

Laitman, J.T. 1985. Evolution of the hominid upper respiratory tract: The fossil evidence. 
Pages 281-286 in P.V. Tobias, ed., Hominid Evolution: Past, Present and Future. Alan R. 
Liss, New York. 

Laitman, J.T., R.C. Heimbuch, & E.S. Crelin. 1 979. The basicranium of fossil hominids as an 
indicator of their upper respiratory systems. Amer. Jour. Phys. Anthropol. 51 : 15-34. 

Leakey, L.S.B., P.V. Tobias, & J.R. Napier. 1964. A new species of the genus Homo from 
Olduvai Gorge. Nature 202:7-9. 

LeMay, M. 1976. Morphological cerebral asymmetry of modem man, fossil man and nonhu- 
man primates. .4nn. N. Y. Acad. Sci. 280:348-366. 

LeMay, M.,M. Billig, & N. Geschwind. 1982. Asymmetries ofthe brains and skulls of nonhu- 
man primates. Pages 263-278 in E. Armstrong, & D. Falk, eds.. Primate Brain Evolution. 
Plenum Press, New York. 

Lieberman, P. 1984. The Biology and Evolution of Language. Harvard University Press, Cam- 

Lieberman, R., E.S. Crelin, & D.H. Klatt. 1972. Phonetic ability and related anatomy ofthe 
newborn and adult human. Neanderthal man, and the chimpanzee. Amer. Anthropol. 

MacLamon, A.M. 1 993. The vertebral canal. Pages 359-390 in A.C. Walker, & R.E.F. Leakey, 
eds., The Nariokotome Homo erectus Skeleton. Harvard University Press, Cambridge. 

Martin, R.D. 1981. Relative brain size and metabolic rate in terrestrial vertebrates. Nature 

■ — . 1 983. Human Brain Evolution in an Ecological Context (52nd James Arthur Lecture 

on the Evolution ofthe Human Brain). American Museum of Natural History, New York. 

. 1986. Primates: A definition. Pages 1-31 in B.A. Wood, L.B. Martin, & P. Andrews, 

eds.. Major Topics in Primate and Human Evolution. Cambridge University Press, Cam- 
bridge, U.K. 
— — . 1989a. Size, shape and evolution. Pages 96-141 in Keynes, M., ed.. Evolutionary 

Studies - A Centenary Celebration ofthe Life of Julian Huxley. Eugenics Society, London, 
. 1989b. Evolution of the brain in early hominids. Ossa 14:49-62. 


. 1990. Primate Origins and Evolution: A Phylogenetic Reconstruction. Chapman 

Hall/Princeton University Press, London/New Jersey. 

. 1993. Primate origins: Plugging the gaps. Nature 363:223-234. 

. 1 996. Scaling of the mammalian brain: the maternal energy hypothesis. News Physiol. 

Sci. 11:149-156. 

Martin, R.D. & A.D. Barbour. 1989. Aspects of line-fitting in bivariate ailometric analyses. 
Folia Primatol. 53:65-81. 

Martin, R.D. & A.M. MacLamon. 1985. Gestation period, neonatal size and maternal invest- 
ment in placental mammals. Nature 313:220-223. 

McHenry, H.M. 1 988. New estimates of body weight in early hominids and their significance 
toencephalizationandmegadontiain"robusf'australopithecines. Pages 133-148 //zGrine, 
F.E., ed., Evolutionaiy History of the "Robust " Australopithecines. Aldine de Gruyter, 
New York. 

. 1992. Body size and proportions in early hominids. Amer. Jour. Phys. Anthropol. 


McNab, B.K. & J.F. Eisenberg. 1989. Brain size and its relation to the rate of metabolism in 
mammals. /Jmer. Nat. 133:157-167. 

Pagel, M.D. & P.H. Harvey. 1 988. How mammals produce large-brained offspring. Evolution 

. 1989. Comparative methods for examining adaptation in primates depend on evolu- 
tionary models. Folia Primatol. 53:203-220. 

. 1990. Diversity in the brain size of newborn mammals: Allometry, energetics or life 

history tactics? Bioscience 40: 1 16-122. 
Purvis, A. 1995. A composite estimate of primate phylogeny. Philos. Trans. Roy. Soc. Lond. B 

Purvis, A. & A. Rambaut. 1995. Comparative analysis by independent contrasts (CAIC): An 

Apple Macintosh application for analysing comparative data. CAB/OS 1 1 :247-25 1. 
Rak, Y. 1986. The Neanderthal: A new look at an old face. Jour. Hum. Evol. 15: 151-164. 
. 1 990. On the difference between two pel vises of Mousterian context from the Qafzeh 

and Kebara caves, Israel. Amer. Jour. Phys. Anthropol. 81:323-332. 
Rak, Y., W.H. Kimbel, & E. Hovers. 1996. On Neandertal autapomorphies discernible in Ne- 

andertal infants: A response to Creed-Miles Jour Hum. Evol. 30:155-158. 
Ringo, J.L., R.W. Doty, S. Dmeter, & P.Y. Simard. 1994. Time is of the essence: A conjecture 

that hemispheric specialization arises from interhemispheric conduction delay. Cereb. 

Cortex 433\-343. 
Sacher, G.A.I 982. The role of brain maturation in the evolution of the primates. Pages 97-1 1 2 

//; E. Armstrong & D. Falk, eds., Primate Brain Evolution. Plenum Press, New York. 
Savage-Rumbaugh, E.S. 1 986. Ape Language: From Conditional Response to Symbol. Oxford 

University Press, Oxford, U.K. 
Schwartz, J.H. & I. Tattersall, 1996. Significance of some previously unrecognized apomor- 

phies in the nasal region of Homo neanderthalensis. Proc. Natl. Acad. Sci. USA 

Sebeok,T.A. & J. Umiker-Sebeok, eds., 1980. Speaking of Apes: A Critical Anthology of Two- 

Way Communication with Man. Plenum Press, New York. 
Seidenberg, M. & L.A. Pettito. 1987. Communication, symbolic communication, and lan- 
guage: Comment on Savage-Rumbaugh, McDonald, Sevcik, Hopkins & Rupert (1986). 

Jour E.xp. Psychol.: Gen. 1 16:279-287. 
Shevlin, M. & M.N.O. Davies. 1997. Alphabetical listing and citation rates. Nature 388:14. 


Springer, M.S., G.C. Cleven, O. Madsen, W. W. de Jong, V.G. Waddell, H.M. Amrine, & M.J. 
Stanhope. 1997. Endemic African mammals shake the phylogenetic tree. Nature 

Stringer, C. & C. Gamble. 1993. In Search of the Neanderthals . Thames & Hudson, London, 

Terrace, H.S. 1 979. Mm.- A Chimpanzee Who learned Sign Language . Eyre Methuen, London, 

Tobias, P.V. 1987. The brain oi Homo habilis: A new level of organization in cerebral evolu- 
tion Jom/*. Hum. Evol. 16:741-761. 

Tregenza, T. 1997. Darwin a better name than Wallace? Nature 385:480. 

Vallois, H.V. 1954. La capacite cranienne chez les Primates superieurs et le 'Rubicon cere- 
bral.' C. R. Acad Sci. Paris 238:1349-1351. 

Zollikofer, C.P.E., M.S. Ponce de Leon, R.D. Martin, & P. Stucki. 1995. Neanderthal com- 
puter skulls. Nature 375:283-285. 

Zollikofer, C.P.E., M.S. Ponce de Leon, & R.D. Martin, (in press). Computer-assisted skull re- 
construction: A bridge between anthropology and medicine. Evol. Anthropol. 


Organization of Semantic 
Knowledge and the Origin of 
Words in the Brain 

Alex Martin 

National Institute of Mental Health 
Laboratory of Brain and Cognition 
Building 10Room4C-104 
10 Center Drive MSC 1366 
Bethesda,MD 20892-1366 

What does it mean to claim that a nonhuman species has words? Recent evi- 
dence from cognitive neuroscience indicates that meaning or semantic information 
about a particular object is represented as a distributed network of discrete corti- 
cal regions. Within this network the features that define an object are stored close 
to the sensory and motor regions of the brain that were active when information 
about that object was acquired. These semantic representations are active when- 
ever the object is perceived and when its name is produced or heard. The organiza- 
tion of semantic information parallels the organization of the sensory and motor 
systems in the primate brain. Evidence of similarities in the way object informa- 
tion is stored in the cerebral cortex of human and nonhuman primates may pro- 
vide a means for assessing the referential status of nonhuman vocalizations. 

Until very recently, our understanding of the organization of language in the brain 
has come from the study of adult humans with deficits in specific language abilities as 
a result of focal brain damage. For example, it has been known since the latter half of 
the 1 9th century that damage to the posterior region of the left temporal lobe (now re- 
ferred to as Wernicke's area) can produce impaired speech comprehension, whereas 
damage to the inferior region of the left frontal cortex (now referred to as Broca's 
area) can produce impaired speech production. These observations have held-up re- 
markably well (Figure 1) and during the hundred years following Broca and Wer- 
nicke we have learned a great deal about how language is represented in the brain 
from the study of language-impaired individuals (for an excellent review and synthe- 
sis see Caplan 1992). 

The recent advent of fianctional brain imaging technologies {e.g., positron emis- 
sion tomography, PET; and functional magnetic resonance imaging, fMRI) have ex- 

The Origin and Diversification o/ Language Memoirs of the California Academy of Sciences 

Editors, N.G. Jablonski & L.C. Aiello Number 24. Copyright ©1998 



Wernicke's aphasia 

Broca's aphasia 

Figure l . The classical language zones. Location of lesions (dark areas) in individual subjects producing 
disorders of speech comprehension (Wernicke's aphasia) and speech production (Broca's aphasia) 
(adapted from Mazzocchi & Vignolo 1 979). The lesion associated with speech comprehension disorders is 
typically in the posterior, superior region of the left temporal lobe; (a) The lesion as sociated with disorders 
of speech production is typically in the inferior region of the left frontal lobe; (b) Comp ass directions: S = 
superior, P = posterior, I = inferior, A = anterior. 

tended our knowledge of brain organization by providing a means to more directly 
observe the brain during the performance of cognitive tasks (see Toga & Mazziota 
1996 for a description of these methods). In this chapter I will describe some recent 
findings on language processes and brain organization. Rather than focusing on the 
formal characteristics of language (i.e., word and sentence processing), these studies 
focus on visual processing of objects as a means of investigating how information 
about object properties is represented in the brain. The results of these studies suggest 
a close link between the sites where information about specific features of objects are 
stored and the sensory and motor systems that were active when that information was 
acquired. I will then turn to a discussion of these and related findings for understand- 
ing the organization of semantic knowledge about concrete objects, and the implica- 
tion of these findings for understanding the origin of word meaning. 



Historically, research on language has focused on grammar and syntax, and the 
apparatus through which language is expressed, namely speech. The available evi- 
dence support the view that grammar and speech are singularly human. As a result, 
focusing on these characteristics of language has served to highlight and to reinforce 
the discontinuity between man and the rest of the animal kingdom. Grammar, as a 
rule-based, generative system provides a powerful tool for communication. Given a 
finite set of rules and tokens (words) an infinite number of meaningful utterances 
(sentences) can be generated, thereby providing a highly efficient solution or adapta- 
tion (Cosmides & Tooby 1995) for solving the problem of intraspecies communica- 
tion. Nevertheless, we should not lose sight of the fact that language has a function 
and this function is to communicate meaning, not to generate grammatically correct 
utterances or to manipulate our vocal cords. What we communicate is our experience 
of the external world, and our internal world — our images, prelinguistic thoughts, 
and feelings. These experiences, in turn, are gained through acting in the world (via 
learned patterns of movement that define our interaction with objects), and through 
perception (via smell, taste, touch, audition, and especially vision). We, as well as our 
closest living relatives, are visual animals and the meaning or semantics of things in 
the world is intimately linked to vision. 

As with language, vision also serves a specific fianction: Building a description, or 
representation of the shape and position of objects (Marr 1982). Thus vision is pri- 
marily about representing what objects look like (their shape, color, texture) and 
where objects are in relation to ourselves and other objects (their location and move- 
ment in space), and language serves to convey these experiences, these representa- 
tions, to ourselves and others. 

Visual Semantics and Words in Monkeys 

In a series of groundbreaking studies, Robert Seyfarth and Dorothy Cheney have 
shown that, in the wild, vervet monkeys emit vocalizations that seem to function as 
words (Cheney & Seyfarth 1990; Seyfarth & Cheney 1992; Marler, this volume). In 
support of this claim they showed that these monkeys produced specific vocalizations 
or alarm calls in response to different predators. One type of call was sounded when a 
martial eagle flew over head, another call was produced when a leopard was spotted 
nearby, and a third call was produced in response to a snake. Not only were the calls 
different, but importantly, other monkeys near enough to hear these calls responded 
differently depending on which call was heard. For example, Seyfarth and Cheney 
observed that the monkeys would look up and hide in the bushes when the martial ea- 
gle call was heard, climb into nearby trees when the leopard alarm call was heard, and 
stand on their hind legs and search the nearby grass when the snake alarm was heard. 
Moreover, these alarm call-specific behaviors were observed in response to taped re- 
cordings of the calls even though no predators were present. As Seyfarth and Cheney 
suggest, the alarm calls functioned as representational or semantic signals (Seyfarth 
& Cheney 1992) and thus share at least one critical feature with human words: They 
are referential. The calls refer to or stand for a particular object in the environment, 
and, based on the behavior of the monkeys that hear the call, they elicit a representa- 


tion of that object in the mind of others. So, as Seyfarth and Cheney would have it, 
alarm calls are like words because they are representational. 

The jury is still out on whether the alarm calls of monkeys fiinction the same way 
words function in human language and in the human mind. (Although, see Hauser 
1996, for suggestions on a research program that would help to answer this question). 
The calls have some characteristics in common with human words, but apparently not 
others (for example, monkeys do not appear to communicate with others with the in- 
tention of influencing the mental state of the listener; Seyfarth & Cheney 1992). The 
issue of whether it is appropriate to consider alarm calls as equivalent to human words 
in all, or even any, meaningful sense will not be settled here. Rather I would like to 
address a related question: What does it mean to claim that a word (or in the case of 
the vervet monkey, an alarm call) is representational? Specifically, what type of pro- 
cesses go on inside the brains of the alarm-call producer and the alarm-call hearer that 
could allow an arbitrary grouping of sounds to "represent" an object? 

Where it is Perceived, it is Learned 
Where it is Learned, it is Stored 

Accumulating evidence from the fields of cognitive psychology and cognitive 
neuroscience suggest a number of conditions that may need to be fiilfilled in order to 
give alarm calls representational status. 

1 ) From an information-processing view, object perception is associated with the 
activation of several qualitatively distinct representations. Chief among these is a 
perceptual representation based on the object's physical features, and a semantic rep- 
resentation based on previously acquired information about the object {e.g., Hum- 
phreys et al. 1988; Martin 1992) (Figure 2). 

2) The semantic representation consists of information about the features and at- 
tributes that define the object. Thus, compared to the perceptual representation, the 
semantic representation is more abstract, containing information about the class of 
objects, rather than the specific object being viewed. For the vervet monkey, the se- 
mantic representation of the martial eagle might include previously learned informa- 
tion about the shape, color and pattern of motion that characterizes these predators 
and distinguishes them from other similar objects {i.e., harmless birds). Additionally, 
the semantic representation would include information about the consequences of 
confronting this predator, which, in turn, produces fear. 

3) Within the brain, the information that defines the semantic representation is not 
localized in any one place. Rather, it is distributed among a number of areas accord- 
ing to the type of object feature being represented. Information about the color of the 
eagle is stored separately from information about its form, which in turn is stored 
separately from information about its pattern of movement. 

4) The locations or sites in the brain where object feature information is stored are 
not distributed arbitrarily. Rather each feature is stored close to the region of the brain 
that mediates perception of that attribute. For example, information about the color of 
an object is stored near the area of the brain that mediates perception of color; infor- 
mation about motion is stored near the area that mediates perception of motion. 

5) The attributes and features critical for uniquely identifying an object {i.e., to 
distinguish it from other objects in the same semantic category; from other animals, 
other tools, other pieces of furniture, etc.) are bound together as a result of having ex- 




-I LU 








< LU 
2 CO 
CO 01 



-I 2 

< h- 

3 < 

H I- 




Q^ LU 

LU q: 


< O 
^ 111 UJ 

< l-o 

^ < -J 

-y o 





CO •=- 

I- o 

LU u 






X -c 

E 2 


perienced them together (i.e., they co-occur in time during perception). This binding 
constitutes a network of discrete cortical regions: a semantic network. 

6) This semantic/cortical network is activated automatically (i.e., obligatorily, 
quickly, and outside of conscious awareness) when an object is viewed. This point 
becomes evident when we consider that "seeing" as we commonly use the term, refers 
to "seeing as." To perceive an object is to identify it as a member of some class (i.e. , a 
"dog," "chair," "word," or even a "shapeless form"). Perception, in this everyday us- 
age of the term, cannot occur without the automatic activation of previously acquired 
information. Typically, for example, it takes less than 700 msec for subjects to name 
line drawings of common objects, even though the drawings have not been seen pre- 
viously by the subjects. How could we quickly recognize or identify an object as be- 
ing of a particular type [e.g., a chair, pencil, or kangaroo) if we did not have prior 
knowledge of chairs, pencils, and kangaroos? 

7) In humans, semantic networks are activated not only when objects are seen, but 
also when the object's name is read, heard, or retrieved in the service of writing and 
speech. The name serves as a powerftil, economical, shorthand description of just 
those features that uniquely define the object. 

Based on these claims, then, to argue that the alarm calls of the vervet monkey 
function like words, is to argue that the semantic representation of "martial eagle" is 
activated not only by perception of the eagle, but also by the sound of the alarm call. 
This is the critical claim. The alarm call represents the martial eagle because both the 
perception of the eagle and the alarm call activate a common semantic network in the 
brain. Presumably, as with the learning of feature-object associations (i.e., learning 
about the form, color, pattern of motion, etc. that define "martial eagle"), the learning 
of the alarm call-object association may be dependent on the coccurrence of environ- 
mental events. In this case, the association between hearing the call and the percep- 
tion of the eagle, and perhaps also the association between hearing the call and the 
perception of the call's affect on the behavior of other monkeys. 

What is the evidence to support these claims? Before turning to that question it is 
necessary to briefly describe the organization of the primate visual system. 

Organization of the Primate Visual System 

The visual system evolved to solve the problem of representing the world. We re- 
main quite far from a formal understanding of how perception is accomplished (i.e., 
understanding the computations performed in enough detail to build a device that 
could accomplish the type of simple visual recognition tasks — such as identifying 
objects from multiple viewpoints — that we accomplish quickly and effortlessly). 
Nevertheless, we have gained considerable knowledge about the locations and fianc- 
tional properties of the brain regions that mediate vision. 

An important starting point for understanding complex brain systems like those 
that mediate vision is the "principle of modular design" (Simon 1962; Marr 1982). 
This principle asserts that complex problems are solved by breaking them down into 
smaller, and presumably more manageable parts. The main idea is that complex sys- 
tems are composed of subsystems that are as nearly independent of one another as the 
overall fiinction of the system will allow. If such independence did not exist then 
even a small change in one part of the system would change the entire system. Thus, 
in order for evolution to occur in a nonmodular scheme, each change would need to be 
accompanied by numerous and simultaneous changes throughout the entire system. 



Modular design allows for the possibility of modifying or creating new subsystems 
without necessitating change in all other subsystems. 

Until the recent advent of functional neuroimaging in man, the bulk of our knowl- 
edge about the modular design of the visual system has come from investigation of 
one of our closest living relatives, the Old World monkey. Studies of the cortical sur- 
face of the macaque have revealed at least 30 separate visual areas occupying nearly 
one half of monkey cortex (Felleman & Van Essen 1 99 1 ). These regions are broadly 
organized into two ftinctional processing streams (Figure 3). An occipitotemporal 
stream that subserves object vision and an occipitoparietal stream that subserves spa- 
tial vision and visual guidance of movements towards objects in space (for recent re- 
view see Ungerleider & Haxby 1994). These processing streams are organized 
hierarchically, with increasingly complex neuronal response properties as one pro- 
ceeds from primary visual areas in occipital cortex to more anterior sites in the tempo- 
ral lobe (for the object processing) and parietal lobe (for spatial processing). 


Figure 3. Schematic diagram of the two cortical visual systems. Visual information is received in pri - 
mary visual cortex in the occipital lobe { V 1 ) and then brought forward along two separate processing path - 
ways. An occipitotemporal pathway that mediates object vision (from VI towards the inferior temporal 
cortex, IT) and an occipitoparietal pathway that mediates spatial vision (from VI towards inferior parietal 
cortex, IP). Each processing stream has extensive feedfoiAvard and feedback connections (bidirectional ar - 
rows), as well as connections between regions assigned to different processing streams. I n the monkey 
over 30 visual processing regions have been identified, only a few of which are shown here. MT is critical 
for motion perception and V4 is critical for perception of form and of color (for details se e Ungerleider & 
Haxby 1994). 


The need for this separation of function becomes apparent when one considers the 
work that each system is required to perform. The object recognition system must be 
designed in such a way as to allow recognition of a particular object as the same object 
regardless of its position in the visual field. Consistent with this requirement, 
whereas neurons in the primary visual cortex (in the occipital lobes) have smal 1 visual 
fields, neurons fiirther up stream (in the temporal lobes) have large visual fields (i.e., 
they respond to a particular object over a large region of space). As a result, informa- 
tion about the exact location of an object is sacrificed in the service of object recogni- 
tion. While this neural architecture is perfectly suited for visual recognition, it is 
particularly ill-suited for the job of object localization. Therefore another system, 
with a different architecture, is needed to keep track of position in space. This job is 
accomplished by the occipitoparietal spatial processing stream. 

Studies of brain-damaged patients have documented a similar organizational 
scheme in the human brain. For example, damage to the parietal lobes (especially the 
right parietal lobe) can produce disorders of spatial cognition, including selective 
deficits in perceiving object locafion and orientation (Newcombe & Ratcliff 1989), 
while temporal lobe lesions can produce deficits in object recognition and object 
naming (especially after left temporal lobe lesions) (Damasio et al. 1989). 

At a more specific level of analysis, modular design extends to functioning within 
each processing stream as well. For example, regions in the lower, or inferior aspect 
of the occipital lobes respond to an object's form and color, whereas other more supe- 
riorly located regions respond to an object's direction and speed of movement 
through space (Desimone & Ungerleider 1989; see Figure 3). Again similar findings 
have been documented in humans. Damage to the inferior region of the occipital lobe 
can produce the syndrome of achromatopsia (acquired color blindness; Damasio et 
al. 1980), whereas a more superiorally located lesion at the junction of the occipital, 
temporal, and parietal lobes can produced akinetopsia (acquired motion blindness; 
Zeki 1991). Thus, within the cortex, the beginning stages of object recognifion are 
mediated by relatively independent neural modules that subserve perception of spe- 
cific features of the visual scene. The previously mentioned perceptual representa- 
fion of an object is mediated by these regions of the brain. 

Investigating the Organization of Semantic Attributes 

The idea that information about the attributes that define an object may be distrib- 
uted among different regions of the brain is not new. Wernicke, for example, argued 
that word forms were stored in one location and received input from visual and other 
sensory modalities in which modality-specific images of the object concept were 
stored (as discussed in Head 1926). More recent formulations of the idea that infor- 
mation about different object attributes and features are stored in separate cortical re- 
gions has been championed, in somewhat different forms, by Elizabeth Warrington 
and her colleagues (Warrington & Shallice 1984), Antonio Damasio (1989), and 
Glynn Humphreys (Humphreys & Riddoch 1987), among others (for review see Saf- 
fran & Schwartz 1994; andsee Allport 1985 for an influential formulation of distrib- 
uted semantic representations from a cognitive view). Evidence for this claim, 
however, has been rather indirect, based almost exclusively on single-case studies of 
patients with unusual and rare syndromes resulting from focal brain damage and dis- 


In order to provide more direct evidence, we set out to investigate semantic repre- 
sentations in the normal human brain using PET (Martin et al. 1995). Our working 
hypothesis was that semantic information about an object is stored as a distributed 
system in which the attributes that define the object are represented in, or near, the 
same tissue that is active during perception. Thus, for example, information about the 
color of an object {e.g., that kangaroos are tan rather than blue) is stored near regions 
active during color perception, while information about the pattern of motion associ- 
ated with an object {e.g., that kangaroos hop rather than gallop) is stored near regions 
active during the perception of motion. 

To explore this hypothesis we focused on knowledge about color and motion for 
two reasons. First, as mentioned previously, lesion studies of both monkeys and hu- 
mans have shown that perception of color and motion are mediated by separate sys- 
tems that include distinct regions of the occipital cortex. Moreover, the anatomical 
location of these regions in man had been identified in earlier PET studies (Corbetta et 
al. 1990; Zeki et al. 1991). In addition, studies of patients with focal cortical lesions 
indicated that retrieval of information about each of these attributes could be selec- 
tively disrupted, thus suggesting that semantic information about color and motion 
also may be stored in separate brain regions. For example, there are reports of patients 
with selective deficits in naming colors (color anomia; Geschwind & Fusillo 1966) 
and in retrieving information about an object's typical color (color agnosia, Luzzatti 
& Davidoff 1994), while other patients have had selective difficulty naming actions 
associated with the use of an object {i.e., a selective deficit producing action verbs; 
Caramazza & Hillis 1991; Damasio & Tranel 1993). 

The design of our experiments was straightforward. Subjects were shown black 
and white line-drawings of common objects. During one PET scan they were asked to 
name the objects, during a second scan they were asked to retrieve the name of a color 
commonly associated with the objects, and during a third scan they were asked to 
name an action commonly associated with the objects. For example, in response to an 
achromatic line drawing of a child's wagon, subjects responded "wagon," "red," and 
"pull" during the object naming, color word generation, and action word generation 
conditions, respectively. A second experiment was also run that was identical to the 
first experiment except that the subjects were presented with the written names of ob- 
jects, rather than pictures (Martin et al. 1995). 

Both experiments yielded similar results. Regardless of whether the stimuli were 
words or pictures, retrieving information about color activated the inferior region of 
the temporal lobe, just anterior to (in front of) the area known to mediate color percep- 
tion, whereas retrieving information about action activated a more superior region of 
the temporal lobe, just anterior to the area known to mediate motion perception (Fig- 
ure 4). 

These results showed that requiring subjects to retrieve previously learned infor- 
mation about object color and motion activated regions of the brain near those that 
mediate perception of those attributes. This occurred even though the pictures and 
words presented to the subjects were colorless and motionless. These data thus pro- 
vided strong evidence for the initial hypothesis: Semantic information about an ob- 
ject appears to be stored as a distributed system in which the attributes that define the 
object are represented near the same tissue that is active during perception. Most im- 
portantly, they suggested that the organization of information storage in the brain fol- 
lows a similar design as the organization of sensory and perhaps motor systems as 
well. Information about an object {e.g., a wagon) is not stored as a single entity in a 
single place. Rather the attributes and features that define the object are stored sepa- 







































o „> ;^ 

z *= 

on c -^' 
■-S ^ oo 

S t^ 



































ID 1^ '«_ >/-l 










5 I .£ 

on «-> c '-^ 

C ^ -^ 73 

^ CO X! « 

're o o c 

■5 "^ 

4^ a> 

E -i- ii. 

.5 J u 










rately. In addition, the sites of storage are near the areas active during the perception 
of, and therefore when we learned about, those attributes. 

More recent work from our laboratory has suggest that brain regions specifically 
associated with perception are themselves not reactivated when perceptually-based 
information is retrieved (Chao & Martin submitted). Thus it appears that retrieving 
information about a specific object attribute like color requires activation of a region 
of the brain situated close to, but not including, the neural circuitry involved in the 
perception of color. This finding is consistent with reports of patients with cortical le- 
sions that can no longer see colors, yet retain the ability to imagine colors (Shuren et 
al. 1 996), and patients who can no longer see objects, yet retain the ability to imagine 
objects (Behrmann et al. 1992). Taken together, the available evidence suggests that 
information about object features are stored near, but not in, the tissue active during 
perception of those features. 

Category-Specific Knowledge 

The PET studies established three important points with regard to earlier claims 
about the organization of semantic information. First, retrieving information about 
different object-associated features is associated with activity in different regions of 
the brain. Second, the locations of these regions are close to, but do not include, areas 
associated with the perception of those features. And, third, these regions are active 
during word retrieval. Retrieving words in appropriate context requires activation of 
a semantic representation. 

To perform these tasks correctly, however, required an appreciable amount of 
work or effort on the part of the subjects. Specifically, they had to focus attention on 
previously acquired information about a specific object feature and then find the right 
word to express that knowledge, all under considerable time pressure (a new stimulus 
appeared every two seconds). In the model of object recognition sketched previously, 
however, semantic representations are activated automatically, without effort, when- 
ever an object is seen. Therefore, the next question became: Could we find evidence 
for the automatic activation of a semantic/cortical network during object identifica- 

Our strategy for pursuing the answer to this question was the same as the one we 
used in the color and action knowledge studies. Specifically, we pitted two different 
categories against each other to determine whether different brain areas became ac- 
tive. This time, however, rather than performing a mentally strenuous and attention- 
demanding task — retrieving words denoting specific attributes of objects, subjects 
performed a simpler, less demanding task — naming objects. In fact, for the condi- 
tions to be discussed below, subjects were not required to overtly produce names, but 
merely to view objects and name them silently to themselves. 

The categories we chose to study were animals and tools (Martin et al. 1 996). As 
with our choice to study knowledge about color and action, this choice of categories 
was motivated by reports of brain-damaged patients. In this case, reports of patients 
with selective difficulty naming and answering questions about living things (War- 
rington &Shallice 1984; Farahe/ a/. 1991; Silveri&Gainotti 1988; Sheridan & Hum- 
phreys 1993), and reports of patients with deficits limited to man-made objects 
(Warrington & McCarthy 1983; Warrington «fe McCarthy 1987; Sacchett & Hum- 
phreys 1992). 


How could such category-specific impairments occur? One approach to this ques- 
tion has been to deny that these selective deficits actually do occur. This is not an un- 
reasonable position given that category-specific disorders are rare, the 
between-category dissociation often not pure {i.e., naming may be abnormal for 
members of both categories, but greater for one than the other), and the findings open 
to alternative interpretations. In particular, it has been argued that the selective deficit 
in naming living things is an artifact of differences in the visual complexity of the 
stimuli used to depict living things and man-made objects (Fennell & Sheridan 1992; 
Stewart et al. 1992). Animals, for example, do tend to have more visually complex 
forms than tools. As a result, pictures of animals may be more difficult to identify 
than pictures of tools; especially for a brain-damaged subject. However, such physi- 
cal differences cannot account for the opposite finding (i.e.. more difficulty naming 
man-made than living objects) and recent studies controlling for visual complexity 
have reduced, if not negated, the explanatory power of the visual-complexity criti- 
cism (Farah et al. 1996). 

Given that category-specific disorders do occur after brain damage, most investi- 
gators have relied on some form of a semantic feature model to explain their occur- 
rence. In general, it has been argued that the critical feature used to differentiate 
members of the category "four-legged animals" is knowledge about physical fea- 
tures. We learn to distinguish animals by their physical characteristics. Moreover, 
the differences between animals can be quite subtle (consider for example, the differ- 
ence between horses, donkeys, and mules, or leopards, tigers, and jaguars). Tools, in 
contrast, while differing in physical form, also have specific functional properties, 
and these functional properties are the critical component of their definition. This dif- 
ference in the types of attributes that define animals and tools can be easily verified by 
consulting a dictionary. This has in fact been done and the results showed that the ra- 
tio of physical properties to functional properties is much greater in the dictionary 
definitions of animals than tools (Farah & McClelland 1991). 

This difference in the attributes and features that define animals and tools suggests 
that the developmental histories of animal and tool learning may differ as well. 
Knowledge about the unique physical features that define each animal would be ac- 
quired primarily through object vision, whereas knowledge about tool differences 
would be acquired through the motor system (patterns of dominant hand movement 
learned through the use of tools) and motion vision (patterns of motion learned 
through observation of tool use by ourselves and others). If these are the kinds of in- 
formation needed to identify objects, then one would predict that naming animals 
would require greater activation of previously acquired information about shape or 
form than would tool naming, whereas tool naming would require activation of infor- 
mation about visual motion and motor movements associated with tool use. 

To investigate this hypothesis we asked subjects to silently name briefly presented 
pictures of objects (Martin et al. 1996). Animal pictures were presented during one 
PET scan and pictures of tools were presented during another PET scan. In addition, 
subjects were scanned while attending to pictures of nonsense object foims, and 
while staring at visual noise patterns. Several findings emerged from this study. 
First, the outer or lateral regions of the left and right occipital lobes were active when 
subjects perceived objects, regardless of whether they were meaningful (animals and 
tools) or meaningless (nonsense object forms) (see Malach et al. 1995 for a similar 
finding). This finding suggests that this region of occipital cortex is critical for per- 
ceiving object form, and therefore associated with the perceptual, but not the seman- 
tic representation of an object (Figure 5). Second, in contrast to nonsense objects, 













-?i"~^'^ 'S-,' t^ 

J"-*^ -: 

-a c 

^ O r- 

3 5 — 

^ !L> CC 

-a 33 ra 

3 &■ o 

c 5 

5 -ii S 


naming meaningful objects (animals and tools) activated the inferior region of the 
temporal lobes {i.e., a more anterior aspect of the object vision pathway) suggesting 
that this portion of the temporal lobe may be the site for stored information about ob- 
ject form. Third, naming meaningful objects (animals and tools) activated a region of 
the inferior frontal lobe known to be associated with speech (Broca's area, Figure 5). 
Fourth, in addition to areas active when naming both animals and tools, other regions 
of the brain were selectively activated by naming objects from one category or the 

In comparison to naming animals, tool naming activated a region of the left tem- 
poral lobe that was nearly identical to the region active in the previously discussed 
studies of action word retrieval (Figure 6). As discussed previously, this region is 
situated just anterior to (in front of) the area known to be active during motion percep- 
tion. This finding provides additional evidence that this region of the left temporal 
lobe may be a critical site for stored information about object motion. In addition, tool 
naming was associated with activation of a region of the left premotor cortex situated 
just anterior to the primary motor cortex that controls right-sided body movement 
(Figure 6). The region of premotor cortex active during tool naming was nearly iden- 
tical to an area previously found to be active when subjects imagined manipulating 
objects with their right hand (Decety et al. 1994). Thus this region of left premotor 
cortex may be the site for stored information about patterns of hand movements asso- 
ciated with tool use. 

In contrast, the only brain region more active for animal naming than tool naming 
was on the inner or medial surface of occipital cortex, greater on the left than on the 
right (Figure 6). This region includes the calcarine cortex which is the first cortical 
area to receive visual information from the eyes. This finding might be viewed as 
supporting the idea that category-specific impairments are simply a byproduct of the 
visual complexity of the pictures, as discussed previously. This explanation, how- 
ever, was eliminated by the results of a separate study that again found greater medial 
occipital lobe activity for animal than for tool naming even though the pictures were 
equated for visual complexity by transfoiTning them into silhouettes (see Martin et al. 
1 996 for details). These results suggest that this early-stage, occipital visual process- 
ing area may be reactivated in top-down fashion by regions higher up in the object vi- 
sion pathway (perhaps via feedback connections from the inferior temporal region 
associated with identifying meaningful objects). Reactivation of the medial occipital 
region may be necessary to uniquely identify an object when relatively subtle differ- 
ences in physical features are the primary means by which the object can be distin- 
guished from other members of its category. Converging evidence for these findings 
has been provided by a recent study of brain-damaged patients with category-specific 
naming deficits (Tranel et al. 1997). 

Object-Associated Affect 

The evidence reviewed so far relates to some of the cognitive aspects of object 
meaning. However, in addition to information about physical and functional fea- 
tures, object meaning can also be emotionally laden. Viewing scenes of accidents, 
surgical procedures, and the like have an aversive component (and have measurable 
affects on the autonomic nervous system), whereas pleasant feelings can be elicited 
by pictures of puppy dogs, flowers, and tranquil environments, etc. Moreover, many 
individuals seems to have an instinctive fear of certain animals (spiders, rats, bats) 



■^ 5 



and situations (public speai<ing, heights) and the negative feehngs associated with 
these experiences can be elicited by viewing pictures. Alarm calls must have an af- 
fective component to serve as warnings. In order to function as a warning the sight of 
a predator {e.g., the martial eagle), and the alarm call that represents it, must elicit an 
emotional response that signals danger and the appropriate avoidance behavior. 
Therefore, in addition to feature information, the semantic representation of the mar- 
tial eagle should include affective information. Similarly, the semantic network acti- 
vated in the brain of the monkey should include regions that were active when the 
martial eagle-fear association was established. 

Neurobiological studies have established that the amygdala, a structure in the me- 
dial region of the anterior temporal lobe, plays a central role in fear conditioning (for a 
recent review see Ledoux 1996). In addition, studies of patients with lesions confined 
to the amygdala (e.g., Adolphs et al. 1994), and functional brain imaging studies of 
normal subjects (Breiter et al. 1 996; Morris et al. 1 996), provide converging evidence 
that the amygdala is involved in the visual recognition of emotional expression. For 
example, in one study (Morris et al. 1996) subjects were shown faces and had to de- 
cide whether the individual depicted was male or female. The faces depicted different 
expressions (happy, fear) and varied according to the intensity of expression (ranging 
from neutral to most fearful or most happy). The data showed that activity in the left 
amygdala was significantly correlated with the intensity of expressed fear. This oc- 
curred even though the task required only gender discrimination, not an emotional re- 
sponse, nor a judgment about emotional expression. This finding is consistent with 
the idea that the affective valence associated with an object is represented near, if not 
in, a region critical for learning object-affect associations. Brain regions associated 
with establishing relationships between objects and emotions may be another compo- 
nent of the semantic network that is automatically engaged whenever an object is seen 
or its name (or alarm call) heard. 

Are Alarm Calls Words? 

In this chapter I have described the locations and functions of some of the brain re- 
gions that underlie semantic knowledge about concrete objects in humans. What is 
the relevance of these findings for assessing the referential status of alarm calls in ver- 
vet monkeys? Imagine that we could perform a functional brain imaging study of an 
alert, vervet monkey (such an undertaking would be fraught with technical difficul- 
ties; however, we will ignore these problems for the sake of our discussion). One pos- 
sible outcome would be that the visual presentation of a martial eagle activates a 
network of cortical regions, and that this same network is activated by the auditory 
presentation of the martial eagle alarm call (excluding, of course, the primary visual 
and auditory processing areas associated with the presentation of the object and the 
call, respectively). Activation of an identical network of brain regions when viewing 
the predator and hearing the alarm call would provide support for the claim that the 
alarm call "stands for" the object. That is, we could argue that the object and the call 
mean the same thing to the monkey because they elicit identical states in the 
monkey's brain. 

At the other extreme, there may be no overlap in the regions activated by the object 
and the alarm call. For example, whereas the visual presentation of the eagle might 
activate a distributed network of cortical areas associated with stored information 
about form and motion, the alarm call might activate a limited circuit comprised of 


auditory cortex and limbic structures associated with learned fear. In this case, the as- 
sociation between the alarm call and the monkey's behavior in the wild would be 
more like a simple, conditioned fear response (although I think there is good evidence 
to reject this extreme point of view; see Cheney & Seyfarth 1990). The call would 
elicit a behavioral response without an intervening stage of cognitive mediation. We 
could then argue that the call no more "referred" to a martial eagle for the monkey, 
than the bell referred to a steak for Pavlov's dog. 

Concluding Comments: Knowledge Primitives 
and the Embodied Mind 

Writing on the structure of categories, the linguist George Lakoff stated that 
"Thought is embodied, that is, the structures used to put together our conceptual sys- 
tems grow out of bodily experience and make sense in terms of it; moreover, the core 
of our conceptual systems is directly grounded in perception, body movement, and 
experience of the physical and social order" (Lakoff 1987: xvi). The evidence and ar- 
guments presented in this chapter support this view. 

The object semantic system discussed here is seen as consisting of learned infor- 
mation about features and attributes that uniquely define an object. This information 
is represented in the brain as a distributed network of discrete regions in which the at- 
tributes that defme the object are stored near the regions active when this information 
was acquired. These include the sensory and motor systems through which we act in. 
and obtain our experience of, the world. It was further argued that these representa- 
tions were active not only during object recognition, but during word recognition and 
production as well. If the alarm calls of the vervet monkeys are referential in the same 
way as human words, then they would be expected to have semantic representations 
that follow a similar organizational scheme. Given the similarity between the organi- 
zation of sensory and motor systems in human and nonhuman primates, the expecta- 
tion of a similarity in the organization and structure of objects semantics does not 
seem to be an unreasonable one. 

An important remaining question is how do words, and alarm calls, get linked to 
semantic representations? The idea of cooccurrence of events has limited explana- 
tory power in and of itself because it does not explain why certain events get linked 
(e.g., words with their referents) and others do not, nor does it explain why this learn- 
ing happens with ease. Clearly, humans are biologically prepared to establish a link 
between auditoiy sounds and object semantics. Perhaps vervet monkeys and other 
nonhuman primates are prepared to establish such links as well. Nevertheless, hu- 
mans are prepared to acquire a seemingly unlimited lexicon. The lexical system is 
both open-ended in capacity and highly flexible in its mapping of words to meaning 
{i.e., the mapping is many to many). Different words can express the same meaning 
and the same word can have different meanings. In contrast, the vervet monkey may 
be prepared to acquire only a limited, genetically detennined, lexicon that is closed 
and rigid in its mappings (one to one). 

The type of semantic information discussed in this chapter may be viewed as se- 
mantic primitives; as the building blocks out of which more refined shades of mean- 
ing could be constructed. The representation of meaning by multiple features stored 
in different brain regions, rather than as a single entity, provides combinatorial power 
for representing an infinite number of concepts using a finite number of features. As 
far as we know, this may be true for both humans and monkeys, but only humans may 


have the additional capacity to hnk a multitude of meanings to arbitrary sounds. I 
have only touched on a few potential candidates for such semantics primitives: 
knowledge about form, color, motion, action, and affective valence. To this list oth- 
ers could undoubtedly be added, including knowledge of time, space, and number. 
Although discussion of these semantic features are outside the scope of this chapter, I 
believe there is evidence from studies of both nonhuman and human primates that 
point to the existence of localized neural mechanisms that could form the basis for 
storing information about each of these concepts as well. 


I thank my colleagues, Leslie Ungerleider, Jim Haxby, Cheri Wiggs, and Francois 
Lalonde. I would also like to thank the organizers and sponsors for inviting me to par- 
ticipate in this exciting symposium. 

Literature Cited 

Adolphs, R., D.Tranei, H.Damasio, & A. Damasio. 1 994. Impaired recognition of motion in fa- 
cial expressions following bilateral damage to human amygdala. Nature 372:669-672. 

Allport, D.A. 1985. Distributed memory, modular subsystems and dysphasia. Pages 32-61 in 
S.K. Newman & R. Epstein, eds., Current Perspectives in Dysphasia. Churchill Living- 
ston, Edinburgh, U.K. 

Behrmann, M., G. Winocur, & M. Moscovitch. 1992. Dissociation between mental imagery 
and object recognition in a brain damaged patient. Nature 359:636—637. 

Breiter, H.C., N.L. Etcoff, P.J. Whalen, W.A. Kennedy, S.L. Raunch, R.L. Buckner, M.M. 
Strauss, S.E. Hyman, & B.R. Rosen. 1996. Response and habituation of the human amyg- 
dala during visual processing of facial expression. Neuron 1 7:875:887. 

Caplan, D. 1992. Language: Stnicture. Processing, and Disorders . MIT Press, Cambridge. 

Caramazza, A. & A. E. Hillis. 1 99 1 . Lexical organization of nouns and verbs in the brain. A^a- 
/t/re 349:788-790. 

Chao, L.L. & A. Martin, (submitted) Cortical representation of perception, naming, and knowl- 
edge of color. 

Cheney, L. & R.M. Seyfarth 1990. How Monkeys See the World: Inside the Mind of Another 
Species. University of Chicago Press, Chicago. 

Corbetta, M., P.M. Miezin, S. Dobmeyer, G.L. Shulman, & S.E. Petersen. 1990. Attentional 
modulation of neural processing of shape, color, and velocity in humans. Science 

Cosmides, L. & J. Tooby. 1995. Fromfunction to structure: The role of evolutionary biology 
and computational theories in cognitive neuroscience. Pages 1 199-1210 in M.S. Gazza- 
niga, ed., The Cognitive Neurosciences. MIT Press, Cambridge. 

Damasio, A.R. & D. Tranel. 1993. Nouns and verbs are retrieved with differently distributed 
neural systems. Proc. Nad. Acad Sci. U.SA . 90:4957^960. 

Damasio, A.R. 1 989. Time locked multiregional retroactivation: A systems level proposal for 
the neural substrates of recall and recognition. Cognition 33:25-62 . 

Damasio, A.R., T. Yamada, H. Damasio, J. Corbett, & J. McKee. 1980. Central achromatopn 
sia: Behavioral, anatomic, and physiologic aspects. Neurolog}' 30: 1064—1071 . 


Damasio, A.R., D. Tranel, & H. Damasio. 1989. Disorders of visual recognition. Pages 
317-332 m F. Boiler & J. Grafman, eds., Handbook of Neiiropsycholog}' vol 2. Elsevier, 
Amsterdam, The Netherlands. 

Decety, J., D. Perani, M. Jeannerod, V. Bettinardi, B. Tadary, R. Woods, J.C. Mazziotta, & F. 
Fazio. 1994. Mapping motor representations with positron emission tomography. Nature 

Desimone, R. & L.G. Ungerleider. 1989. Neural mechanisms of visual processing in monkeys. 
Pages 267-299 m F. Boiler & i. Grafman, eds.. Handbook of Neiiropsycholog)' vol 2. El- 
sevier, Amsterdam, The Netherlands. 

Farah, M.J. & J.L. McClelland. 1991. A computational model of semantic memory impair- 
ment: Modality specificity and emergent category specificity. Jour. Exp. Psychol. 

Farah, M.J., M.M. Meyer, & P. A. McMullen. 1996. The living/nonliving dissociation is not an 
artifact: Giving an a priori implausible hypothesis a strong test. Cog. Neiiropsychol. 

Farah, M.J. , P. A. McMullen, & M.M. Meyer. 1991 Can recognition of living things be selec- 
tively impaired? Newopsychologia 29:185-193. 

Felleman, D.J. & D.C. Van Essen. 1991. Distributed hierarchical processing in the primate 
cerebral cortex. Cereb. Cortex 1:1-47. 

Funnell, E. & J. Sheridan. 1992. Categories of knowledge? Unfamiliar aspects of living and 
non-living things. Cog. Neuropsychol. 9:135-154. 

Geschwind, N. & M. Fusillo. 1966. Color naming defects in association with alexia. Arch. 
Neurol. \5:U1-\A6. 

Hauser, M.D, \996. The Evolution of Communication. MIT Press, Cambridge, 760 pp. 

Head, H. 1 963 (first published 1 926). Aphasia and Kindred Disorders of Speech. Hafner Pub - 
lishing Company, New York. 

Humphreys, G.W. & M.J. Riddoch. 1987. On telling your fruit from vegetables: A considera- 
tion of category-specific deficits after brain damage. Trends Neurosci. 10:145-148. 

Humphreys, G. W., M.J. Riddoch, & P.T.Quinlan. 1988. Cascade processes in picture identi- 
fication. Cog. Neuropsychol. 5:67-103. 

Lakoff G. 1987. Fire, Women, and Dangerous Things: What Categories Reveal About the 
Mind. University of Chicago Press, Chicago. 

LeDoux, J. E. 1996. The Emotional Brain. Simon & Schuster, New York. 

Luzzatti, C. & J. Davidoff 1994. Impaired retrieval of object color knowledge with preserved 
color naming. Neuropychologia 32:933-950. 

Malach, R., J.B. Reppas, R.R. Beson, K.K. Kwong, H. Jiang, W.A. Kennedy, P.J. Ledden, T.J. 
Brady, B.R.Rosen, & R.B.H.Tootell. 1995. Object-relatedactivity revealed by functional 
magnetic resonance imaging in human occipital cortex. Proc. Natl. Acad. Sci. USA 

Marr, D. 1982. Vision. W.H. Freeman and Company, San Francisco, California. 

Martin, A. 1992. Semantic knowledge in patients with Alzheimer's disease: Evidence for de- 
graded representations. Pages 1 1 9- 1 34 in L. Backman, ed.. Memory Functions in Demen- 
tia. Elsevier North-Holland, Amsterdam, The Netherlands. 

Martin, A., J.V.Haxby, F.M. Lalonde, C.L.Wiggs, & L.G. Ungerleider. 1995. Discrete corti- 
cal regions associated with knowledge of color and knowledge of action. Science 

Martin, A., C.L. Wiggs, L.G. Ungerleider, & J.V.Haxby. 1 996. Neural correlates of category- 
specific knowledge. Nature 379:649-652. 

Mazzocchi, F. & L.A. Vignolo. 1979. Localisation of lesions in aphasia: Clinical-CT correla- 
tions in stroke patients. Cortex 1 5:627-1 72. 


Morris, J.S., CD. Frith, D.I. Perrett, D. Rowland, A.W. Young, A.J. Cadler & R.J. Dolan. 
1 996. A differential neural response in the human amygdala to fearful and happy facial ex- 
pressions. Nature 383: 812-815. 

Newcombe, F. & G. Radcliff. 1989. Disorders of Visuospatial Analysis. Pages 333-356 in F. 
Boiler, & J. Grafman, eds.. Handbook of Neuropsychology vol 2. Elsevier, Amsterdam, 
The Netherlands. 

Sacchett, C. & G.W.Humphreys. 1992. Calling a squirrel a squirrel but a canoe a wigwam: A 
category-specific deficit for artefactual objects and body parts. Cog. Neuropsvchol. 

Saffran, E. M. & M.F.Schwartz. 1994. Ofcabbages and things: Semantic memory from a neu- 
ropsychological perspective — a tutorial review. Pages 507-536 in C. Umilta & M. Mos- 
covitch, eds.. Attention & Performance XV. MIT Press, Cambridge. 

Seyfarth,R.M.& D.L. Cheney 1992. Meaning and mind in monkeys. Sci.Amer. 267:122:128. 

Sheridan, J. & G.W. Humphreys. 1993. A verbal-semantic category-specific recognition im- 
pairment. Cog. Neuropsvchol. 10:143-184. 

Shuren, J.E., T.G. Brott, B.K.Schefft & W. Houston. 1996. Preserved color imagery in an 
achromatopsic. Neuropsychologia 34:485-489. 

Silveri, M.C. & G. Gainotti. 1988. Interaction between vision and language in category- 
specific semantic impairment. Cog. Neuropsvchol. 5:677-709. 

Simon, H. A. 1962. The architecture of complexity. Proc. Amer. Philos.Soc. 106:467-482. 

Stewart, F., A.J. Parkin, & N.M. Hunkin. 1992. Naming impairments following recovery 
from herpes simplex encephalitis: Category specific? Quart. Jour. Exper. Psychol. 

Toga, A.W. & J.C. Mazziotta. 1996. Brain Mapping: The Methods. Academic Press, Inc., San 
Diego, California. 

Tranel, D., H. Damasio & A.R. Damasio. 1997. A neural basis for the retrieval of conceptual 
knowledge. Neuropsychologia 35: 1 3 19-1 328. 

Ungerleider, L.G. & J.V. Haxby. 1994. 'What' and 'where' in the human brain? Curr. Opinion 
Neurobiol. 4:157-165. 

Warrington, E. K. & R. McCarthy. 1983. Category specific access dysphasia. Brain 

Warrington, E.K, & T. Shallice. 1984. Category-specific semantic impairments. Brain 

Warrington, E.K. & R. McCarthy. 1987. Categories of knowledge: Further fractionation and 
an attempted integration. Brain 1 1 0: 1273-1296. 

Zeki, S. I99I. Cerebral akinetopsia (visual motion blindness). Brain 1 14:81 1-824. 

Zeki, S., J.D.G. Watson, C. J. Lueck, K.J. Friston, C. Kennard, & R.S.J. Frackowiak. 1 991 . A 
direct demonstration of functional specialization in human visual cortex. Jour. Neurosci. 


Neanderthals, Modern Humans 
and the Archaeological Evidence 
for Language 

Paul Mellars 

Department of Archaeology 
University of Cambridge 
Downing Street 
Cambridge CB2 3DZ, UK 

There seems to be widespread agreement that the full panoply of 'Upper Paleo- 
lithic' culture in Eurasia, with its rich technology, art, ceremonial and symbolic 
components, must reflect the presence of essentially '•modern' language patterns, 
broadly similar to those of present-day populations. The argument is developed 
here that the behavior associated with the earlier Middle Paleolithic/Neanderthal 
populations could well reflect a much simpler language structure, of the kind that 
Bickerton and others have described as 'protolanguage.' Arguably the most rel- 
vant features in this context are the apparent absence of explicit 'symbolic' behav- 
ior in the archaeological records of the Neanderthals; their apparently simpler 
patterns of strategic and long-range planning behavior; and simpler and less mor- 
phologically structured tool inventories, which may reflect more limited linguistic 
vocabularies and less 'categorical' forms of mental conceptualization. The much 
more sharply defined cultural and technological traditions which emerged during 
the Upper Paleolithic may also be highly significant for the development of lan- 
guage. If these contrasts are accepted, we must presumably look for the origins of 
essentially modern language in the ancestral African populations, from which the 
modern populations of Homo sapiens apparently emerged. It is argued that some 
strong indications of this may be discernible in the archaeological records from Af- 
rica and southwest Asia between 50,000 and 110,000 years ago. Finally, a critical 
distinction is drawn between cognitive pof<?/ffia/ and behavioral expression, which 
may help to explain some of the apparent 'anomalies' in the archaeological records 
of human development over the period of the archaic-to -modern human transi- 

Few issues have been more debated recently than the origins of language, and few 
issues are more problematic (e.g^., Parker & Gibson 1990; Bickerton 1990, 1995; Lie- 
berman 1991 ; Aiello & Dunbar 1993; Gibson & Ingold 1993; Pinker 1994; Aitchison 
1996; Dunbar 1996; Noble & Davidson 1996, etc.). The problems are obvious. First, 
what do we mean by 'language' — which is now generally seen as an extremely com- 

Thc Origin and Diversification of Language Memoirs of the California Academy of Sciences 

Editors, N.G. Jablonski & L.C. Aiello Number 24. Copyright ©1998 


plex and multifaceted communication system, different elements of which could well 
have emerged at different points in human evolution, conceivably in response to dif- 
ferent environmental and social stimuli? What were the kinds of stimuli which pro- 
vided the essential selective pressures for the evolution of language at different stages 
in our primate and early human past? Indeed, how far is language a biologically and 
genetically programmed componentof our behavior (in Tinker's [1994] terms, a 'lan- 
guage instinct') as opposed to a basic element of 'cultural technology' which 
emerged at some point (or points) as part of the broader patterns of cultural and tech- 
nological evolution? Finally, what possible sources of behavioral evidence — as re- 
flected in the archaeological records of human behavior — can we invoke to attempt 
to document the origins and emergence of language? In other words, what exactly are 
the likely behavioral and archaeological correlates of varying levels of complexity of 
language in different behavioral spheres? 

The complexity and intransigence of these issues are, to put it mildly, daunting. 
My own instinct in these situations is to move from the known to the unknown, much 
as one might tackle a complicated jigsaw puzzle. In other words I prefer to start at the 
point where no one would seriously question that complex, essentially modem lan- 
guage patterns must have existed, and then to consider what precedes this. 

For reasons I hope will become apparent as the paper proceeds a convenient point 
at which to begin the discussion is at the time of the apparent replacement of the 
'archaic' Neanderthal populations of Europe by new and apparently intrusive popula- 
tions of anatomically modem humans (Stringer & McKie 1996; Stringer & Gamble 
1993; Mellars 1996a). This will allow us to focus on a number of critical features of 
the behavioral and associated archaeological aspects of this transition which I will ar- 
gue may have a direct bearing on the nature and complexity of language, and may pro- 
vide some of the key clues as to how specific aspects of archaeological data may relate 
to specific aspects of language development. In the later part of the discussion I will 
look at some of the possible implications of this evidence for the earlier stages of lan- 
guage development. 

The Human Revolution in Europe 

In Europe the period centered on 35^0,000 BP (in conventional radiocarbon 
terms) witnessed two dramatic changes. On the one hand there was the effective re- 
placement of the earlier, archaic Neanderthal populations which had occupied the 
continent for at least the previous 200,000 years by populations of physically or ana- 
tomically modem humans, closely similar in most anatomical respects to ourselves 
(Stringer & McKie 1996; Stringer & Gamble 1993; Krings et al. 1997). And on the 
other hand there was a range of conspicuous changes in the archaeological records of 
human behavior which collectively define the transition from the Middle Paleolithic 
(or 'Mousterian') to the Upper Paleolithic periods (see Table l:Mellars 1973, 1989a, 
1989b, 1991, 1996a; Farizy 1990;Knechte/fl/. 1993). This transition took place at a 
point approximately midway during the last glacial period at a time when the more 
northerly parts of Europe, together with the Alps, the Pyrenees and other more local- 
ized mountainous areas were covered by substantial ice caps. Temperatures over this 
period were oscillatory but for much of the time were probably between 5°C and 1 0°C 
lower than those of the present day (Van Andel & Tzedakis 1996). Landscapes over 
large areas of Europe were dominated by predominantly open tundra or steppe-like 
vegetation, supporting large herds of reindeer, horse, bison and, in some areas, mam- 


Table l . Principal archaeological features of the 'Upper Paleolithic Revolution' in Europe. 

1 . New 'punch' blade and bladelet technology 

2. New forms of stone tools 

3. More imposed form in tool manufacture 

4. Rapid changes in tool forms 

5. Complex, shaped, bone/antler/ivoiy artifacts 

6. Personal ornaments (beads, perforated teeth, etc.) 

7. Incised/decorated bonework 

8. Representational art 

9. Long-distance transport of marine shells 

10. Musical instruments 

1 1 . Increased population size? 

1 2. Increased social group size? 

13. More 'stractured' occupation sites 

14. More specialized, strategically organized hunting? 

15. More organized raw material distribution 

moth. These would have provided a rich, if in some contexts sHghtly unpredictable, 
food supply for the contemporaneous human groups (Mellars 1985; Mithen 1990). 

Following almost a century of debate there now seems to be increasing agreement 
among both anthropologists and archaeologists that over at least the greater part of 
Europe the transition from Neanderthal to anatomically modem (or 'Cro-Magnon') 
populations must reflect a population replacement event, with the anatomically and 
behaviorally modem populations expanding fairly rapidly from east to west across 
Europe over the time range from ca. 43,000 to 35,000 BP (Stringer & Gamble 1993; 
Hublin 1990; Mellars 1992; Stringer &McKie 1996;Krings£'/a/. 1997;butsee Wol- 
poffet al. 1994 and Clark 1997 for conflicting views). Several separate lines of evi- 
dence have recently served to reinforce this pattem: The demonstrable presence of 
essentially anatomically modem human remains in both Israel and parts of Africa by 
around 100,000 BP — i.e., at least 50-60,000 years before their first appearance in 
Europe; the extraordinarily abmpt nature of the anatomical interface between the fi- 
nal Neanderthal and early anatomically modem humans — as reflected for example 
by the anatomical contrasts between the late Neanderthal remains fi"om Saint-Cesaire 
and Le Moustier and the early Cro-Magnon forms from Stetten, Mladec, Cro- 
Magnon etc.; and the rapidly increasing corpus of both nuclear and mitochondrial 
DNA studies of present-day human populations in different parts of the world which 
point fairly consistently to a relatively recent origin of the modem human genotype 
(probably within the last 200,000 years), and its most likely origin at some point in 
Africa (Aitken era/. 1992; Stonekinge/a/. 1992; Harpending era/. \993:Cannetal. 
1994; Stringer & McKie 1996; Tishkoff er al. 1996; Howell 1996; Brauer et al. 
1997). Recently, the same case has been greatly strengthened by the recovery of 
'ancient' mitochondrial DNA from the original Neanderthal skeleton itself (Krings et 
al. 1997). As we shall see, the archaeological evidence from Europe could be used to 
argue equally forcibly for an episode of population replacement at this point in the ar- 
chaeological succession, as opposed to a process of gradual, in situ, local evolution 
from the final Mousterian to the earliest Upper Paleolithic populations (Mellars 1992, 

The archaeological records spanning the period of the replacement of Neanderthal 
by anatomically modem populations reflect what I and many other archaeologists 
{e.g., Pfeiffer 1982; White 1993; Noble & Davidson 1996; Mithen 1996a) would re- 
gard as the most radical episode of cultural, technological and general behavioral 


change in the entire history of the European continent — at least since the initial colo- 
nization of Europe by early Homo populations around a million years ago (Table 1). 
The term 'Upper Paleolithic Revolution' is now commonly used to describe this tran- 
sition. As I have discussed in more detail elsewhere, clear and well documented pat- 
terns of change at this point can be documented in at least seven or eight separate 
features of the archaeological material (Mellars 1973, 1989a, 1989b, 1991. 1996a; 
Kozlowski 1990): 

1) The appearance of much more widespread 'blade' and 'bladelet' as opposed to 
'flake'-based technologies, probably involving the use of new 'punch' techniques of 
blade production; 

2) The appearance of a wide range of entirely new forms of stone tools, some of 
which clearly reflect new patterns of technology in other related spheres (such as skin 
or bone working, or the construction of new forms of hunting weapons) but others 
which seem to reflect an entirely new component of conceptual or visual form and 
standardization in the production of stone tools (Mellars 1989b; see below); 

3) The effective explosion of bone, antler and ivory technology, involving not 
only the appearance of many new techniques for shaping these materials (sawing, 
grooving, grinding, polishing, perforating, etc.) but a remarkably wide range of new 
and tightly standardized tool forms (Figure 1) — again apparently reflecting shifts in 
the complexity of several other related areas of technology (Knecht 1993); 

4) The appearance of the first reliably documented beads, pendants and other 
items of personal decoration (Figure 2) — frequently in large numbers, and manufac- 
tured in a range of often difficult materials (e.g., ivory and steatite) by a variety of 
manufacturing techniques (White 1989, 1993); 

5) The transportation of sea shells and other materials over remarkable distances 
of up to 400-600 km — as for example between the Atlantic coasts and Rhone Valley 
in France, or between the Adriatic coast and the Danube Valley in Austria (Taborin 
1993; Gamble 1986); 

6) The appearance of the first unmistakable sound-producing instruments, in the 
form of the bird-bone flutes recorded from early Upper Paleolithic levels at sites such 
as das Geissenklosterle in Central Europe and Isturitz in the Pyrenees; 

7) Most dramatically of all, the sudden appearance of explicitly 'artistic' activity, 
in a remarkable variety of forms. From western France we now have simple outlines 
of animals engraved on stone blocks, together with engravings of apparently female 
sex organs and at least one example of a carved bone phallus (Delluc & Delluc 1978). 
From Vogelherd, das Geissenklosterle, Hohienstein-Stadel and Stratzing in Central 
Europe we have a range of carved ivory or stone statuettes of various animal or human 
figures (Figure 3), together with one extraordinaiy composite carving of a male hu- 
man figure equipped with a lion's head (Figure 4) (Hahn 1972, 1993; Bahn 1994). 
And from eastern France dates of 30-32,000 BP have recently been announced for the 
series of remarkable drawings of horses, aurochs, rhinos, mammoths and other ani- 
mals in the cave of Grotte Chauvet in the Ardeche (Chauvet etal. 1995; Clottes 1 996). 
Alongside these more explicitly 'artistic' creations we have a variety of more geo- 
metric representations, ranging from regularly spaced lines and crosses engraved on 
the surfaces of bone tools to more complex arrangements of dots or liner incisions 
which Alexander Marshack ( 1 99 1 ) has been tempted to interpret as early forms of nu- 
merical notation or even lunar calendars. 

The above are the more easily documented archaeological aspects of the so-called 
'Upper Paleolithic revolution' in Europe, all clearly documented in the archaeologi- 
cal records in several areas of Europe well before 30,000 BP, and in most cases as 



early as 35-40,000 BP (Table 1 ). The first appearance of all these innovations seems 
to be closely associated with the earliest appearance of anatomically modern popula- 
tions, in a way which suggests that the behavioral innovations appeared with these 
new populations (Gambier 1989, 1993). In addition to the various features listed 
above there are at least strong hints of closely associated changes in several other as- 

FIGURE 1 . Eai-ly Upper Paleolithic bone, antler and ivory artifacts from sites in Central Europe, ca. 30- 
40,000 BP. 



FIGURE 2. Carved ivory beads from Aurignacian levels at the Spy Cave, southern Belgium, ca . 30-35,000 

pects of human behavior at broadly the same point in the archaeological sequence — 
including a sharp increase in the numbers of occupied sites in some of the best docu- 
mented regions (suggesting a significant increase in human population densities), the 
appearance of larger and more highly structured occupation sites (in some cases with 
clear living structures); more large-scale and systematic distribution of raw materials; 
and in at least certain areas a sharp increase in the degree of specialization in animal 
exploitation patterns, which at least hints at the emergence of more sharply focused, 
and probably more 'strategically' organized hunting activities (Mellars 1973. 1989a, 
1996a; Stringer & Gamble 1993; Knecht era/. 1993). Since the interpretation of some 
of these features remains more controversial, it may be best for the present discussion 



to focus on the most easily demonstrated 'facts' of the archaeological record listed in 
the preceding paragraphs. 

The overall impression which emerges therefore is of a fundamental restructuring 
in almost all of the archaeologically visible aspects of human behavior at this point in 
the archaeological records of Europe, which appears to correlate closely if not pre- 
cisely with the anatomical transition from archaic {i.e.. Neanderthal) to fully modem 
populations. If the preceding speculations are correct, this revolution would have ex- 
tended through almost all spheres of behavior, ranging from several different dimen- 
sions of technology, through 'aesthetics,' subsistence patterns, social organization 
and demography. As several authors have pointed out {e.g.. Pfeiffer 1982; White 
1989, 1993;Mellars 1989a, 1991;Knight 1991;DonaIdl991;Knighte^ a/. 1995:No- 
ble & Davidson 1996) the most dramatic single demonstration of this behavioral 
revolution is in the explicitly 'symbolic' spheres ofart, personal ornamentation, cere- 
monial burial practices, and apparently music. It is this pattern that has led many of us 
to characterize the so-called Upper Paleolithic revolution as a preeminently symbolic 
revolution — or in John Pfeiffer's (1982) words 'symbolic explosion.' It is the rami- 
fications of this perspective that I will explore below. 

The critical question of course is what relevance, if any, does this 'behavioral 
revolution' or 'symbolic explosion' have for studies of language origins and develop- 
ment in early populations? It is here that one must make a flindamental, but in my 

Figures. Series of animal figures, cai^ved from mammoth ivory, from the early Aurignacian levels of the 
Vogeiherd Cave, southern Germany. 



view and that of most other archaeologists and anthropologists, entirely valid as- 
sumption — namely, that the extraordinarily complex, highly structured, visually 
symbol-laden, artistically creative and remarkably 'modern' patterns of culture so ex- 
plicitly reflected in the archaeological records of Upper Paleolithic populations in 

FIGURE 4. Cai-ved mammoth-ivoi^ human figure with a Hon's head, from the early Aurignacian levels o f 
the Hohlenstein-Stadel Cave, southern Gemiany. 


Europe would be inconceivable in the absence of essentially modem, highly struc- 
tured language. This argument could no doubt be presented in a number of different 
ways and from a variety of perspectives. It could be argued, for example, that the pat- 
terns of technology, subsistence strategies, symbolism, etc. documented in many 
early Upper Paleolithic contexts are at least as complex — and in some cases appar- 
ently more complex — than those documented among many of the technologically 
'simpler' hunter-gatherer communities of the recent ethnographic past (such as the 
Tasmanians, or some Bushmen groups), all of which are known to have possessed 
highly complex, entirely modem language pattems. Other more involved arguments 
have been constructed relating complex syntax to the creation of complex artistic and 
representational images (Davidson & Noble 1989, 1993; Noble & Davidson 1996). 
Leaving aside these more specific arguments, I doubt whether any modem archae- 
ologist, anthropologist or cognitive scientist would seriously dispute that the full 
panoply of typically Upper Paleolithic culture must reflect the presence of essentially 
modem language pattems among the earliest anatomically modem populations of 
Europe by at least, say, 35-40,000 years ago (Donald 1991; Bickerton 1990, 1995; 
Pinker 1994; Gibson 1996; Mithen 1996a, 1996b; Dunbar 1996, etc.). 

The critical issue in this context therefore hinges on the existence and/or nature of 
language among the preexisting Neanderthal populations of Europe — and it is on 
this issue that most of the debate over the past 10-15 years has centered. Some 
authors {e.g., Binford 1989; Lieberman 1991) have argued that language in any 
strictly defined sense was probably lacking among the Neanderthal and earlier popu- 
lations of Europe. Others, including Bickerton (1990, 1995) and myself (1989a, 
1996a, 1996b), have taken a less extreme view, while nevertheless accepting that 
some major shift in the basic complexity or stmcture of language almost certainly was 
associated with the Neanderthal to modern human transition. Others of course have 
taken the view that there is no reason to infer any major contrast between the nature of 
Neanderthal and modem language (Ragir 1985; Gibson 1996; Mithen 1996b). What 
follows is an attempt to explore what in my view are the most potentially relevant as- 
pects of the available archaeological evidence to come to grips with this question. 

At the outset it seems to me highly unlikely that the Neanderthals possessed no lin- 
guistic abilities. As the paper by Peter Marler in the present volume makes clear, 
many species of primates, and indeed other groups of mammals and birds, have been 
shown to have relatively complex repertoires of vocal expressions, which seem to 
convey clear meaning to other members of the groups, even if any form of multiple, 
sequential symboling or 'syntax' is lacking (see also Parker & Gibson 1990; Pinker 
1994; Cheney & Seyfarth 1990, 1996). The Neanderthals of course were separated 
from our closest primate ancestors by at least five million years of biological and be- 
havioral evolution, had brains around three times the size of any primates, and de- 
monstrably vastly more complicated repertoires of behavior in many different 
spheres (Stringer & McKie 1996; Mellars 1996a). As Mithen (1996a, 1996b) and 
others have argued, it seems inconceivable that Neanderthals — and indeed preced- 
ing populations of Homo erectus and even Homo habilis — would not have had sys- 
tems of vocal communication far more complex than those of any of the living 

At the same time however, it would be totally irrational to argue that Neander- 
thals, simply because they had brains the same size as ours and survived as recently as 
30-35,000 years ago, must have had linguistic and other cognitive capacities identical 
to those of modem populations. In brains, as in other spheres, size is not everything, 
and many earlier studies have demonstrated that brain size alone — or even allomet- 


ric relationships between brain size and total body size — are notoriously poor indica- 
tors of relative cognitive capacities (though the so-called 'neocortex ratio' which 
relates brain volume to the area of the surrounding neocortex may given a better indi- 
cation: Aiello & Dunbar 1993; Aiello 1996; Mellars 1996: 368). Clearly, there are 
many potential functions for large brains quite apart from language (Passingham 
1989; Byrne 1996). The crucial point to recognize is that if most of the recent recon- 
structions of the patterns of human evolution and phylogeny do have any validity, 
then the line of evolution which led to the European Neanderthals is likely to have 
been separated from that which led to the evolution of anatomically modem humans 
over a span of at least 300,000 and probably closer to 500,000 years (Krings et al. 
1997; also Stringer &McKie 1996; Howell 1996; Brauererr//. 1997) — implying a 
cumulative evolutionary divergence of these two lineages over a total of at least half a 
million years. To argue that there can have been no significant change in the neuro- 
logical complexity of brain structures over this period would plainly be absurd. To ar- 
gue that the evolution of the brain had effectively come to an end before the 
evolutionary divergence of the European Neanderthals from the African ancestors of 
anatomically modem populations would be equally absurd, since this would imply 
that biological evolutionary processes effectively came to an end at precisely the 
point when most fomis of human interpersonal and intergroup behavioral and social 
competition are likely to have become most intense (Humphrey 1976; Byme & 
Whiten 1988; Dunbar 1996), Clearly, there is no evolutionary or biological justifica- 
tion for assuming that Neanderthals must have had brains identical to ours. 

If we turn now to the strictly behavioral and archaeological aspects of the ques- 
tion, I would suggest that there are three features of the available archaeological evi- 
dence, in particular, which could be of critical relevance to the nature or stmcture of 
Neanderthal language (Table 2; Mellars 1991, 1996a, 1996b): 

Table 2. Likely behavioral correlates of a iinguistic revolution.' 

1 . Symbolic explosion 

2. Increased organizational complexity 

3. Increased social complexity — especially kinship/descent structures 

4. More long-term 'strategic' planning 

5. Shaiper and more prescribed cultural/ethnic divisions 

6. Emergence of complex ideologies, mythologies etc. 

7. Increased 'categorical' concepts — e.g., in tool production 

The first and most obvious is the virtual lack of convincing evidence for explicitly 
symbolic thinking or behavior among the Neanderthals. The whole topic of symbol- 
ism is of course plagued by semantic and terminological debates and has generated a 
lively literature (c/:,Hodder 1982; Chase 1991;Byers 1994, e/c; see Chase 1991 and 
Mellars 1996a:369 for a discussion of different meanings of symbolism). Without 
wishing to get too enmeshed in these debates, the most striking feature of the archaeo- 
logical records of the European Neanderthals is the effective lack of convincing and 
clearly documented fomis oi explicit symbolism, in the form of clearly representa- 
tional or even 'geometric' art, obvious items of personal adornment or decoration, or 
demonstrably 'ritualistic' or 'ceremonial' burial practices (see Chase & Dibble 1987; 
Gargett 1989; Chase 1991; Mellars 1996a:369-83, 1996b for full documentation of 
these points). It is tme that Neanderthals occasionally used coloring pigments for 
some as yet unknown purpose, sporadically collected unusual or intriguing fossil 
shells (though with no evidence for their use for personal ornaments) and certainly 


buried human corpses in simple, unadorned graves in their cave or rock-shelter living 
sites. But none of this in my view necessarily implies anything clearly symbolic in the 
activities represented (Mellars 1996a, 1996b; Chase & Dibble 1987; Chase 1991), 
and certainly pales into virtual insignificance when set beside the dramatic explosion 
of decorative, ornamental, artistic and other explicitly symbolic activities and arti- 
facts recorded from scores of early Upper Paleolithic sites across the length and 
breadth of Europe (White 1989. 1993). 

The linkages between visual symbolism and language can no doubt be argued in 
many ways (see for example Davidson and Noble 1989, 1993; Donald 1991; Knight 
1991;Mithen 1996a; Noble & Davidson \996, etc.). No one would question however 
that language is, par excellence, a highly symbolic activity and, as Merlin Donald 
(1991) and others have argued, one might reasonably predict that any sudden intro- 
duction of highly elaborate and symbolic forms of linguistic communication would 
be accompanied by a coiresponding increase in the range and complexity of visual 
and ceremonial symbolism. In other words it would not be surprising if the 'symbolic 
explosion' in archaeological terms coincided with a 'linguistic explosion' in cogni- 
tive and communication terms. 

The second, more direct correlation I have proposed between the archaeological 
record and language lies in the essentially new element of clearly 'imposed form' and 
'visual standardization' in artifact production which I believe represents one of the 
most striking general features of the archaeological transition from the Middle to the 
Upper Paleolithic (Mellars 1989b, 1991, 1996b). The essence of this idea is that 
while Middle Paleolithic craftsmen were unquestionably expert flint knappers and 
clearly invested a good deal of time and effort into producing /imc//o/7fl//v efficient 
and varied tool forms, they seem with a few rare and debatable exceptions (notably bi- 
faces: see Mellars 1996a: 135-6; Gowlett 1996) to have shown surprisingly little con- 
cern with the overall visual appearance of the tools. As a result. Middle Paleolithic 
stone tools seem to grade almost imperceptibly — and to an archaeological taxono- 
mist extremely irritatingly — between 'convergent sidescapers' and 'points.' be- 
tween 'bifacial racloirs' and 'handaxes.' between 'notches' and 'denticulates,' and so 
on (see also Chase & Dibble 1987; Dibble 1989). Anyone who has tried to teach 
Mousterian typology to students will be acutely aware of the problem of attempting to 
neatly categorize or classify Middle Paleolithic stone tools into clearly separated 
'types' or discrete morphological forms. 

By contrast. Upper Paleolithic stone tool types not only show a vastly greater 
range of morphological diversity than Middle Paleolithic tools (diversity which, as 
discussed below, changes rapidly over both space and time) but also a much more ob- 
vious degree of repetition and standardization in the visual appearance of the tools. 
In most cases this was achieved by flaking away a good deal of the original blade or 
flake blank until some relatively standardized and normally highly distinctive visual 
form was imposed on the tool. As I have commented elsewhere (Mellars 1991 :66), it 
is as though Upper Paleolithic flint workers were saying "this is an end-scraper: I use 
it as an end-scraper, I call it an end-scraper and it must therefore look like an end- 
scraper," whereas Middle Paleolithic groups seem to have been content with the 
maxim "who cares what this tool looks like as long as it fiinctions adequately for the 
job in hand." This same element of visually imposed form and morphological stan- 
dardization is arguably even more conspicuous in many of the major types of Upper 
Paleolithic bone, antler and ivory tools, as well as in various forms of personal orna- 
ments, artistic motifs and even in the clearly 'structured' form of many Upper Paleo- 
lithic occupation sites (Mellars 1991, 1996a:381-3) 


To cut a potentially rather complicated argument short (Meliars 1991; 1996a: 
133-6, 381-3), my contention is that this kind of greatly increased emphasis on the 
visual morphology and appearance of tools, and the sharply defined separation of the 
different tool forms into relatively standardized and clearly separated visual catego- 
ries, is probably exactly what one would anticipate if Upper Paleolithic groups had a 
much more complex and highly structured vocabulaiy for the different artefacts 
forms. The contention here is simply that words and names are essentially devices for 
breaking up what are often continuously varying shapes or concepts into a range of 
discrete, categorical entities — much as we divide the continuously varying colors of 
the rainbow into "red," "orange," "yellow," "green," etc. In other words one can ar- 
gue that where a name exists, there must presumably be some distinctive visual or 
mental image(or 'mental template') to go with it. If I use the word 'spoon' for exam- 
ple, this immediately evokes not only the notion of something which can be used to 
eat soup, but a distinctive visual image, with a rounded bowl, elongated, splayed han- 
dle, curved profile, etc. In the present context, all of this would make sense if, for ex- 
ample. Neanderthal groups had just one or two names for stone tools, whereas Upper 
Paleolithic groups had perhaps ten or twenty. At the very least it could be used to ar- 
gue for a greatly increased linguistic vocabulary over the period of the Middle-Upper 
Paleolithic transition. When we add to this the even greater degree of complexity of 
bone, antler and ivory tools in the Upper Paleolithic, various forms of personal orna- 
ments and decorative items, artistic representations, etc., the scale of contrast purely 
at the level of the complexity of vocabularies in the two periods could be fairly dra- 

One final and much more general aspect of the archaeological records of the Mid- 
dle and Upper Paleolithic which has been stressed by Binford, Whallon and others as 
potentially highly relevant to the nature and complexity of language is the overall de- 
gree of complexity reflected in different spheres of the behavior, organization and 
strategic 'planning' of Neanderthal and anatomically modem groups. As Whallon 
(1989) in particular has argued, the emergence of highly structured language, with 
complex syntax, past and fiiture tenses, subordinate clauses, subjunctives, e/c. would 
almost inevitably have had a profound on the capacity for human groups to organize 
and structure all aspects of their activities, ranging from the planning and coordina- 
tion of hunting expeditions, through the construction of elaborate myths, belief sys- 
tems and ideologies, to the patterns of social roles and relationships between 
individuals in the groups — including for example the recognition of complex sys- 
tems of kinship and descent (see also Donald 1 99 1 ; Knight 1 99 1 ; Rodseth et al. 1 99 1 ; 
Bickerton 1990; Dunbar 1996; Meliars 1996c; etc.). Binford (1989) has emphasized 
the role of language m what he refers to as 'long-range planning' — the capacity to 
conceptualize and organize activities (including more long-term plans for tool pro- 
duction and 'curation,' or the long term use of occupation sites) over periods of more 
than a few hours or a few days. In this context the evidence for apparently more 
highly specialized and strategically organized hunting patterns in the Upper Paleo- 
lithic, more highly organized systems for the long-range procurement and distribu- 
tion of raw materials, and larger and more structured occupation sites could all be 
highly significant (Meliars 1996a). Indeed, Whallon (1989) and Soffer (1994) have 
argued that the appearance of fully complex language could have been the critical fac- 
tor which allowed the first permanent occupation of some of the more extreme peri- 
glacial environments in central and eastern Europe which seem to have been largely if 
not entirely avoided by Neanderthal groups. 


There is one further aspect of the archaeological transition from the Middle to the 
Upper Paleolithic which is potentially highly relevant to the emergence of language 
but which to my knowledge has never been specifically discussed in these terms. This 
is the very much sharper degree of both chronological and spatial variations in tech- 
nology documented throughout the European Upper Paleolithic in comparison to that 
in the Middle Paleolithic. The archaeological reality of this contrast has of course 
been commented on frequently in the earlier literature {e.g., Isaac 1972; Gamble 
1986; Sackett 1988; Klein 1989) and is to some extent a commonplace of the Paleo- 
lithic record. This is emphatically not to suggest that no significant variations can be 
detected in Middle Paleolithic/Neanderthal technologies in Europe — both chrono- 
logical and spatial — as I myselfhave argued at length (eg., 1969, 1992, 1996a). The 
point is simply that variations of this kind are not only reflected in far more striking 
and idiosyncratic terms in the Upper Paleolithic — that is by a welter of visually strik- 
ing and distinctive 'type-fossil' forms in both stone tools and bone/antler implements 
(Mellars 1989b) — but also that the individual chronological and spatial units of 
variation change over a much more rapid time-scale in the Upper than in the Middle 
Paleolithic. In the classic Upper Paleolithic sequence of western Europe for example 
we can now recognize at least 1 5-20 discrete technological stages over a total period 
of ca. 30,000 years, most of them defined by effectively unique type-fossil forms 
which change at intei-vals of every few thousand years (de Sonneville-Bordes 1961; 
Bordes 1984; Sackett 1988). In at least many stages of the Upper Paleolithic se- 
quence similar complexity can be documented in the geographical distribution of dif- 
ferent industries. Thus the various stages of the Solutrian and most of the classic 
variants of the Magdalenian are restricted to western Europe (Smith 1966; Bosinski 
1990); the Noaillian is confined essentially to southern France and northern Italy 
(David 1985); a variety of distinctive regional variants can be recognized in different 
areas of the west European Solutrian (Smith 1964, 1966); and a similar range of re- 
gional variants in the central and eastern European Gravettian(Otte 1981; Kozlowski 
& Kozlowski 1979). Whatever degree of chronological and regional variation can be 
recognized within the technology of Neanderthal communities in Europe is exceeded 
by a least an order of magnitude in the technologies of Upper Paleolithic groups. 

The generally held view among Paleolithic specialists is that this greatly increased 
degree of technological and typological variation in the European Upper Paleolithic 
is in some way reflecting both a much more sharply defined pattern of ethnic divisions 
among Upper Paleolithic groups — apparent in a strong component of 'style' in the 
associated artefacts — and the emergence of strong cultural, social and ethnic tradi- 
tions, which led to the clearly defined inheritance of these technological and stylistic 
features from one generation to the next (Isaac 1972; Jochim 1983; Gamble 1986; 
Sackett 1988; Mellars 1989b; Shennan 1996). In the context ofthe present discussion 
I would suggest that both of these features are precisely what one would predict if 
complex, highly structured language emerged for the first time in the Upper Paleo- 
lithic. There are three main strands to the argument. The first is that [as Donald 
( 1991 ) and others have argued] language is probably essential to the effective trans- 
mission of tightly defined rules of complex social and cultural behavior from one gen- 
eration to the next, and is therefore at least a strong catalyst to the emergence of 
sharply defined cultural traditions, if not an essential prerequisite. The second is that 
language is arguably the most important single factor in unifying the culture and be- 
havior of specific social groups, and therefore maintaining this kind of ethnicity in the 
material record (Wiessner 1983, 1984). Thethird, of course, is that linguistic differ- 
ences, once established, become by far the most powerftil and effective means of rein- 


forcing and maintaining social divisions between distinct ethnic groups — as for 
example all the literature on documented tribal divisions in recent hunter-gatherer 
communities clearly reveals {e.g., Peterson 1976). In short, if there was indeed a 'rev- 
olution' in the complexity and structure of language over the period of the Neander- 
thal to modem human transition in Europe, one should perhaps expect to see as an 
automatic concomitant of this a much shaiper degree of divergence and rapid change 
in the associated material cultures. Moreover it would not be unreasonable to suggest 
that at least some of the geographical and chronological divergences in material cul- 
ture documented throughout the Upper Paleolithic sequence may be reflecting fairly 
directly corresponding divergences in language patterns (see Renfrew, this volume). 

The inference I am drawing from all the preceding lines of evidence is that 
whereas Neanderthal communities in Europe almost certainly had some form of lan- 
guage, and while this is likely to have been far more complex and highly structured 
than that of any of the present-day primates. Neanderthal language patterns most 
probably were still radically simpler in certain fundamental respects than those of the 
succeeding Upper Paleolithic populations. I have suggested that two of the critical 
differences may have been a much more limited vocabulary or lexicon, and probably 
a much more rudimentary pattern of grammar and syntax, which severely restricted 
the range and complexity of verbal communications. Lieberman, of course, has 
drawn essentially the same conclusion based on certain anatomical details of the Ne- 
anderthal skull and associated larynx (Lieberman 1989, 1990), though these interpre- 
tations have been disputed by others (e.g., Ahrensburg 1989). The implication in 
other words is that the Neanderthal — sand probably earlier populations of Homo 
erectiis and possibly Homo habilis — possessed some form of 'protolanguage.' One 
of the main challenges at present is to construct explicit models of exactly what form 
these evolutionary 'protolanguages' might have taken, since I believe it is only by 
constmcting explicit models for alternative patterns of language development that we 
will be able to test these potential models effectively against the archaeological evi- 
dence. At present I am aware of only one really explicit attempt of this kind. This is 
embodied in Derek Bickerton's (1990, 1995) suggestion that the development of lan- 
guage in the course of human evolution is likely to have paralleled that of the ontoge- 
netic development of language in young children — i.e., that at least in language 
development 'phylogeny may recapitulate ontogeny.' On this basis he has outlined a 
pattern of 'protolanguage' which he suggests may have resembled that of 1 V2-2 year 
old children in containing only very simple, 'here-and-now' sentences, lacking any 
complex syntax, and virtually lacking the ability to refer to objects or events 
'displaced' in time or space from the present. He argues from both the patterns of lan- 
guage development in young children and from similar transformations between 
pidgin and Creole languages in recent language-contact situations that the transition 
from 'proto' to 'full' language is likely by its nature to have been relatively abrupt, 
with few if any potentially intermediate stages (Bickerton 1 990: 1 64-74). Of course, 
all of this is controversial. If true it would argue for a remarkably abrupt, essentially 
quantum-leap event in the evolutionary development of language, presumably in- 
volving some fairly dramatic genetic mutation in the course of human evolution. 
Nevertheless the hypothetical pattern which Bickerton has suggested for the nature of 
his 'protolanguage' could well correspond fairly closely with some of the specula- 
tions outlined above for the language of the European Neanderthals. 

As a final point we should perhaps turn to Mithen's recent suggestion that the es- 
sence of the contrast between Neanderthal as opposed to fully modem cognition lay 
in the transition from what he refers to as 'modular' or 'domain-specific' to 'genera- 


lized' intelligence (Mithen 1996a, 1996b). The central argument, drawing on the 
work of Tooby & Cosmides (1992) and others, is that separate, highly specialized 
components of cognition — including language — emerged fairly gradually in the 
course of human evolution, in response to specific selective pressures acting on 
equally specific areas of behavior — tool manufacture, food acquisition, social rela- 
tionships, communication, etc. Only later, at the time of the classic Middle-to-Upper 
Paleolithic transition, did these different domains of intelligence finally coalesce to 
generate fiilly modem, much more fluid and flexible forms of intelligence. These are 
all plausible suggestions, which would fit well with many aspects of the archaeologi- 
cal records of the Neanderthals as well as with much of the recent thinking by evolu- 
tionary psychologists on the evolutionary development of intelligence. With regard 
to Mithen's speculations on the origins of language, however, three specific questions 
spring to mind: 

1) How plausible is it that the kind of relatively sudden integration of different, 
previously separate spheres of intelligence into a single 'generalized' intelligence 
could occur without some equally direct and fairly dramatic impact on the structure 
and complexity of language? 

2) If this kind of direct impact on language should be seen as an almost inevitable 
concomitant of the processes he is proposing, in what sense is he challenging the no- 
tion that the 'Upper Paleolithic revolution' was also in some sense a linguistic revolu- 

3) Thirdly, could we not in fact reverse the direction of the cause and effect rela- 
tionships that Mithen has in mind and suggest that it could well have been the evolu- 
tionary emergence of fully modem linguistic stmctures which allowedi\\Q integration 
of the previously separated modular cognitive domains in the human mind? 

We could pose a broadly similar set of questions in relation to Bickerton's recent 
arguments (outlined above) that the evolutionary emergence of fially modem lan- 
guage must, by its nature, have been a relatively sudden, 'catastrophic' event, rather 
than a slow and gradual shift in increasing linguistic complexity. For all the reasons 
which have been spelled out by Bickerton himself -(1990, 1995), Whallon (1989), 
Donald ( 199 1 ), Knight ( 199 1 ), Pinker ( 1 994) and others, any sudden transformation 
in the complexity and structure of language would almost inevitably have fairly radi- 
cal consequences for virtually all aspects of human organization and behavior — 
ranging from technology, through subsistence strategies, social structures, communi- 
cation patterns, ideological and mythological beliefs etc. The critical question in this 
case is exactly where, in the archaeological records of the European Paleolithic, 
might one identify this kind of behavioral watershed, \^ not over the period of the 
Middle-Upper Paleolithic transition (Mellars 1996a:391)? 


In the preceding sections I have argued that in the archaeological records of 
Europe there was probably a radical shift in the overall complexity and structure of 
language associated closely with the archaeological transition from the Middle to the 
Upper Paleolithic, and with the associated transition from Neanderthal to fully ana- 
tomically modem humans. I have tried to identify what may be the most relevant as- 
pects of human behavior for inferring the emergence of more complex and highly 
structured language, and discussed how these might be reflected in the surviving ar- 
chaeological records (Table 2). Following the policy of working from the known to 


the unknown, this has hopefully put at least some of the basic elements of the linguis- 
tic jigsaw into place. 

Exactly what relevance all of this has for the more distant evolutionary origins of 
language is of course an entirely separate question. As discussed in the introduction, 
it is now clear that the transition from Neanderthal to fully modem populations in 
Europe almost certainly represents a major population replacement event, with the 
biologically and behaviorally modem populations dispersing fairly rapidly across 
Europe from some extemal source. Although debates continue, several different 
strands of the evidence point increasingly to Africa as the most likely point of origin 
of the genetically modem populations (Harpending et al. 1993; Cann et al. 1994; 
TishkoffeM/. 1996; Stringer & McKie 1996; Howell 1996). The evidence suggests 
that some of these populations may initially have spread from Africa as early as 
100,000 years ago — although the possibility of more than one dispersal event has 
sometimes been discussed (Lahr & Foley 1994; Foley & Lahr 1997). But in any 
event, the populations which spread across Europe around 40,000 years ago are likely 
to have had their biological and behavioral origins well beyond the bounds of Europe, 
and well before the date of 40,000 BP. 

The obvious question therefore arises as to how far we can identify evidence for 
the initial emergence of some of the more culturally and cognitively advanced pat- 
tems of behavior — including language — in areas beyond Europe, and especially 
within the hypothetical homeland of the anatomically modem populations in, or close 
to, Africa. Here we immediately encounter the problem that the intensity of archaeo- 
logical research on African sites within the relevant time range {i.e., between say 
50,000 and 100,000 BP) has been much less than in Europe, so that the degree of de- 
tail and resolution of the archaeological records is much less. Nevertheless there are 
at least three features of the available archaeological evidence from Africa and 
closely adjacent areas which could be held to provide at least some strong hints that 
the emergence of more complex, broadly 'Upper Paleolithic'-like behavioral pattems 
— and most probably language — may be significantly earlier in these areas than in 

1) The first is that remarkable archaeological phenomenon known as the 'Howi- 
eson's Poort' industries, now documented from a range of sites in southern Africa to 
the south of the Zambezi (Deacon 1989; Clark 1992). Precise dating of these indus- 
tries is still slightly controversial, but most estimates place them at around 50-75,000 
BP, certainly long before the emergence of typically Upper Paleolithic technologies 
in Europe (Gmn et al. 1990). The remarkable feature of the Howieson's Poort indus- 
tries is their strikingly Upper Paleolithic-like appearance (Figure 5), including not 
only typical blade technology but also classic forms of end scrapers and burins, and a 
range of careftilly shaped, semi-microlithic trapeze and crescent shaped forms which 
not only reflect a high degree of standardization and visually "imposed form," but 
which almost certainly indicate the presence of complex, multicomponent hunting 
projectiles at this early date (Deacon 1989). Apparently associated with this industiy 
at the site of Klasies River Mouth was at least one specimen of a careftilly shaped 
bone point and two specimens of regularly notched bones, which again would be 
much more at home in an Upper than a Middle Paleolithic context in Europe (Singer 
& Wymer 1982). Overall the African Howiesons Poort industries could be seen as 
evidence of the emergence of at least the major technological components of typically 
Upper Paleolithic culture at least 10-20,000 years earlier in southem Africa than in 
either Europe or Asia (Mellars 1989a:367-9). 




FIGURE 5. Stone artifacts from the Howiesons Poort levels (507-75,000 BP) at Kiasie's River Mouth, 
South Africa, showing a range of microlithic crescents, triangles, end scrapers and other characteristically 
Upper Paleolithic forms. 

2) The second feature is the striking abundance of coloring pigments recorded at 
many African sites within the general time range of 40-100,000 BP. As Knight, 
Power, and Watts (1995) have recently documented, the evidence comes in the form 
not only of large quantities of red and yellow ocher recorded from many Middle Stone 
Age sites, but also the presence of clear 'pencils' or 'crayons' of ocher, and a number 
of stone slabs heavily smeared with ocher which appear to represent coloring palettes. 
While fragments of ocher are by no means unknown in Neanderthal contexts in 
Europe (Mellars 1996a), the evidence does seem to suggest a more intensive use of 



coloring materials in Africa than in contemporaneous European contexts. It is tempt- 
ing of course to suggest that this might reflect a much more enterprising use of pig- 
ments in these early African sites than among the European Neanderthals, 
conceivably involving the earliest stages of symbolic or 'artistic' representation. Un- 
fortunately, any direct evidence of how the coloring materials were employed is at 
present lacking. 

3) Thirdly, moving just beyond the bounds of Africa, there are the remarkable 
burials associated with the essentially anatomically modem skeletons from the two 
sites of Skhul and Qafzeh in Israel, both now securely dated to around 90-100,000 BP 
(Bar-Yosef 1992, 1994, 1996). The unique feature of these burials, in contrast to 
those of the European Neanderthals, is the presence in each case of unmistakable 
grave offerings (Figure 6), in the form of a complete boar's jaw reputedly 'clasped in 
the arms' of one of the skeletons at Skhul, and a large fallow deer antler placed imme- 
diately on top of one of the burials at Qafzeh (Defleur 1993; Vandermeersch 1970, 

Figure 6. Burial of an anatomically modem skeleton accompanied by a large fallow deer antler, from the 
site of Qafzeh, Israel, dated to ra . 1 00,000 BP. 


1976). Once again, the evidence could be seen as a reflection of a significantly more 
symbolic or 'ceremonial' approach to burial ritual in these Israeli sites at this early 
date than in anything at present recorded at the same time in Europe. These discover- 
ies of course are located only immediately beyond the limits of northeast Africa and 
are generally seen as evidence for a brief, temporary incursion of anatomically mod- 
ern human populations into southwest Asia around the beginning of the last glacial 
period (Bar- Yosef 1992). 

The evidence cited above remains limited, but is nevertheless consistent with the 
hypothesis of a significantly earlier emergence of various forms of explicitly sym- 
bolic behavior in areas either in or closely adjacent to Africa than in other parts of the 
world. In terms of possible criteria for the emergence of increasingly complex and 
highly structured language discussed earlier, this could be fully consistent with cur- 
rent models for an essentially African origin for the early evolution of genetically 
modem populations, implied by recent studies of both the anatomical and genetic evi- 
dence (Harpending £>/<://. 1993: Stringer &McKie 1996; Howell 1996). In language, 
as in other aspects of our genetic and biological make-up, the evolutionary source of 
'modem' human populations could very well be African. 

Cognitive Capacity and Cultural Performance: 
The 'Sapient Paradox' 

There is one final point which should be raised here, since this seems to have 
emerged as a source of considerable conftision in some of the recent literature, and in- 
deed surfaced in the discussions at the present conference. This is the need to make a 
clear and fundamental distinction between the notions o^ cognitive potential and be- 
havioral performance in discussions of the emergence of fully modern behavior and 
associated language in human evolution (c/^Mellars 199 1:70; Renfrew 1996). There 
is an obvious and unfortunate asymmetry in the nature of the evidence here. While it 
is obviously (and by definition) true that no individual or human group can perform 
above the level of their biological and cognitive capacities, it is equally self-evident 
that people can, and very frequently do, perform below their full capacities. In other 
words, while some pattems of behavior in the archaeological record can easily be held 
up as concrete proof that the necessary cognitive capacities for that fomi of behavior 
existed in the populations in question, the absence of these pattems of behavior need 
not imply that the essential capacities for the behavior were lacking. The absence of 
particular pattems of behavior or cultural expression in particular societies could be 
due to a whole range of factors: the lack of a specific need or stimulus for that behav- 
ior in particular economic, social or environmental contexts; the lack of adequate raw 
materials to support particular technological processes; or indeed the simple fact (or 
historical accident) that particular aspects of behavior or technology had not yet been 
developed or 'invented' by the societies in question. Clearly, the lack of advanced 
metal technology or knowledge of mathematics in recent hunter-gatherer communi- 
ties in no way reflects any lack of the basic cognitive capacities for these behaviors in 
the societies in question, still less any cognitive inferiority to modem industrialized 
societies. Similarly the failure to use computers in ancient Rome did not reflect any 
lack of mental capacities to develop or operate computers. In all these cases technol- 
ogy and other aspects of behavior for various reasons either had not 'caught up' with 
the full mental capacities of the societies in question, or was simply not necessary, ap- 


propriate, or (in some cases) possible in the specific social, economic and environ- 
mental contexts of the societies involved. 

These points are all very obvious but may help to resolve some of the recurrent 
confusions in recent discussions of the emergence of fully modem behavior. A full 
discussion of all the relevant issues would no doubt require a separate paper. But rec- 
ognition of these points could well explain, for example, while the early anatomically 
modem populations represented at Skhul and Qafzeh in Israel (at around 1 00,000 BP) 
were still employing technologies that we describe in archaeological terms as essen- 
tially 'Middle' as opposed to 'Upper' Paleolithic (Bar-Yosef 1992. 1994). Lithic 
technology of course represents only a minute fraction of the total behavioral and 
adaptive repertoire of any human group, and it could well be that in many other 
spheres the behavior and organization of the Skhul/Qafzeh populations was essen- 
tially 'modem' (c/ Lieberman & Shea 1994). In this context, as noted above, the 
presence of what appear to be demonstrably ceremonial burial rites at both sites could 
well be highly significant (Figure 6). Similar factors almost certainly explain why 
some of the most impressive aspects of complex bone, antler and ivory technology, 
personal omamentation, and various forms of art are far more conspicuous in the ar- 
chaeological records of the European Upper Paleolithic than in those of the Middle 
East, Africa and parts of Asia. As several authors have pointed out {e.g.. Oswalt 
1976; Price & Brown 1985), a variety of different environmental and related eco- 
nomic and demographic factors tend to favor more complex patterns of social organi- 
zation and associated material culture among societies occupying highly stressed and 
economically unpredictable arctic or periglacial environments than in those occupy- 
ing less-demanding temperate or tropical environments. And the absence of highly 
developed blade technology in many contexts simply reflects the heavy requirements 
which blade technology imposes on both the quality and nodule-size of available raw 
material supplies. Even in some classic European Upper Paleolithic contexts blade 
technology may be virtually lacking in certain situations where good raw materials 
are not locally available — as for example at El Castillo and some other sites in north- 
em Spain or in many of the Upper Paleolithic industries of the west Italian coast 
(Strauss & Heller 1988; Cabrera Valdes 1984;Mussi 1992). How far similar consid- 
erations may explain the lack of classic blade-based technologies over large areas of 
eastern and southeast Asia and Australasia is an interesting point for speculation 
(Mellars 1989a:377; Schick & Toth 1993; Lourandos 1996) — especially if we re- 
member that any technology has to adapt to the overall levels of environmental con- 
straints in any region, and not just to small, isolated 'windows' of environmental 

The purpose of this discussion is to emphasize that the absence of highly devel- 
oped, classically Upper Paleolithic-like technologies in many contexts need not in 
any sense imply the lack of fully modem cognitive capacities, nor the absence of fiilly 
developed language. The asymmetry of the evidence for cognitive potential as op- 
posed to behavioral performance must be kept clearly in mind here. In my view the 
emergence of what appear to be typically Upper Paleolithic technologies at Klasie's 
River Mouth and other early sites in southern Africa may well provide strong evi- 
dence for the emergence of fully modem cognitive and language capacities in this 
area at a substantially earlier date than in Europe. The absence of the full panoply of 
Upper Paleolithic technologies in certain other contexts demonstrably associated 
with anatomically modem humans on the other hand need not imply a lack of these 
capacities, for all the reasons outlined above. The fact remains that the most dramatic 
event in the history of European populations is reflected in the classic 'Upper Paleo- 


lithic revolution,' which occurred closely if not precisely in association with the ap- 
pearance of the first anatomically modem populations. I would suggest that in this 
case the archaeological evidence almost certainly is reflecting some form of cognitive 
and linguistic revolution, and that this revolution was introduced into Europe by the 
arrival of new populations, most probably from an ultimately African source. 

Literature Cited 

Aiello, L.C. 1996. Hominine preadaptations for language and cognition. Pages 89-99 in P. 
Mellars & K. Gibson, eds.. Modeling the Early Human Mind. McDonald Institute for Ar- 
chaeological Research, Cambridge, U.K. 

Aiello, L.C, & R. Dunbar. 1993. Neocortex size, group size and the evolution of language. 
Clin: Anthropol. 34:184-193. 

Aitken, M., C. Stringer, & P. Mellars, eds. 1 992. The Origin of Modern Humans and the Impact 
ofChronometric Dating. Royal Society (Philosophical Transactions of the Royal Society, 
series B, 337, no. 1280), London, U.K. 

Aitchison,J. 1996. The Seeds of Speech: Language Origin and Evolution . Cambridge Univer- 
sity Press, Cambridge, U.K. 

Arensburg, B. 1989. New skeletal evidence concerning the anatomy of Middle Paleolithic 
populations in the Middle East:The Kebara skeleton. Pages 165-171 in P. Mellars & C. 
Stringer, eds.. The Human Revolution: Behavioral and Biological Perspectives on the Ori- 
gins of Modern Humans. Princeton University Press, Princeton. 

Bahn. P.G. 1994. New advances in the field of ice age art. Pages 121-132 in M.H. Nitecki & 
D.V. Nitecki, eds.. Origins of Anatomically Modern Humans. Plenum Press, New York. 

Bar-Yosef O. 1992. The role of western Asia in modem human origins. Pages 193-200 in 
M.J. Aitken, C.B. Stringer, & P. A. Mellars, eds.. The Origin of Modern Humans and the Im- 
pact ofChronometric Dating. Royal Society (Philosophical Transactions of the Royal So- 
ciety, series B, 337, no. 1280), London, U.K. 

. 1994. The contribution of southwest Asia to the study of the origins of modem hu- 
mans. Pages 23-66 in M.T. Nitecki & D. V. Nitecki, eds, Origins of Anatomically Modern 
Humans. Plenum Press, New York. 

. 1 996. Modem humans. Neanderthals and the Middle/Upper Paleolithic transition in 

western Asia. Pages 1 75-1 90 in M. Piperno, ed.,77/e Lower and Middle Paleolithic: Collo- 
quia. XIII International Congress of Prehistoric and Protohistoric Sciences, Forii, Italy. 

Bard, E., B. Hamelin, R.G. Fairbanks, & A. Zindler. 1990. Calibration of the '"C timescale 
over the past 30,000 years using mass spectrometric U-Th ages from Barbados corals. yVa- 
mre 354:405-410. 

Bickerton, D. 1990. Language and Species. University of Chicago Press, Chicago. 

. 1995. Language and Human Behavior. University of Washington Press, Seattle. 

Binford, L.R. 1989. Isolating the transition to cultural adaptations: An organizational ap- 
proach. Pages 18-41 in E. Trinkaus, ed., The Emergence of Modern Humans: Biocultural 
Adaptations in the Later Pleistocene. Cambridge University Press, Cambridge, U.K. 

Bordes, F. 1 984. Legons sur le Paleolithique: vol. 2: Le Paleolithique en Europe. Centre Na- 
tional de la Recherche Scientitlque (Instimt du Quatemaire, Universite de Bordeaux I, Ca- 
hiers du Quaternaire no. 7), Paris, France. 

Bosinski, G. 1990. Homo Sapiens. Editions France, Paris, France. 

Brauer, G., Y. Yokoyama, C. Falgueres, & E. Mbua. 1 997. Modern human origins backdated. 
Nature 386:337. 


Byers, A.M. 1994. Symboling and the Middle-Upper Paleolithic transition: a theoretical and 
methodological critique. Curr. Anthropoi 35:369-399. 

Byrne, R.W. 1996. Relating brain size to intelligence in primates. Pages 49-56 /« P. Mellars& 
K. Gibson, eds. Modeling the Early Human Mind. McDonald Institute for Archaeological 
Research, Cambridge, U.K. 

Byrne, R.W. & A. Whiten. 1 988. Machiavellian Intelligence: Social Expertise and the Evolu- 
tion of Intellect in Monkeys, Apes and Humans. Oxford University Press, Oxford, U.K. 

. 1992. Cognitive evolution in primates: evidence from tactical deception. Man 


Cabrera Valdes, V. 1984. El Yacimento de la Cueva de "El Castillo " (Puente Viesgo, San- 
tander). Bibliotheca Praehistorica Hispana, vol. XXII, Madrid, Spain. 

Cann, R.L., O. Rickards, & J. Koji Lum. 1 994. Mitochondrial DNA and human evolution: Our 
one lucky mother. Pages 135-148 in M.H. Nitecki & D.V. Nitecki, eds.. Origins of Ana- 
tomically Modern Humans. Plenum Press, New York. 

Chase, P.O. 1991. Symbols and Paleolithic artifacts: style, standardization, and the imposition 
of arbitrary form. Jour. Anthropoi. Archeol. 10:193-214. 

Chase, P.G.& H.L. Dibble. 1987. Middle Paleolithic symbolism: A review of current evidence 
and interpretations. Jour. Anthropoi. Archeol. 6:263-296. 

Chauvet, J.-M., E.B. Deschamps,&C. Hillaire. 1995. LaGrotteChauvet. Seuil, Paris, France. 

Cheney, D.L.&R.M. Seyfarth. 1990. How Monkeys see the World: Inside the Mind of Another 
Species. University of Chicago Press, Chicago. 

— . 1996. Function and intention in the calls of non-human primates. Pages 59-76 in 

W.G. Runciman, J. Maynard Smith, & R.I. Dunbar, eds.. The Evolution of Social Behav- 
iour Patterns in Primates and Man. Oxford University Press (Proceedings of the British 
Academy, vol. 88), Oxford, U.K. 

Clark, J. D. 1992. African and Asian perspectives on the origins of modem humans. Pages 
201-216 in M. Aitken, C.B. Stringer, & P. A. Mellars, eds. The Origin of Modern Humans 
and the Impact ofChronometric Dating. Royal Society (Philosophical Transactions of the 
Royal Society, series B, 337, no. 1280), London, U.K. 

Clark, G.A. 1997. The Middle-Upper Paleolithic transition in Europe: an American perspec- 
tive. Norweg. Archaeol. Rev. 30:25-53. 

Clottes, J. 1996. Recent studies on Paleolithic art. Cambr. Archaeol. Jour. 6: 1 79-1 89. 

David, N.C. 1985. E.xcavation of the Abri Pataud, Les Eyzies (Dordogne) France: TheNoail- 
lian (level 4) Assemblage and the Noaillian Culture in Western Europe. Peabody Museum, 
Harvard University, Cambridge. 

Davidson, 1. & W. Noble. 1989. The archaeology of perception: Traces of depiction and lan- 
guage. Curr. Anthropoi. 30:125-155. 

. 1993. Tools and language in human evolution. Pages 363-387 in K.R. Gibson & T. 

Ingold, eds.. Tools, Language and Cognition in Human Evolution . Cambridge University 
Press, Cambridge, U.K. 

Deacon, H. 1989. Late Pleistocene palaeoecology and archaeology in the southern Cape, 
South Africa. Pages 547-564 in P. Mellars & C. Stringer, eds. The Human Revolution: Be- 
havioral and Biological Perspectives on the Origins of Modern Humans. Princeton Univer- 
sity Press, Princeton. 

Defleur, A. 1993. Les Sepultures Mousteriennes. CNRS, Paris, France. 

Delluc, B. & G. Delluc. 1978. Les manifestations graphiques aurignaciennes sur support 
rocheux des environs des Eyzies (Dordogne) Gallia-Prehistoire 21:21 3^38. 

Dibble, H.L. 1989. The implications of stone tool types for the presence of language in the 
Lower and Middle Paleolithic. Pages 415-432 in P. Mellars & C. Stringer, eds.. The Hu- 


man Revolution: Behavioral and Biological Perspectives on the Origins of Modern 
Humans. Princeton University Press, Princeton. 

Donald, M. 1 99 1 . Origins of the Modern Mind: Three Stages in the Evolution of Culture and 
Cognition. Harvard University Press, Cambridge. 

Dunbar, R. 1996. Grooming. Gossip and the Evolution of Language. Faber & Faber, London, 

Farizy, C, ed. 1990. Paleolithique Moyen Recent et Paleolithique Superieur Ancien en 
Europe. Memoires du Musee de Prehistoire d'lle de France, Nemours, France. 

Foley, R. & M.M. Lahr. 1997. Mode 3 technologies and the evolution of modem humans. 
Cambr. Archaeol. Jour. 7:3-36. 

Gambier, D. 1989. Fossil hominids from the eariy Upper Paleolithic (Aurignacian) of France. 
Pages 194-21 1 in P. Mellars & C. Stringer, eds.. The Human Revolution: Behavioral and 
Biological Perspectives on the Origins of Modern Humans. Princeton University Press, 

. 1993. Les hommes modemes du debut du Paleolithique superieur en France: bilan 

des donnees anthropologiques et perspectives. Pages 409-430 in V. Cabrera Valdes, ed.. El 
Origen del Hombre Moderno en el Suroeste de Europa. Madrid: Universidad Nacional de 
Educacion a Distancia, Spain. 

Gamble, C. 1986. The Paleolithic Settlement of Europe. Cambridge University Press, Cam- 
bridge, U.K. 

Gargett, R.H. 1989. Grave shortcomings: The evidence for Neanderthal burial. Curr. Anthro- 
pol. 30: 1 57-90. 

Gibson, K. R. 1996. The biocultural human brain, seasonal migrations, and the emergence of 
the Upper Paleolithic. Pages 33-46 in P. Mellars & K. Gibson, eds.. Modeling the Early Hu- 
man Mind. McDonald Institute for Archaeological Research, Cambridge, U.K. 

Gibson, K.R. & T. Ingold. 1993. Tools, Language and Cognition in Human Evolution . Cam- 
bridge University Press, Cambridge, U.K. 

Gowlett, J.A.J. 1 996. Mental abilities of early Homo: Elements of constraint and choice in rule 
systems. Pages 191-215 in P. Mellars & K. Gibson, eds.. Modeling the Early Human Mind. 
McDonald Institute for Archaeological Research, Cambridge, U.K. 

Griin, R. & C. Stringer. 1 99 1 . Electron spin resonance dating and the evolution of modem hu - 
mans. Archaeometry 33:\53-\99. 

Hahn, J. 1972. Aurignacian signs, pendants, and art objects in Central and Eastem Europe. 
World Archaeol. 3:252-266. 

. 1 977. Aurignacien: Das Altere Jungpaldolithikum in Mittel- und Osteuropa. Funda- 

menta Series A9, Koln, Germany. 

1 993. Aurignacian art in Central Europe. Pages 229-24 1 in H. Knecht, A. Pike-Tay & 

R. White, eds., Before Lascaux: The Complex Record of the Early Upper Paleolithic. CRC 
Press, Boca Raton. 

Harpending, H., S. Sherry, A. Rogers, & M. Stoneking. 1993. The genetic structure of ancient 
human populations. Curr Anthropol. 34:483^96. 

Hayden, B. 1993. The cultural capacities of Neanderthals: A review and reevaluation. Jour. 
Hum. Evol. 24:113-146. 

Hodder, I. 1982. Symbols in Action: Ethnoorchaeological Studies of Material Culture. Cam- 
bridge University Press, Cambridge, U.K. 

Howell, F.C. 1996. Thoughts on the study and interpretation of the human fossil record. Pages 
1-45 in W.E. Meikle, F.C. Howell, & N.G. Jablonski,eds., Contemporary Issues in Human 
Evolution. California Academy of Sciences, Memoir 21, San Francisco. 

Humphrey, N. 1976. The social functions of intellect. Pages 303-317 /« P.P. G. Bateson& R. 
A. Hinde, eds., Growing Points in Etholog\\ Cambridge University Press, Cambridge, U.K. 


Isaac, G.LI. 1972. Early phases of human behaviour: Models in Lower Paleolithic archae- 
ology. Pages 167-199 in D.L. Clarke, ed.. Models in Arcliaeolog)'. Methuen, London, 

Jochim, M. 1983. Paleolithic cave art in ecological perspective. Pages 212-219 in G.N. Bai- 
ley, ed.. Hunter-gatherer Economy in Prehistoiy: A European Perspective. Cambridge 
University Press, Cambridge, U.K. 

Klein, R.G. 1 989. The Human Career: Human Biological and Cultural Origins. University of 
Chicago Press, Chicago. 

Knecht, H. 1993. Splits and wedges: Thetechniquesand technology of early Aurignacian ant- 
ler working. Pages 137-162 in H. Knecht, A. Pike-Tay, & R. White, eds.. Before Lascaux: 
The Complex Record of the Early Upper Paleolithic. CRC Press, Boca Raton. 

Knecht, H., A. Pike-Tay, & R. White, R. eds. 1993. Before Lascaux: The Complex Record of 
the Early Upper Paleolithic. CRC Press, Boca Raton. 

Knight, CD. 1991 . Blood relations: Menstruation and the Origins of Culture. Yale University 
Press, New Haven. 

Knight, C, C. Power, & I. Watts. 1995. The human symbolic revolution: A Dai^v^'inian ac- 
count. Cambr. Archaeol. Jour. 5:75-1 14. 

Kozlowski, J.K. 1990. A multi-aspectual approach to the origins of the Upper Paleolithic in 
Europe. Pages 419-437 in P. Mellars, ed.. The Emergence of Modern Humans: an ar- 
chaeological perspective. Edinburgh University Press, Edinburgh. 

Kozlowski, J.K. & S.K. Kozlowski. 1979. Upper Paleolithic and Mesolithic in Europe: Tax- 
onomy and Prehistoiy. Polska Akademia Nauk, Cracow, Poland. 

Krings, M.,A.Stone,R.W. Schmitz, H.Krainitzki, M.Stoneking, & S. Paabo. 1997. Neander- 
thal DNA sequences and the origin of modem humans. Cell 90: 1 9-30. 

Lahr, M.M., & R. Foley. 1 994. Multiple dispersals and modem human origins. Evol. Anthro- 
pol. 3:48-60. 

Lieberman, D.E. & J.J. Shea. 1994. Behavioral differences between archaic and modem hu- 
mans in the Levantine Mousterian. .4mer. Anthropol. 96:300-322. 

Lieberman, P. 1989. The origins of some aspects of human language and cognition. Pages 
391-414 /« P. Mellars &C. Stringer, eds.. The Human Revolution: Behavioral and Biologi - 
cal Perspectives on the Origins of Modern Humans. Princeton University Press, Princeton. 

. 1991. Uniquely Human: The Evolution of Speech, Thought, and Selfless Behavior. 

Harvard University Press, Cambridge. 

Lourandos, H. 1996. Continent of Hunter-Gatherers: New Perspectives in Australian Prehis- 
tory. Cambridge University Press, Cambridge, U.K. 

Marshack, A. 1991. The Roots of Civilization. Moyer Bell, Mount Kisco, New York. 

Mellars, P. A. 1969. The chronology of Mousterian industries in the Perigord region of south- 
west France. Proceed. Prehist. Soc. 35:134-171. 

. 1973. The character of the Middle-Upper Paleolithic transition in southwest France. 

Pages 255-276 in C. Renfrew, ed., The Explanation of Culture Change: Models in Prehis- 
tory. Duckworth, London, U.K. 

— . 1 985. The ecological basis of social complexity in the Upper Paleolithic of southwest- 
ern France. Pages 271-297 in T.D. Price & J. A. Brown, eds.. Prehistoric Hunter- 
Gatherers: The Emergence of Cultural Complexity. Academic Press, Orlando. 

1989a. Major issues in the emergence of modern humans. Curr. Anthropol. 


. 1989b. Technological changes across the Middle-Upper Paleolithic transition: Tech- 
nological, social, and cognitive perspectives. Pages 338-365 in P. Mellars & C. Stringer, 
eds.. The Human Revolution: Behavioral and Biological Perspectives on the Origins of 
Modern Humans. Princeton University Press, Princeton. 


. 1991. Cognitive changes and the emergence of modem humans in Europe. Cambr. 

Archaeol. Join: 1 : 63-76. 
. 1 992. Archaeology and the population-dispersal hypothesis of modem human origins 

in Europe. Pages 225-234 /// M. Aitken, C.B. Stringer, & P. A. Mellars, eds.. The Origin of 
Modern Humans and the Impact ofChronometric Dating. Royal Society (Philosophical 
Transactions of the Royal Society, series B, 337, no. 1280), London, U.K. 
. 1996a. The Neanderthal Legacy: An Archaeological Perspective from Western 

Europe. Princeton University Press, Princeton. 
. 1 996b. Symbolism, language and the Neanderthal mind. Pages 1 5-32 in P. Mellars & 

K. Gibson, eds., Modeling the Early Human Mind. McDonald Institute for Archaeological 
Research, Cambndge, U.K. 
. 1996c. The emergence of biologically modem populations in Europe: A social and 

cognitive 'revolution'? Pages 179-201 in W.G. Runciman, J. Maynard Smith, & R.l. Dun- 
bar, eds., The Evolution of Social Behaviour Patterns in Primates and Man. Oxford Uni - 
versity Press (Proceedings of the British Academy, vol. 88), Oxford, U.K. 

Mithen, S. 1 990. Thoughtful Foragers: A Study of Prehistoric Decision Making. Cambridge 
University Press, Cambridge, U.K. 

. 1 996a. The Prehistoiy of the Mind: A Search for the Origins of Art, Religion and Sci- 
ence. Thames & Hudson, London, U.K. 

. 1 996b. Domain-specific intelligence and the Neanderthal mind. Pages 2 1 7-229 in P. 

Mellars & K. Gibson, eds.. Modelling the Early Human Mind. McDonald Institute for Ar- 
chaeological Research, Cambridge, U.K. 

Mussi, M. 1 992. Popoli e Civil ta ' dell 'Italia Antica: Volume Decimo: II Paleolitico e il Meso- 
litico in Italia. Biblioteca di Storia Patria, Bologna, Italy. 

Nitecki, M.H. & D.V. Nitecki, eds. 1994. Origins of Anatomically Modern Humans. Plenum 
Press, New York. 

Noble, W. & I. Davidson. 1996. Human Evolution. Language and Mind: A Psychological and 
Archaeological Inquir}'. Cambridge University Press, Cambridge, U.K. 

Oswalt, W.H. 1976. An Anthropological Analysis of Food-getting Technology. John Wiley, 
New York. 

Otte. M. \9S\. Le Gravettien en Europe Centrale. Dissertationes Archaeologicae Gandenses, 
No. 20, Bmges, Belgium. 

Passingham, R.E., 1989. The origins of human intelligence. Pages 123-136 /T? R. Durant, ed.. 
Human Origins. Clarendon Press, London, U.K. 

Peterson, N.,ed. 1976. Tribes and Boundaries in Australia . Australian Institute for Aboriginal 
Studies, Canberra, U.K. 

Pfeiffer, J.E. 1 982. The Creative Explosion: An Inquiiy into the Origins of Art and Religion . 
Harper & Row, New York. 

Pinker, S. 1994. The Language Instinct: The New Science of Language and Mind. Penguin, 
London, U.K. 

Price, T.D. & J. A. Brown, eds. 1985. Prehistoric Hunter-Gatherers: The Emergence of Cul- 
tural Complexity. Academic Press, Orlando. 

Ragir, S. 1985. Retarded development: The evolutionary mechanism underlying the emer- 
gence of language. Jour Mind Behav. 6:45\-46S. 

Renfrew, C. 1996. The sapient behaviour paradox: How to test for potential? Pages 11-14 in P. 
Mellars & K. Gibson, eds.. Modelling the Early Human Mind. McDonald Institute for Ar- 
chaeological Research, Cambridge, U.K. 

Rodseth, L., R.W. Wrangham. A.M. Harrigan, & B.B. Smuts. 1 99 1 . The human community as 
a primate society. Curr. Anthropol. 32:221-254. 


Sackett, J. 1988. The Mousterian and its aftermath: A view from the Upper Paleolithic. Pages 
4 1 3^26 in H.L. Dibble & A. Montet-White, eds.. Upper Pleistocene Prehistory of Western 
Eurasia. University Museum of Pennsylvania, Monograph 54, Philadelphia. 

Schick, K.D. & N. Toth. 1993. Making Silent Stones Speak: Human Evolution and the Dawn 
of Technology. Weidenfeld & Nicolson, London, U.K. 

Shennan, S. 1996. Social inequality and the transmission of cultural traditions in forager socie- 
ties. Pages 365-379 in J. Steele & S. Shennan, eds.. The Archaeolog}' of Human Ancestiy. 
Routledge, London, U.K. 

Singer, R. & J. Wymer. 1982. The Middle Stone Age at Klasies River Mouth in South Africa. 
University of Chicago Press, Chicago. 

Smith, P.E.L. 1964. The Solutrean culture. Sci. Amer 21 1:86-94. 

. 1966. Le Solutreen en France. Delmas, Bordeaux, France. 

Soffer, O., 1994. Ancestral lifeways in Eurasia — the Middle and Upper Paleolithic records. 
Pages 1 01-1 09 in M.H. Nitecki & D.V. Nitecki, eds.. Origins of Anatomically Modern Hu- 
mans. Plenum Press, New York. 

Sonneville-Bordes, D. de. 1960. Le Paleolithique Superieur en Pergord. Delmas, Bordeaux, 

Stoneking, M.,S.T. Sheiry, A.J. Redd,& L. Vigilant. 1992. New approaches to dating suggest 
a recent age for the human DNA ancestor. Pages 1 67-1 76 in M. Aitken, C.B. Stringer & 
P. A. Mellars, eds., The Origin of Modern Humans and the Impact of Chronometric Dating. 
Royal Society (Philosophical Transactions of the Royal Society, series B, 337, no. 1 280), 
London, U.K. 

Straus, L.G. & C. W. Heller. 1 988. Explorations in the twilight zone: The Upper Paleolithic of 
Vasco-Cantabrian Spain andGascogny. Pages 97- 134 in J.F. Hoffecker& C.A. Wolf eds.. 
The Early Upper Paleolithic: evidence from Europe and the Near East. British Archaeo- 
logical Reports S437, Oxford, U.K. 

Stringer, C. & C. Gamble. 1 993. In Search of the Neanderthals: Solving the Puzzle of Human 
Origins. Thames & Hudson, London, U.K. 

Stringer, C. & R. McKie. 1996. African Exodus: The Origins of Modern Humanity. Jonathan 
Cape, London, U.K. 

Taborin, Y. 1993. Shells of the French Aurignacian and Perigordian. Pages 21 1-229 /'// H. 
Knecht, A. Pike-Tay, & R. White, eds.. Before Lascaux: The Complex Record of the Early 
Upper Paleolithic. CRC Press, Boca Raton. 

Tishkoff, S.A., E. Dietzsch, W. Speed, A.J. Pakstis, J.R. Kidd, K. Cheung, B. Bonne Tamir, 
A.S. Santachiara-Benerecetti, P. Moral, M. Krings, S. Paabo, E. Watson, N. Risch, T. 
Jenkins & K.K. Kidd. 1996. Global patterns of linkage disequilibrium attheCD4 locus and 
modem human origins. Science 271 : 1380-1 387. 

Tooby, J. & L. Cosmides. 1992. The psychological foundations of culture. Pages 19-136 in 
J.H. Barkow, L. Cosmides, & J. Tooby, eds.. The Adapted Mind. Oxford University Press, 
Oxford, U.K. 

Van Andel, T. & P.C. Tzedakis. 1996. Palaeolithic landscapes of Europe and environs, 
150,000-25,000 years ago. Quat. Sci. Rev. 15:481-500. 

Vandermeersch, B. 1970. Une sepulture mousterien avec offrandes decouverte dans la grotte 
de Qafzeh. Comptes Rendus de I 'Academic des Sciences de Paris 270:298-30 1 . 

. 1976. Les sepultures neanderthaliennes. Pages 725-727 in H. de Lumley, ed.. La 

Prehistoire Frangaise. Centre National de la Recherche Scientifique, Paris, France. 

Wiessner, P. 1983. Style and social information in Kalahari San projectile points. Amer Antiq- 
uity 48:253-276. 

. 1 984. Reconsidering the behavioral basis for style: a case study among the Kalahari 

San. Jour Anthropol. Archaeol. 3:190-234. 


Whallon, R. 1989. Elements of cultural change in the later Palaeolithic. Pages 433-454 in P. 
Mellars & C. Stringer, eds.. The Human Revolution: Behavioral and Biological Perspec- 
tives on the Origins of Modern Humans. Princeton University Press, Princeton. 

White, R. 1989. Production complexity and standardization in early Aurignacian bead and 
pendant manufacture: evolutionary impHcations. Pages 366-390 in P. Mellars & C. 
Stringer, eds.. The Human Revolution: Behavioral and Biological Perspectives on the Ori- 
gins of Modern Humans. Princeton University Press, Princeton. 

. 1993. Technological and social dimensions of "Aurignacian-age" body ornaments 

across Europe. Pages 277-300 in H. Knecht, A. Pike-Tay, & R. White, eds., Before La- 
scaitx: The Complex Record of the Early Upper Paleolithic. CRC Press, Boca Raton. 

Wolpoff, M.H., A.G. Thome, F.H. Smith, D.W. Prayer, & G.G. Pope. 1994. Multiregional 
evolution: a worldwide source for modern human populations. Pages 175-199 in M.H. 
Nitecki, & D.V. Nitecki, eds.. Origins of Anatomically Modern Humans. Plenum Press, 
New York. 


The Evolution of the Human 
Language Faculty 

Steven Pinker 

Department of Brain and Cognitive Sciences 
Massachusetts Institute of Technology 
Cambridge, N4A 02139 

Language is the remarkable faculty by which humans convey thoughts to one 
another by means of a highly structured signal. Language works by two principles: 
A dictionary of memorized symbols, that is, words, and a set of generative rules or- 
ganized into several subsystems, that is, grammar. The machinery of language ap- 
pears to have been designed to encode and decode prepositional information for 
the purpose of sharing it with others. Language is universally complex and devel- 
ops reliably throughout the species, partly independently of general intelligence. I 
suggest that language is an adaptation for sharing information. It fits with many 
other features of our zoologically distinctive "informavore" niche, in which people 
acquire, share, and apply knowledge of how the world works to outsmart plants, 
animals, and each other. 

If an alien biologist studying the human species were to observe us in this room to- 
day, it would be struck by the fact that you are sitting quietly doing nothing but listen- 
ing to a man make noises as he exhales. This remarkable habit of our species is called 
language. You are listening to my exhaling noises because I have coded information 
into some of the acoustic properties of that noise which your brains have the ability to 
decode. By this means I am able to transfer ideas from inside my head to inside your 

In this paper I will say a few words about the design of the human language faculty 
— the tricks behind the ability to transfer information by making noise — and about 
whether language is an adaptation, and if so, what it is for. (For more detailed discus- 
sion of many of these points, see Pinker 1994.) 

Reprinted with minor changes from Pinker, S. 1997. Language as a psychological adaptation. 
Pages 162-180 in G.R. Bock & G. Cardew. eds.. Characterizing Human Psychological 
Adaptations. Ciba Foundation Symposium 208. Wiley, Chichester, U.K., by permission of the 
author and the Ciba Foundation. 

The Origin and Diversification oj Language Memoirs of the California Academy of Sciences 

Editors. N.G. Jablonski &. L.C. Aiello Number 24, Copyright © 1 998 


The Design of Human Language 

What is the secret behind my abihty to cause you to think specific thoughts by 
means of the vocal channel? There is not one secret, but two, and they were identified 
in the 19th century by continental linguists. 


The first principle was articulated by Ferdinand de Saussure and lies behind the 
mental dictionary, our finite memorized list of words. A word is an arbitrary symbol, 
a connection between a signal and an idea shared by all members of the community. 
The word duck doesn't look like a duck, walk like a duck or quack like a duck. How- 
ever, I can use it to convey the idea of a duck because we all have, in our developmen- 
tal history, formed the same connection between the sound and the meaning. 
Therefore, I can bring the idea to mind virtually instantaneously simply by making 
that noise. If instead I had to shape the signal to evoke the thought in your mind using 
some sensible connection between its form and its content, every word would require 
the amusing but inefficient contortions of a game of Charades. 

The symbols underlying words are bidirectional. Generally if I can say something, 
I can understand it, and vice versa. When children learn words, their tongues are not 
molded into the right shape by parents, and they do not need to go through a process of 
being rewarded for successive approximations to the target sound for every word they 
learn. Instead, children have an ability upon hearing somebody else use a word to 
know that they in turn can use it to that person or to a third party and expect to be un- 

Another noteworthy feature of the mental dictionary is its size. One can use dic- 
tionary sampling techniques to estimate how many independent memorizations must 
have taken place to install a typical adult vocabulary. For example, if you take the 
largest dictionary you can find, pick the third word down on the right hand side of 
every tenth page, put the word in a multiple-choice test say, correct for guessing, and 
then multiply the performance on the test by the size of the dictionary, you obtain esti- 
mates for a typical high school student of about 60,000 independent words, probably 
twice that number for a highly literate high school student. The learning rate for the 
smaller estimate works out to be about one word every 90 minutes, starting at the age 
of one. Considering that every one of these words is as arbitrary as a telephone 
number or a date in history, it is remarkable that children pick one up every hour and a 
half, at the same time as they are struggling over multiplication tables, the date of the 
Treaty of Versailles, and so on. 


Of course we don't just learn individual words. We combine them into strings 
when we talk, and that leads to the second trick behind language, grammar. The prin- 
ciple behind grammar was articulated by Wilhelm von Humboldt as "the infinite use 
of finite media." Inside everyone's head there is a finite algorithm — it has to be finite 
because the head is finite — with the ability to generate an infinite number of potential 
sentences, each of which corresponds to a distinct thought. For example, our knowl- 
edge of English incorporates rules that say "a sentence may be composed of a noun 
phrase (subject) and a verb phrase (object)" and "a verb phrase may be composed of a 


verb, a noun phrase (object), and a sentence (complement)." That pair of rules is [re- 
cursive]: an element is introduced on the right hand side of one rule which is also on 
the left hand side of the other rule, creating the possibility of a loop that could generate 
sentences of any size, such as "I think that she thinks that he said that I wonder 
whether ...." We can generate an infinite number of sentences, and each sentence ex- 
presses a distinct thought. Therefore this system gives us the ability to put an unlim- 
ited number of distinct thoughts into words, and other people to interpret the string of 
words to recover the thoughts. 

Grammar can be thought of as a discrete combinatorial system, which can be op- 
posed to a blending system. In a blending system all the possibilities that you get 
when you combine the ingredients lie on a continuum between the two endpoints. For 
example, if you mix red paint and white paint you obtain various degrees of pink 
paint, which differ in an analogue fashion. In a discrete combinatorial system, such as 
DNA, atoms, molecules, and sentences, each one of the combinations can be qualita- 
tively different from the other combinations and from the ingredients. Therefore, 
there is an infinite number of thoughts, but the set is infinite in a discrete way as op- 
posed to a continuous way. 

Grammar can express a remarkable range of thoughts because our knowledge of 
languages resides in an algorithm that combines abstract symbols, such as noun and 
verb, as opposed to concrete concepts, such as man and dog or eater and eaten. This 
gives us an ability to talk about all kinds of wild and wonderful ideas. We can talk 
about a dog biting a man, or, as in the journalist's definition of "news," a man biting a 
dog. We can talk about aliens landing at Harvard. We can talk about the universe be- 
ginning with a big bang, or the ancestors of native Americans immigrating to the con- 
tinent over a land bridge from Asia during an Ice Age, or Michael Jackson marrying 
Elvis's daughter. All kinds of novel ideas can be communicated because our knowl- 
edge of language is couched in abstract symbols like "noun" and "verb" which can 
embrace a vast set of concepts, and then can be combined freely into an even vaster set 
of propositions. How vast? In principle it is infinite; in practice it can be crudely esti- 
mated by assessing the number of word choices possible at each point in a sentence 
(roughly, 10) and raising it to a power corresponding to the maximum length of a sen- 
tence a person is likely to produce and understand, say, 20. The number is lO"'^ or 
about a hundred million trillion sentences. 

Let me say a bit more about the design of the human grammatical system. Gram- 
mar is not just a single pair of rules, as I listed above, but hundreds or thousands of 
rules, which fall into a number of subsystems. The most prominent is syntax, the com- 
ponent of language that combines words into phrases and sentences. One of the tools 
of syntax is word order. That is the tool that allows us to distinguish Man bites dog 
from Dog bites man. Order is the first thing people think of when they think about 
syntax, but in fact it is a relatively superficial manifestation; there are complex princi- 
ples that underlie its ability to convey ideas. 

Far more important than linear word order is constituency. A sentence has a hier- 
archical structure. It can be represented as a bracketed string of phrases embedded 
within phrases, which allow us to convey complex propositions of ideas embedded 
inside ideas. A simple demonstration of the brain's ability to parse sentences into hi- 
erarchical phrase stiuctures is an (unintentionally) ambiguous sentence published in 
TV Guide: On tonight 's program Dr. Ruth will discuss sex with Dick Cavett. We have 
a single string of words in a particular order, but it has two very different meanings. 
The intended meaning was that Dick Cavett is who you discuss sex with, and the alter- 
native meaning is that sex with Dick Cavett is what you discussed. In this particular 


sentence they are, unfortunately, expressed by the same string of words; the different 
interpretations correspond to the different phrase structure bracketings our brain can 
impose on that string. Thankfully, not all sentences are as blatantly ambiguous as that 
one, and the brain's ability to compute phrase structure is put to use in recovering the 
single meaning that the speaker intended. Much as in other symbolic systems that en- 
code logical information, such as arithmetic, propositional calculus, and computer 
programming, it is important in linguistic communication to get the parentheses right, 
and that's what phrase structure is for. 

Syntax also involves predicate-argument structure, the component of language 
that encodes the information that a logician would express as a predicate, that is, a re- 
lationship among a set of participants, and its arguments, that is, the particular partici- 
pants in a given instance of that relationship. In particular, what your grammar school 
teacher called the predicate of the sentence, namely the verb, is similar in many ways 
to what a logician would call a predicate. To understand a sentence you cannot merely 
pay attention to the order of words, or even just group them. You have to look up in- 
formation associated with the predicate. A simple demonstration the pair of sentences 
Man fears dog and Man frightens dog. The word man is the subject of both of these 
sentences, but the semantic role that man is playing in the first sentence is different 
from the role it is playing in the second sentence: In one case the man is causing the 
fear; in the other the man is being affected by fear. That shows that in understanding a 
sentence you have to look up information stored with the mental dictionary entry of 
the verb and see whether it says "my subject is the one doing the fearing" or "my sub- 
ject is the one causing the fear." 

A fourth trick of syntax is the operation called the transformation, which is associ- 
ated most strongly with Noam Chomsky's theory of transformational grammar. After 
you have generated a hierarchical tree structure into which the words of a sentence are 
plugged, a further set of operations can mangle the order of words in precise ways. For 
example, the sentence Do^zi^ bitten by man contains the verb bite, which ordinarily re- 
quires a direct object. But in this sentence the object is missing from its customary lo- 
cation; it has been moved to the front of the sentence. This gives us a means of shifting 
the emphasis and the quantification of a given set of participants in a relationship. 
The sentence Man bites dog and Dog is bitten by man both express the same informa- 
tion about who did what to whom — a man does the biting, a dog gets bitten — but one 
of them is a comment about the man and the other is a comment about the dog. Simi- 
larly, sentences in which a phrase is replaced by a [wh]-word and moved to the front 
of a sentence, such as Who did the dog bite?, allow the speaker to seek the identity of 
one of the participants in a given interaction. Transformations, therefore, give us a 
layer of meaning above and beyond who did what to whom. That layer emphasizes or 
seeks information about one participant or another, while keeping the actual event 
that one is talking about constant. 

Syntax, for all that complexity, is only one component of grammar. All languages 
have a second combinatorial system, morphology, in which bits of words are assem- 
bled to produce whole words. In English we don't have much morphology, com- 
pared with other languages, but we have some. The noun duck comes in two forms — 
duck and ducks — and the verb quack in four — quack, quacks, quacked, and quack- 
ing. In other languages morphology plays a much greater role. In Latin, for example, 
there is a rich inflectional system which plays an important role in expression. By 
placing suffixes onto the ends of nouns, one can convey information about who did 
what to whom, allowing one to scramble the left-to-right order of the words for em- 
phasis or style. For example, Canis hominem mordet and Hominem canis mordet 


have the same non-newsworthy meaning, and Homo canem mordet and Canem homo 
mordet have the same newsworthy meaning. 

In addition to syntax and morphology, language comprises a third combinatorial 
system — a third layer of assembly of elements into larger elements by rules, called 
phonology. The rules of phonology govern the sound pattern of a language. In no lan- 
guage do people form words by associating them directly with articulator/ gestures 
like a movement of the tongue or lips. Instead, a fixed set of articulator/ gestures is 
combined into sequences, each sequence defining a word. The combinations are gov- 
erned by phonological rules that people have to acquire as they acquire a language. In 
English, for example, we know that bluck is not a word but could be a word, whereas 
nguck is not a word and could not be a word because the English rules of word forma- 
tion don't allow the consonant ng at the beginning. In other languages, ng can be 
placed there. Interestingly, whereas syntax and morphology are semantically compo- 
sitional, that is, you can predict the meaning of the whole by the meanings of the ele- 
ments and the way they are combined, this is not true of phonology. You cannot 
predict the meaning of duck from the meaning ofd, the meaning ofu, and the meaning 
of A'. The combinatorial system called phonology simply allows us to have large vo- 
cabularies, for example, 1 00,000 words, without having to pair each one of them with 
a different simple noise coming out of the mouth. 

Phonology also consists of a set of adjustment rules which, after the words are de- 
fined and combined into phrases, smooth out the sequence of aiticulatory gestures to 
make them easier to pronounce and comprehend. One of those rules in English causes 
us to pronounce the same morpheme -ed for the past tense in three different ways de- 
pending on what it is attached to. In jogged it is pronounced as d. In walked it is pro- 
nounced as a t, thanks to a rule that keeps the consonants at the end of a word either all 
voiced (larynx buzzing) or all unvoiced. And \n patted the suffix is pronounced with 
the neutral schwa vowel before it, thanks to a rule that inserts a vowel to separate two 
J-like sounds. Therefore, even though the actual morpheme is the same in all cases, 
that is, d, there are rules that fiddle with the pronunciation pattern before it is articu- 
lated. These adjustments are not just peoples' effort to be clear, or for that matter their 
tendency to be lazy, as they put words together. There is a set of regulations for each 
language that dictate when you are allowed to be lazy and when you are not, and they 
are partly arbitrary in that you acquire them as you acquire the sound pattern of a lan- 
guage. (An accent is what happens when someone applies the phonological adjust- 
ment rules of one language to the content of another language.) Phonological rules 
have the fiinction of helping people achieve a mixture of clarity and ease of pronun- 
ciation, but they are a distinct part of one's knowledge of language. 

Interfaces of Language With Other Parts of the Mind 

Grammar is only one component of language. It has to look toward the rest of the 
mind in three different directions. Grammar has to be connected to the ear, so that we 
can understand; to the mouth, so that we can articulate; and to the rest of the mind, so 
that we can say sensible things in the context of a conversation. Each requires an inter- 

The first interface is the speech articulation or articulator/ phonetic system. One 
of the salient properties of this system is the actual anatomy of the vocal tract, which 
seems to have evolved in the human lineage in the service of the language. Darwin 
pointed to the fact that every mouthful of food we swallow has to pass over the tra- 
chea, with some chance of getting lodged in it and causing death by choking. The hu- 


man vocal tract has a low larynx by mammalian standards; this placement 
compromises a number of physiological functions but allows us to articulate a large 
range of vowel sounds. Because our larynx is so low in the throat it gives room for the 
tongue to move both back and forth and up and down independently. This defines a 
two-dimensional space in which the tongue can move. Because there are two resonant 
cavities, defined by the position of the tongue with respect to the throat, on one hand, 
and the mouth, on the other, we can produce a two-dimensional space of vowel 
sounds, which multiplies out the number of distinct discriminable signals we can ar- 
ticulate. One can argue that given the physiological cost, that is, the risk of death by 
choking, there must have been a corresponding benefit in our evolutionary history, 
presumably the benefit of rapid, expressive communication. 

The second interface is speech comprehension. Information being received by the 
ear has to be unpacked into a meaning. One of the remarkable features of speech com- 
prehension is the way in which the brain can unpack a stream of sound into its compo- 
nent words, which are not physically demarcated in the sound stream by little silences 
analogous to the small spaces that separate words on a printed page. When we hear 
language as a string of words we are the victims of an illusion. We realize this only 
when we hear speech in another language: to our ears it sounds like a continuous rib- 
bon of sound, which is exactly what it is, physically speaking. Another demonstration 
bringing this feat to our attention is a kind of wordplay seen in doggerel such as 
"Mairzey doats and dozey doats" (Mares eat oats and does eat oats) and "Fuzzy 
Wuzzy was a bear," which are designed to exploit the fact that speech is not a discrete 
chain of words separated by silence. 

Another remarkable feature of speech comprehension is the rate at which informa- 
tion can be conveyed. A rapid talker can convey about 40 phonemes per second, and 
even a more leisurely talker can reach 10 to 20 phonemes per second. Twenty cycles 
per second is the lower limit of pitch perception in humans. We hear 20 beats per sec- 
ond not as 20 rapid events but as a low tone or a buzz. Clearly, when we are listening 
to speech at 25 phonemes per second we are not registering 25 separate auditoiy 
events, because that is neurologically impossible. There must be some sort of multi- 
plexing or compression of the infoimation. in which the phonemes are superimposed 
in the process of speaking and the brain has to unpack them in the process of under- 

Finally, language has an interface with more general inference systems. The de- 
coding of the literal information conveyed by words and grammar is just the first step 
of a long chain of inference by which we try to guess what the speaker wants us to 
think he or she is trying to say. We engage in this process of inference even in under- 
standing the simple sentences. A nice example from the linguist Jim McCawley 
shows the knowledge we must apply to something as simple as assigning referents to 
pronouns such as he and she. Imagine a dialogue in which Marsha says "I'm leaving" 
and John says "Who is he"? We all know who "he" is, not by any information that is 
explicitly encoded in the sentence, but by our knowledge about human behavior. 
Those expectations are brought to bear in understanding a sentence, in conjunction 
with the particular rules of grammar. 

Is Language an Adaptation? 

With that summary of the design of the human language faculty in mind, we can 
now turn to the question: Is language an adaptation? Darwin wrote, "Man has an in- 


stinctive tendency to speak, as we see in the babble of our young children, while no 
child has an instinctive tendency to bake, brew or write." This is probably the first 
statement that human language is an adaptation. What are the alternatives, and what is 
the evidence? 

One alternative is that language is not an adaptation itself, but is a manifestation of 
more general cognitive abilities, some form of "general intelligence," in which case 
general intelligence would be the adaptation, not language. There is a reasonable 
amount of evidence against this possibility. 

First, language is universal across societies and across all neurologically normal 
people within a society. There may be technologically primitive peoples, but there are 
no primitive languages. And the language of uneducated, working class, and rural 
speakers has been found to be systematic and rule-govemed, though the rules may be- 
long to a dialect that did not have the good fortune of becoming the standard dialect of 
Britain and its former colonies. 

Second, languages conform to a universal design. The languages of the highlands 
of New Guinea, for example, use computational machinery that is identical to that de- 
scribed earlier in this paper, even though that machinery was motivated by an exami- 
nation of English and other European languages. 

A third kind of evidence was alluded to by Darwin: the ontogenetic development 
of language. There is a uniform sequence of stages that children pass through all over 
the world. That sequence culminates in masteiy of the local language, despite the 
computational difficulty of programming a computational system to take in a finite 
sample of sentences from a couple of speakers (the child's parents) and induce a 
grammar for the infinite rest of the language. Moreover, children's speech patterns, 
including their errors, are highly systematic, and often confomi to linguistic univer- 
sals for which there was no direct evidence in parents' speech. 

A fourth kind of evidence also comes from the study of language acquisition. If 
children are thrown together without a model language, such as in a multilingual plan- 
tation or, if the children are deaf, a school that does not have people using sign lan- 
guage, the children will develop a systematic, rule-govemed language of their own, a 
phenomenon called creolization. 

A fifth kind of evidence is that language and general intelligence, to the extent we 
can make sense of that term, seem to be doubly dissociable in neurological and ge- 
netic disorders. In aphasias and in a developmental syndrome (probably genetic in 
origin) called Specific Language Impairment, intelligent people can have extreme 
difficulties speaking and understanding. Conversely, in what clinicians informally 
call "chatterbox syndromes," severely retarded children may talk a stream of fully 
grammatical English but with a content that is highly childlike or is confabulated, 
bearing no relation to the world. 

A different alternative to the hypothesis that language is an adaptation is the possi- 
bility that language indeed is a separate system from general intelligence, but that it 
evolved by nonselectionist mechanisms. Perhaps, on this view, language evolved all 
at once as the product of a macromutation, or as a byproduct of some other evolution- 
ary development such as evolving a large head. The main reason to doubt this theory 
is the standard argument for the operation of natural selection, the argument from 
adaptive complexity. The information processing circuitry necessary to produce, 
comprehend and learn language must involve a great deal of organization and detail. 
As with other complex biological systems that accomplish improbable feats, this cir- 
cuitry is unlikely to have evolved by something as crude as a single mutation or some 
other evolutionary force that is insensitive to what the circuit accomplishes. 


What Did Language Evolve For? 

If language is an adaptation, what is it an adaptation for? Note that asking this 
question is different from asking what language is typically used for, especially what 
it is used for at present. The question concerns the engineering design of language and 
the extent to which it informs us about the selective pressures that shaped it. 

What is the machinery of language trying to accomplish? The system looks as if it 
was put together to encode and decode digital prepositional information — who did 
what to whom, what is true of what, when, where and why — into a signal that can be 
conveyed from one person to another. 

It is not difficult to think of why it would have been a good thing for a species with 
the rest of our characteristics to evolve the ability to do this (Pinker 1997). The struc- 
tures of grammar are well suited to conveying information about technology, such as 
which two things should be put together to produce a third thing; about the local envi- 
ronment, such as where things are and which people did what to whom; and about 
one's own intentions, such as "If you do this, I will do that," which convey relation- 
ships of exchange and of dominance, as in threats. 

Gathering and exchanging information of this kind is, in turn, integral to the larger 
niche that modem Homo sapiens has filled, which George Miller has called the in- 
formavore niche and which John Tooby and Irven DeVore previously called the cog- 
nitive niche (Tooby & DeVore 1987). Tooby and DeVore have assembled a theory 
that tries to explain the list of properties of the human species that a biologist would 
consider zoologically unusual, such as our extensive manufacture of and dependence 
on complex tools, our wide range of habitats and diets, our long childhoods and long 
lives, our hypersociality, and our division into groups or cultures each with a set of 
distinctive local variations in behavior. Their explanation is that the human lifestyle 
is a consequence of a specialization for overcoming the evolutionary fixed defenses 
of plants and animals by cause-and-effect reasoning, which is driven by intuitive 
theories about various domains of the world, such as objects, forces, paths, places, 
manners, states, substances, hidden biochemical essences, and other people's beliefs 
and desires. 

The information captured in these intuitive theories is reminiscent of the informa- 
tion that the machinery of grammar is designed to convert into strings of noises. It is 
probably not a coincidence that what is special about humans is that we outsmart other 
animals and plants by cause-and-effect reasoning, and language seems to be a way of 
converting information about cause-and-effect and action into a signal. 

An unusual feature of information is that it can be duplicated without loss. If I give 
you a fish, I don't have the fish, as we know from sayings like "you can't eat your cake 
and have it." Information, however, can be both eaten and had: If I tell you how to 
fish, it is not the case that I now lack the knowledge of how to fish because I've given it 
away to you; we can both have it. There is a brilliant, eccentric computer programmer 
associated with MIT, Richard Stallman, who started a free software foundation based 
on the idea that no one should charge for software. While it's perfectly reasonable, he 
argues, for a baker to charge for bread, since there's only a finite amount of fiour and 
once it is given to someone then someone else cannot have it. But once software is de- 
veloped and can be copied, Stallman argues, there is no reason that it should not be 
free. If it is given to one person, that does not mean that someone else cannot have it 
(with the exception of the floppy disk itself)- 


Tooby and DeVore (1987) have pointed out that in a species like ours that lives on 
information, it is quite natural that in conjunction with evolving the ability to gather 
this information we evolved a means to exchange it. Having language multiplies the 
benefit of knowledge. Knowledge is not only useful to oneself as a way of figuring 
out, for example, how to build snares to catch rabbits, but it is also useful as a trade 
good: I can exchange it with somebody else at a low cost to myself and hope to get 
something in return. It can also lower the original acquisition cost. I can learn about 
how to catch a rabbit from someone else's trial and error; I don't have to go through it 

Language, therefore, fits with other features of the informavore niche. The zoo- 
logically unusual features of Homo sapiens can be united by the idea that humans 
have evolved an ability to encode information about the causal structure of the world 
and to share it among themselves. Our hypersociality makes sense because informa- 
tion is a particularly good commodity of exchange that makes it worth people's while 
to hang out together. Our long childhood is an apprenticeship — before we go out in 
the world, we spend a lot of time learning what eveiyone else around us has figured 
out already. The existence of culture can be seen as a kind of pool of local expertise. 
Many traditions develop locally because many of the requirements to deal with the 
various aspects of the world have been acquired by other people, resulting in a net- 
work of information sharing that is close to what sociologists and anthropologists 
have called "culture." Humans have long lifespans because once you've had an ex- 
pensive education you might as well make the most out of it by having a long lifespan 
during which the expertise can be put to use. Humans inhabit a wide range of habitats 
because we don't have knowledge that is highly specialized, such as how to catch a 
rabbit; our knowledge is more abstract, such as how living things work and how ob- 
jects bump into each other. That machinery for construing the world can be applied to 
many kinds of environments; it is not specific to a particular ecosystem. 

People have occasionally raised objections to the hypothesis that language is an 
adaptation for sharing information. One objection is that organisms are competitors, 
so sharing information is in fact costly by virtue of -the advantages it gives to one's 
competitors. If I teach someone to fish, they may overfish the local lake, leaving no 
fish for me. The argument, however, just boils down to the standard problem of the 
difficulties facing the evolution of cooperation or altruism, and the solution in the 
case of language is the same. By sharing information with our kin, we help copies of 
our genes inside those kin, and when it comes to sharing information with nonrela- 
tives, if we ensure that we inform only those people who return the favor, we both gain 
the benefits of trade. Certainly we do use our faculties of social cognition to ration our 
conversation with those with whom we have established a nonexploitative relation- 
ship; hence the expression "to be on speaking terms." 

A second objection is that language may be used to deceive, so perhaps language 
evolved as a means of manipulation rather than as means of communication. The an- 
swer to the objection is. once again, that language surely coevolved with our faculties 
of social cognition. We apply those faculties as we listen to others; we are constantly 
vigilant for whether we are being lied to. And I find it hard to imagine any coherent ac- 
count by which language evolved to allow us to manipulate others. Unlike signals 
with the physiological power to manipulate another organism directly, such as loud 
noises or chemicals, the signals of language are impotent unless the recipient actively 
applies complicated neural machinery to decode them. It is impossible to use lan- 
guage to manipulate someone who doesn't understand the language, so if language is 
an adaptation to manipulate others, how could it have gotten off the ground? What 


would have been the evolutionary incentive for the designated targets to evolve those 
exquisitely complex mental algorithms for unpacking the speech wave into words 
and assembling them into trees? Like a shop owner who makes sure he is not around 
when the gangster selling "protection"' comes by, or a negotiator who remains incom- 
municado until a deadline passes, hominids in the presence of the first linguistic ma- 
nipulators would have done best by refusing to allow their nascent language systems 
to evolve further, and language evolution would have been over before it began. 

Literature Cited 

Pinker, S. 1994. The Language Instinct. Harper Collins, New York. 

.1997. How the Mind Works. Norton, New York. 

Tooby, J. & I. DeVore. 1 987. The reconstruction of hominid evolution through strategic mod- 
eling. Pages 183-237 in W.G. Kinzey, ed., The Evolution of Human Behavior: Primate 
Models. SUNY Press, New York. 


The Origin and Dispersal of 
Languages: Linguistic Evidence 

Johanna Nichols 

Department of Slavic Languages and Literatures 
University of California 
Berkeley, CA 94720 

Tracing the descent of languages and reconstructing language families has pro- 
duced some of the great achievements of linguistic science in the 19th and 20th cen- 
turies, including classification of most of the world's languages as well as detailed 
linguistic reconstructions and fairly precise dates of origin for great language 
families such as Indo-European, Uralic, Semitic, Bantu, Dravidian, Chinese, Al- 
gonquian, Mayan, Austronesian, and others. Specific cultural reconstructions fol - 
low from reconstructed vocabulary and dates, and consequently many of these 
families have been rather securely linked to places of origin and some even to ar- 
cheological cultures. But linguistic descent cannot be traced back more than about 
10,000 to 12,000 years at the very most, and secure reconstruction has not suc- 
ceeded beyond about 6000 years, since regular grammar change and vocabulary 
loss gradually remove the critical evidence for ancestral vocabulary and grammar 
and the diagnostics that prove relatedness. Though this fade-out threshold is an 
absolute obstacle to reconstructing the ultimate ancestor(s) of the world's lan- 
guages, it can be turned to advantage, for it provides a temporal threshold enabling 
linguists to identify descent lineages of an approximately identical absolute age. 
This in turn makes possible various kinds of comparison which, together with 
some basic principles of linguistic geography, yield information on the geography 
and time frame of ancient migrations and dispersals. The resultant picture, 
though still spare and approximate, is much richer than a wordlist and a family 
tree, which are all one could hope to gain from tracing descent far back. 

Happily for present purposes, so little is known about the origin and dispersal of 
human language that everything we know can be outlined, at least in general form, in 
a single article. Let us begin with the notion of language family. There are about 6000 
different languages spoken on Earth, and in the century and a half since the relevant 
comparative methods were developed linguists have succeeded in working out a ge- 
netic classification of almost all of them — rough and provisional in some cases but 
refined and precise in most.' The 6000 languages fall into about 300 families, which 
further comparative and descriptive work may eventually succeed in reducing to 
about 200 lineages. These families vary in their sizes and ages, but each is genetically 
discrete, and the oldest ones that we can trace are generally in the range of about 6000 

The Origin and Divcrsificatiim <>/ Language Memoirs of the California Academy of Sciences 

EditorsrN.G. Jablonski & L.C. Aiello Number24, Copyright ©1998 


years old. Descent of course goes back farther, but traceable descent fades out rap- 
idly. This is because languages gradually change over time, and after a few millennia 
the kind of evidence that can prove family relatedness, or help reconstruct the ances- 
tral language, erodes beyond recognition. An average rate of vocabulary loss from a 
standard word list of 100 or 200 items has been computed at about 20% per millen- 
nium, and dates of separation can be calculated for pairs of languages by determining 
the number of cognate items they share from that list. (This technique is known as 
glottochronology. For a recent textbook presentation see Trask 1996:361; for a re- 
finement that produces remarkably accurate estimates of time depth, Embleton 1 986, 
1 99 1 . ) After 6000 years of separation, two languages are expected to exhibit only 7% 
shared cognates; and 7% represents the lowest number of resemblant items that can 
safely be considered distinct from chance (for the latter figure see Nichols, 
submitted). Hence the 6000-year age of the oldest securely reconstructable language 
families is also the age after which, on binary tests, the incidence of cognates slips be- 
low the level of significance." 

Thus, though there are a few exceptions (some mentioned below), 6000 years can 
be taken as a rule-of-thumb average for the oldest securely traceable language fami- 
lies. Now, anatomically modem humanity is about 100,000 years old, and, since the 
modem anatomy includes a speech tract dedicated to speech production and aural and 
cognitive capacities dedicated to speech decoding, it is safe to estimate that human 
language as we know it is at least 100,000 years old. Six thousand years just barely 
scratches the surface of this language prehistory. Therefore, there is no hope of recov- 
ering information about language origins by tracing linguistic descent. 

Furthermore, even if descent could be traced back beyond 6000 or so years, much 
important information would be missing. It is well known that many of the world's 
language families have gone extinct. From the ancient Near East alone, in just the last 
two to three thousand years Sumerian, Elamite, the Hurrian-Urartean line, and Hattic 
have vanished without descendants or demonstrable kin. There is no reason to doubt 
that extinction has always gone on, usually caused by language shift (for this term see 
again note 1 ). Therefore, even if descent could be traced far back, we would have de- 
scent lines only for existing and attested languages, which would give a distorted pic- 
ture of early language. As will be discussed below, at regional and even continental 
scales, the languages of large areas tend to bear a certain areal resemblance to one an- 
other, and this means that some of the potential stmctural diversity of the world's lan- 
guages has been lost to us, ironed out by pressure toward areal conformity. Some of 
the world's continents show evidence of bottlenecks and founder effects in their lin- 
guistic populations,' and this tells us that the accidental survival and proliferation of 
one language, and the accidental loss of another, can snowball into a large-scale effect 
on statistical frequencies of types of languages over the course of millennia. For all of 
these reasons, tracing the world's existing languages back to their source — even if it 
were possible — would not give us an accurate picture, either genetic or stmctural, of 
the languages spoken by the earliest humans. In view of these problems, different 
means of comparison that bypass descent entirely have begun to take shape in the last 
two decades. This paper summarizes some of their findings. 

Family Trees 

Languages fall into families, families are branches of older families, and so on. 
Henceforth the oldest descent groups that can be securely demonstrated and to which 



classic comparative-historical methods can be applied to yield a reconstructed pro- 
tolanguage and a family tree will be caWedstocks. A stock, then, is a maximal demon- 
strable clade or descent group, and as mentioned above the oldest stocks average 
about 6000 years old. The generic term family will refer here to descent groups of all 
ages and all kinds (including stocks, their branches, and subbranches). 

Figure 1 shows examples of some family trees, illustrated for most of the stocks of 
northern Eurasia. They are schematic, showing major branches but not individual 
languages. They show only branches that have survived to the present; this simplifies 
the known history of Indo-European, which has early written records, so as to make it 
comparable to the majority of language families that have little or no written history 
and are known only from their extant daughter languages. In all the diagrams the an- 
cestor is at the top and the descendants (daughter branches or daughter languages) at 
the bottom. They are drawn to a uniform time scale, so that each is traced back for the 
approximately 6000 years that linguistic science can trace descent. Consider the first 
tree, that of Indo-European. Proto-Indo-European broke up and began to disperse 
about 5500-6000 years ago, so in the diagram the ancestral point (or root of the tree) 
and the initial branching are at the veiy top. Proto-Indo-European broke over the 
course of about a millennium, into a large number of daughter branches, of which 
eight survive (Celtic, Romance or Italic. Greek, Albanian, Armenian, Gennanic, 
Balto-Slavic, Indo-Iranian; extinct early branches include Tocharian of Central Asia 
and the Anatolian branch which included Hittite). The daughter branches continued 
to differentiate, some rather elaborately, but the diagram does not show these lower 
branchings. Another family that broke up about 6000 years ago is Uralic, whose mod- 
em descendants include Hungarian, Finnish, Estonian, and the Samoyedic languages 
of western Siberia. Unlike Indo-European, Uralic underwent a modest initial binaiy 





Yukagir Turkic Mongol 





Yeni- Japa 



seian nese 




Korean Nivkh Ainu 

Figure l . Family trees for the linguistic stocks of northern Eurasia to a constant scale, with approximate 
internal branching structure corresponding to major branchings over the last ca. 6000 years. An example 
language is named in brackets for each nonisolate stock. Only surviving stocks and branches are shown. 
Major branches are shown, but not their further subdivisions. Lower pails of trees in particular are highly 


split and the subsequent history of most branches is occasional binary splits. Another 
stock with an initial split at about 6000 years ago is Nakh-Daghestanian (Northeast 
Caucasian), which underwent an initial binary split followed by more elaboration of 
the eastern branch. 

Most of the trees have much less high-level structure. Kartvelian, whose descen- 
dants include Georgian, underwent its first split some 4000 to 4500 years ago, so the 
first millennium or two of its traceable existence was as a single language, shown as a 
straight vertical line. Similar structures with relatively late branching are found in 
Chukchi-Kamchatkan, Eskimo-Aleut, and Abkhaz-Circassian (Northwest Cauca- 
sian). An extreme version of this kind of history is the family tree of Basque or 
Yukagir or Nivkh or Ainu, in which there is no branching at all until the quite recent 
development of some dialect differentiation and the family tree is simply a straight 
line for most of the histoiy. (Again, the diagrams show only surviving branches; ex- 
tinct sisters are attested for Basque and Ket, and presumably all family trees without 
branching are the result of extinction of all but one daughter.) For a language like 
Zuni, with no significant dialectal divergence, the family tree is a straight line. Lan- 
guages with no branching, or only recent dialect branching, in their family trees are 
known as isolates. For both isolates and families with only a few millennia of branch- 
ing history, the fact that there are no demonstrable outside kin justifies extending the 
descent line back 6000 years, the assumption being that if there are demonstrable kin, 
the larger grouping is a stock and of approximately typical stock-like age, and if there 
are no demonstrable kin, then we know the structure of the family tree up to at least the 
typical stock lifespan. Isolates and small families with shallow branching dominate 
Figure 1 , and this is the situation worldwide: Fully one-third to one-half of the world's 
language stocks are isolates or near-isolate shallow families. (Again, isolate status is 
presumably caused by extinction of sisters rather than by failure to proliferate; if ac- 
tual rather than surviving branches could be shown, the diagrams in Figure 1 would 
probably all be extremely bushy.) 

For some of these stocks, deeper connections are sometimes posited. Indo- 
European and Uralic are thought by many to be related, and Turkic, Mongol, and Tun- 
gusic are widely believed to be related. Proof, however, has not been offered, 
whereas for true stocks the genetic connection can be proved. As nonlinguists are 
generally not familiar with the criteria for proving that a set of languages form a fam- 
ily, it may be helpfiji to review them here. There are two kinds of criteria: genetic 
markers and threshold frequencies of strongly resemblant words in a carefully con- 
trolled lexical list. Both kinds of criteria rest on sufficient frequency of sufficiently 
close resemblances within a consistently defined grammatical or lexical set of items. 

Genetic Markers 

A good genetic marker is any structural feature that more or less single-handedly 
suffices to prove genetic relatedness. Essentially, a genetic marker is a feature or set 
of features whose chances of independent occurrence are so infinitesimal, and whose 
likelihood of diffusion is so low, that if the feature is found to recur in more than one 
language the most parsimonious explanation is that it is inherited from a common an- 
cestor. Figure 2 shows the approximate pre- 1600 distribution of the widespread Al- 
gonquian language family, which extends from the high plains to most of the eastern 
seaboard. The first native American language family encountered by English colo- 
nists, it contributed to American English a good number of words such as skunk. 



^'\ R^>^ 


Figure 2. The range of the Algonquian family and the Yurok and Wiyot languages of California. The pro ■ 
nominal prefixes that prove their relatedness are: 

Proto- Wiyot Yurok 

Algonguian I II 

1 St person 














?we- 5; ?u- 

moose, sachem, wigwam, moccasin, and place names such as Manhattan. Massachu- 
setts, Connecticut, Passamaquoddy, Winnepesaukee, Punxsutawney, and Chicago. 
The foundations of an Algonquianist descriptive and comparative-historical tradition 
had been laid when Sapir (1913) proposed that the Algonquian family was distantly 
related, as one branch of a tree, to two languages of coastal northern California, Yurok 
and Wiyot. Given the vast jumble of languages and language families along the 
American Pacific coast (to be discussed below), the claim to have unearthed exactly 
two languages related to Algonquian was like finding two needles in a haystack. But 
Sapir's find was later confirmed by Haas (1958) and the theoretical groundwork of 
the discovery expounded by Goddard ( 1975). Also shown on Figure 2 is the evidence 
that made it possible to detect unerringly, and in short order, the relatedness of Yurok, 


Wiyot, and the Algonquian family, which make up what is now known as the Algic 
stock: The personal pronominal prefixes. (Yurok and Wiyot, incidentally, are not 
close sisters. They may each be independent branches of Algic, on a par with the Al- 
gonquian family, or they may have had a period of common existence followed by a 
long period of independent existence.) 

The pronominal prefix paradigm is the person agreement system for verbs and 
possessed nouns. The categories are first person (T, me, my'), second ('you, your'), 
third ('he, she, it; him, her; his, her, its'), and indefinite ('someone; someone's'), and 
the distinctive elements are the four different initial consonants *n-, *k-, *w-, and 
*m-. This four-member paradigm suffices to establish genetic relatedness among the 
languages exhibiting it, as seems to have been realized by Sapir and Haas in their 
demonstration of Algic linguistic unity and as is firmly realized now and noted in all 
commentaiy on the case. (The history of the question and the definitive explanation 
of method are given inGoddard 1975; as noted there, the Yurok Isg J- and indefinite 
b- can be shown by internal reconstruction to descend from *n- and *w- respectively.) 
The statistical grounds for regarding a four-consonant paradigm as diagnostic of ge- 
netic relatedness are explained in Nichols (1996, submitted). Briefly, the logic is as 
follows. There are some 6000 languages on Earth; hence in a random draw any one 
language has a one-in-six-thousand chance of turning up. Hence 1/6000, or anything 
on the order of 1/10,000 (or 0.0001), represents the probability ofoccuirence of any 
random language. A secure conventional threshold of statistical significance is 0.01, 
and the random language's probability can be multiplied by this to yield 0.000001 , or 
one in a million, a probability threshold which represents the likelihood o^ any par- 
ticular language turning up and can be taken as a statistical definition of the unique in- 
dividual language. In practical terms, this means that a linguistic subsystem with the 
tiny one-in-a-million or less chance of occurrence is something so unlikely to recur 
randomly that all tokens of such a system can confidently be assumed to descend from 
a single unique ancestor. 

The four-consonant system of Algic pronominals meets this threshold. Since the 
world average for consonant phonemes is about 25 (this figure is based on my 220- 
language worldwide sample), any given consonant has 0.04 (1/25) chance of occur- 
rence in any particular position in a randomly chosen word or form. The Algic pro- 
nominal system is a closed four-member opposition of person categories, and any one 
of the person categories has a 1/4 or 0.25 chance of occurrence in any position in the 
system. The Algic system with exactly those four consonants in exactly those four 
person categories (the full system of consonants in the respective paradigm positions 

attested in its entirety in each of the Algic branches) has an overall probability of 0.04 
X 0.04X 0.04 X 0.04 X 0.25 x 0.25 x 0.25 x 0.25 - 0.00000001 or one-in-a-hundred 

For this kind of computation it is important that the various categories be strictly 
defined. If the consonants are defined generically (for instance, if instead of requiring 
specifically n the comparison can use any voiced dental, or if instead of specifically k 
it can use any velar obstruent, etc.; the ranges of variation must be set up in advance), 
the probability increases somewhat (a generic consonant has 0.07 probability of oc- 
currence), though not enough to vitiate the Algic comparison.'' What is essential is 
that the four person categories constitute a strict paradigm or subparadigm in each of 
the language families, that they occur in a fixed order with respect to the four conso- 
nant markers, and that the full system be attested in its entirety. Section (1) below 
shows some hypothetical pronominal systems which could be compared to the Algic 
one, some of which match well and indicate that the languages in question are daugh- 


ters or sisters of Algic, and some of which do not match. Language (a) has the identi- 
cal system suffixed rather than prefixed, and language (b) has generic consonantal 
resemblances; both of these languages qualify as probable Algic kin. Languages (c) 
and (d) have the same four consonants but (partly or entirely) in different functions. 
Language (e) has three of the consonants in partly different functions and a different 
marker for one of the functions (the cross-linguistically very common zero marking 
of third person). Language (f) has two of the Algic consonants in the same two func- 
tions, but different markers for the other two; in this language, the Algic paradigm is 
attested only partially. Language (g) has a singular/plural opposition in the person 
categories, and the Algic markers are randomly distributed between singular and plu- 
ral. None of the non-qualifying languages are demonstrably Algic, but (f) and (g), 
with their first person n and second person k. are sufficiently suggestive of relatedness 
that a comparativist would take a second look at those languages in search of firmer 

( 1 ) The Algic pronominal system and some qualifying and nonqualifying matches 
to it. (a), (b), etc.: different languages. Sg.= singular, Pl.= plural. 





















































Sections (2)-(4) below show further examples of subparadigms or small para- 
digms that meet the threshold and suffice to demonstrate genetic relatedness. 
Sections (2)-(3) are from language families whose relatedness is amply evident. Both 
have generic resemblances of consonants and vowels. Section (4) was used by 
Greenberg (1960) to demonstrate genetic relatedness of Afroasiatic, the world's old- 
est proven genetic grouping. Greenberg's demonstration explicitly used the para- 
digm as sufficient evidence of relatedness, and I believe it was the first time that a 
genetic marker was explicitly used as sole evidence in an argument for genetic relat- 

(2) Partly suppletive third person forms of 'be' in Indo-European. 











sotu {i.e., *soNtu) (u = short w; N = generic nasal) 






asanzi {i.e., *esti, *asanti) 


(3) Germanic suppletive adjective 'good.' 

positive comparative 

English good better 

German gut besser 

Norwegian god bedre 

Gothic goths batiza 

(4) Gender-number suffixes in Afroasiatic determiners (following Greenberg 

Sing. PI. 
Masc. -n 

} -n 


Shared cognates, or lexical matches more generally, can be useful as genetic 
markers provided the resemblances are close enough in both phonology and meaning, 
numerous enough in a standard wordlist, and represented frequently enough in the set 
of languages under comparison. Rates of vocabulary loss and threshold frequencies 
were discussed above. Tests using frequencies of lexical sharings to demonstrate ge- 
netic relatedness have been designed by Ringe (1992 and later works; some fme- 
tuning of the mathematics is in progress), Oswalt ( 1 99 1 ), and Bender ( 1 969). A text- 
book presentation of Oswalt's test is given in Trask (1996:368, 372). 

Stock Density 

Recall that there are some 300 linguistic stocks on Earth, a figure which is still be- 
ing refined as undescribed languages receive description, and which may eventually 
be reduced to around 200 older lineages, but which is unlikely to be further reducible. 
Classificatory methodology and the genetic classification of most of the world's lan- 
guages are among the great accomplishments of scientific linguistics, and the next 
sections of this paper show that the genetic classification can be applied to solving 
nongenealogical problems as well. The world's stocks, since they are genetically dis- 
crete up to the time depth of the fade-out point, provide a ready-made worldwide lin- 
guistic sample for comparative work of all kinds. The work reported here samples 
stocks by taking one well-described language from each major (or initial) branch of 
each stock. (For isolates and near-isolate shallow families, of course, there is only 
one branch.) Such a sample should contain some 400 to 500 languages, but so far 
mine has only 220, partly because data gathering is not complete but mostly because 
there are many stocks for which not even one daughter language has a satisfactory 
published description. Figure 3 shows the sample in its current state. 

Figure 3 is also a map of genetic diversity of languages (as it was in precolonial 
times, in that only indigenous languages, including some now extinct, are used in the 



-n E 

re a 

H S 

o ■ = 

__ = o 

IT'S • 

OJ Q. 

Ob ry, -73 

re jj _u 

— i« re 


<L> i: i. 

O E 

u: < 


sample for the Americas, the Pacific, and Siberia). As the map shows, genetic diver- 
sity is not evenly distributed over the Earth: Linguistic stocks are densely bunched up 
in some areas and much more sparsely spread out in others. (This uneven distribution, 
incidentally, would be equally visible regardless of whether it was languages, shallow 
families like Romance or Slavic or Algonquian, deep families like Oceanic or Mayan 
or Iranian, or stocks that were sampled.) The reasons for the unevenness have been 
established in broad general terms (Austerlitz 1980; Birdsell 1953; Nichols 1990; 
Mace & Pagel 1995; Nichols 1997a): Density is favored by tropical latitude, parklike 
and savannah vegetation, mild and nonarid climate, low population density, and sim- 
ple social structure; it is disfavored by high latitude, continuous forest, grassland or 
desert, arid or highly seasonal climate, high population density, and complex social 
structure, especially state or empire. Stock density is the ratio of language stocks to 
square miles (or kilometers) of land (this metric was proposed by Austerlitz 1980). 
As will be discussed again below, wherever stock density is sparse, this is the conse- 
quence of the spread of one or a few languages or families. Note that there are no 
properly linguistic factors that account for stock density; just as there is nothing in the 
linguistic type or structure of particular languages that leads them to develop into ei- 
ther elaborate or minimal family trees, so likewise there is nothing in the type or struc- 
ture of languages that causes them to either bunch up or spread out. 

The Age of Human Language 

The first benefit of an exhaustive classification of the world's languages into strict 
stocks is the possibility of computing the age of human language. Recall from Figure 
1 that linguistic stocks have different shapes due to different rates and extents of pro- 
liferation, and from the discussion of Figure 3 that a rapid or slow rate of proliferation 
is not a 1 inguistic matter and not inborn in a language fam i ly but caused by geographic 
and economic factors. Let us now consider what I will call initial branching of lan- 
guage stocks: the number of distinct branches at the very top of the tree or close to it 
(when the tree is extended back to the approximately 6000-year limit of reconstructa- 
bility). Indo-European has ten known initial branches, all of which had diverged 
within about a thousand years of the initial dispersal.'' This history, however, is quite 
unusual. Several stocks separate initially into two branches {e.g., Uralic and North- 
east Caucasian in Figure 1), and the majority are either isolates (which have no 
branching) or shallow families (which have only a single branch for the first few mil- 
lennia of their traceable lives). This same situation obtains worldwide. Several years 
ago I surveyed all the language stocks of the Northern Hemisphere (for which com- 
parative work has been most thoroughly and consistently done) and found that the av- 
erage number of initial branches ranges from 1.4 to 1.6. [The exact figure depends on 
how certain questions of language classification are resolved, and on whether Indo- 
European is included. The multiple initial branching of Indo-European is so unusual 
that excluding it from the sample drives the average down appreciably. This survey is 
described in Nichols (1990).] 

Let us use the figure of 1.5 as the average number of initial branches of stocks. 
This figure, together with the approximately 6000-year age of stocks, makes it possi- 
ble to compute rates of linguistic divergence and, from that, estimate the age of a 
group of stocks which are assumed (even if only hypothetically) to descend from a 
single ancestor. This is done by dividing the number of stocks by 1 .5, dividing the re- 
sult by 1 .5, and so on until the result is less than 2; then counting the number of divi- 


sions performed and multiplying by 6000. This represents the number of stock life 
spans it takes to get back to approximately one ancestral stock. Table 1 shows two hy- 
pothetical applications of this procedure, and Table 2 shows the ages it gives for vari- 
ous numbers of stocks. 

This kind of calculation is very rough (because the branching rates and stock life- 
span are estimates, and because ages are even multiples of 6000) but usefiil. It gives a 
way of estimating dates of first settlement of areas, and those in turn can be used to 
raise hypotheses about the actual linguistic histories of areas. Some figures are shown 
in Table 3. (Criteria for determining the number of stocks in each area are discussed 
in Nichols 1997a; for Africa and northern Eurasia the higher figures are conservative 
counts of stocks as defined above, and the lower figures are less conservative because 
they count some plausible but unproven older-than-stock groupings. A full listing of 
the world's stocks and their representative languages in my sample will be available 
in a website now under design.) Consider Africa, where modem humanity probably 
arose some 1 00,000 years ago: The age of the African linguistic population, had it di- 
verged from a single ancestor, would be at most about 42,000 years. Now, the African 
languages are not thought to all diverge from a single ancestor unique to them; though 
quite possibly some of the African stocks form one or more deep groupings, they are 

Table 1. Computing the age ofa set of linguistic stocks. The average stock has 1.5 initial branches. Assum- 
ing a stock age of 6000 years, div ide the number of stocks by 1 .5 (the average number of initial branches per 
stock) and repeat until the result is less than 2. Count the number ofsteps and multiply by 6000. (Decimals 
rounded to two places. ) The result is the time it would take to derive this many stocks from a single ancestor 
(//they share a single ancestor, which the calculation does not establish). 

Number of stocks in the set: 

n 30 



12^ 1.5 = 8 

30^1.5 = 20 


8 j 1.5 



5.33^ 1.5 

13.33^ 1.5 


3.56 -f 1.5 

8.89^ 1.5 


2.37^ 1.5 

5.93^ 1.5 


3.95^ 1.5 



5 X 6000 = 30,000 

7 X 6000 = 42,000 

Table 2. Estimated ages for various numbers of stocks. Calculations as in Table 1 . 

Number stocks 

Age in years 

Number stocks 

Age in years 


























ings, they are not all to be subsumed under anything younger than the ancestor of all 
human languages. That is, the age of the African linguistic population can be pre- 
sumed on archeological grounds to be at least 100,000 years, while the age computed 
from the stock divergence half-life is 42,000 years. The great discrepancy between 
the linguistic and archeological ages of Africa indicates that there has been much ex- 
tinction of linguistic diversity there, and/or loss by emigration. The same is true of 
northern Eurasia, which (at least at its southern periphery) has been inhabited by ana- 
tomically modem humans nearly as long as Africa, but where the linguistic age is 
even less than for Africa. For both of these continents the geography — high ratio of 
interior to coast, much of the interior either arid (Africa) or continental and seasonally 
cold (Eurasia) — disfavors stock diversity {cf. again Figure 3), and there has been 
much extinction as a result of the language spreads favored by this geography. 

A somewhat similar picture obtains for Australia, where the number of indigenous 
language stocks yields a linguistic age less than the archeological age. (For early 
dates in Australia see Roberts e/fl/. 1990, Roberts & Jones 1994.) On the other hand, 
the linguistic age of New Guinea, and of Australia-New Guinea together, is 60,000 
years, close to the archeological date of 50,000+ years. (Australia and New Guinea 
were a single landmass during the last glaciation, when water was locked up in gla- 
ciers and sea levels were low, exposing the continental shelf Hence their early hu- 
man settlement histoiy is shared.) That Australia has a distinctly younger linguistic 
age than New Guinea is a consequence of Australia's geography with its sizable arid 
interior which fosters language spreads and extinctions. Still, even for New Guinea 
the linguistic age is misleading: Multiple colonization is known to have occurred, 
probably from the very beginning (see e.g.. White 1996; Nichols 1997b), hence the 
languages cannot be presumed to have descended from a single unique ancestor. If 
the modem linguistic diversity is what would have descended from only one initial 
colonization, then even here some extinction has evidently taken place. 

The case of the Americas is different. Here the linguistic age is much greater than 
the archeological age. As will be discussed below, the discrepancy is due in part to 
multiple colonization, but even when that is factored in the linguistic age is greater 
than the known archeological age. 

Thus comparing linguistic and archeological ages of continents can be infomia- 
tive in various ways, but it cannot straightforwardly indicate actual colonization 
dates; linguistic dates are most useful when they are either much younger than ar- 
cheological dates and therefore give clear evidence of extinction of linguistic lines (as 
in Africa) or much older than archeological dates, thereby indicating multiple coloni- 
zation (as in the Americas). 

This much holds for linguistic ages of continent-sized populations of languages, 
where dates of first colonization and frequency of colonizations complicate the pic- 
ture. But what of the age of human language in general? Table 3 shows that the lin- 
guistic age of the entire world population of linguistic stocks is only 72,000 to 78,000 
years — much too young, given that modem language is probably at least as old as 
anatomically modem humanity. Here, as with Africa, the discrepancy is due to pat- 
tems of extinction. A more accurate estimate of the world's linguistic age is shown in 
the last two entries in Table 3, where the number of stocks is the number that would be 
expected if the entire traditionally inhabited world were populated with language 
stocks at the density attested in the Americas or — most accurately — New Guinea, 
which has the world's highest stock density. The outcome is that to populate the tra- 
ditionally inhabited world with language families at the density that can be reached 
when circumstances are favorable — that of New Guinea — would take some 


Number stocks 

Africa (low) 


Africa (high) 


Northern Eurasia 


New Guinea 




New Guinea and Australia 


North America 


North and Central America 


Entire World 



Tables. Estimated ages forsome actual linguisticpopulations. Calculations as in Ta ble 1 . Descentofeach 
population from a single ancestor is assumed for the calculation, but is not supported by linguistic evidence 
or (with the possible exception of the entire world) generally believed by linguists. Ar chaeological age is 
the oldest generally accepted dated site or human remains; as the criteria are conservative and older sites 
may exist but not have been found, aichaeological age is a minimum age of habitation. 

Linguistic Age in Ky Archaeological age in Ky 

36 100 

42 100 

36 90"^ 

60 50+ 

36 50+ 

60 50+ 

48 13 

54 13 

66 13 

World (low) 200 72 

World (high) 300 78 

World at New World density 466 84 

World at New Guinea density ^12,000 132 

132,000 years. This is then the linguistic age of the world. If the world's language 
stocks descend from a single ancestor, then that ancestor began to diversify at least 
132,000 years ago.** This computation does not tell us whether the world's languages 
descend from a single ultimate ancestor. But if they do, then that ancestor language 
began to disperse well before the anatomically modern physical type began to spread. 

Ages for Some Possible Ancient Lineages 

Although the maximum age for demonstrated stocks is about 6000 years, there are 
a few groupings of stocks that are widely regarded as probable sisters whose genetic 
unity is too ancient to be amenable to standard reconstruction. The clearest example 
is Afroasiatic, a group of African stocks whose genetic unity is proved by several dif- 
ferent genetic markers (see Greenberg 1960, 1963; Newman 1980) but for which 
regular correspondences and reconstruction cannot be demonstrated to the satisfac- 
tion of the field. [Two recent comprehensive proposals for correspondences and re- 
constructions are Orel & Stolbova (1995) and Ehret (1995), but neither has won 
general acceptance.] The surviving branches of Afroasiatic number from four to 
seven depending on how questions of subgrouping are resolved: Berber, whose inter- 
nal differentiation is very recent; Chadic, an old family; Semitic, a very old family of 
stock-like age; Cushitic, now often divided into two or three stock-like branches 
(northern Cushitic, southern Cushitic, and isolate Beja); and probably Omotic. (An- 
cient Egyptian and its descendant Coptic represent another branch of Afroasiatic 


which, however, has not survived and therefore does not figure in the age computa- 
tion.) Various proposals unite Berber with one more branches (Chadic, Semitic, 
Cushitic) (proposals and overviews of Afroasiatic include Bender 1975, 1997;Diak- 
onoff 1988; Newman 1980; Greenberg 1963). 

By the half-life computation, depending on the branching structure assumed, the 
age of Afroasiatic is from 12,000 years for four branches to 24,000 years for seven. 
The calculation assumes that Afroasiatic has a normal branching rate and a typical 
higher-level branching structure (though of course it cannot identify the actual 
branches and propose a specific subgrouping): If there are seven stock-level branches 
now, 6000 years ago there were 4.67, i.e., under five, and 1 2,000 years ago there were 
3. 1 1 , /.e., about three; if there are four branches now, 6000 years ago there were 2.67 
and 12,000 years ago there were 1.78, i.e., under two. Because the calculation can 
only measure age in 6000-year increments, any adjustment in branching structure 
(such as division of Cushitic or union of Berber with another branch) has a drastic ef- 
fect on the computed age for a relatively small set of stocks like this one. Thus the re- 
sults of linguistic comparative reconstruction bear heavily on the computable age of 
Afroasiatic. Positing fewer subgroups that unite more stocks lowers the age; positing 
numerous binary splits raises the age. Uniting any of the branches into true stocks 
will lower the age. The strongest case for a great age for Afroasiatic would come from 
a demonstration that it has a complex hierarchical internal branching structure; this 
would show that the computation of a regular half-life is plausible. 

This discussion of Afroasiatic shows that the half-life with its large increments is a 
very crude measure of age for relatively recent events, but that combined with close 
comparative work on branching substructure its accuracy can be improved. In addi- 
tion, note that even with questions of higher branching structure unresolved it is clear 
that the age of Afroasiatic is likely to be closer to 12,000 years than to 6000, and per- 
haps over 12,000. Thisisby far the oldest securely demonstrated linguistic lineage on 
Earth. (Recall, however, that Afroasiatic has so far eluded reconstruction, that there 
is much debate about its higher-level branching structure, and that it is by far the 
clearest case of an older-than-stock grouping. These considerations justify using 
6000 years as the best estimate of the average fade-out point.) Some proposed but un- 
proven groupings of comparable age are discussed below. 

Monogenesis of Language? 

Do the world's languages in fact ultimately descend from a single ancestor, a 
Proto-World as it is often called? Or were there several ancestral languages? The lat- 
ter situation — polygenesis — would have obtained if premodem language evolved 
into modem language separately more than once in more than one distinct population, 
or if the evolution of modem language took place gradually in a human population 
large enough to consist of several different language communities. These are two 
very different scenarios: The first implies that the evolutionary process occuired 
more than once, each time in a single speech community, while the second implies 
that the evolutionary process occurred only once, but affected a population of several 
speech communities simultaneously. Monogenesis — one-time-only development 
of modem language out of just one ancestral premodem language — implies either a 
unique gradual evolution of just one premodem language into a modem language, or 
sudden appearance of modem language in one individual or household or clan with a 
sharp discontinuity from whatever preceded and from whatever was in use next door. 


These are very different scenarios: The first involves a unique occurrence of a com- 
mon sort of gradual evolutionary change, while the second implies a fairly drastic sal- 
tatory change. Thus we have, in all, four possible scenarios. The two polygenetic 
scenarios both imply multiple ancestors to modem languages, and the two monoge- 
netic scenarios both imply a single ancestor. 

There is of course no way of establishing the number of ultimate ancestral lan- 
guages by tracing descent, but a rough estimate of the number — or at least a well- 
founded conjecture as to whether the ancestors were one or many — can be derived 
from linguistic geography and the study of contact-induced language change. Tropi- 
cal foraging societies tend to be small in favorable geographical and climatic condi- 
tions, averaging at most about 500 individuals in what I will call the ethnolingiiistic 
unit, the tribe-like group speaking a single language. (For the figure of 500 see Bird- 
sell 1953; for comparison linguistic figures see Nichols 1990. The smallest groups 
are found where cultures are simplest, population density is thin, and the environment 
is tropical and nonarid, factors that obtained for early humans.) Therefore, an esti- 
mate of the total population size for earliest//omo sapiens will make it possible to es- 
timate an approximate number of small to average-sized ethnolinguistic units. The 
following argument is something of a thought experiment using this and other plausi- 
ble assumptions about paleosociolinguistics. 

Though I know of no precise figures on the matter, in a society as small as 500 in- 
dividuals some intermarriage with neighboring ethnolinguistic units is likely, espe- 
cially if there are large-scale internal exogamous groupings such as clans, sections, or 
moieties that restrict one's choice of marriage partners beyond the usual close kinship 
constraints. In all cases that I am aware of where ethnolinguistic groups are very 
small, intermarriage with neighboring groups is common, even systematic. Intermar- 
riage often means bilingualism — fairly uniform society-wide bilingualism if there 
are standing intermarriage patterns between two societies, and more diverse 
household-by-household bilingualism or multilingualism where intermarriage pat- 
terns involve more societies and are more varied. Now, systematic and long-term bi- 
lingualism or multilingualism can entail convergence between the languages 
involved (for convergence see Thomason & Kaufman 1 988; case studies of multilin- 
gualism in small societies include Ross 1996; Heath 1978). Thus, from small group 
size we can infer likely bilingualism and multilingualism, and from that we can infer 
likely convergence at least among neighboring languages and perhaps over the entire 
early human range, depending on its size and layout. 

There are several different hypotheses about the population size, location, and ter- 
ritorial range of the ancestral modern humans. On the one most widely held, modem 
humans emerged over 100,000 years ago in the riverine and lacustrine environment 
of eastern Africa, spreading into the Old World tropics and eastern Mediterranean 
area sometime thereafter, perhaps around 80,000 years ago and certainly by 50,000 
years ago. Let us now combine this scenario with what is known about linguistic ge- 
ography and language contact. The range within which our ancestors evolved was 
roughly linear, somewhat over 2000 miles long and a few hundred miles wide (from 
the vicinity of Lake Turkana down the Rift Valley and continuing to southernmost 
Africa). These are dimensions comparable to New Guinea, which harbors great di- 
versity of language families and has some distinct intemal linguistic subareas as well 
as a certain amount of overall areality in features which spread easily, such as word 
order and sound systems. 

There is good reason to believe that modem language evolved gradually within a 
sizable population, rather than abruptly in a small advanced social circle. The modem 


brain and speech apparatus appear to have evolved slowly, gradually, and relatively 
uniformly over the entire species. Modem language depends for its proper transmis- 
sion on both the inherited language faculty and proper exposure at the right age, and 
the set of inherited apparatus, inherited learning capacity, and actual learning could 
not arise spontaneously and abruptly. In addition, abrupt appearance of full-blown 
modem language in one individual or family in a society speaking a more primitive 
form of language would have been an evolutionary saltation that would have frac- 
tured the society and compromised its viability. For all of these reasons it is safe to as- 
sume that at no time during the long and gradual emergence of modem language were 
there marked differences within the species as to the degree of modemity of language. 

Estimates of the size of the earliest modem human population at about 100,000 
years ago range from a few tens of thousands to a million or more. '" Even the smallest 
of these figures would allow for a good number of distinct ethnolinguistic groups, and 
the territorial range over which early modem humans are found virtually guarantees 
considerable linguistic diversity. The closest historically attested analogs to this 
combination of low population density, riverine-lacustrine environment with savan- 
nah interface, tropical climate, and foraging economy are probably northem Austra- 
lia and (albeit horticultural) lowland New Guinea and parts of northem South 
America. These areas all accommodate considerable numbers of language families. 

On the other hand, the probable small size of the early ethnolinguistic groups sug- 
gests intermarriage and multilingualism and hence convergence at least locally; here, 
too, the analogs I am familiar with (Australia, New Guinea) are marked by a good deal 
of multilingualism and some degree of local to regional linguistic convergence. As 
the range was more or less linear, convergence pattems would have been chainlike, 
with the languages at the far northem and southern ends of the chain greatly distinct 
from each other in structure as well as descent." 

Taking into account all of the factors discussed here, it is likely that the human lin- 
guistic population prior to the expansion out of Africa consisted of perhaps a hundred 
distinct languages falling into perhaps ten distinct stock-like genetic lineages and 
showing several distinct areal patterns of typological resemblance and probably a 
good deal of typological diversity over the entire range.'" To populate the tradition- 
ally inhabited world at New Guinea densities from an ancestral population of 10 
stocks would require 108,000 years, a time frame reaching back to the eve of the dis- 
persal of anatomically modern humans. 

Thus modem language in its earliest existence was probably a population of lan- 
guages with internal genetic and structural diversity and some local and general areal- 
ity. When the spread of modem humans out of Africa began, the northern part of the 
ancestral linguistic range must have been a disproportionate contributor of emigrant 
linguistic populations, but nonetheless over the millennia the languages that spread 
out represented a good swath of the range of ancestral linguistic diversity, both ge- 
netic and stmctural. 

Homo erectus appeared nearly two million years ago and soon thereafter spread 
over Africa and southem Europe and Asia, in all probability carrying primitive lan- 
guages which continued to diversify and diverge for over a million years. Presumably 
Africa continued to be a center of slow biological and linguistic spread thereafter; the 
distinctive physiology of modem humans, following the same origin and trajectory 
much later, made the spread trajectory visible. Modern language, and descent linea- 
ges of languages, must have followed the same trajectory, but the spread of modem 
language need not have coincided exactly with the spread of modem physiology and, 
as argued above, may have preceded it. 


Thus the east African linguistic population within which modem language 
evolved is likely to have been a genetic and structural subset of world linguistic diver- 
sity of its time, and this means that modern language, though polygenetic, is not maxi- 
mally polygenetic: Most of the linguistic lineages of premodern humanity died out 
when modem humanity and modem language spread. Whether the range of modem 
stmctural diversity of languages is restricted as a result of this ancient skewing is un- 
known and probably unknowable. What is clear is that when modem humans spread 
across southern Eurasia and beyond between about 70,000 and 50,000 years ago the 
languages of the previous inhabitants at the periphery had been diversifying for at 
least the few hundred thousand years of archaic modem humans and perhaps the en- 
tire nearly two-million-year lifespan of Homo erectiis. This was the greatest degree of 
genetic diversity that has ever existed in human and hominid language, and it went ex- 
tinct without a trace. 

The Spread of Languages Over the World 

Sample and Historical Markers 

This section will summarize research published in detail elsewhere (Nichols 
1992, 1995a, 1997a; Nichols & Peterson 1996) and show that the relative frequencies 
of diagnostic stmctural features in large areally-based sets of languages can reveal 
fundamental affinities between some of these populations, and that these in tum point 
to shared geographical origins. This approach bypasses descent entirely and instead 
traces nongenealogical affinities between large geographically-based groupings of 
language families. It cannot trace the origins of individual families very well, but it 
can trace the settlement of continents, explain the worldwide geographic distribution 
of language families, and reach very far back into prehistory. 

For purposes of interarea comparison, the sample is divided into 1 8 subcontinent- 
sized areas: Southem Africa, northem Africa, the Caucasus and the languages of an- 
cient Mesopotamia, Europe, inner Asia, northeastem Asia (= eastem Siberia, 
roughly), southeast Asia, coastal New Guinea and Melanesia, interior and southem 
New Guinea, northem coastal Australia, southem and interior Australia, the Ameri- 
can Pacific northwest (from Alaska to Oregon), California, the Great Basin and 
Plains, eastem North America, Mesoamerica, westem South America, and eastem 
South America. These areas are defined in purely geographical temis and set up so as 
to contain roughly equal numbers of sample languages. Various stmctural features 
are surveyed in the sample languages, and the 18 areas are compared in terms of the 
relative frequencies of those features in the area's languages. The 1 8 sample areas are 
shown in Figure 4. 

The database of sample languages and sampled stmctural features was designed in 
order to seek out statistically significant differences in features across areas, differ- 
ences that would point to geographical origins for populations of languages much as 
frequencies of blood types, mitochondrial DNA lineages, e/c, in human populations 
point to geographical origins for those populations. (In both cases it is individual 
lineages within the population, and not the entire population as such, that have geo- 
graphical origins, but reaching statistical significance requires comparing popula- 
tions.) The stmctural features chosen for this first comparison were ones that 
observation had suggested were relatively persistent in language families, of rela- 
tively low frequency worldwide, not readily diffused, and not likely to arise spontane- 




ously. The presence of such a feature in a language or language population is likely to 
be due to inheritance or direct, close contact and unlikely to be due to universals. indi- 
rect or superficial contact, or accident. If a usably large set of such features can be 
found, statistically significant differences in their frequencies from area to area are 
likely to point to shared origins or shared population history. Such features can be 
called historical markers, and a long-term goal of historically-oriented comparative 
grammar is to amass as large as possible a set of reliable historical markers. Fortu- 
nately, the dozen or so prospective historical markers pursued in this survey have all 
proven useful in their cross-linguistic statistical patterning, and comparative ty- 
pological work (Nichols 1992, 1995b) has shown that they are reasonably independ- 
ent of each other, persistent in language families, not prone to spontaneous 
innovation, and reliably extractable from published grammatical descriptions. 

Sections (5)-(7) below illustrate some of the historical markers used in the work 
reported here. (It should be cautioned that the inventory of markers is biased toward 
morphological ones.) Sections (5)-(6) are examples from Ingush, a language of the 
Northeast Caucasian family spoken in southern Russia, illustrating ergativity, the 
identical grammatical coding of subject of intransitive verb and direct object of in- 
transitive, while the subject of a transitive verb is coded differently. In (5), the verb 
'gave' is transitive; the subject Miiusaa-z is in the ergative case, an oblique case spe- 
cialized for marking subjects of transitive verbs; the direct object (Xxc/za 'money' is in 
the nominative case, the basic case or citation form, which has no suffix; and the verb 
agrees in gender (shown by its prefix) with the direct object. In (6) the verb is intransi- 
tive; Miaisaa is subject and in the nominative case, and the verb agrees with it in gen- 
der. Ergativity is a low-frequency feature worldwide (only some 25% of the 
languages in the 220-language sample exhibit it to some degree), and Figure 5 shows 
that its distribution is skewed: It is very frequent in the Caucasus and Australia, fairly 
well attested in highland New Guinea and central Eurasia, and rare elsewhere. These 
differences are statistically significant, and therefore cannot be assumed to be due to 
random chance or language universals. 

(5) Ingush Muusaa-z shii voV-a axcha d-alar 

Musa (V)-ERG his son-DAT money (D) D-gave 
'Musa gave his son money' 

(Case name abbreviations: ERGative, DATive; D = gender prefix; 
(V), (D) = gender of nouns.) 

(6) Ingusb Muusaa aara - v-ealar 

Musa (V) out V-went 
'Musa went out' 

Ingush is a dependent-marking language in which grammatical relations such as 
subject, object, and possessor are marked primarily by cases on the nouns and pro- 
nouns. In (7), in contrast, taken from Abkhaz of the Northwest Caucasian family 
(spoken in western Georgia), nouns bear no cases, the verb agrees elaborately with 
the subject, indirect object, and direct object, and in general heads of constituents 
agree with their dependents; Abkhaz is a head-marking language. Head-marking lan- 
guages are not frequent worldwide, except in the Americas where they abound (Fig- 
ure 6). Head marking is another historical marker. 





c — 

^ aii 

S3 .S 

-I ^ 

£ O 


(7) Abkhaz a - xaca a - pPi°8s a - sq°'a - la - y - te - yt' 

the man the woman the book it to her he give PAST 
'The man gave the woman a book' 

Identical stems in singular and plural forms of personal pronouns provide another 
historical marker. Here first person pronouns have been surveyed as the single best 
indicator of this tendency. Section (8) shows identical stems in two languages of 
California. In Yawlumni (Yawelmani Yokuts), the plural forms are derived from the 
singular forms by suffixation. In Wintu, there is no singular/plural distinction but 
only a person distinction, so it can be said that singular and plural forms are com- 
pletely identical. Identical singular/plural stems are uncommon worldwide yet fre- 
quent in a few areas (Figure 7). 

Section (8) also illustrates personal pronoun systems with first persons and sec- 
ond person w. Such systems again are rare worldwide but common along an extended 
stretch of Pacific coast in the Americas (Figure 8; Nichols & Peterson 1996). 

(8) Personal pronouns in selected native American languages, sg. = singular, pi .= 







sg. - pi. 







en, e- 

nun, nuku 





min, mi- 

man. matu 

The rest of this section summarizes recent work that makes use of these historical 
markers in order to trace the spread of languages around the Pacific Rim. The modem 
distribution of historical markers is the only evidence we have of the ancient spread of 
languages from an origin probably in the vicinity of southeastern Asia to New 
Guinea, Australia, and Oceania on the one hand and to northern Asia, Beringia, and 
the New World on the other, and for the chronology of those spreads. 

Only the general fact of language spread can be inferred from the distribution of 
the markers. Just how a language has spread — by migration of the speaker popula- 
tion, demographic expansion of the speaking population, shift of the speech commu- 
nity to the language of a prestigious or powerful neighbor, or some combination of 
these — cannot be inferred. Nor can it be inferred whether a given language has in- 
herited a given historical marker or acquired it through close direct contact. All that 
can be assumed is that statistically significant sharing of historical markers between 
languages or language families points to a common geographical origin of the lan- 
guages or families, and statistically significant sharing of historical markers between 
two or more populations points to a common geographical origin of some component 
of the populations. The sociolinguistics of the transmission scenario and the spreads 
cannot be reconstructed from the distribution of the markers, but this is no obstacle to 
reconstructing the abstract fact of language spread and the trajectories and chronolo- 
gies of spread. 

The method used here has been explicated previously (Nichols 1992: Ch. 6, 
1995a, 1997b, c; Nichols & Peterson 1996). Fifteen or more historical markers are 
traced across the sample languages and their relative frequencies (languages showing 







the feature as a percent of the languages of the area) are counted in each of the 1 8 large 
areas listed above (or some similar geographically-based breakdown of a dozen or 
more areas). They are also plotted on a map. Any skewings in their cross-area! fre- 
quencies that are statistically significant are taken to be non-accidental and indicators 
of a common origin. An origin and spread scenario are reconstructed to the extent 
possible by comparing the linguistic distribution to what is known of prehistory from 
archeological and other sources. 

Worldwide Longitudinal Raniving of Areas 

As tabulated in Nichols (1997c), for each of the historical markers the 18 areas 
were ranked by frequency of the marker in the area, then either the rank steps or the 
distance in steps from the first-ranked area was totaled for each of the areas. '^ The re- 

Table 4. Sum of rankings of areas, based on frequencies of historical markers in them. Lineariz ation is 
schematic; a space means a relatively large distance between areas. An asterisk (* ) marks ar eas that are out 
of place, and the notes to the right comment on this. 


Europe is at the far western periphery of the 


North Africa 
South .A.frica 
♦Australian interior 
Noitheast Asian coast 
*New Guinea interior 

Inner Asia 

* Australia coast 

Africa is western and peripheral 
Africa is western and peripheral 
Interior Australia is in the western Old World 

Interior New Guinea is in the western Old 

Coastal Australia is in Asia 

Southeast Asia 

Eastern North America 
Basin - Plains 
Western South America 
Eastern South America 
Alaska - Oregon 
*New Guinea coast 


Coastal New Guinea is in western America 

(With Mesoamerica) at the far eastern 


suits are shown in Table 4: The relative frequencies of historical markers worldwide 
correspond rather well to geographical longitude, with Europe and Africa at one ex- 
treme (which can be considered the western edge) and the Americas at the other 
(which can be considered the eastern edge). This basically linear distribution, run- 
ning from what can be called a western pole in the vicinity of Africa to an eastern pole 
in the Americas, is a schematic geography of the traditionally inhabited world from 
the initial dispersals of modem humans and modem languages up to the end of the 
first millennium AD, when the Viking era brought the first known transgression of the 
westem pole and the first known colonization of the Americas from the western Old 

Variants of this computation, using a different inventory of historical markers, dif- 
ferent scales of areal breakdown, a somewhat different inventory of sample lan- 
guages, and/or (as mentioned in Note 13) different bellwether areas for determining 
whether rank sorts would be ascending or descending, yield essentially the same re- 
sults (the main difference between the variants is some jockeying for last place among 
the American subcontinents). The results are thus reasonably robust. The strongest 
determinant of the relative frequencies of stmctural features in language populations 
is longitude, and the cline of frequency rankings of areas is also an approximate sche- 
matic linear map of the premodern distribution of our species and our languages. The 
salient departures from literal geographical longitude in Table 4 — the westward dis- 
placement of Australia and interior New Guinea to westem Eurasia and the eastward 
displacement of coastal New Guinea to the southem Americas — are discussed im- 
mediately below. 

Coastal Spread Around the Pacific Rim 

Based on their frequencies of historical markers, the language areas around the 
Pacific fall into three large groups which, although each is spread over several conti- 
nents and two hemispheres, exhibit a degree of internal cohesion and extemal conti- 
nuity that suggests a single historical origin for each of the three and a regular 
chronological continuity between them. I will call these larger areas provinces. The 
most clearly demarcated province can be called the Pacific Rim province, as it clus- 
ters at or near the coastline all around the Pacific. Figures 7 and 8 above, and Figure 
9-12 below, show the diagnostic historical markers of the Pacific Rim province: iden- 
tical singular and plural pronominal stems, n : m personal pronoun roots, numeral 
classifiers, verb-initial word order, tones, and possessive classification. These 
markers cluster around the Pacific coast of New Guinea (but not Australia), Asia, 
North America, and Southem America. Numeral classifiers, n : m pronoun roots, and 
possessive classification are found almost exclusively around and near the Pacific 
Rim. Tones and verb-initial order are well represented in the westem Old World, but 
in Asia, Australasia, and the New World they again cluster near the coast. Identical 
singular and plural pronoun stems and n : m pronoun systems have a sparse coastally 
oriented distribution in Asia and Australasia but a more abundant coastally oriented 
distribution in the Americas, perhaps suggesting founder effects. Taken together, the 
five markers show continuity and an overall coastal orientation almost entirely 
around the Pacific Rim. 

The next province can be called the Pacific Hinterland. It is not clearly distinct 
from the Pacific Rim province, but is marked by a few features that are common in the 
Rim province and also extend farther inland and to northern Australia. These fea- 
tures are head-marking moiphology, shown in Figure 6, and gender or other agree- 










5 -= 






merit classes in nouns, shown in Figure 13. Three other features appear to have this 
distribution, but they have not yet been fully surv eyed: Reduplicated plurals; exten- 
sive prefixation; and causativization as a regular derivational process in the verbal 

More distinct in its linguistic properties, but more diffuse geographically, is the 
Pacific Interior province consisting of interior Australia, interior New Guinea, east- 
em South America, and to some extent eastern North America. It is distinguished by 
ergativity (Figure 5). rarity of head-marking morphology and frequency of 
dependent-marking morphology (Figure 14), systematic marking of singular/plural 
(or singular/dual^plural) oppositions on nouns, minimal consonant systems (often 
limited to a single manner of articulation), and high frequency of derived intransitiv- 
ity in the verbal lexicon (these last three have not been fully surveyed and are not 
mapped here). 

The structural affinities between interior Australia, interior New Guinea, and the 
eastern New World explain some of the departures from literal geographical longi- 
tude in the linear typological geography of Table 4. The interiors of Australia and 
New Guinea are closer to each other in Table 4 than either is to its own coast, and both 
are displaced far to the west, a distribution that reflects both the unity of the Interior 
province and the fact that many of its distinctive markers are also common in western 
Eurasia and in Africa. Coastal New Guinea belongs to the Pacific Rim pro\ ince and 
is drawn far to the east in the linear map. ending up as a neighbor of the American Pa- 
cific coastal areas. The northern coast of Australia is part of the Pacific hinterland and 
lies well to the east of the Australian interior but well to the west of the Pacific Rim. 

The maps in Figures 5-14 show that all three provinces are cut off. as it were, in 
the north; the Eskimo-Aleut. Chukchi-Kamchatkan. and Tungusic language families 
that dominate northern Siberia find most of their typological affinities to the west, 
among the languages of central Eurasia, and their considerable weight in the sample 
draws the Northeast Asian sample area somewhat to the west. 

These distributions can point to geographical origins of languages and to the his- 
tory of linguistic colonization around the Pacific. Some basic assumptions need to be 
made about Pacific colonization. The human populations that colonized the lands 
around the Pacific can be assumed to have emanated ultimately from eastern Asia: 
People moved from mainland southeast Asia through insular southeast Asia to colo- 
nize Australasia some 50,000 years ago, and from coastal and or interior northern 
Asia via Beringia to colonize the Americas. (This assumption seems to be the re- 
ceived view in archeology and human genetics.) The linguistic populations of the 
colonized areas can also be presumed to derive from eastern Asia; this much can be 
assumed even without knowing the sociolinguistic and historical specifics of the laiv 
guage spreads. I also assume that any linguistic population in the right part of eastern 
Asia would have been positioned to see some of its descendants move northward and 
ultimately into the Americas and others move southward and ultimately into Austra- 

On these assumptions, the following interpretation of historical marker distribu- 
tions can be given. The Interior province is dominated by descendants of a very early 
wave of colonization, a wave that began in some critically positioned part of eastern 
Asia and is now best represented in the most distant and presumably last-reached 
parts of very separate areas: the east, south, and interior of Australia, the interior of 
New Guinea, lowland and eastern South America, and to a lesser extent eastern North 
America. In both Australasia and the New World, this province includes a great range 
of climatic and geographical conditions, settlement of which required a variet>' of 



E S 

G ~, 

3 ^ 





a an 

50 CO 

c E 

ft- c 

2 O 

00 <u 

.S a. 


secondary adaptations on the part of the colonizers (whose entry points were the 
northwest coast and offshore islands of then-united Australia-New Guinea, and of 
Alaska). The longer the residence in the colonized area, the greater the chance of 
spread to the distant points and extreme environments; hence descendants of earliest 
immigrants are disproportionately represented in these far reaches. The movement of 
first-wave languages into the colonized continents probably involved a relatively 
high proportion of language spread by human migration, as the lands were minimally 
inhabited or uninhabited at first. 

The Pacific Hinterland province is a slightly expanded coastally oriented area. It 
suggests a later colonization impulse which was coastal and retained a primarily 
coastal orientation throughout a sizable spread along the northern coasts of Australia 
and New Guinea and the entire Pacific region of North, Central, and South America. 
This is a more recent stratum which has not penetrated far into the interior of Austra- 
lia, New Guinea, or the Americas. The Pacific Rim province is still more recent; it is 
more strictly coastal, geographically more restricted (not reaching Australia at all), 
and sharply delimited by the feature distributions on the maps in Figures 7-12. Its ge- 
ography and the sharpness of its linguistic profile mark it as a recently formed area. In 
the case of both the Pacific Hinterland and the Pacific Rim provinces, the coastal ori- 
entation was retained perhaps because coastally adapted immigrants remained 
coastal and carried their languages with them, or perhaps because immigrants entered 
a coastal cultural and linguistic area that favored difflision and language spread 
through shift. 

Finally, after the Pacific Rim population had entered the Americas, Proto- 
Eskimo-Aleut and Proto-Chukchi-Kamchatkan entered the coastal sphere in the 
north and spread, cutting off the continuity of the Asian and American Pacific Rim 
populations and representing the first appearance of the interior Eurasian linguistic 
type in America (Figure 15). 

The Ages of the Circum-Pacific Spreads 

A reliable though approximate relative and absolute chronology can be worked 
out for the Pacific Rim province. On the near side it is bounded in the Americas by the 
spread of the Eskimo-Aleut family in the north. The Eskimo-Aleut language family 
lacks the historical markers of the Pacific Rim province, and its structural type is 
strongly reminiscent of the languages of interior Siberia and central Asia: It is erga- 
tive, suffixing, case-using, consistently distinguishes number in nouns and pronouns, 
and lacks tones and numeral classifiers. The family is probably some 4000 to 5000 
years in age (Woodbury 1984) and almost certainly dispersed in the vicinity of south- 
western Alaska where its Aleut and Eskimoan branches now meet, but linguistic 
sources are of the unanimous opinion that it is a relatively recent entrant to the New 
World from Siberia. Similar to Eskimo-Aleut in type is the Chukchi-Kamchatkan 
family, comparable in age to Eskimo-Aleut and indigenous to coastal Siberia, origi- 
nating probably between the bases of the Chukotkan and Kamchatkan peninsulas. 
(For this family and a map of the Beringian languages see Krauss 1988. For historical 
connections ofEskimo-Aleut and Chukchi-Kamchatkan see Fortescue 1998.) Thus it 
appears that as of about 5000 years ago the Pacific Rim province had been separated 
in the north by an entering wedge of languages of interior Eurasian type, led by 
Eskimo-Aleut and followed by Chukchi-Kamchatkan. 

In Australasia there is a clear date for the beginning of the Pacific Rim stratum 
(Nichols 1997b). The two earlier provinces are found in both Australia and New 




Guinea, while the Pacific Rim province is limited to New Guinea. Australia and New 
Guinea were sundered by the postglacial sea-level rise, a process which began about 
16,000 years ago, was substantially complete 1 1,000 years ago, and was fully com- 
plete 8000 years ago when the remaining land bridge was flooded. The Pacific Rim 
province in Australasia must have formed after Australia and New Guinea had begun 
to separate, as it is lacking in Australia but well represented in New Guinea; but 
probably not long thereafter, as this province is an intensified late phase of the Hinter- 
land province, which does appear in northern coastal Australia. Thus the formation of 
the Australasian Pacific Rim province may have begun with movements of seafarers 
to northwestern coastal New Guinea as early as 16,000 years ago. The last immigra- 
tions belonging to this province were colonizations of northern coastal New Guinea 
by Austronesian speakers beginning 4000 years ago; the spread of Austronesian lan- 
guages through Melanesia and out into the Pacific continued until recent times (for 
Austronesian see e.g., Ross 1988; Pawley& Ross 1993; Kirch 1996). The Austrone- 
sian family dispersed about 6000 years ago from the vicinity of Taiwan and spread, in 
occasional colonization episodes rather than a continuous stream, over insular South- 
east Asia to reach Melanesia and New Guinea some two millennia later. Though the 
early Austronesians carried agriculture and their predecessors probably did not, other 
aspects of the history of its spread may well be typical of what must have occuired 
from time to time in the 12,000 years from the formation of the Pacific Rim province 
in Australasia to the arrival of Austronesians in New Guinea (and indeed of the his- 
tory of overseas colonization in Melanesia and Australasia for the last 50,000 years). 
Importantly for the present argument, the Austronesian languages in New Guinea are 
still mostly coastal and exhibit some of the Pacific Rim features (verb-initial word or- 
der, numeral classifiers, m in second person pronouns). 

The relative chronology of entries of the earlier strata must be as follows. The In- 
terior province formed first, as a result of early colonizations, and the Pacific Hinter- 
land province formed later and perhaps not greatly earlier than the Pacific Rim 
province. More time probably elapsed between the Interior and Hinterland entries 
than between the Pacific Hinterland and Pacific Rim strata. An absolute chronology 
for the earlier strata is less certain. The Pacific Hinterland is most clearly visible in 
northern Australia near where the postglacial sea-level rise first cut off Australia from 
New Guinea beginning about 1 6,000 years ago; thus this stratum is older than 1 6,000 
years, but (in view of its clarity) probably not greatly older. 

The stock half-life calculation presented above provides several different ways of 
estimating an age for the Pacific Rim population of languages. The crudest way is to 
count the number of stocks in the population and compute the age of the group in the 
event that it is a single genetic grouping (derived from a single entrant to the Americas 
or New Guinea, or a single Siberian or Southeast Asian dispersant). In my sample the 
number of language stocks in the American Pacific areas (Alaska-Oregon, Califor- 
nia, Mesoamerica, and western South America) is 44, and there are proposals for 
deeper genetic connections which, if proven, would reduce the number to 39 or less. 
The age of such a group would be 48,000 years — much too early for any plausible en- 
try to the Americas, let alone a relatively late entry. Of course, it is quite likely that 
some of the languages in the Pacific Rim zone survive from earlier entries and spreads 
and do not belong to the Pacific Rim population but happen to share their territory. 
Reducing the size of the population by one-quarter to one-half to accommodate this 
possibility yields ages of 42,000 and 36,000 years, still much earlier than any archeo- 
logical evidence of habitation and implausibly early for a relatively recent entry. 
These results indicate that the Pacific Rim population of America is not a single ge- 


netic lineage, since all consequences of assuming deep genetic unity are implausible, 
but they do not tell us the age or the actual number of genetic lineages. 

Personal pronoun systems with n in the first person singular and m in the second 
person singular are one of the markers of the Pacific Rim group. Since personal pro- 
nouns are easily inherited and not often borrowed, let us hypothesize that the lan- 
guages with n : m pronouns may be ancient sisters and compute the age of their 
hypothetical family. (Recall, though, that the n : m paradigm is not a sufficient ge- 
netic marker. Relatedness of all the n : m languages can be hypothesized but not as- 
sumed.) The languages with n : m pronouns in the western Americas represent from 
12 to 17 stocks (depending on exactly where the Pacific Rim boundary is placed, and 
on whether some possible but unproven deep groupings are taken to be stocks). This 
points to an age of 30,000 to 36.000 years, again too early for these stocks to be de- 
scendants of a single entrant to the Americas (given that the Pacific Rim population is 
a relatively recent formation), although not implausibly early for an ancestor in Sibe- 
ria or northern Asia. 

The conclusion to be drawn from these various calculations is that the Pacific Rim 
population is highly unlikely to be a set of sister languages descending from a single 
immigrant into the Americas. The signature historical markers that cluster in this 
population must result from a mixture of inheritance and acquisition (from substra- 
tum or borrowing). Let us now consider the ages for two proposed but unproven deep 
groupings within the population: the Hokan and Penutian groups centered in Califor- 
nia and Oregon. The Hokan group (see Langdon and Jacobsen 1996; Kaufman 1988) 
contains most or all of Karok, Chimariko. Shasta, the Achomawi-Atsugewi family, 
Washo, Yana, the Pomoan family, Salinan, the Yuman-Cochimi family, Seri, and the 
Tequistlatec-Jiqaque family (listed from north to south as they run from northern 
California to southern Mexico), and perhaps Esselen and/or Chumashan of coastal 
California, a total of about 12 branches. Each branch is an isolate or small family, and 
as relatedness between any of them has not been demonstrated each can be regarded 
as its own stock. Then the age of the family — if it is a family — is on the order of 
30,000 years. 

Penutian consists of a dozen families for which some further subgrouping is usu- 
ally posited: The Miwok-Costanoan and Yokuts families of central California; 
Maiduan of northern California and Klamath-Modoc, Sahaptian, and Molala of Ore- 
gon; Wintun of northern California and Coos, Siuslawan, and Alsean of coastal Ore- 
gon, and Takelma-Kalapuyan and Chinookan also of Oregon (this subgrouping from 
Callaghan 1997; DeLancey & Golla 1997; Golla 1997; see also Goddard 1996). If 
each of these four groupings were a stock — and this has not been proven — then Pe- 
nutian would be 12,000 years old. Without these intermediate groupings it would be 
of the same age as Hokan, since both groups contain about a dozen stocks. 

Thus, if both Hokan and Penutian are genetic groups, then the age of the Pacific 
Rim population (which includes them) is such as to accommodate the dispersals of 
two ancient families, one perhaps 12,000 years old and one probably older. These 
were separate events which occurred within the larger, and therefore presumably still 
older. Pacific Rim population. By this metric too, the age of the Pacific Rim popula- 
tion is greater than the firm archeological age for settlement of the Americas. 

The Pacific Rim province in New Guinea contains — depending on how certain 
questions of classification are resolved and exactly where its inland boundary is 
placed — some 1 5-20 stocks. If they all descend from a single ancestor, the age of the 
set is 30,000 to 36,000 years. Descent from a single ancestor, however, is implausi- 
ble: Austronesian is one of these stocks, it is known to have originated from abroad. 


Table 5. Combined immigration and diversification. A constant stock diversification rate of 1.5 is as- 
sumed. Age = age of the population. Rate = period ( in years) within which there is an average of one sui^viv- 
ing linguistic colonization. Numbers of stocks rounded to whole numbers. 

Age Rate Number stocks 

12,000 2000 10 

4000 6 

16,000 2000 14 

4000 8 

20,000 2000 21 

4000 1 1 

and its history is probably typical. A combination of the regular branching rate and an 
average immigration rate will give an approximate age for a set of stocks derived by a 
combination of immigration and subsequent diversification. Table 5 shows some fig- 
ures for various colonization rates. 

New Guinea has been colonized by languages from different branches of the East- 
ern Malayo-Polynesian branch: by Oceanic languages in the east and by others in the 
west. This amounts to colonization by one stock (Austronesian) in approximately 
4000 years (the age of the Eastern Malayo-Polynesian branch). Though it is admit- 
tedly risky to base a model on a single example, it can at least be concluded that a colo- 
nization rate of one stock immigrant in 4000 years is plausible. On the other hand, the 
Austronesian immigration also shows that independent entries could occur in two re- 
gions: The west (where colonizations came from Halmahera) and the east (where they 
came from New Britain and New Ireland). The early Austronesians introduced agri- 
culture where they spread, and agriculture can lower stock densities. Therefore, 
while today the Melanesian islands to both the west and the northeast of New Guinea 
are almost entirely Austronesian-speaking, in pre-Austronesian times stock diversity 
was probably greater, so colonizations at those two entry points might have come 
from entirely different families. Hence two immigrations per 4000 years, or one per 
2000, may be a better estimate of the pre-Austronesian rate. 

Table 5 shows that, for the 2000-year immigration rate, an age of 16,000 years for 
the province is expected. Of course, some minority of the languages in the province 
must descend from earlier immigrants, so the number of stocks originating in the Pa- 
cific Rim colonization phase must be fewer than the 1 5-20 actually in the province — 
perhaps 12-15. Assuming that one-quarter of the stocks are earlier survivors lowers 
the age of the province, though it is still probably older than 12,000 years. Though 
this measure is highly approximate, it yields an age for the Australasian Pacific Rim 
province that is reasonably consistent with the date of 16,000 years or less computed 
above based on the postglacial sea-level rise. 

To summarize this survey of language origins and language spreading around the 
Pacific Rim, the presence of the same three strata in the Americas and Australasia 
suggest that a single southeast Asian source fed the growing linguistic populations of 
both areas as languages moved out into the Pacific and others moved north to Siberia 
and thence to Alaska. Initially, when the spread began, the structural typology of the 
source area seems not to have differed greatly from that of the rest of the world 
(which, at that time and for modem human languages, was the Eurasian tropics and 
Africa). Hence the Interior Province areas belong structurally in the western Old 


World (Table 4). But then a very distinctive structural type made itself felt and cre- 
ated the second and third strata, giving rise to the markers of the Pacific Rim popula- 
tion. The Pacific Rim colonization thrust reached Australasia perhaps as early as 
16,000 years ago and was active until recent times when it produced Austronesian 
colonizations. It reached the Americas (spreading coastally in Asia up to Beringia and 
then down the American Pacific coast) sufficiently long ago to have fonned a large 
and diverse population stretching the entire length of North and South America by 
about 5000 years ago when the interior Siberian linguistic type, represented by 
Eskimo-Aleut, severed Pacific Rim linguistic continuity. The internal age of the Pa- 
cific Rim province in the Americas is great: To derive its total of over 40 stocks from 
regular diversification and immigration (with a single entry point and a 2000-year pe- 
riodicity, and assuming 10 of those stocks are remnants of the earlier population) 
would take 24.000 years, and to derive them from a single ancestor would take even 


The calculations given here have indicated that modem human language is over 
130,000 years old if monogenetic and over 100.000 years old if polygenetic with 10 
separate ancestors: both figures are highly approximate minima. Considerations of 
paleodemography and linguistic geography support polygenesis. Languages (and 
modem humans) spread out of the original small range to extend from Africa to south- 
east Asia, and the eastern edge of this range contributed the initial colonizing lan- 
guages to Australasia (colonized overwater from mainland Asia, with coastal 
landfalls) and also the Americas (colonized overland via Beringia, probably 
coastally). When this colonization began — over 50,000 years ago on archeological 
evidence from Australia — there was a gradual west-to-east clinal distribution of 
structural features in languages. Occasional additional colonizations continued on 
both fronts, and the combination of immigration and divergence created the tremen- 
dous linguistic diversity of New Guinea and the Americas. A datable linguistic event 
is the appearance of the Pacific Rim structural type in the east Asian staging area for 
both colonization fronts. That type must have begun spreading out in both directions 
some 20,000 years ago and perhaps earlier. It reached Australasia about 12,000- 
16,000 years ago, with the waning of the glaciers, and has continued to spread, colo- 
nize, and recolonize in insular southeast Asia, Melanesia, and Oceania. It reached the 
Americas at some undetermined time (early, though relatively late in the settlement 
history of the Americas) and continued occasional immigration until, with the 
Eskimo-Aleut spread about 5000 years ago, a very different structural type entered 
the pipeline. The origin of language was vastly earlier than the rise of the Pacific Rim 
type, but long-standing language spreading in various directions from Asia has given 
that type impact on half of the areas in the sample used here. 

This brief survey shows that nongenealogical comparison can tell us a good deal 
about when and where modem language arose and about the proximate and ultimate 
major geographical contributors to large populations of languages. Far from consti- 
tuting an obstacle to reconstmcting language origins, the fade-out point of about 6000 
years for tracing linguistic descent makes it possible to design a uniform sample, de- 
termine sizes of linguistic populations, compute genetic density on a consistent basis, 
estimate an average rate of divergence through branching, and estimate rates of lin- 
guistic immigration. These in tum have made it possible to identify and date some sa- 


lient phases in the origin and dispersal of the world's languages. On the most reahstic 
estimates, modern language as we know it is at least as old as modern humanity. Its 
origin was not a single point on a map or a single event in time, but a gradual process 
that unfolded in several different speech communities of varying degrees of discrete- 
ness. A continuous history from that origin to the present day can be traced in the 
spreads of originally local developments, of which this paper has described the main 
developments around the Pacific. 


'in linguistics, genetic refers to classification by descent, or cladistic analysis. It is not liter- 
ally genetic, since linguistic descent does not involve genes. Linguistic descent and biological 
descent are completely independent of each other, in that every individual carries the genes of 
his or her ancestors but by no means eveiyone speaks the language of their ancestors. This is 
self-evidently true for a large immigrant nation like the United States and for languages of the 
colonial period, but it is more common in other societies than is generally recognized by non- 
linguists. The French, for instance, carry genes inherited from the Gauls but speak a descen- 
dant of the language of the Romans. The Bulgarians carry, at least in some part, the genes of the 
Bulgar Turks who entered the Balkan peninsula in the eighth century, but they speak the Slavic 
language which the Bulgars soon thereafter took over from their neighbors. Language shift — 
the process where a society or speech community shifts, usually over the course of a few gen- 
erations, to another language % is common in societies of all types and sizes. In fact the 
number of speech communities whose languages go back in unbroken inheritance, without 
shifts, for more than a few millennia is probably small. 

"Comparing more than two languages can reduce the number of resemblants needed to 
demonstrate nonchance resemblance — provided a significant number of the languages com- 
pared participate in each putative cognate set. Threshold numbers are given in Nichols ( 1 997) 
{e.g., each set of putative cognates with generic resemblances must occur in at least 4 out of 6 
languages compared, or 5 out of 10, or 15 out of 50). To my knowledge these criteria have 
never actually been reached in multiple (or mass) comparisons; every multiple comparison I 
have seen offers too few attestations per cognate, and/or too loosely defined resemblances, for 
the results to be considered nonchance. 

"'A bottleneck is any substantial reduction in the number of continuing descent lines, as oc- 
curs in colonization, where only a few of the world's individuals or languages enter the new 
land to proliferate. The entering population is the founder or founding population , and founder 
effects occur when the inheritable traits of the founders show up among the descendants not in 
their expected or worldwide frequencies but in unusual frequencies reflecting their accidental 
idiosyncratic frequency in the founding population. 

''Using generic consonants is more appropriate, as we need to match Wiyot A7/ to Algon- 
quian k, and the Yurok glottalization to nonglottalization in the other branches. The overall 
probability of the whole paradigm, using the 0.07 probability for the generic consonants, is 
0.07 X 0.07 X 0.07 X 0.07 x 0.25 x 0.25 x 0.25 x 0.25 = 0.000000094, about nine in a hundred 
million or one in ten million. 

^In the earlier work on Algic, both Sapir and Haas realized the importance of the pronomi- 
nals but apparently believed that cognate vocabulary and regular sound correspondences were 
also an essential ingredient in the proof AsGoddard{ 1975) shows, thepronominalsare suffi- 
cient evidence and working out cognates and con-espondences are secondary to demonstrating 
relatedness, no matter what Sapir and Haas may have believed about what they were doing. 

The firm initial branches are: Anatolian, Tocharian, Celtic, Armenian, Greek, Albanian, 
Italic, Germanic, Balto-Slavic, Indo-Iranian. (Of these, Anatolian and Tocharian are now ex- 
tinct.) Poorly attested ancient Indo-European languages such as Phrygian, Thracian, and Illy- 
rian may or may not be separate branches. 


For these computations I have taken the traditionally inhabited world to be what it was dur- 
ing the last glaciation, subtracting the area under glaciers and adding an estimate of the exposed 
continental shelf from southeast Asia to Australia-New Guinea, where the exposed shelf is 
likely to have been well populated. 

Two caveats are in order. First, both comparative work and description in New Guinea are 
just beginning; comparative work may reduce the number of stocks, but on the other hand field 
work continues to turn up new families and isolates. If description can possibly keep up with 
extinction in New Guinea, it is possible that in some ten to twenty years we will have a different 
number of stocks for New Guinea, but at present it appears likely that newly discovered stocks 
will cancel out newly discovered genetic connections. Second, the initial branching rate of 
stocks in New Guinea, as in high-diversity zones generally, is low. Lower branching rates yield 
higher linguistic ages. 

The literature on the modem human dispersal is vast. Recent overviews include Howell 
(1996), White (1996). 

'°An overview is found in Howell ( 1 996). Harpending et al. ( 1 993), Sherry et al. ( 1 994), 
Relethford and Haipending (1995) posit a population bottleneck and a reduction to perhaps 
only a few thousand individuals, followed by population growth during the global expansion. 
This reduction followed the evolution of modem humans, and the reduced population had sev- 
eral subgroups, of which the largest is now best represented in sub-Saharan African popula- 

If as suggested in Note 10 about half of the population formed a single gene pool in the 
south, the size of the pool — a thousand to a few thousand individuals — suggests several lan- 
guages while its biological genetic unity suggests much intermarriage and hence multilingual- 
ism and convergence. Migration frequencies estimated for the other populations by 
Harpending, Sherry, Rogers, and Stoneking. are far less than what is required for multilingual - 
ism and linguistic convergence, hence these ethnicities may have been more discrete. 

'The bottleneck discussed in Notes 1 and 1 1 would have reduced the diversity but would 
still have left very different genetic lineages and stmctural types at distant parts of the range. It 
could easily have removed individual languages and lower branches while leaving most of the 
higher-level families with at least some representation on Earth. 

The rank-sorting of areas was either ascending or descending so as to put one or another 
bellwether area in the top half of the ranking. The choice of bellwether area has relatively little 
impact on the outcome. In the counts reported in Nichols ( 1 997d), Africa was used as the bell- 
wether area. 

Numeral classifiers are particles or affixes that are mandatory in phrases with numerals; 
the particular classifier used depends on the quantified noun, and classifiers often form elabo- 
rate shape-classified sets. Possessive classification is the analog, in head-marking languages, 
to declension classes in languages like the classical Indo-European ones. Declension classes 
are sets of case and/or number allomorphs, lexically determined by the noun and more or less 
orthogonal to gender and other classifications. Possessive classification is allomorphic sets of 
person-number possessive affixes, lexically determined by the possessed noun. It is rare and 
incompletely surveyed (for the semantics of the classes see Croft 1994), but appears to have a 
strictly Pacific Rim distribution. 

Literature Cited 

Austerlitz, R. 1980. Language-family density in North America and Eurasia. Ural-Aliaische 

Jahrbiicher 52:1-10. 
Bender, M.L. 1969. Chance CVC correspondences in unrelated languages. Language 

. Upside-down Afrasian. Afrikanistische Arbeitspapiere 50:19-34. 


— . 1975. Omotic: A New Afroasiatic Language Family. (University Museum Series, 3.) 

Southern Illinois University, Carbondale. 

Birdsell, J.B. 1953. Some environmental and cultural factors influencing the structuring of 
Australian Aboriginal populations. Amer. Nat. 87:171-207. 

. 1 957. Some population problems involving Pleistocene man. Cold Spring Harbor 

Symp. Quant. Biol. 22:47-70. 

Callaghan, C.A. 1997. Evidence for Yok-Utian. Intl. Jour. Aner. Ling. 63:18-64. 

Croft, William. 1994. Semantic universals in classifier systems. Word AS.\A5-\1\ . 

DeLancey, S. & V. Golla. 1 997. The Penutian hypothesis: retrospect and prospect. Intl. Jour. 
Amer. Ling. 63:171-202. 

Diakonoff, l.M. 1988. Afrasian Languages. Nauka, Moscow, Russia. 

Durie, M. & M.D. Ross, eds. 1 996. The Comparative Method Reviewed: Regularity and Ir- 
regularity in Language Change. Oxford University Press, New York. 

Ehret, C. 1995. Reconstructing Proto-Afroasiatic (Pro to- Afrasian): Vowels. Tone, Con- 
sonants, and vocabulary. (University of California Publications in Linguistics 126.) Uni- 
versity of California Press, Berkeley. 

Embleton, S. 1986. Statistics in Historical Linguistics. Brockmeyer, Bochum, Germany. 

. i 99 1 . Mathematical methods of genetic classification. Pages 365-388 in S. M. Lamb, 

& E.D. Mitchekk, eds., Sprungfrom Some Common Source: Investigations into the Prehis- 
tory of Languages. Stanford University Press, Stanford. 

Fortescue, M . 1 998. Language Relations across the Bering Strait: Reappraising the A rcheo - 
logical cmd Linguistic Data. Cassell Academic, London, U.K. 

Goddard, I. 1975. Algonquin, Wiyot, and Yurok: Proving a distant genetic relationship. Pages 
249-62 in M. D. Kinkade, K. L. Hale, & O. Werner, eds.. Linguistics and Anthropology: In 
Honor of C. F. Voegelin. de Ridder, Lisse, The Netherlands. 

. 1 996. The classification of the native languages of North America. Page 290 in Ives 

Goddard, ed. Handbook of North American Indians, vol. 1 7: Languages. Smithsonian In - 
stitution, Washington, DC. 

Golla, V. 1997. The Alsea-Wintuan connection. Intl. Jour. Amer. Ling. 63:157-70. 

Greenberg,J.H. 1960. An Afro-Asiatic pattern of gender and number agreement. Jour. Amer. 
Orien.Soc. 80:317-21. 

. 1963. The Languages of Africa. Indiana University Press, Bloomington. 

Haas, M.R. 1958. Algonkian-Ritwan: the end of a controversy. Intl. Jour. Amer. Ling. 

Harpending, H. C.,S.T. Sherry, A. R. Rogers, & M. Stoneking. 1993. Thegenetic structure of 
ancient human populations. Curr. Anthropol. 34:483-96. 

Heath, J. 1978. Linguistic Diffusion in Arnhem Land. Australian Institute of Aboriginal Stud- 
ies, Canberra, Australia. 

Howell, F.C. 1 996. Thoughts on the study and interpretation of the human fossil record. Pages 
1-39 in W. E. Meikle, F. C. Howell, & N. G. Jablonski, eds.. Contemporary Issues in Hu- 
man Evolution, California Academy of Sciences Memoir 2 1 . California Academy of Sci- 
ences, San Francisco. 

Kaufman, T. 1988. A research program for reconstructing Proto-Hokan: First groupings. Pages 
50-168 in S. DeLancey, ed., Papers from the Hokan-Penutian Language Workshop, 1988. 
Department of Linguistics, University of Oregon, Eugene. 

Kirch, P.V. 1996. The Lapila Peoples: Ancestors of the Oceanic World. Blackwell, Cam- 

Krauss, M.E. 1988. Pages 144-51 in W. W. Fitzhugh& A. Crowell,ed., A/am'/o/7g»e5 — An- 
cient Tales. Crossroad.'^ of Continents: Cultures of Siberia and Alaska. Smithsonian Insti- 
tution, Washington, D.C. 


Lamb, S.M., & E.D. Mitchell, eds. 1991 . Spntngfrom Some Common Source: Investigations 
into the Prehistory' of Languages. Stanford University Press, Stanford. 

Langdon,M.&W.H.Jacobsen,Jr. 1996. Pages 129-130 m V. GoUa, ed.. Report on the Spe- 
cial Hokan Session in Albuquerque, July 1 995. Proceedings of the Hokan-Penutian Work- 
shop. July 8-9, 1994 and July 5-6. 1995. (Survey of California and Other Indian 
Languages, Report 9.) Survey of California and Other Indian Languages, University of 
California, Berkeley. 

Mace, R. & M. Pagel. 1 995. A latitudinal gradient in the density of human languages in North 
America. Proc. R. Soc. Lond B 261 : 1 1 7-21 . 

Newman, P. 1980. The Classification ofChadic within Afroasiatic. Universitaire Pers, Lei- 
den, The Netherlands. 

Nichols, J. 1990. Linguistic diversity and the first settlement of the New World. Language 

. 1992. Linguistic Diversity in Space and Time. University ofChicago Press, Chicago. 

. 1995a. The spread oflanguage around the Pacific Rim. Evol. Anthropol. 3:206-15. 

. 1 995b. Diachronically stable structural features. Pages 337-356 in H. Andersen, ed.. 

Historical Linguistics 1993. Papers from the Eleventh International Conference on His- 
torical Linguistics. John Benjamins, Amsterdam-Philadelphia. 

. 1996. The comparative method as heuristic. Pages 39-7 1 //? M. Durie& M.D. Ross, 

eds.. The Comparative Method Reviewed: Regularity and Irregularity in Language 
Change. Oxford University Press, New York. 
. 1 997a. Modeling ancient population structures and movement in linguistics. Annual 

Rev. Anthropol. 26:359-84. 
. 1997b. Sprung from two common sources: Sahul as a linguistic area. Archaeology 

and Linguistics: Global Perspectives on Ancient Australia, P. McConvell & N. D. Evans, 
eds. Oxford University Press, Melbourne, Australia. 
. 1997c. The geography of language origins. Pages 267-278 in Proceedings of the 

22nd Annual Meeting. Berkeley Linguistics Society, Berkeley. 
. (submitted) Of needles and hay.stacks: Searches and heuristics in comparative 

method. MS under review. 

Nichols, J. & D. A. Peterson. 1996. The Amerind personal pronouns. Language 72:336-71. 

Orel, V. & O. Stolbova. 1995. Hamito-Semitic Etymological Dictionaiy. Brill, Leiden, The 

Oswalt, R.L. 1991 . A method for assessing distant linguistic relationships. Pages 389-404 in 
S.M. Lamb & E.D. Mitchell, eds., Sprungfrom Some Common Source: Investigations into 
the Prehistofy of Languages. Stanford University Press, Stanford.. 

Pawley, A. & M.D. Ross. 1993. Austronesian historical linguistics and culture history. Ann. 
Rev. Anthropol. 22:425-59. 

Relethford, J.H. &H.C. Harpending. 1995. Ancient differences in population size can mimic a 
recent African origin of modem humans. Curr. Anthropol. 36:667-76. 

Ringe, D.A., Jr. 1992. On calculating the factor of chance in language comparison. Trans. 
Amer. Philos. Soc. No. 82. 

Roberts, R.G., R.Jones, & M.A.Smith. 1990. Thermoluminescence dating of a 50,000-year- 
old human occupation site in northern Australia. Nature 345: 1 53-56. 

Roberts, R.G. & R. Jones. 1994. Luminescence dating of sediments: new light on the human 
colonization of Australia. Austral. Abor. Stud. 1994:2-17. 

Ross, M.D. 1988. Proto Oceanic and the Austronesian Languages of Western Melanesia. (Pa- 
cific Linguistics C-98.) Department of Lmguistics, Research School of Pacific Studies, 
Australian National University, Canberra, Australia. 


. 1 996. Contact-induced change and the comparative method: cases from Papua New 

Guinea. Pages 180-217 in M. Durie & M.D. Ross, eds., The Comparative Method Re- 
viewed: Regularity and Irregularity in Language Change. Oxford University Press, New 

Sapir, E. 1913. Wiyot and Yurok, Algonkin languages of California. Ainer. Anthropol. 

Shen", S.T., A.R. Rogers, H. Harpending, H. Soodyall,T. Jenkins, & M. Stoneking. 1 994. Mis - 
match distributions of mtDNA reveal recent human population expansions. Hum. Biol. 

Thomason, S.G. & T. Kaufman. 1988. Language Contact, Creolization. and Genetic Linguis- 
tics. University of California Press, Berkeley. 

Trask, R.L. 1996. Historical Linguistics. Arnold, London, U.K. 

Woodbury, A.C. 1984. Eskimo and Aleut languages. Pages 49-63 in W. C. Sturtevant, ed., 
Handbookof North .American Indians, vol. 5: Arctic. Smithsonian Institution, Washington, 

White, J. P. 1996. Paleolithic colonization in Sahul land. Pages 303-308 in T. Akazawa & E.J. 
E. Szathmary, eds.. Prehistoric Mongoloid Dispersals. Oxford University Press, Oxford, 


The Origins of World Linguistic 
Diversity: An Archaeological 

Colin Renfrew 

The McDonald Institute for Archaeological Research 
Downing Street 
Cambridge CB2 3DZ. UK 

The potential convergence of historical linguistics, prehistoric archaeology and 
molecular genetics offers hope of a new synthesis, in which the events and pro- 
cesses underlying the geographical distribution of human diversity, and specifi- 
cally linguistic diversity, may be clarified. One would hope and perhaps imagine 
that the archaeological record would also be informative about the early emer- 
gence of linguistic ability as a speciflcally human attribute, and in particular about 
the time and context when a well-developed capacity for language first emerged. 
But so far the relationships between material culture and linguistic ability have 
proved difficult to establish. Early aspects of hominid diversity, such as the differ- 
ing distributions of handaxe and pebble-tool cultures in earlier Paleolithic times, 
should be noted, as well as the very limited distribution of 'naturalistic' cave art 
and mobiliary art in Upper Paleolithic times. These are instances of geographical 
variability, for which linguistic correlates might conceivably be claimed. It is ar- 
gued, however, that the most significant process underlying the 'spread zone' ver- 
sus 'residual zone' pattern evident in the modern geographical distribution of 
language families was agricultural dispersal, with a time depth no greater than 
about ten thousand years. 

It is suggested that population history is the factor common to the three poten- 
tially converging disciplines, and that recent and ongoing developments in molecu - 
lar genetics are likely soon to clarify some outstanding questions. 

At first sight the languages of the world today can teach us little about the origins 
of language, or even about the origins of linguistic diversity. The diversity is certainly 
there: It is estimated that there are some 6.000 different languages spoken in the world 
today. But the time depth is lacking. We have no direct knowledge of any language 
prior to the invention of writing in the Near East and then Egypt some 5 ,000 years ago, 
and then in China some 3,500 years ago, and then more widely from 3,000 years ago. 

Fortunately, however, the languages of today contain much information relevant 
to their own histories. The resemblances or affinities between two languages, in terms 

The Origin and Diversification of Language Memoirs of the California Academy of Sciences 

Editors, N.G. Jablonski & L.C. Aiello Number 24. Copyright © 1 998 



of their phonology, morphology and vocabulary, can indicate very clearly that they 
are related. The discipline of historical linguistics uses such insights in a systematic 
way to group such related languages into what may be termed 'language families.' 
Like the taxonomies of biology, such classifications may go beyond arranging like 
with like in terms of superficial appearance (phonetic classification). In certain cir- 
cumstances they may reasonably allow inferences about family relationships, sug- 
gesting that some branches are more closely related than others, and permit inferences 
about hypothetical ancestors. Such inferences may allow the emergence of a pattern 
which reflects the historical process of descent: A genetic classification. 

For over a century archaeologists and historical linguists have struggled to use the 
archaeological data for the human occupation of the continents of the Earth to give 
some insights into the formation of these language families. At times scholars have 
confused linguistic groups with supposed racial groups, and in the earlier half of this 
century notions of racial supremacy brought about an unhappy episode both in the 
historical sciences and in world history. But today it should be possible to avoid such 
fallacious equations. It is necessary also to avoid simplistic assumptions about how 
languages change, and indeed about how 'peoples' move. 

It is my belief that a new synthesis is today possible, using three disciplines which 
are in a sense formally independent, since they use different classes of data: historical 
linguistics, prehistoric archaeology and molecular genetics. Modem linguistics is to- 
day increasingly taking a worldwide view, studying the whole wide range of lan- 
guages, and not focusing to the exclusion of others upon the languages of Europe and 
Western Asia. From this global view new insights are emerging. 

Molecular genetics, as applied to living human populations, is now the fastest 
growing field of study about the human past. For just as the languages of the world to- 
day contain information about their own past, so do the genes within human popula- 
tions. Studies of nuclear and mitochondrial DNA are allowing the similarities and 
differences between individuals and between groups to be studied, and, as in the lin- 
guistic case, phonetic classifications may in favorable cases lead to genetic classifica- 
tions (Figure 1). 

There is , moreover, the hope that the recovery of ancient DNA from preserved hu- 
man remains may give direct information about the genetic composition of individu- 


PhenetJc dendrogram 

Phylogenetic tree 



Time before 


Units (arbitrary order) • = living forms 

O = ancestral forms ("fossils") 
Figure l . The potential isomorphism between phonetic dendrogram and genetic descent. 


als long dead, which will offer a control and a test on the genetic inferences which 
have been based upon the study of living materials. 

Prehistoric archaeologists are thinking in new ways about culture change and cul- 
ture process, and models of change are being constructed which are more appropriate 
to our understanding of early cultural dynamics. Moreover archaeology offers a much 
better grasp upon time depth than do the other two disciplines which join with it in the 
new synthesis. Radiometric dating methods do now reliably allow the provision of 
reasonably accurate dates, in calendar years, for the deposits and the artifacts which 
the archaeologist discovers. 

In reality it is not easy to bring together these three different classes of data: the lin- 
guistic, the molecular genetic and the archaeological. There is perhaps a tendency in 
each discipline to favor its own data set, and, it has to be said, to draw preferentially 
upon the other disciplines to support current inferences, in such a way that the argu- 
ment often becomes a circular one. Certainly it cannot be too firmly asserted that lan- 
guages in themselves tell us nothing about the genetic composition of their speakers, 
and genes carry no direct linguistic implications: There are, in practice, strong corre- 
lations, but (as I shall hope to emphasize) these arise from historical contingencies, 
not from any general equivalence between languages and genes. 

The first central point which I wish to make is that there was in fact only one his- 
torical reality. People lived upon the face of the Earth: Specific languages were spo- 
ken at particular times at definite places. Those people had genetic compositions of 
specific kinds, and to the extent that their descendants survive, those DNA patterns 
albeit transformed and modified, survive to influence and indeed determine the DNA 
of those descendants. Those same people had settlements and used tools, and ex- 
ploited plants and animals, and remains of all these things persist in the archaeologi- 
cal record to this day. 

The new synthesis will not come about until all three classes of data are used with 
respect. Most archaeologists of my acquaintance are very hazy indeed about the pat- 
terns visible when one studies language families, or nuclear and mitochondrial DNA. 
Most molecular geneticists are prone to make hasty conclusions not only about the 
population histories which might have given rise to the genetic patterns which they 
observe, but also about possible linguistic or archaeological correlations. I have to say 
that I have found rather few historical linguists who are willing to recognize the ar- 
chaeological or molecular genetic data at all: One sometimes has the impression that 
languages are supposed to evolve and change of their own accord without any human 

Nowhere are these shortcoming, these failures in bringing different classes of data 
to bear, more evident than in the matter of chronology, of dating. Archaeologists, as I 
have said, do have sure dating methods, but they are used for dating artifacts, and arti- 
facts in themselves tell us nothing of language or of genes. Molecular geneticists use 
a variety of frameworks of inference for estimating genetic rates of change, and there 
is no doubt that these are continually being refined. But it is sometimes difficult to es- 
cape the conclusion that the foundations are insecure. The recent suggestion by Paabo 
(1996) that estimated mutation rates for mitochondrial DNA may have been up to an 
order of magnitude in error certainly gives cause for reflection. 

But if molecular genetics have problems with dates (or rates) these are as nothing 
compared to those of the historical linguist. Most historical linguists today reject the 
techniques of "glottochronology" as formulated by the Swedish (which assumed an 
approximately constant rate of word loss in all languages at all times). 


But while few linguists will go so far as Joseph Greenberg in developing an alter- 
native formulation (defining a core vocabulary, and assuming an exponential rather 
than a linear decay process), many still find it possible to offer approximate dates for 
the formation or dispersal of this or that protolanguage. Some of these rely upon the 
archaeology for the dating of vocabulary features — and many are the pitfalls which I 
could catalogue in the supposed dating of the 'wheel' or the 'horse' in Indo-European 
studies, from Gustav Kossinna to Jared Diamond. But some historical linguists make 
the claim that they operate by principles of dating which go beyond those of glotto- 
chronology or of linguistic paleontology. I have never been able to discover what 
these principles are, and have always suspected that they depend upon circular rea- 
soning. The only effective solution must be the bringing into proper perspective of the 
genetic, the linguistic and the archaeological data, and that is no easy task. 

Today I would like to indicate the growing body of evidence that suggests that 
there is indeed some meaningful patterning in the distribution of the world's language 
families. I shall seek to suggest that there are indications, which we can recognize, of 
language groupings which have been in place geographically for well over ten thou- 
sand years, before the end of the last ice age, the termination of the Pleistocene period. 
These languages and families would therefore document the conclusion of the disper- 
sal of our own species. Homo sapiens sapiens, during the Pleistocene period. And I 
would like to indicate that there are other more recent patterns which we can recog- 
nize, many of them associated with the inception of farming, and then the widespread 
radiation of agricultural and farming technologies, which began to take place some 
10,000 years ago. There is a growing body of archaeological evidence which can be 
brought to bear upon these matters. But what makes the current situation so interest- 
ing is that there is a considerable flow of data from molecular genetics which bears 
very strongly upon these hypotheses. It can tell us nothing about early language di- 
rectly, as noted above. But it can tell us a great deal about population history. It is in- 
deed the concept of population history that lies at the heart of the new synthesis, since 
it is of direct relevance in each of the three intersecting fields with which we are deal- 
ing (Figure 2). In each case it is the processes of inference, the models of change, that 
require fiirther scrutiny. 

The Origins of Cultural Diversity and 
the Question of Language Origins 

When did linguistic competence develop among our ancestors? Here I should first 
like to draw attention to some of the limitations of the view of language origins that 
are currently so widely held. Although it is in many ways plausible that there should 
be a correlation between the emergence of anatomically modem humans, of new as- 
pects of material culture (such as the blade industries of northern Europe, first seen at 
the onset of the Upper Paleolithic) and the development of full speech capacity, there 
is in fact little concrete evidence that such was the case. There are, moreover, earlier 
indications of regional diversity in the archaeological record which should certainly 
be noted, even if linguistic correlates are not proposed for them. 

At first sight, indeed, the pattern of world linguistic diversity today casts little light 
upon that deep and challenging question of the origins of language. I had almost be- 
come resigned, at symposia on early language, to being the man at the end: The one 
who deals with the past ten or twenty millennia. This may seem a formidable time 
depth to the conventional historical linguist. But with the assumption, almost univer- 






Figure 2. The intereecting fields of the new synthesis. 

sally held, that all human groups were anatomically modem by at least 40,000 years 
ago, and all speaking fiilly developed languages at that time, by the time I would have 
anything to offer, the process of initial language formation, that is to say the develop- 
ment of linguistic ability is already complete. To put the matter in more colloquial 
terms, it is all over bar the shouting. 

But I have recently come to feel that some of the widely accepted conclusions 
about the early history of language are not so well established as they might seem. The 
data are inevitably few, and some aspects of the archaeological record which do show 
some degree of coherent patterning are perhaps being undervalued. 

Today many archaeologists, geneticists and linguists accept what I should like to 
call the "Michelangelo effect." They accept that anatomically modern humans ap- 
peared in Europe some 40,000 years ago. They accept, quite rightly, that all anatomi- 
cally modern humans today (and of course we all are) have fully developed linguistic 
capacities, which are therefore to be taken as characteristic of Homo sapiens sapiens. 
And they note that the transition from Middle to Upper Paleolithic industries in the ar- 
chaeological record at that time show the emergence of new tool forms, notably blade 
industries, and worked bone and antler, indicative of the skills associated with the 
new species. Add to this the wonderfiil appearance of cave art in France and Spain 
only some 5,000 years later, and you have a "package" which, with the eye of faith, 
can be seen to constitute a single "Human Revolution" scarcely less miraculous than 
the Creation of Adam. 

The prominent exponents of this view, such as Lewis Binford or Paul Mellars are 
critical, skeptical about the abilities of the Neanderthals who preceded our own spe- 
cies, questioning, for instance, the alleged evidence for deliberate burial in Middle 
Paleolithic times. Sometimes I wonder if we are not in face of some miraculous new 
"Creation," and that wonderful image from the Sistine Chapel ceiling comes into my 



mind. By the "'Michelangelo effect," then, I mean the assertion that there was a revo- 
lutionary emergence at one and the same time of anatomically modem humans, of full 
linguistic capacities, of the ability to think conceptually, of a much wider range of so- 
cial and cultural behaviors (manifested in new lithic industries and soon in cave art). 

But as these scholars are well aware, the emergence of anatomically modern hu- 
mans took place much earlier, presumably in Africa around 1 50,000 years ago, and is 
seen already in southwestern Asia at the Qafzeh Cave around 100,000 years ago. 

Here I want to draw your attention to just two interesting questions. First the abil- 
ity to conceptualize. And second the matter of regional diversity. Each has been taken 
by some scholars as an indication of linguistic ability. But both are clearly evident in 
the archaeological record of the Middle Paleolithic period, hundreds of thousands of 
years before the emergence of modem humans and of the widely assumed develop- 
ment along with them of advanced linguistic capacities 

The Handaxe 

For some time scholars have debated the significance of those striking core tools, 
called 'handaxes'(Figure 3), which are found over a time range of at least half a mil- 
lion years in Afi-ica, western Asia and Europe, the characteristic product of our imme- 
diate prehuman ancestor, Homo erectus. Although some scholars, such as Davidson 
and Noble (1993) have regarded them as simply worked-out cores, the byproduct of 
the manufacture of flake tools, most have taken the view that these are deliberately 
made tools {e.g., Toth & Schick 1993). 

I have recently been impressed by the arguments of the philosopher John Searle 
(1995) that many of the most important foundations of our social life are what he 
terms "institutional facts," conceptual formulations by which society functions (such 

. 1 inch , 
Figure 3. Handaxe from site ofSwanscombe, Kent. Based on Oakley (1967:72). 


as rank, power, gender (as opposed to sex), value, wealth, money, etc.) They require 
human institutions for their existence, and indeed institutions of a kind which we 
would normally assume to require the cognitive capacities of modem humans. These 
may be contrasted with 'brute facts,' often facts of nature, which exist quite independ- 
ently of language or of any other institution. They are not dependent upon linguistic 
conventions or on the fairly sophisticated tacit understandings and conventions 
among individuals that form the basis for our social lives. It may certainly be argued 
that the archaeological record can give unequivocal indications of 'institutional facts' 
and of the concepts associated with them: The stone cubes of the Indus Valley civili- 
zation around 2000 BC as indicative of modular systems of weighing and of counting 
are a case in point (Renfrew 1982). 

What is interesting here, however, is that Searle very clearly associates what he 
terms 'agentive functions' — seen for instance in the production of a specific artifact 
form designed to fulfill a particular purpose — with such institutional facts. He gives 
the example of a screwdriver, stressing the distinction between features which are in- 
trinsic to nature and others which are relative to the intentionally of observer, users, 
etc. But I would argue that in this case what is good for the screwdriver is good also 
for the handaxe, so long as we are willing both to recognize it ourselves as a distinc- 
tive tool type and also to infer or assume that its makers also saw it in that way, as a 
purposive product not just the unintended consequence of making flakes. 

Such analyses are helpful. They help to show us that we are dealing here with 
well-defined concepts. But we should remember, as Thomas Wynn reminds us 
( Wynn 1 993), that this need not necessarily imply the existence of a spoken language. 
Merlin Donald in his Origins of the Modern Mind (Donald 1 99 1) regards the era of 
Homo erectus as the mimetic phase of hominid cognitive evolution, preceding the 
mythic phase (with its stress on language) of early Homo sapiens sapiens. And Bloch 
(1991) has reminded us that culture is not necessarily linguistic. We should note the 
importance of imitation, or serial memorization, in so much mimetic learning. This 
point is well made in a diagram by David Clarke (1968) (Figure 4). 

We see here that no speech is needed for one individual to teach another how to 
make a handaxe, nor indeed would speech be needed to teach how to use it. Speech is 
not needed to show a young child how to sit on a chair, and the concept "chair" does 
not need the verbal fomiulation of the word "chair" for its understanding and commu- 
nication. To say this is simply to make a general remark about the understanding and 
communication of agentive functions. 

This point can be used in two diametrically different ways. Those scholars who 
see a long and slow evolution of language as likely, will argue that vocal utterance is 
likely to have accompanied the formulation of the concept {e.g., of the handaxe) and 
its practical embodiment through manufacture. They will take it that vocal means will 
have been found already in the Lower Paleolithic period to accompany such 'instit- 
utional facts.' They will argue that there will already have been signs, and perhaps 
sounds (or 'words') indicative of 'handaxe' or of 'fire' (when that came to be pro- 
duced deliberately). These can be regarded, within the perspective developed by 
Searle, as conceptual formulations of a kind which many of us would imagine require 
a degree of cognitive sophistication difficult to achieve without the use of language. 

An alternative view is that the Lower and Middle Paleol ithic do indeed represent a 
'mimetic' phase, and that the production of well-defined tool forms does not require 
any very sophisticated conceptualizing power. Moreover such conceptualization as 
was needed need not have been language dependent. The real breakthrough will have 



Figure 4. Learning by mimesis (possibly without language). Based on Clarke ( 1 968: 1 82, figure 39). 

come with the formation of developed linguistic powers, perhaps 40,000 years ago 
and that these were exclusively restricted to our own species. 

Regional Diversity 

Regional diversity in behavior within a single species must be regarded as cultural 
behavior — that is to say learned behavior passed on locally within a regional tradi- 
tion — if it does not have a genetic origin (which would imply regional diversity in 
the genome). For that reason local cultural traditions have excited the attention of ar- 
chaeologists as displaying an aspect of behavior which is particularly characteristic of 
humans, although it is true that some other species do show patterns of regionally spe- 
cific learned behavior. It seems to be the case, however, that some aspects of regional 
diversity seen in the Paleolithic period have not received sufficient consideration: 
They too may have a bearing upon the emergence of developed linguistic ability. 

It may be that when we speak in global terms of the Pleistocene and refer to Nean- 
derthals or to 'Upper Paleolithic humans' we are making the same categoiy error of 
dealing in gross and unrefined categories which today we are ready to criticize when 
we read anthropological works of a century ago which speak of 'primitive man' or 
'primitive art.' It may be this global generalization which has obscured some of the 
significant regional variation in the archaeological record. It is seen already in the 
Lower Paleolithic, and some aspects persist right through into the Holocene period, 
so that there are features here which could conceivably have a bearing upon our un- 
derstanding of quite recent linguistic diversity. 



One of the earliest known instances of regional diversity is the disparity which ex- 
isted around half a million years ago, following the dispersal out of Africa of Homo 
erectus , in the distribution of lithic artifacts. We have already spoken of the handaxes 
which are such a frequent feature of the Lower Paleolithic of western Europe, and of 
the same period in Afl"ica. It is notable, however, that in southeast Asia, where re- 
mains of the same hominid species are found, and in some other areas including East- 
em Europe, there are no handaxes. Instead we find the 'pebble tool tradition' which in 
fact resembles in some ways the lithic industries found in Africa during the earlier, 
pre-erectus period, and generally attributed to that earlier ancestor Homo hahilis 
(Figure 5). 

It is particularly interesting that this same regional divide continues right up into 
the Upper Paleolithic period, and indeed in southeast Asia even later, where the Ho- 
abhinian lithic tradition is found. Blade industries which we associate with the Euro- 
pean Upper Paleolithic are not a universal feature. Although I am myself persuaded 
by the genetic evidence for an out-of-Africa origin for our own species, it has to be 
said that the 'multiregional' hypothesis, with a local hominid evolution from erectus 
to sapiens would sit more easily with the above inteipretation of the lithic sequence. 
Presumably the analysis has to be undertaken at a much finer level before it can bear 
such weighty constructions. But the point here is that there are regional diversities in 
culture that must have an important place in the evolutionary history. If this long- 
standing differentiation does not have a genetic origin (and few would suggest that 
the tendency to make handaxes rather than pebble tools is genetically determined) 
does it not reflect a long-standing cultural tradition which might also be of linguistic 

Figure 5. Map of handaxe vs. pebble industries: spatially distinct traditions of tool making, both associ 
ated with Homo erectus about 1 million to 100,000 years ago. Based on Burenhult (1993:64). 



The second case of regional localization which deserves emphasis is that of cave 
art. The casual reader might have the impression that the remarkable painted caves of 
France and Spain have their counterparts over much of the world. Did we not learn re- 
cently, and it is indeed a remarkable discovery, that there is rock art in Australia going 
back some 40,000 years and perhaps more, and that there may be an unbroken tradi- 
tion of rock art in that subcontinent from that time until the present? 

A closer analysis will, however, distinguish between the Franco-Cantabrian cave 
art and the rock shelter art which is indeed a very widespread feature globally, but in- 
dicative of more recent hunter-gatherers and perhaps of pastoralists. In the first place 
the Franco-Cantabrian cave art is not (as we know it) a rock shelter art: It is found 
deep in limestone caverns which are difficult of access now and were difficult of ac- 
cess then. And in the second place there is a coherent style in the Aurignancian and 
Magdalenian art of France and Spain which deserves to be defined closely in a way 
that avoids such terms as 'naturalistic' or 'figurative,' but yet which indicates how 
these images differ in the impressions which they give from those of most rock art 
styles. In short they are different. This was a special phase in a limited region which 
began 35,000 years ago, and is without known parallel elsewhere. The point is em- 
phasized for us by the 'Venus' figurines and other carvings associated with the 
Gravettian culture in much the same area, but with an eastward extension to Siberia 
(Figure 6). These again are, I believe, v/ithout parallel for the Paleolithic period. 
There is no virtue in our basing our discussions about early language on the assump- 
tion that the human abilities which produced these visual symbols were universally 
available to our species from 40,000 years ago onwards. 

Once again we have a notable case of regional diversity, this time in the display (in 
the archaeological record) of the outstanding conceptual abilities implicit in the pro- 
duction of Franco-Cantabrian cave art and in the small Gravettian sculptures. Such 

Figure 6. The distiibutions of Franco-Cantabrian cave art (hatched area) and Gravettian "Venus" tlgu 
rines which are restricted to Europe ca . 29,000-22,000 years ago. 


abilities cannot, however, be claimed as a universal feature of our species in the Upper 
Paleolithic period: Their documentation is very restricted geographically. Did these 
evident conceptual abilities have no linguistic counterpart? Might not this imply 
some degree of linguistic diversity? 

Linguistic and Cultural Evolution 

In discussing the 'Sapient Paradox' (Renfrew 1996b), that is to say, the disparity 
between the assumed fully human status (anatomically and linguistically) of our spe- 
cies by some 100,000 to 80,000 years ago, and the considerable delay before we see 
anything very remarkable in its behavior, I suggested that in our analysis we should 
indeed lay more stress on praxis, on actual activity and behavior, and upon the art- 
ifacts which document it and which often motivated it. 

Biologists are used to speaking, among living species, of the genotype (defined by 
the genes) and the phenotype (reflecting the living organism). But in human affairs it 
is the 'prototype,' representing the actual behavior, which interests us. rather than the 
physical foirn with all its potential for interesting behavior but perhaps with little to 
show for it. 

I begin to feel that we should not assume modern or complex behavior until we 
have proof of it. That is why the discussion of the limited distribution of rupestrian 
and mobiliary art is so illuminating. These were not human universals in the Upper 
Paleolithic, but very restricted in their scope. 

How long did it take language to evolve? Let us imagine that there is a gene or re- 
lated set of genes which govern the capacity for complex linguistic behavior: Such an 
assumption is indeed made among nearly all evolutionary biologists. Whether or not 
we embrace the "Michelangelo effect" we can recognize that fully modem linguistic 
ability, common to all living human groups, represents a significant mutation or se- 
ries of mutations from what went before. Let us further imagine that this mutation or 
set of mutations occurred in a given local area and restricted time period. It is widely 
believed today that enhanced linguistic facility would confer adaptive advantage, and 
that the gene would diffuse successfully through the population. But the interesting 
question is: What happens next? 

Language is a social phenomenon: It is spoken between people. It implies not only 
a shared vocabulary, but shared understandings and concepts, shared 'institutional 
facts. ' When and how did these arise, and how long did that take? Should we expect to 
see behavioral correlates in the archaeological record? 

At first sight we might imagine that these new, gifted individuals in the population 
would soon be forging their own language. It might at first be a 'private' language, 
like that shared by the Bronte siblings, but it would soon become public. The insight 
that it might grow among siblings and cousins is probably right, since it would be 
these individuals who would share the new genetic endowment. But the Bronte paral- 
lel may not be a sound one. It is one thing to imitate what is already done in a different 
way: Quite another to originate «/> initio. The extensive literature on 'stimulus diffu- 
sion' reminds us that it is easier to reinvent the wheel, when you have seen it, than to 
invent it in the first instance. 

Do we, in those special cases where symbols were being used with particular rich- 
ness and profusion, as with the Gravettian figures of 20,000 years ago, and the 
Franco-Cantabrian paintings from 35,000 years ago, see cases where conceptual and 
symbolic virtuosity may be reflecting the development of new symbolic skills in oral 


and hence verbal communication as well as visual communication — i.e., language 

The point which I am making here is that we should not make the mistake of as- 
suming that the behavior of hunter-gatherers in recent years is a sound guide to that of 
the hunter-gatherers of 30,000 years ago. The former have the benefit of the same 
length of time of cultural and social evolution as do farmers and city-dwellers. Indeed 
is it to be concluded that, over the past 40,000 years, there has been continuing selec- 
tion of those individuals who are linguistically able, and against those with less devel- 
oped linguistic capacities? 

There may be a significant distinction to be drawn between the genetic mutations 
(genotypic) which led to linguistic capacity or abilit}' (phenotypic) on the one hand, 
and the actual achievement, through the evolutionary development and use of lan- 
guage on the other hand (praktotypic) of those linguistic skills which we associate 
with the exponents of all known languages. Whether the indigenous Tasmanians, 
who at the time of their encounter with Europeans were accounted to have the most 
impoverished known material culture among hunter-gatherers, had the same sophisti- 
cated language skills which we associate with surviving hunter-gatherers is unfortu- 
nately a question which history has left unanswered. 

The Language Farming Dispersal Hypothesis 

It is appropriate now to change the focus of discussion from the question of the ori- 
gins of linguistic ability to that of the origins of linguistic diversity or capacity. Of 
course one would expect the latter to be related to, indeed dependent upon, the former. 
But many linguists would hold that it is not practicable to undertake linguistic recon- 
structions beyond a time depth of some five or six thousand years. 

They would argue that the processes of linguistic divergence (including word re- 
placement and loss) operate in such a way that a language at one time point will have 
been so transformed over the millennia in the course of evolution and transmission to 
its daughter languages that no trace of the parent will be evident through the study of 
the offspring. The existence of such a linguistic time barrier of five or six thousand 
years is currently coming under question. But caution is still appropriate and it may 
still be appropriate to refrain at this time ft^om linking questions of the origins of lan- 
guage (as a human capacity) with those of the origins of languages (as reflected in 
contemporary linguistic diversity). 

Several linguists have remarked on the marked difference in the nature of the geo- 
graphical distribution of language families. Austerlitz (1980) has contrasted patterns 
of language-family density, and Nichols (1992) made the usefial distinction between 
what she terms linguistic 'spread zones' and 'residual zones.' 

When a language family displays the 'spread zone' pattern it displays what the 
linguist may term a low genetic density {i.e., a limited number of linguistic units un- 
related by descent from a common language ancestor) over a sizable area, and a rela- 
tively shallow time depth. The 'residual zone' pattern shows more language families, 
greater linguistic diversity within the language family, and greater antiquity of the 
linguistic stocks there. The Caucasus is a good example of a 'residual zone,' and is 
sometimes regarded as a linguistic refiigium. 

Here I wish to reiterate the hypothesis, already formulated in rather different form 
(Renfrew 1992,1996a) that most of the world's language families which show a 
'spread zone' distribution are the result of a process of farming dispersal. The same 


point has been persuasively developed by Peter Bellwood in a number of influential 
papers {e.g., Bellwood 1996). 

It would seem that the distinction is a fundamental one, which has also a chrono- 
logical significance. All the "residual zone" language distributions have been in place 
for at least 10.000 years, since before the end of the last glaciation (Class A in Table 
1 ). The early settlement dates now available for Australia and for New Guinea (i.e., 
the 'Indo-Pacific' languages) explain their place in category L It should be noted, 
however, that only the North Australian languages show the characteristics of a 'resi- 
dual zone': For reasons as yet not well understood the remaining languages of Austra- 
lia (the so-called Pama-Nyungan group) show the characteristics of a 'spread zone.' 

All the "spread zone" distributions are the product of dispersals within the past 
1 0,000 years (Class B in Table 1 ). The most significant cases, those listed in category 
II are indeed fanning dispersals. But the role of northern climate-sensitive adjust- 
ments by hunter-gatherers after 10,000 BP should be noted (category III) and of 
course that of long distance maritime colonization over the past five centuries (cate- 
gory V). Only in a very few cases is it appropriate to ascribe the distribution of an en- 
tire language family to the process of "elite dominance" (category IV). This may 
however be the effective mechanism for much of the later distribution of the Altaic 
language family (made possible by the development of horse riding), as well as for the 
Indo-Iranian branch of the Indo-European family. 

It should be noted that in the table some language families are indicated within sin- 
gle quotes: This is intended to indicate that by many linguists they are not regarded as 
true language families (in the linguistically genetic sense of sharing a common origin 
from an ancestral protolanguage). They may instead simply represent language areas 
where previously unrelated languages have come to share a number of characteris- 

TaBLE 1 . "Residual Zone" (Class A) and "Spread Zone" (Class B) Language Families (Renfrew 1 996a, 
with modifications). 

The present distribution of each language area is accounted for by one of the following five processes: 

Class A: pleistocene 

I. Initial colonization prior to 12,000 BP: 

'Khoisan', 'Nilo-Saharan' (plus later 'aquatic' expansion). Northern Caucasian, South Caucasian, 
"Indo-Pacitlc" (plus later farming changes). North Australian, 'Amerind." Localised ancest ral 
groups of II and 111 (below) 

Class B; post-pleistocene 

I I . Farm ing dispersal after 10,000 BP: 

Niger-Kordofanian (specifically the Bantu languages), Afroasiatic, Indo-European, Elamo- 
Dravidian, Early Altaic, Sino-Tibetan, Austronesian, Austroasiatic. 

III. Northern, climate-sensitive adjustments after 10,000 BP: 
Uralic-Yukaghir, Chukchi-Kamchatkan, Na-Dene, Eskimo-Aleut 

IV. Elite dominance: 

Indo-lranian, Later Altaic, Southern Sino- Tibetan (Han) 

V Long-distance maritime colonization since 1400 AD: 

Mainly Indo-European (English, Spanish, Portuguese. French). 


tics. It should be noted that all of these, like the others in category I show the 'residual 
zone' pattern. 

In earlier discussions of farming dispersal I have emphasized the demographic ef- 
fects of the spread of farming, and have emphasized the "demic diffusion" model ini- 
tially proposed by Ammerman and Cavalli-Sforza. It is indeed the case that many 
farming economies have been propagated through the gradual expansion of the farm- 
ing population, and this point has on a number of occasions been very well argued by 
Bellwood {e.g., 1996). However it is appropriate also to recognize the point made by 
Zvelebil (1996) and others, that the techniques of farming may well be taken up by the 
indigenous hunter-gatherer population. In such a case one might well have farming 
transmitted, but without a significant degree of gene flow, or of language replace- 

It is, however, worth developing what one might term the "substitution model" for 
language replacement in the wake of farming dispersal. Here one may draw upon the 
model developed by Zvelebil (1996:325) but with a rather different purpose (Figure 


Zvelebil here envisages a three-phase availability model for the transition to farm- 
ing, where the indigenous population acquires the new domesticates and the tech- 
niques of the farming economy, while itself remaining genetically little altered. Rhret 
(1988) has, however, described a situation such as this where the local group, for- 
merly hunter-gatherers, take up the language of the farmers along with their economy 
and technology, while retaining their own ethnic and more particularly genetic iden- 
tity. The process of acculturation which Zvelebil envisages allows time for many ele- 
ments of the speech of the farmers to be acquired along with the necessary 
technology. This then would be a case of language replacement, although the case is 
very different from that of demic diffusion, and there is plenty of room for features of 
the indigenous language, whether lexical or morphological, to survive in the new lan- 
guage which the indigenous inhabitants adopt, or rather form by their adoption of so 
much of the farmers' speech and technology. In such cases, where the new speech is 
acquired by substitution rather than by demic diffusion, we should expect the newly- 
emerging language family to show more diversity, and to owe more to the language 
being replaced (which would form a more evident substrate) than in the demic diffu- 
sion case. 

There are many instances of language families whose distribution may be as- 
signed to a language-farming dispersal process, whether by colonization (demic dif- 
fusion) or by acculturation (substitution). A number of them have been conveniently 
listed by Bellwood (1996:469) (Table 2). 

Bellwood, in a number of articles, has very persuasively dealt with the Austrone- 
sian languages. The case of the Niger-Kofdofanian languages (notably the Bantu lan- 
guages) has been well set out by a number of authors including Phillipson ( 1 977). In a 
number of papers I have suggested that the source for the farming dispersal which car- 
ried the Indo-Euuroean protolanguage to Europe was the Anatolian lobe of the south- 
western Asian zone of agricultural origins. The northeastern lobe, towards 
Turkmenia, may well have transmitted farming techniques to the region where the 
proto-Altaic languages were spoken, which later gave raise to the Altaic expansion. 
The southwestern lobe was the startmg point for the Elamo-Dravidian languages. 
And it has always seemed to me that the only coherent process underlying the Afroa- 
siatic linguistic unity must be the dispersal of sheep and goats, and later cereals, from 
the Jordan-Palestine area across to North Africa, although the story there may be a 
long and complicated one. 





100 - 

90 - 

80 - 


60 - 


40 - 


20 - 


Availability phase: 

foraging principal 
means of subsist- 
ence Domesticates 
and cultigens <5% 
of all remains. 

Substitution phase: 

fanming strategies 
developed while 
foraging strategies 

Domesticates and 
cultigens 5-50% of 
total remains 

Consolidation phase;. 

farming principal | 
mode of subsistence^ 
Foraging loses | 

economic, organizat-| 
ional and ideological j 
significance. I 

Domesticates and I 
cultigens >50% of I 
total remains. ' 


Figure 7. Linguistic adjacency acceptance: The adoption by hunter-_gatherersofthe language of neighbor- 
ing cultivators through conduct-mduced language shift. Based on Zvelebil ( 1 996: 325), with additions. 

Glover and H igham ( 1 996) have recently reconsidered the origins of rice cultiva- 
tion in southeast and east Asia considering the possibility that rice cultivation began 
first on the AssaniA'unnan border area, from which agriculturists expanded down the 
Yangtze River, reaching the early farming site of Hemudu by 5000 BC and standing 
at the head of the proto-Austro-Tai linguistic grouping. Movement from the nuclear 
area westwards via the Brahmaputra River into eastern India would have brought the 
Proto-Munda languages. Southward expansion down the Mekong River would have 
been responsible for the dispersal of Proto-Mon-Khmer, and down the Red River for 
Proto-Viet. These conclusions are at present hypothetical, but they do associate the 
early dispersal of rice cultivation with several language families of southeast Asia. 

The effects of the development of farming in New Guinea was significant enough 
for that island, but the economy did not in this case prove an expansive one, partly per- 
haps because it was associated with highland areas. For that reason it is in many ways 
appropriate to see upland New Guinea as a 'residual zone.' Farming there is not the 
product of some dispersal process, but an indigenous development. 

In the case of the Americas there were indeed more localized dispersal processes 
underlying the distributions of the language families indicated in Table 2. But there 
were no continent- wide effects, mainly because the relevant plant domesticates (there 


Table 2. Regions of eaiiy agricultural development and farming dispersals (based on Bellwood 1996). 

Region of early agriculture Associated language families 

sub-Saharan Africa Niger-Kordofanian 

Southwest Asia Elamo-Dravidian 


China (north) Sino-Tibetan 

China (south) Austroasiatic 

Hmong-Mien (Miao-Yao) 

New Guinea Many Papuan families 

Mexico Otomanguean 


Andes/Upper Amazon Chibchan/Paezan 


being few animals) were not so economically decisive as the cereal crops of south- 
western Asia had been in that continent, and were indeed still to prove to be in the 
Americas once transported thither by the European colonists. 

This discussion may be consolidated into some three hypotheses: 

Hypothesis 1. Most language families of 'spread zone' type are the product and re- 
sult of faiTning dispersals. 

Hypothesis 2. Most farming dispersals had clearly identifiable linguistic concomi- 
tants, often resulting in language families of 'spread zone' type. 

Hypothesis 3. Most linguistic configurations of 'residual zone' type are the prod- 
ucts of initial population dispersals into previously uninhabited territories, taking 
place (apart from those in the northern periarctic zone) prior to 10,000 BP. 

As I see the position, from the standpoint of the hoped-for synthesis heralded 
above, this approach has the strength that it recognizes that linguistic replacement 
generally takes place in pace with some degree of population replacement, or at the 
least in situations of marked technical change and acculturation. Languages do not 
change of their own accord in some purely linguistic dimension: Linguistic change is 
a social phenomenon, with appropriate social and economic correlates which, in fa- 
vorable cases, should be visible in the archaeological record. 

The Genetic Dimension 

One of the great reasons for optimism about the new synthesis is that there is a 
steady flow of genetic data which may ultimately serve to resolve some outstanding 
controversies. As noted earlier, there is no expectation of any immediate correlation 
between genetic and linguistic data. Attempts to equate trees of genetic descent with 


those of notional linguistic evolution are to a large extent misleading. This is partly 
because the pace of linguistic evolution is so much greater than genetic. But in par- 
ticular it overlooks the historical dimension. Had there been no major episodes of lan- 
guage replacement, mainly powered by farming dispersals, most of the world's 
language families might well be of 'residual zone' type, having been in position for 
tens of thousands of years. There might then be case for comparing genetic with lin- 
guistic affinity. In reality, however, the genetic map, like the linguistic, has been radi- 
cally modified by these dispersal processes. So indeed there is certainly a strong 
correlation between the language family of a particular ethnic group and its genetic 
composition, but that is because both are the product of replacement in relatively re- 
cent times, as the result of a dispersal process. 

This point may be illustrated effectively by the work of Excoffier and his col- 
leagues, using classical genetic markers, in their examination of African ethnic 
groups, classified by language family (see Renfrew 1992b:464). They examined the 
frequency of the different gamma globulin alleles in a number of African sampling 
populations which were chosen on a tribal {i.e., in effect a linguistic) basis. When they 
classified these in ternis of similarity and difference (by producing a phonetic dendro- 
gram of the kind seen in Figure 1) it turned out that the classification achieved by 
looking at the genetic markers in fact grouped together the Afroasiatic speaking 
tribes, the Niger- Kordofanian speaking tribes, etc. In other words, the genetic charac- 
teristics (as determined from the blood samples) were a good predictor of linguistic 
affinity also. 

Molecular genetics may well be in a position to support or contradict a number of 
the hypotheses set out above. For as I stressed at the outset, the common ground be- 
tween the three disciplines of historical linguistics, molecular genetics and prehis- 
toric archaeology is population history. But the interpretive frameworks are not yet 
fully developed m molecular genetics, nor. as noted earlier, are mutation rates yet 
well established. 

To give an example of the changing fortunes of molecular genetic arguments it 
may be interesting to compare various results relating to the language farming hy- 
pothesis for Europe. In previous papers I have argued that the early dissemination of 
the Proto-Indo-European language m Europe was the product of a farming dispersal 
of the type discussed above, with Anatolia (the northwestern lobe of the southwestern 
Asiatic nuclear fanning area) as the starting pomt. An earlier proposal to account for 
the spread of the Indo-European languages was the mounted warrior horseman hy- 
pothesis — the suggestion that the proto-Indo-European language originated in what 
is now the Ukraine and was carried westwards in the fourth millennium BC in a mili- 
tary conquest activated by the early domestication in that region of the horse. This 
still has its adherents, but its standing has diminished recently with the realization 
that claims for the early domestication of the horse have been exaggerated, and with 
the confirmation that horse riding for military purposes came to Europe only after the 
introduction of the horse and chariot, not long before 1000 BC (Renfrew, in press). 

One of the earliest investigations of the genetic background to the demic diffijsion 
hypothesis for the origins of farming in Europe was offered by Cavalli-Sforza and his 
colleagues (1993), who undertook a principal components analysis of classical ge- 
netic markers among living populations of Europe. The first principal component 
(representing 28% of the variability) showed clear clines from southeast to northwest. 
This was interpreted by them as giving clear indications of the demic diffusion pro- 
cess accompanying the dispersal of agriculture (Figure 8). 


This did not necessarily support the language-farming hypothesis, since it was 
still possible that the dispersal of Proto-Indo-European could be a later event, but it 
was nonetheless taken as support for the importance of the fanning dispersal phe- 
nomenon itself 

A statistical analysis using analogous genetic data for North Africa and other areas 
of Eurasia beyond Europe (Barbujani et al. 1994) produced comparable results for 
Europe and the Near East and (less confidently) for North Africa, thus supporting the 
importance of farming dispersals in southwestern Asia and thus, it was argued, to the 
language-farming hypothesis. 

An interesting study using data from mitochondrial DNA of living populations in 
Europe, recently undertaken by Richards, Sykes and their colleagues (1996) led to 
the identification of a number of mitochondrial lineages (Figure 9). Estimated diver- 
gence times for these suggested that most of them could be dated back into the Upper 
Paleolithic period. It was concluded that the population composition of Europe was 
largely constituted at this early date, and that, while later arrivals could be recognized 
in the data, these were of relatively minor importance. 

It was, however, a weakness of the study by Richards and his colleagues that they 
did not have mtDNA samples from eastern Europe, and few from Anatolia. For if the 
mitochondrial DNA lineages of Anatolia around 7000 BC were already much like 

Figure 8. Major areas of primary domestication of selected principal food plants and distribu tions of se- 
lected language families whose extent is here ascribed to agricultural dispersal. The ar eas of primai^ crop 
domestication are numbered: 1 -sorghum/millet; 2-wheat/barley; 3-millet; 4-Asian rice; 5-taro/sweet po - 
tato. Southeast Asian language families indicated by letters are D-Daic; A-Austroasiatic. (Note: the Indo- 
European and Elamo-Dravidian distributions reflect the hypothetical agincultural dispersals and do not 
show the subsequent spread of the Indo-lranian languages. The agricultural dispersal underlying the Aus - 
tronesian family distribution is believed to have originated in southeast Asia but was based subsequently 
on yam, taro and tree fruits). 



N \ Group 3A 

» 293X-C 

FIGURE 9. Network diagram of mtDN A haplotypes used to suggest that the living population of Europe is 
mainly of Paleolithic origin although it is suggested that group 2a is descended from Neolithic immigrants 
(based on Richards el al. 1 996: ! 92). 


those of Europe, even a significant demic diffusion process might not show up 
strongly in the genetic composition of modem populations. As Comas and his col- 
leagues (1996) remark, following a study of mtDNA and the population history of 
Turkey in relation to Upper Paleolithic and Neolithic dispersals, 

It is intrinsically difficult to separate the genetic effects of these two diachronic 
waves, which had very similar geographic origins and expansion paths. However it is 
possible that the reduced population size during the Upper Paleolithic allowed drift to 
act deeply on gene frequencies but had little effect on sequence diversity, as it is likely 
that the European population did not suffer any narrow bottleneck. In this case, the ef- 
fect of the expansion of farming (that is, a sharp increase in mobility and population 
size) on gene frequencies could have been deep, transforming a random variation pat- 
tern into a cline, but would have had few consequences on mtDNA sequence diversity, 
which would reflect more ancient events (p. 1076). 

The interpretive story does not end there, however. For recently Haeseler, Sajan- 
tila and Paabo (1996; see also Watson et al. 1996) have emphasized a different ap- 
proach to the genetic data, which lays more emphasis upon the history of whole 
populations rather than just of individual lineages. They stress that the study of pair 
wise sequence distributions (in mitochondrial DNA from recent populations under 
study) can illuminate population history. They show that different patterns are to be 
observed in groups which at an earlier time have expanded notably in size (as a result 
of agricultural expansion) in comparison with those whose size has remained con- 
stant over time, as is the case with hunter-gatherers. Paabo (1996) has gone further 
and suggested that the mutation rate between different regions of the mitochondrial 
DNA is variable, and may have been underestimated. In this case the divergence 
times estimated by Richards and his colleagues would be serious overestimates, and 
the patterning in their data might after all be consistent with the demic diffusion 
model for agricultural dispersal in Europe. 

I have set out these various approaches to the problem because I think that they 
give a very clear idea of the dynamism of the current situation. Within a few years that 
situation will have become very much clearer, and the genetic evidence will have 
made a very significant contribution, as it is already beginning to do, to our under- 
standing of population history. I do not doubt that this will also have significant impli- 
cations for historical linguistics, as indeed it will for prehistoric archaeology. 


In the preceding section I have sought to indicate that the application of molecular 
genetics, which has only recently been brought effectively to bear upon some of these 
issues, is likely to transform our understanding of human population history. If it does 
we can expect to learn much more about the operation of various social processes, 
such as human migration, population replacement or demic diffusion, which are di- 
rectly relevant to language history. In particular the language-farming hypothesis, 
that a significant component of modem language diversity may be explained as the 
consequence of farming dispersals, may be open to testing. Our picture of the origins 
of modem world linguistic diversity is likely to develop rapidly over the next decade 
or so. 

Whether there will be comparable progress in the question of the much earlier ori- 
gins of language as a general attribute of our species is more doubtful. There are 
grounds foroptimism that molecular genetics will indeed contribute further to our un- 


derstanding of the early history of human evolution. But no secure framework of in- 
ference has yet been constructed such as would permit one to infer the capacity for 
language from material culture, or indeed from the fossil remains of our early ances- 
tors. In consequence the patterns of veiy early cultural diversity discussed in the first 
part of this paper may continue to tantalize us without offering any secure grounds for 
identifying or dating the emergence of speech by means of the archaeological record. 

Literature Cited 

Austerlitz, R. 1980. Language family density in North America and Eurasia Ural-Altaisches 
Jahrbucher 52:1-10. 

Barbujani, G., A. Pilastro, S. de Domenico, & C. Renfrew. 1994. Genetic variation in North 
Africa and Eurasia: Neolithic demic diffusion vs. Paleolithic colonization. Amer. Jour. 
Phys. Anthropol. 95:137-54. 

Bellwood, P. 1996. The origins and spread of agriculture in the Indo-Pacitlc region: Gradual- 
ism and diffusion or revolution and colonization. Pages 465^98 in D.R. Harris, ed.. The 
Origins and Spread of Agriculture and Pastoralism in Eurasia UCL Press, London, U.K. 

Bloch, M. 1 991 . Language, anthropology and cognitive science. Man 26: 1 83-98. 

Burenhult, G. 1993. The First Humans . Harper Collins, New York. 

Cavalli-Sforza, L.L., P. Menozzi, & A. Piazza . 1993. Demic expansion and human evolution. 
Science 259:639-646. 

Clarke, D.L. 1968. Analytical Archaeology. Methuen, London, U.K. 

Comas, D., F. Calafell, E. Mateu, L. Perez-Lezaun, & J. Bertranpetit. 1996. Geographic varia- 
tion in human mitochondrial DNA control region sequence: the population history of Tur- 
key and its relationship to the European populations, Molec. Biol. Evol. 13:1067-1077 

Davidson I. & W. Noble. 1993, Tools and language in human evolution. Pages 363-388 in 
K.R. Gibson & T. Ingold, eds.. Tools, Language and Cognition in Human Evolution. Cam - 
bridge University Press, Cambridge, U.K. 

Donald, M. 1991. Origins of the Modern Mind.Harvard Um versify Press, Cambridge. 

Ehret, C. 1988. Language change and the material correlates of language and ethnic shift. An- 
tiquity 62:564-573. 

Glover, I. C. & C.F.W. Higham. 1996. New evidence for nee cultivation in south southeast 
and east Asia, Pages 413-441 in D.R. Harris, ed.. The Origins and Spread of Agriculture 
and Pastoralism in Eurasia. UCL Press, London, U.K. 

von Haeseler, A., A. Sajantila, & S. Paabo. 1996. The genetical archaeology of the human ge- 
nome. Nature Genetics 14:135-140. 

Nichols, J. 1992. Linguistic Diversity in Space and Time University of Chicago Press, Chi- 

Oakley, K.P. 1967. Man the Tool-Maker. British Museum (Natural History), London, U.K. 

Paabo, S. 1996. Mutational hot spots in the mitochondrial microcosm. Amer. Jour. Hum. 
Genet. 59:493-496. 

Phillipson, D. W. 1977. The spread of the Bantu languages. Sci. Amer. 236: 106-i 14. 

Renfrew, C. 1982. Towards an Archaeology of Mind. Cambridge University Press, Cam- 
bndge, U.K. 

. 1992a. World languages and human dispersals: A minimalist view. Pages 1 1-68 /// 

J. A. Hall & I.e. Garvie, eds.. Transition to Modernity: Essays on Power, Wealth and Be- 
lief. Cambridge University Press, Cambridge, U.K. 

.1992b Archaeology, genetics and linguistic diversity, Man 27:445-78. 


. 1996a. Language families and the spread of farming. Pages 70-92 //; D.R. Hairis, ed.. 

The Origins and Spread of Agriculture and Pastoral ism in Eurasia. UCL Press, London, 
. 1996b. The Sapient Paradox: How to test for potential. Pages 11-14 in P. Meliars& K. 

Gibson, eds.. Modeling the Early Human Mind. McDonald Institute, Cambridge, U.K.. 
. in press. All the King's Horses: Assessing cognitive maps in later prehistoric Europe, 

in S. Mithen, ed., forthcoming. 

Richards, M., H. Corte-Real , P. Forster , V. Macaulay , H. Wilkinson-Herbots, A. Demaine , S. 
Papiha, R. Hedges , H.-J. Bandelt, & B. Sykes. 1996. Paleolithic and neolithic lineages in 
the European mitoochondrial gene pool Amer. Jour. Hum. Genet. 59:185-203. 

Searle, J. 1995. The Construction of Social Reality. Allen Lane, Hamiondsworth, U.K. 

Toth N. & K. Schick. 1 993. Early stone industries and inferences regarding language and cog- 
nition. Pages 346-362 in K.R. Gibson & T. Ingold, eds.. Tools, Language and Cognition in 
Human Evolution. Cambridge University Press, Cambridge, U.K. 

Watson E., K. Butler, R. Aman, G. Weiss, A. von Haeseler, & S. Paabo. 1996. mtDNA se- 
quence diversity in Africa. Amer. Jour. Hum. Genet. 59:437^44. 

Wynn, T. 1993. Layers of thinking in tool behavior. Pages 389^06 m K.R. Gibson & T. In- 
gold, eds.. Tools, Language and Cognition in Human Evolution. Cambridge University 
Press, Cambridge, U.K. 

Zvelebil, M. 1996. The agricultural frontier and the transition to farming in the circum-Baltic 
region. Pages 323-345 in D.R. Harris, ed.. The Origins and Spread of Agriculture and Pas- 
toralism in Eurasia. UCL Press, London, U.K. 



Acheulian tool tradition 27 

influence on language dispersal 171, 174, 182, 184 

rice cultivation 185 
alarm calls 84-85 

bird 5 

mammal 5 

meaning of 6, 71 

vervet monkeys 4-6, 71, 74, 84 
allometric analysis 45, 47; see also brain, size of 
art, emergence of 92 
Australopithecus africanus 49 

body mass 50 

brain size 50, 52 



amygdala 84 

Broca's area 37, 43, 69, 82 
development of 54 
enlargement of 35, 49 

in relation to body size 52; see also brain, scaling of 

in relation to language 30 

in relation to tool making 52 
evolution of 35 
imaging technologies 69 
lateralization 37, 42-43 
metabolic expense of 25 
occipital cortex 80 
organization of 70 

semantic knowledge 70, 76-77, 79-80, 85 
parietal lobes 76 

PET (positron emission tomography) studies of 77,79 
prefrontal cortex 3 1 
scaling of 35,44,49,52 
semantic networks 74 
semantic representation in 72, 74, 76-77 
size of 25 

"cerebral rubicon" 42 

allometric analysis 45 

in fossil hominins 49-50 

in modem apes 50 

in primates 47, 49 

196 INDEX 

explanations of 52 
in relation to body size 44, 59 
in relation to gestation period 54, 58-59 
in relation to group size 28, 61, 63 
in relation to language 36-37, 63 
in relation to metabolic rate 53-54, 59; see also maternal 

energy hypothesis 
neocortex ratio 28, 61, 98; see also brain, scaling of 
temporal lobes 75, 82 
visual system, see primate visual system 
Wernicke's area 37, 43, 69 
Broca's area, see brain 

cave art 92, 175, 180 

linguistic ability of 30 
cognitive evolution 177 
communication, animal 1 

alarm calls, see alarm calls 

audience effects 9 

food calls, see food calls 

functional reference 4 

intentionality 9 

lexicoding 1 1 

meaning of 2 

deception 5 
presence of syntax 15 
symbolic content 2^ 

phonocoding 11-13 

presence of syntax 9. 11 

signaling 8 


birdsong 11-13 
cranial capacity 42; see brain, size of 


decoration 175; see also cave art 
personal 92, 98 



of early //owo 25 

quality of, in relation to cognitive evolution 29 

relation to interpersonal relationships 29 

analysis of living human populations 188, 190 
Neanderthals 188; see Neanderthals, DNA 

INDEX 197 

domestication 187; see also agriculture 
of horses 187 


ethnolinguistic groups 141-142 

of early Homo sapiens 1 4 1 

farming 182; see agriculture 
feeding behavior 

relation to cognitive evolution 28 
food calls 4 

bird 5 

mammal 5 

meaning of 6 


gestation period 59; see life-history characteristics 
glottochronology 128 
grammar 71, 118-119 

generative 71 

in Neanderthals 102 

universal 37, 120 
grooming 28, 36, 63 

in relation to group size 28 

vocal 28 
group size 28, 31 

in relation to evolution of language 27-28 


handaxes 27, 99, 176-177, 179; see also Acheulian tool tradition; tools, stone 
Hoabhinian tool tradition 179 
hominin evolution 

mosaic nature of 41 

origin of hominins 41 
hominin evolution 42; see also Homo 
Homoerectus 43,97, 142. 176 

languages of 142-143 

speech capabilities 43 
Homo ergaster 25; see also Homo, early 

body proportions 28 

diet 25,28 

facial expressions 30 

198 INDEX 

jaw(s) 26 

larynx (vocal tract) 26, 30 

linguistic abilities 30 

Locomotion 28 

mimetic ability 28 
Homo habilis 97; see Homo, early 
Homo neanderthalensis 44; see also Neanderthals 
Homo sapiens 43, 174 

emergence of 176 

informavore niche 117, 125 

size of early populations 141 
Homo, early 42,97, 179 

anatomical requirement for speech 26 

linguistic capabilities 64 

sexual dimorphism in 42 
Homo, modem 

anatomical distinctions 37 

behavioral distinctions 37 
human evolution, see hominin evolution 41 


archaeological evidence of 90, 97, 99, 101, 103, 176, 191 

as an adaptation 1 17, 122-125 

development of 102, 123 

dispersal of 127 

distinctive features of 23 


in relation to agricultural dispersals 190; see also agriculture 

in relation to environmental parameters 141 
evolution of 117, 141-142, 181 

relation to tool making 27 
families 127-128,171-172,183 

family trees 128-130 

residual zones 182,185 

spread in relation to agriculture 171; see also agriculture 

spread zones 182-183,186 
in Upper Paleolithic populations 97,99-101, 104 
Neanderthal 102; see also Neanderthals, linguistic abilities of 
mental dictionary concept 118 
origin of 89, 104, 127, 182 
languages, see linguistic stocks 
age of 136-139 

compared to archaeological estimates 138 
descent of 127-128, 141 

in relation to environmental parameters 136 

in relation to group size 141 

INDEX 199 

extinct 130, 134 
fade-out threshold 127 
genetic diversity of 134 
genetic markers in 130, 133, 166 

pronomial prefixes 132, 166 
shared cognates 134 
initial branching of 136 
isolates 130 
monogenesis of 140 
shallow families 130 
spread of, see linguistic stocks, spread of 
structural features 143 

historical markers 145, 148, 151, 157, 163 
ergativity 145 
head-marking 145, 152 
numeral classifiers 152,167 
larynx (vocal tract), human 21, 26, 43, 122 

Neanderthal 24,43, 102 
lexicoding 11; see also communication, animal 
life-history characteristics 

of early hominins 29 
limbic system 27 
linguistic provinces 152 

Pacific Hinterland 152, 160 
Pacific Rim 152, 157, 163, 165 
linguistic stocks 134 
Africa 138 
Afroasiatic 139 

age of 140 
American 138, 157 
Australia 138, 152, 160 
density of 136 
half-life calculation 140,162 
maximum age 139 
New Guinea 138, 141, 162. 164, 185 
spread of 143, 148, 164-165 

around the Pacific rim 152, 157, 160 

timing 160 
by demic diffusion 184,187 
in relation to agriculture 162; see also agriculture 
into the Americas 157, 160, 162-163, 185 
out of Africa 142 
Lower Paleolithic 177 


maternal energy hypothesis 36, 53-54, 61 

challenges to 57-58 
metabolic turnover 54 

200 INDEX 

in relation to metabolic rate 57 
metabolism 53; see also brain, size of 
Middle Paleolithic 89,99,177; see also Neanderthals 

technology 101, 108 
mimesis 28, 177 

vocal 28 

emergence of 103, 177 

theory of 3 1 
molecular genetics 173,190 

correlation with linguistic data 173, 186 

in the study of human prehistory 172; see also primate evolution 
morphology (linguistic) 120; see also words, morphology 
Mousterian tool tradition 90 

tool typologies 99 
music 95 
musical instruments 92 


Neanderthals 175 

anatomical distinctions of 44 

brain size 24, 35, 44 

culture of 23 

divergence from anatomically modem humans 24, 35, 43, 98 

DNA 44,91 

jaws, in relation to speech 24 

larynx (vocal tract) 24, 102 

linguistic abilities of 31, 97, 102 

replacement of 90 

symbolic behavior in 98 

tools 99 

naming of tool types 100 
neuroimaging, 75; see also brain, imaging technologies 


Olduwan tool tradition 27 

Paranthropus 25 

phonocoding, see communication, animal 1 1 

phonology 121 

pigments 98 

use of 106 
primate evolution 37-38 

INDEX 201 

bias in the primate fossil record 40 

dates of divergence judged from the fossil record 39 

divergence from other mammals 41 

molecular data bearing on 39^1 

origin of primates 40-41 
primate visual system 71, 74-75 

modular design of 74 

object naming 79-80 

of macaques 75 

visual pathway(s) 82 

visual semantics 71 
protolanguage 89, 102, 174, 183 


reproductive strategies 

of early hominins 29 
residual zones 186; see also language, families, residual zones 

sapient paradox 107,181 

semantic network 74; see also brain, semantic representation 

semantic representation 84; see also brain, semantic representation 

signaling 8; see also communication, animal 

social intelligence 31 

in relation to language 28-31 
speech 21 

comprehension of 69, 122 

evolution of 26 

origin of, compared to language 23 

production of 23, 69, 121 
spread zones, 186; see also languages, families, spread zones 
symbolic thinking, evidence of 23, 26, 89, 95, 106 

anatomically modem humans 107 

Neanderthals 23 
syntax 23,29-30,71, 119-120 

in Neanderthals 97, 102 

tongue, human 21, 122 

tool-making 92, 108 


materials for 108 
mental template 100 
naming of tool types 99-100 

202 INDEX 

stone 35, 50, 92; see also handaxes; Neanderthals, tools; Olduwan, 
Acheulian and Mousterian tool traditions99 


Upper Paleolithic 89,99, 104, 171 
hunting patterns 100 

populations 101; see also language, in Upper Paleolithic populations 
revolution 92, 95, 103, 108; see also cave art, decoration, personal, and 

as a linguistic revolution 103, 109, 174 
technology 101, 108 

vervet monkeys, see alarm calls 4 
vision, see primate visual system 71 

loss of 128 
vocal tract, see larynx (vocal tract) 26 

modulation of 27, 30 
vocalizations, see communication, animal 1 1 


Wernicke's area 37; see also brain 
Word(s) 118 

loss of 182 

morphology 120-121 

order of 119 

^^ :? 

vf^mu, if.^.'^P, ^'^ SCIENCES LIBRARY DUP 

3 1853 10002 7486 


aca d emy I 


Distributed by 

the University of California Press 

Printed in the United States of America