ON THE NATURE AND NURTURE OF LANGUAGE

Elizabeth Bates

University of California, San Diego

Support for the work described here has been provided by NIH/NIDCD-R01-DC00216 (“Cross-

linguistic studies in aphasia”), NIH-NIDCD P50 DC1289-9351 (“Origins of communication

disorders”), NIH/NINDS P50 NS22343 (“Center for the Study of the Neural Bases of Language

and Learning”), NIH 1-R01-AG13474 (“Aging and Bilingualism”), and by a grant from the John

D. and Catherine T. MacArthur Foundation Research Network on Early Childhood Transitions.

Please address all correspondence to Elizabeth Bates, Center for Research in Language 0526,

University of California at San Diego, La Jolla, CA 92093-0526, or bates@crl.ucsd.edu.

2

ON THE NATURE AND NURTURE OF LANGUAGE

Elizabeth Bates

Language is the crowning achievement of thehuman species, and it is something that all normalhumans can do. The average man is neither aShakespeare nor a Caravaggio, but he is capable offluent speech, even if he cannot paint at all. In fact, theaverage speaker produces approximately 150 words perminute, each word chosen from somewhere between20,000 and 40,000 alternatives, at error rates below0.1%. The average child is already well on her waytoward that remarkable level of performance by 5 yearsof age, with a vocabulary of more than 6000 words andproductive control over almost every aspect of soundand grammar in her language.

Given the magnitude of this achievement, and thespeed with which we attain it, some theorists haveproposed that the capacity for language must be builtdirectly into the human brain, maturing like an arm or akidney. Others have proposed instead that we havelanguage because we have powerful brains that can learnmany things, and because we are extraordinarily socialanimals who value communication above everythingelse. Is language innate? Is it learned? Or, alterna-tively, does language emerge anew in every generation,because it is the best solution to the problems that wecare about, problems that only humans can solve?These are the debates that have raged for centuries in thevarious sciences that study language. They are alsovariants of a broader debate about the nature of the mindand the process by which minds are constructed inhuman children.

The first position is called “nativism”, defined asthe belief that knowledge originates in human nature.This idea goes back to Plato and Kant, but in moderntimes it is most clearly associated with the linguistNoam Chomsky (see photograph). Chomsky’s viewson this matter are very strong indeed, starting with hisfirst book in 1957, and repeated with great consistencyfor the next 40 years. Chomsky has explicated the tiebetween his views on the innateness of language andPlato's original position on the nature of mind, asfollows:

"How can we interpret [Plato's] proposal inmodern terms? A modern variant would be thatcertain aspects of our knowledge and understandingare innate, part of our biological endowment,genetically determined, on a par with the elementsof our common nature that cause us to grow armsand legs rather than wings. This version of theclassical doctrine is, I think, essentially correct."(Chomsky, 1988, p. 4)

He has spent his career developing an influentialtheory of grammar that is supposed to describe theuniversal properties underlying the grammars of everylanguage in the world. Because this Universal Grammar

is so abstract, Chomsky believes that it could not belearned at all, stating that

“Linguistic theory, the theory of UG[Universal Grammar]... is an innate property of thehuman mind.... [and].... the growth of language [is]analogous to the development of a bodily organ”.

Of course Chomsky acknowledges that Frenchchildren learn French words, Chinese children learnChinese words, and so on. But he believes that theabstract underlying principles that govern language arenot learned at all, arguing that “A general learningtheory ... seems to me dubious, unargued, and withoutany empirical support”.

Because this theory has been so influential inmodern linguistics and psycholinguistics, it is impor-tant to understand exactly what Chomsky means by“innate.” Everyone would agree that there is somethingunique about the human brain that makes languagepossible. But in the absence of evidence to thecontrary, that “something” could be nothing other thanthe fact that our brains are very large, a giant all-purpose computer with trillions of processing elements.Chomsky’s version of the theory of innateness is muchstronger than the “big brain” view, and involves twologically and empirically separate claims: that languageis innate, and that our brains contain a dedicated,special-purpose learning device that has evolved forlanguage alone. The latter claim is the one that isreally controversial, a doctrine that goes under variousnames including “domain specificity”, “autonomy” and“modularity”.

The second position is called “empiricism”, definedas the belief that knowledge originates in theenvironment, and comes in through the senses. Thisapproach (also called “behaviorism” and “associa-tionism”) is also an ancient one, going back (at least) toAristotle, but in modern times it is closely associatedwith the psychologist B.F. Skinner (see photograph). According to Skinner, there are no limits to what ahuman being can become, given time, opportunity andthe application of very general laws of learning.Humans are capable of language because we have thetime, the opportunity and (perhaps) the computingpower that is required to learn 50,000 words and theassociations that link those words together. Much ofthe research that has taken place in linguistics,psycholinguistics and neurolinguistics since the 1950’shas been dedicated to proving Skinner wrong, byshowing that children and adults go beyond their input,creating novel sentences and (in the case of normalchildren and brain-damaged adults) peculiar errors thatthey have never heard before. Chomsky himself hasbeen severe in his criticisms of the behaviorist approachto language, denouncing those who believe that

3

language can be learned as “grotesquely wrong”(Gelman, 1986).

In their zealous attack on the behaviorist approach,nativists sometimes confuse Skinner’s form ofempiricism with a very different approach, alternativelycalled “interactionism”, “constructivism,” and “emer-gentism.” This is a much more difficult idea than eithernativism or empiricism, and its historical roots are lessclear. In the 20th century, the interactionist orconstructivist approach has been most closely associatedwith the psychologist Jean Piaget (see photograph).More recently, it has appeared in a new approach tolearning and development in brains and brain-likecomputers alternatively called “connectionism,” “paral-lel distributed processing” and “neural networks”(Elman et al., 1996; Rumelhart & McClelland, 1986),and in a related theory of development inspired by thenonlinear dynamical systems of modern physics (Thelen& Smith, 1994). To understand this difficult butimportant idea, we need to distinguish between twokinds of interactionism: simple interactions (black andwhite make grey) and emergent form (black and whiteget together and something altogether new and differenthappens).

In an emergentist theory, outcomes can arise forreasons that are not obvious or predictable from any ofthe individual inputs to the problem. Soap bubbles areround because a sphere is the only possible solution toachieving maximum volume with minimum surface(i.e., their spherical form is not explained by the soap,the water, or the little boy who blows the bubble). Thehoneycomb in a beehive takes an hexagonal formbecause that is the stable solution to the problem ofpacking circles together (i.e., the hexagon is notpredictable from the wax, the honey it contains, norfrom the packing behavior of an individual bee—seeFigure 1). Jean Piaget argued that logic and knowledgeemerge in just such a fashion, from successiveinteractions between sensorimotor activity and astructured world. A similar argument has been made toexplain the emergence of grammars, which represent theclass of possible solutions to the problem of mapping arich set of meanings onto a limited speech channel,heavily constrained by the limits of memory, perceptionand motor planning. Logic and grammar are not givenin the world, but neither are they given in the genes.Human beings discovered the principles that compriselogic and grammar, because these principles were thebest possible solution to specific problems that otherspecies just simply do not care about, and could notsolve even if they did. Proponents of the emergentistview acknowledge that something is innate in thehuman brain that makes language possible, but that“something” may not be a special-purpose, domain-specific device that evolved for language and languagealone. Instead, language may be something that we dowith a large and complex brain that evolved to servethe many complex goals of human society and culture(Tomasello & Call, 1997). In other words, language is

a new machine built out of old parts, reconstructed fromthose parts by every human child.

So the debate today in language research is notabout Nature vs. Nurture, but about the “nature ofNature,” that is, whether language is something that wedo with an inborn language device, or whether it is theproduct of (innate) abilities that are not specific tolanguage. In the pages that follow, we will explorecurrent knowledge about the psychology, neurology anddevelopment of language from this point of view. Wewill approach this problem at different levels of thesystem, from speech sounds to the broader com-municative structures of complex discourse. Let usstart by defining the different levels of the languagesystem, and then go on to describe how each of theselevels is processed by normal adults, acquired bychildren, and represented in the brain.

I. THE COMPONENT PARTS OFLANGUAGE

Speech as Sound: Phonetics and PhonologyThe study of speech sounds can be divided into two

subfields: phonetics and phonology . Phonetics is the study of speech sounds as physical

and psychological events. This includes a huge body ofresearch on the acoustic properties of speech, and therelationship between these acoustic features and the waythat speech is perceived and experienced by humans. Italso includes the detailed study of speech as a motorsystem, with a combined emphasis on the anatomy andphysiology of speech production. Within the field ofphonetics, linguists work side by side with acousticalengineers, experimental psychologists, computerscientists and biomedical researchers.

Phonology is a very different discipline, focused onthe abstract representations that underlie speech in bothperception and production, within and across humanlanguages. For example, a phonologist may concen-trate on the rules that govern the voiced/voicelesscontrast in English grammar, e.g., the contrast betweenthe unvoiced “-s” in “cats” and the voiced “-s” in “dogs”.This contrast in plural formation bears an uncannyresemblance to the voiced/unvoiced contrast in Englishpast tense formation, e.g., the contrast between anunvoiced “-ed” in “walked” and a voiced “-ed” in“wagged”. Phonologists seek a maximally general setof rules or principles that can explain similarities ofthis sort, and generalize to new cases of word formationin a particular language. Hence phonology lies at theinterface between phonetics and the other regularitiesthat constitute a human language, one step removedfrom sound as a physical event.

Some have argued that phonology should not existas a separate discipline, and that the generalizationsdiscovered by phonologists will ultimately be explainedentirely in physical and psychophysical terms. Thistends to be the approach taken by emergentists. Othersmaintain that phonology is a completely independentlevel of analysis, whose laws cannot be reduced to anycombination of physical events. Not surprisingly, this

4

tends to be the approach taken by nativists, especiallythose who believe that language has its very owndedicated neural machinery. Regardless of one’sposition on this debate, it is clear that phonetics andphonology are not the same thing. If we analyze speechsounds from a phonetic point of view, based on all thedifferent sounds that a human speech apparatus canmake, we come up with approximately 600 possiblesound contrasts that languages could use (even more, ifwe use a really fine-grained system for categorizingsounds). And yet most human languages use no morethan 40 contrasts to build words.

To illustrate this point, consider the followingcontrast between English and French. In English, theaspirated (or "breathy") sound signalled by the letter “h-”is used phonologically, e.g., to signal the differencebetween “at” and “hat". French speakers are perfectlycapable of making these sounds, but the contrast createdby the presence or absence of aspiration (“h-”) is notused to mark a systematic difference between words;instead, it is just a meaningless variation that occursnow and then in fluent speech, largely ignored bylisteners. Similarly, the English language has a binarycontrast between the sounds signalled by “d” and “t”,used to make systematic contrasts like “tune” and“dune.” The Thai language has both these contrasts,and in addition it has a third boundary somewhere inbetween the English “t” and “d”. English speakers areable to produce that third boundary; in fact, it is thenormal way to pronounce the middle consonant in aword like “butter”. The difference is that Thai uses thatthird contrast phonologically (to make new words), butEnglish only uses it phonetically, as a convenient wayto pronounce target phonemes while hurrying from oneword to another (also called “allophonic variation”). Inour review of studies that focus on the processing,development and neural bases of speech sounds, it willbe useful to distinguish between the phonetic approach,and phonological or phonemic approach.

Speech as Meaning: Semantics and theLexicon

The study of linguistic meaning takes place withina subfield of linguistics called semantics . Semanticsis also a subdiscipline within philosophy, where therelationship between meaning and formal logic isemphasized. Traditionally semantics can be divided intotwo areas: lexical semantics , focussed on themeanings associated with individual lexical items (i.e.,words), and propositional or relational seman-t ics , focussed on those relational meanings that wetypically express with a whole sentence.

Lexical semantics has been studied by linguistsfrom many different schools, ranging from the heavilydescriptive work of lexicographers (i.e., “dictionarywriters”) to theoretical research on lexical meaning andlexical form in widely different schools of formallinguistics and generative grammar (McCawley, 1993).Some of these theorists emphasize the intimaterelationship between semantics and grammar, using a

combination of lexical and propositional semantics toexplain the various meanings that are codified in thegrammar. This is the position taken by many theoristswho taken an emergentist approach to language,including specific schools with names like “cognitivegrammar,” “generative semantics” and/or “linguisticfunctionalism”. Other theorists argue instead for thestructural independence of semantics and grammar, aposition associated with many of those who espouse anativist approach to language.

Propositional semantics has been dominatedprimarily by philosophers of language, who areinterested in the relationship between the logic thatunderlies natural language and the range of possiblelogical systems that have been uncovered in the last twocenturies of research on formal reasoning. Aproposition is defined as a statement that can be judgedtrue or false. The internal structure of a propositionconsists of a predicate and one or more arguments ofthat predicate. An argument is an entity or “thing” thatwe would like to make some point about. A one-placepredicate is a state, activity or identity that we attributeto a single entity (e.g., we attribute beauty to Mary inthe sentence “Mary is beautiful”, or we attribute“engineerness” to a particular individual in the sentence“John is an engineer.”); an n-place predicate is arelationship that we attribute to two or more entities orthings. For example, the verb "to kiss" is a two-placepredicate, which establishes an asymmetric relationshipof “kissing” to two entities in the sentence “John kissesMary.”, The verb "to give" is a three-place predicatethat relates three entities in a proposition expressed bythe sentence “John gives Mary a book..” Philosopherstend to worry about how to determine the truth orfalsity of propositions, and how we convey (or hide)truth in natural language and/or in artificial languages.Linguists worry about how to characterize ortaxonomize the propositional forms that are used innatural language. Psychologists tend instead to worryabout the shape and nature of the mental representationsthat encode propositional knowledge, with develop-mental psychologists emphasizing the process by whichchildren attain the ability to express this propositionalknowledge. Across fields, those who take a nativistapproach to the nature of human language tend toemphasize the independence of propositional orcombinatorial meaning from the rules for combiningwords in the grammar; by contrast, the variousemergentist schools tend to emphasize both thestructural similarity and the causal relationship betweenpropositional meanings and grammatical structure,suggesting that one grows out of the other.

How Sounds and Meanings Come Together:Grammar

The subfield of linguistics that studies howindividual words and other sounds are combined toexpress meaning is called grammar. The study ofgrammar is traditionally divided into two parts:morphology and syntax.

5

Morphology refers to the principles governing theconstruction of complex words and phrases, for lexicaland/or grammatical purposes. This field is furtherdivided into two subtypes: derivational morpho-logy and inflectional morphology .

Derivational morphology deals with theconstruction of complex content words from simplercomponents, e.g., derivation of the word “government”from the verb “to govern” and the derivationalmorpheme “-ment”. Some have argued that derivationalmorphology actually belongs within lexical semantics,and should not be treated within the grammar at all.However, such an alignment between derivationalmorphology and semantics describes a language likeEnglish better than it does richly inflected languageslike Greenlandic Eskimo, where a whole sentence mayconsist of one word with many different derivational andinflectional morphemes.

Inflectional morphology refers to modulations ofword structure that have grammatical consequences,modulations that are achieved by inflection (e.g.,adding an “-ed” to a verb to form the past tense, as in"walked") or by suppletion (e.g., substituting theirregular past tense “went” for the present tense “go”).Some linguists would also include within inflectionalmorphology the study of how free-standing functionwords (like "have", "by", or "the", for example) areadded to individual verbs or nouns to build up complexverb or noun phrases, e.g., the process that expands averb like “run” into “has been running” or the processthat expands a noun like “dog” into a noun phrase like“the dog” or prepositional phrase like “by the dog”.

Syntax is defined as the set of principles thatgovern how words and other morphemes are ordered toform a possible sentence in a given language. Forexample, the syntax of English contains principles thatexplain why “John kissed Mary” is a possible sentencewhile “John has Mary kissed” sounds quite strange.Note that both these sentences would be acceptable inGerman, so to some extent these rules and constraintsare arbitrary. Syntax may also contain principles thatdescribe the relationship between different forms of thesame sentence (e.g., the active sentence “John hit Bill”and the passive form “Bill was hit by John”), and waysto nest one sentence inside another (e.g., “The boy thatwas hit by John hit Bill”).

Languages vary a great deal in the degree to whichthey rely on syntax or morphology to express basicpropositional meanings. A particularly good exampleis the cross-linguistic variation we find in means ofexpressing a propositional relation called t ransi t iv i ty(loosely defined as “who did what to whom”). Englishuses word order as a regular and reliable cue to sentencemeaning (e.g., in the sentence "John kissed a girl", weimmediately know that "John" is the actor and "girl" isthe receiver of that action). At the same time, Englishmakes relatively little use of inflectional morphology toindicate transitivity or (for that matter) any otherimportant aspect of sentence meaning. For example,there are no markers on "John" or "girl" to tell us who

kissed whom, nor are there any clues to transitivitymarked on the verb "kissed". The opposite is true inHungarian, which has an extremely rich morphologicalsystem but a high degree of word order variability.Sentences like “John kissed a girl” can be expressed inalmost every possible order in Hungarian, without lossof meaning.

Some linguists have argued that this kind of wordorder variation is only possible in a language with richmorphological marking. For example, the Hungarianlanguage provides case suffixes on each noun thatunambiguously indicate who did what to whom,together with special markers on the verb that agreewith the object in definiteness. Hence the Hungariantranslation of our English example would be equivalentto “John-actor indefinite-girl-receiver-of-action kissed-indefinite). However, the Chinese language poses aproblem for this view: Chinese has no inflectionalmarkings of any kind (e.g., no case markers, no form ofagreement), and yet it permits extensive word ordervariation for stylistic purposes. As a result, Chineselisteners have to rely entirely on probabilistic cues tofigure out "who did what to whom", including somecombination of word order (i.e., some orders are morelikely than others, even though many are possible) andthe semantic content of the sentence (e.g., boys aremore likely to eat apples than vice-versa). In short, itnow seems clear that human languages have solved thismapping problem in a variety of ways.

Chomsky and his followers have defined UniversalGrammar as the set of possible forms that the grammarof a natural language can take. There are two ways oflooking at such universals: as the intersect of all humangrammars (i.e., the set of structures that every languagehas to have) or as the union of all human grammars(i.e., the set of possible structures from which eachlanguage must choose). Chomsky has alwaysmaintained that Universal Grammar is innate, in a formthat is idiosyncratic to language. That is, grammar doesnot “look like” or behave like any other existingcognitive system. However, he has changed his mindacross the years on the way in which this innateknowledge is realized in specific languages like Chineseor French. In the early days of generative grammar, thesearch for universals revolved around the idea of auniversal intersect. As the huge variations that existbetween languages became more and more obvious, andthe intersect got smaller and smaller, Chomsky beganto shift his focus from the intersect to the union ofpossible grammars. In essence, he now assumes thatchildren are born with a set of innate options that definehow linguistic objects like nouns and verbs can be puttogether. The child doesn’t really learn grammar (in thesense in which the child might learn chess). Instead,the linguistic environment serves as a “trigger” thatselects some options and causes others to wither away.This process is called “parameter setting”. Parametersetting may resemble learning, in that it helps toexplain why languages look as different as they do andhow children move toward their language-specific

6

targets. However, Chomsky and his followers areconvinced that parameter setting (choice from a largestock of innate options) is not the same thing aslearning (acquiring a new structure that was never therebefore learning took place), and that learning in thelatter sense plays a limited and perhaps rather trivial rolein the development of grammar.

Many theorists disagree with this approach togrammar, along the lines that we have already laid out.Empiricists would argue that parameter setting really isnothing other than garden-variety learning (i.e., childrenreally are taking new things in from the environment,and not just selecting among innate options).Emergentists take yet another approach, somewhere inbetween parameter setting and learning. Specifically, anemergentist would argue that some combinations ofgrammatical features are more convenient to processthan others. These facts about processing set limits onthe class of possible grammars: Some combinationswork; some don’t. To offer an analogy, why is it that asparrow can fly but an emu cannot? Does the emu lack“innate flying knowledge,” or does it simply lack arelationship between weight and wingspan that iscrucial to the flying process? The same logic can beapplied to grammar. For example, no language has agrammatical rule in which we turn a statement into aquestion by running the statement backwards, e.g.,

John hit the ball” --> Ball the hit John?

Chomsky would argue that such a rule does notexist because it is not contained within UniversalGrammar. It could exist, but it doesn’t. Emergentistswould argue that such a rule does not exist because itwould be very hard to produce or understand sentences inreal time by a forward-backward principle. It mightwork for sentences that are three or four words long, butour memories would quickly fail beyond that point e.g.,

The boy that kicked the girl hit the ball that Peterbought -->

Bought Peter that ball the hit girl the kicked thatboy the?

In other words, the backward rule for questionformation doesn’t exist because it couldn’t exist, notwith the kind of memory that we have to work with.Both approaches assume that grammars are the way theyare because of the way that the human brain is built.The difference lies not in Nature vs. Nurture, but in the“nature of Nature,” i.e., whether this ability is built outof language-specific materials or put together from moregeneral cognitive ingredients.

Language in a Social Context: Pragmaticsand Discourse

The various subdisciplines that we have reviewedso far reflect one or more aspects of linguistic form,from sound to words to grammar. Pragmatics isdefined as the study of language in context, a fieldwithin linguistics and philosophy that concentratesinstead on language as a form of communication, a toolthat we use to accomplish certain social ends (Bates,

1976). Pragmatics is not a well-defined discipline;indeed, some have called it the wastebasket of linguistictheory. It includes the study of speech acts (ataxonomy of the socially recognized acts ofcommunication that we carry out when we declare,command, question, baptize, curse, promise, marry,etc.), presupposit ions (the background informationthat is necessary for a given speech act to work, e.g.,the subtext that underlies a pernicious question like“Have you stopped beating your wife?”), andconversational postulates (principles governingconversation as a social activity, e.g., the set of signalsthat regulate turn-taking, and tacit knowledge of whetherwe have said too much or too little to make a particularpoint).

Pragmatics also contains the study of discourse.This includes the comparative study of discourse types(e.g., how to construct a paragraph, a story, or a joke),and the study of text cohesion , i.e., the way we useindividual linguistic devices like conjunctions (“and”,“so”), pronouns (“he”, “she”, “that one there”), definitearticles (“the” versus “a”) and even whole phrases orclauses (e.g., “The man that I told you about....”) to tiesentences together, differentiate between old and newinformation, and maintain the identity of individualelements from one part of a story to another (i.e.,coreference relations).

It should be obvious that pragmatics is aheterogeneous domain without firm boundaries.Among other things, mastery of linguistic pragmaticsentails a great deal of sociocultural information:information about feelings and internal states,knowledge of how the discourse looks from thelistener’s point of view, and the relationships of powerand intimacy between speakers that go into calculationsof how polite and/or how explicit we need to be intrying to make a conversational point. Imagine aMartian that lands on earth with a complete knowledgeof physics and mathematics, armed with computers thatcould break any possible code. Despite these powerfultools, it would be impossible for the Martian to figureout why we use language the way we do, unless thatMartian also has extensive knowledge of human societyand human emotions. For the same reason, this is onearea of language where social-emotional disabilitiescould have a devastating effect on development (e.g.,autistic children are especially bad on pragmatic tasks).Nevertheless, some linguists have tried to organizeaspects of pragmatics into one or more independent“modules,” each with its own innate properties (Sperber& Wilson, 1986). As we shall see later, there has alsobeen a recent effort within neurolinguistics to identify aspecific neural locus for the pragmatic aspect oflinguistic knowledge.

Now that we have a road map to the componentparts of language, let us take a brief tour of each level,reviewing current knowledge of how information at thatlevel is processed by adults, acquired by children, andmediated in the human brain.

7

II. SPEECH SOUNDSHow Speech is Processed by Normal Adults

The study of speech processing from apsychological perspective began in earnest after WorldWar II, when instruments became available thatpermitted the detailed analysis of speech as a physicalevent. The most important of these for researchpurposes was the sound spectrograph. Unlike the morefamiliar oscilloscope, which displays sound frequenciesover time, the spectrograph displays changes over timein the energy contained within different frequency bands(think of the vertical axis as a car radio, while thehorizontal axis displays activity on every station overtime). Figure 2 provides an example of a soundspectrogram for the sentence “Is language innate?”—oneof the central questions in this field.

This kind of display proved useful not only becauseit permitted the visual analysis of speech sounds, butalso because it became possible to “paint” artificialspeech sounds and play them back to determine theireffects on perception by a live human being. Initiallyscientists hoped that this device would form the basis ofspeech-reading systems for the deaf. All we would haveto do (or so it seemed) would be to figure out the“alphabet”, i.e., the visual pattern that corresponds toeach of the major phonemes in the language. By asimilar argument, it should be possible to createcomputer systems that understand speech, so that wecould simply walk up to a banking machine and tell itour password, the amount of money we want, and soforth. Unfortunately, it wasn’t that simple. As it turnsout, there is no clean, isomorphic relation between thespeech sounds that native speakers hear and the visualdisplay produced by those sounds. Specifically, therelationship between speech signals and speechperception lacks two critical properties: linearity andinvariance.

Linearity refers to the way that speech unfolds intime. If the speech signal had linearity, then therewould be an isomorphic relation from left to rightbetween speech-as-signal and speech-as-experience in thespeech spectrogram. For example, consider thesyllable “da” displayed in the artificial spectrogram inFigure 3. If the speech signal were linear, then the firstpart of this sound (the “d” component) should corre-spond to the first part of the spectrogram, and thesecond part (the “a” component) should correspond tothe second part of the same spectrogram. However, ifwe play these two components separately to a nativespeaker, they don’t sound anything like two halves of“da”. The vowel sound does indeed sound like a vowel“a”, but the “d” component presented alone (with novowel context) doesn’t sound like speech at all; itsounds more like the chirp of a small bird or a speakingwheel on a rolling chair. It would appear that ourexperience of speech involves a certain amount ofreordering and integration of the physical signal as itcomes in, to create the unified perceptual experience thatis so familiar to us all.

Invariance refers to the relationship between thesignal and its perception across different contexts. Eventhough the signal lacks linearity, scientists once hopedthat the same portion of the spectrogram that elicits the“d” experience in the context of “di” would alsocorrespond to the “d” experience in the context of “du”.Alas, that has proven not to be the case. As Figure 3shows, the component responsible for “d” looks entirelydifferent depending on the vowel that follows. Worsestill, the “d” component of the syllable “du” looks likethe “g” component of the syllable “ga”. In fact, theshape of the visual pattern that corresponds to aconstant sound can even vary with the pitch of thespeaker’s voice, so that the “da” produced by a smallchild results in a very different-looking pattern from the“da’ produced by a mature adult male.

These problems can be observed in clean, artificialspeech stimuli. In fluent, connected speech theproblems are even worse (see word perception, below).It seems that native speakers use many different parts ofthe context to break the speech code. No simple“bottom-up” system of rules is sufficient to accomplishthis task. That is why we still don’t have speechreaders for the deaf or computers that perceive fluentspeech from many different listeners, even though suchmachines have existed in science fiction for decades.

The problem of speech perception got “curiouserand curiouser” as Lewis Carroll would say, leading anumber of speech scientists in the 1960’s to proposethat humans accomplish speech perception via a special-purpose device unique to the human brain. For reasonsthat we will come to shortly, they were also persuadedthat this “speech perception device” is innate, up andrunning in human babies as soon as they are born. Itwas also suggested that humans process these speechsounds not as acoustic events, but by testing the speechinput against possible “motor templates” (i.e., versionsof the same speech sound that the listener can producefor himself, a kind of “analysis by synthesis”). Thisidea, called the Motor Theory of Speech Perception, wasoffered to explain why the processing of speech isnonlinear and invariant from an acoustic point of view,and why only humans (or so it was believed) are able toperceive speech at all.

For a variety of reasons (some discussed below)this hypothesis has fallen on hard times. Today we finda large number of speech scientists returning to the ideathat speech is an acoustic event after all, albeit a verycomplicated one that is hard to understand by looking atspeech spectrograms like the ones in Figures 2-3. Forone thing, researchers using a particular type ofcomputational device called a “neural network” haveshown that the basic units of speech can be learned afterall, even by a rather stupid machine with access tonothing other than raw acoustic speech input (i.e., no“motor templates” to fit against the signal). So theability to perceive these units does not have to beinnate; it can be learned. This brings us to the nextpoint: how speech develops.

8

How Speech Sounds Develop in ChildrenSpeech Perception. A series of clever

techniques has been developed to determine the set ofphonetic/phonemic contrasts that are perceived bypreverbal infants. These include High-AmplitudeSucking (capitalizing on the fact that infants tend tosuck vigorously when they are attending to aninteresting or novel stimulus), habituation anddishabituation (relying on the tendency for small infantsto “orient” or re-attend when they perceive an interestingchange in auditory or visual input), and operantgeneralization (e.g., training an infant to turn her headto the sounds from one speech category but not another,a technique that permits the investigator to map out theboundaries between categories from the infant’s point ofview). (For reviews of research using these techniques,see Aslin, Jusczyk, & Pisoni, in press). Although thetechniques have remained constant over a period of 20years or more, our understanding of the infant’s initialrepertoire and the way that it changes over time hasundergone substantial revision.

In 1971, Peter Eimas and colleagues published animportant paper showing that human infants are able toperceive contrasts between speech sounds like “pa” and“ba” (Eimas, Siqueland, Jusczyk, & Vigorito, 1971).Even more importantly, they were able to show thatinfants hear these sounds categorically. To illustratethis point, the reader is kindly requested to place onehand on her throat and the other (index finger raised) justin front of her mouth. Alternate back and forth betweenthe sounds “pa” and “ba”, and you will notice that themouth opens before the vocal chords rattle in making a“pa”, while the time between vocal chord vibrations andlip opening is much shorter when making a “ba.” Thisvariable, called Voice Onset Time or VOT, is acontinuous dimension in physics but a discontinuousone in human perception. That is, normal listeners heara sharp and discontinuous boundary somewhere around+20 (20 msec between mouth opening and voice onset);prior to that boundary, the “ba” tokens sound very muchalike, and after that point the “pa” tokens are difficult todistinguish, but at that boundary a dramatic shift can beheard. To find out whether human infants have such aboundary, Eimas et al. used the high-amplitude suckingprocedure, which means that they set out to habituate(literally “bore”) infants with a series of stimuli fromone category (e.g., “ba”), and then presented with newversions in which the VOT is shifted gradually towardthe adult boundary. Sucking returned with a vengeance(“Hey, this is new!”), a sharp change right at the sameborder at which human adults hear a consonantalcontrast.

Does this mean that the ability to hear categoricaldistinctions in speech is innate? Yes, it probably does.Does is also mean that this ability is based on an innateprocessor that has evolved exclusively for speech?Eimas et al. thought so, but history has shown thatthey were wrong. Kuhl & Miller (1975) made adiscovery that was devastating for the nativist approach:Chinchillas can also hear the boundary between

consonants, and they hear it categorically, with aboundary at around the same place where humans hearit. This finding has now been replicated in variousspecies, with various methods, looking at manydifferent aspects of the speech signal. In other words,categorical speech perception is not domain specific, anddid not evolve in the service of speech. The ear did notevolve to meet the human mouth; rather, human speechhas evolved to take advantage of distinctions that werealready present in the mammalian auditory system.This is a particularly clear illustration of the differencebetween innateness and domain specificity outlined inthe introduction.

Since the discovery that categorical speechperception and phenomena are not peculiar to speech,the focus of interest in research on infant speechperception has shifted away from interest in the initialstate to interest in the process by which children tunetheir perception to fit the peculiarities of their ownnative language. We now know that newborns can hearvirtually every phonetic contrast used by humanlanguages. Indeed, they can hear things that are nolonger perceivable to an adult. For example, Japaneselisteners find it very difficult to hear the contrastbetween “ra” and “la”, but Japanese infants have notrouble at all with that distinction. When do we lose(or suppress) the ability to hear sounds that are not inour language? Current evidence suggests that thesuppression of non-native speech sounds beginssomewhere between 8-10 months of age—the point atwhich most infants start to display systematic evidenceof word comprehension. There is, it seems, no suchthing as a free lunch: In order to “tune in” to language-specific sounds in order to extract their meaning, thechild must “tune out” to those phonetic variations thatare not used in the language.

This does not mean, however, that children are“language neutral” before 10 months of age. Studieshave now shown that infants are learning somethingabout the sounds of their native language in utero!French children turn preferentially to listen to French inthe first hours and days of life, a preference that is notshown by newborns who were bathed in a differentlanguage during the last trimester. Whatever the Frenchmay believe about the universal appeal of theirlanguage, it is not preferred universally at birth.Something about the sound patterns of one’s nativelanguage is penetrating the womb, and the ability tolearn something about those patterns is present in thelast trimester of pregnancy. However, that “something”is probably rather vague and imprecise. Kuhl and hercolleagues have shown that a much more precise formof language-specific learning takes place between birthand six months of age, in the form of a preference forprototypical vowel sounds in that child’s language.Furthermore, the number of ‘preferred vowels” is tunedalong language-specific lines by six months of age.Language-specific information about consonants appearsto come in somewhat later, and probably coincides with

9

the suppression of non-native contrasts and the abilityto comprehend words.

To summarize, human infants start out with auniversal ability to hear all the speech sounds used byany natural language. This ability is innate, but it isapparently not specific to speech (nor specific tohumans). Learning about the speech signal begins assoon as the auditory system is functional (somewhere inthe last trimester of pregnancy), and it proceedssystematically across the first year of life until childrenare finally able to weed out irrelevant sounds and tuneinto the specific phonological boundaries of their nativelanguage. At that point, mere speech turns into reallanguage, i.e., the ability to turn sound into meaning.

Speech production. This view of the develop-ment of speech perception is complemented by findingson the development of speech production (for details seeMenn & Stoel-Gammon, 1995).

In the first two months, the sounds produced byhuman infants are reflexive in nature, “vegetativesounds” that are tied to specific internal states (e.g.,crying). Between 2-6 months, infants begin to producevowel sounds (i.e., cooing and sound play). So-calledcanonical or reduplicative babbling starts between 6-8months in most children: babbling in short segments orin longer strings that are now punctuated by consonants(e.g., "dadada"). In the 6-12-month window, babbling“drifts” toward the particular sound patterns of thechild’s native language (so that adult listeners candistinguish between the babbling of Chinese, Frenchand Arabic infants). However, we still do not knowwhat features of the infants’ babble lead to thisdiscrimination (i.e., whether it is based on consonants,syllable structure and/or the intonational characteristicsof infant speech sounds). In fact, some investigatorsinsist that production of consonants may be relativelyimmune to language-specific effects until the secondyear of life.

Around 10 months of age, some children begin toproduce "word-like sounds", used in relatively consistentways in particular contexts (e.g., "nana" as a soundmade in requests; "bam!" pronounced in games ofknocking over toys). From this point on (if notbefore), infant phonological development is stronglyinfluenced by other aspects of language learning (i.e.,grammar and the lexicon). There is considerablevariability between infants in the particular speechsounds that they prefer. However, there is clearcontinuity from prespeech babble to first words in anindividual infant’s “favorite sounds”. This findingcontradicts a famous prediction by the linguist RomanJakobson, who believed that prespeech babble andmeaningful speech are discontinuous. Phonologicaldevelopment has a strong influence on the first wordsthat children try to produce (i.e., they will avoid the useof words that they cannot pronounce, and collect newwords as soon as they develop an appropriate“phonological template” for those words). Conversely,lexical development has a strong influence on thesounds that a child produces; specifically, the child’s

“favorite phonemes” tend to derive from the sounds thatare present in his first and favorite words. In fact,children appear to treat these lexical/phonologicalprototypes like a kind of basecamp, exploring the worldof sound in various directions without losing sight ofhome.

Phonological development interacts with lexicaland grammatical development for at least three years.For example, children who have difficulty with aparticular sound (e.g., the sibilant "-s") appear topostpone productive use of grammatical inflections thatcontain that sound (e.g., the plural). A rather differentlexical/phonological interaction is illustrated by manycases in which the “same” speech sound is producedcorrectly in one word context but incorrectly in another(e.g., the child may say "guck" for “duck”, but have notrouble pronouncing the “d” in “doll”). This is due, wethink, to articulatory facts: It is hard to move the speechapparatus back and forth between the dental position (in“d”) and the glottal position (in the “k” part of “duck”),so the child compromises by using one position only(“guck”). That is, the child may be capable ofproducing all the relevant sounds in isolation, but findsit hard to produce them in the combinations required forcertain word targets.

After 3 years of age, when lexical and grammaticaldevelopment have "settled down", phonology alsobecomes more stable and systematic: Either the childproduces no obvious errors at all, or s/he may persist inthe same phonological error (e.g., a difficulty pronoun-cing “r” and “l”) regardless of lexical context, for manymore years. The remainder of lexical development from3 years to adulthood can generally be summarized as anincrease in fluency, including a phenomenon called“coarticulation”, in which those sounds that will beproduced later on in an utterance are anticipated bymoving the mouth into position on an earlier speechsound (hence the “b” in “bee” is qualitatively differentfrom the “b” in “boat”).

To summarize, the development of speech as asound system begins at or perhaps before birth (inspeech perception), and continues into the adult years(e.g., with steady increases in fluency and coarticulationthroughout the first two decades of life). However,there is one point in phonetic and phonologicaldevelopment that can be viewed as a kind of watershed:8-10 months, marked by changes in perception (e.g.,the inhibition of non-native speech sounds) and changesin production (e.g., the onset of canonical babbling and“phonological drift”). The timing of these milestonesin speech may be related to some important events inhuman brain development that occur around the sametime, including the onset of synaptogenesis (a “burst”in synaptic growth that begins around 8 months andpeaks somewhere between 2-3 years of age), togetherwith evidence for changes in metabolic activity withinthe frontal lobes, and an increase in frontal control overother cortical and subcortical functions (Elman et al.,1996). This interesting correlation between brain andbehavioral development is not restricted to changes in

10

speech; indeed, the 8-10-month period is marked bydramatic changes in many different cognitive and socialdomains, including developments in tool use,categorization and memory for objects, imitation, andintentional communication via gestures (see Volterra,this volume). In other words, the most dramaticmoments in speech development appear to be linked tochange outside the boundaries of language, furtherevidence that our capacity for language depends onnonlinguistic factors. This bring us to the next point: abrief review of current evidence on the neural substratesof speech.

Brain Bases of Speech Perception andProduction

In this section, and in all the sections on the brainbases of language that will follow, the discussion willbe divided to reflect data from the two principalmethodologies of neurolinguistics and cognitiveneuroscience: evidence from patients with unilaterallesions (a very old method) and evidence from theapplication of functional brain-imaging techniques tolanguage processing in normal people (a brand-newmethod).

Although one hopes that these two lines ofevidence will ultimately converge, yielding a unifiedview of brain organization for language, we should notbe surprised to find that they often yield differentresults. Studies investigating the effects of focal lesionson language behavior can tell us what regions of thebrain are necessary for normal language use. Studiesthat employ brain-imaging techniques in normals cantell us what regions of the brain participate in normallanguage use. These are not necessarily the same thing.

Even more important for our purposes here, lesionstudies and neural imaging techniques cannot tell uswhere language or any other higher cognitive functionis located, i.e., where the relevant knowledge "lives,"independent of any specific task. The plug on the wallbehind a television set is necessary for the television’snormal function (just try unplugging your television,and you will see how important it is). It is also fair tosay that the plug participates actively in the process bywhich pictures are displayed. That doesn’t mean thatthe picture is located in the plug! Indeed, it doesn’teven mean that the picture passes through the plug onits way to the screen. Localization studies arecontroversial—and they deserve to be! Figure 4displays one version of the phrenological map of Galland Spurzheim, proposed in the 18th century, and stillthe best-known and most ridiculed version of the ideathat higher faculties are located within discrete areas ofthe brain. Although this particular map of the brain isnot taken seriously anymore, modern variants of thephrenological doctrine are still around, e.g., proposalsthat free will lives in frontal cortex, faces live in thetemporal lobe, and language lives in two places on theleft side of the brain, one in the front (called Broca’sarea) and another in the back (called Wernicke’s area).As we shall see throughout this brief review, there is no

phrenological view of language that can account forcurrent evidence from lesion studies, or from neuralimaging of the working brain. Rather, language seemsto be an event that is staged by many different areas ofthe brain, a complex process that is not located in asingle place. Having said that, it should also be notedthat some places are more important than others, even ifthey should not be viewed as the “language box.” Thedancer’s many skills are not located in her feet, but herfeet are certainly more important than a number of otherbody parts. In the same vein, some areas of the brainhave proven to be particularly important for normallanguage use, even though we should not conclude thatlanguage is located there. With those warnings inmind, let us take a brief look at the literature on thebrain bases of speech perception and production.

Lesion studies of speech. We have knownfor a very long time that injuries to the head can impairthe ability to perceive and produce speech. Indeed, thisobservation that first appeared in the Edmund SmithSurgical Papyrus, attributed to the Egyptian Imhotep.However, little progress was made beyond that simpleobservation until the 19th century. In 1861, Paul Brocaobserved a patient called “Tan” who appeared tounderstand the speech that other people directed to him;however, Tan was completely incapable of meaningfulspeech production, restricted entirely to the singlesyllable for which he was named. This patient died andcame to autopsy a few days after Broca tested him. Animage of that brain (preserved for posterity) appears inFigure 5. Casual observation of this figure reveals amassive cavity in the third convolution of the leftfrontal lobe, a region that is now known as Broca’sarea. Broca and his colleagues proposed that thecapacity for speech output resides in this region of thebrain.

Across the next few decades, European investigatorsset out in search of other sites for the language faculty.The most prominent of these was Carl Wernicke, whodescribed a different lesion that seemed to be responsiblefor severe deficits in comprehension, in patients who arenevertheless capable of fluent speech. This region (nowknown as Wernicke’s area) lay in the left hemisphere aswell, along the superior temporal gyrus close to thejunction of the temporal, parietal and occipital lobes. Itwas proposed that this region is the site of speechperception, connected to Broca’s area in the front of thebrain by a series of fibres called the arcuate fasciculus.Patients who have damage to the fibre bundle onlyshould prove to be incapable of repeating words thatthey hear, even though they are able to producespontaneous speech and understand most of the speechthat they hear. This third syndrome (called “conductionaphasia”) was proposed on the basis of Wernicke’stheory, and in the next few years a number ofinvestigators claimed to have found evidence for itsexistence. Building on this model of brain organizationfor language, additional areas were proposed to underliereading (to explain “alexia”) and writing (responsible for“agraphia”), and arguments raged about the relative

11

separability or dissociability of the emerging aphasiasyndromes (e.g., is there such a thing as alexia withoutagraphia?).

This neophrenological view has had its critics atevery point in the modern history of aphasiology,including Freud’s famous book “On aphasia”, whichridicules the Wernicke-Lichtheim model (Freud,1891/1953), and Head’s witty and influential critique oflocalizationists, whom he referred to as “The DiagramMakers” (Head, 1926). The localizationist view fell onhard times in the period between 1930-1960, theBehaviorist Era in psychology when emphasis wasgiven to the role of learning and the plasticity of thebrain. But it was revived with a vengeance in the1960’s, due in part to Norman Geschwind’s influentialwritings and to the strong nativist approach to languageand the mind proposed by Chomsky and his followers.

Localizationist views continue to wax and wane,but they seem to be approaching a new low point today,due (ironically) to the greater precision offered bymagnetic resonance imaging and other techniques fordetermining the precise location of the lesionsassociated with aphasia syndromes. Simply put, theclassical story of lesion-syndrome mapping is fallingapart. For example, Dronkers (1996) has shown thatlesions to Broca’s area are neither necessary norsufficient for the speech output impairments that defineBroca’s aphasia. In fact, the only region of the brainthat seems to be inextricably tied to speech outputdeficits is an area called the insula, hidden in the foldsbetween the frontal and temporal lobe. This area iscrucial, but its contribution may lie at a relatively lowlevel, mediating kinaesthetic feedback from the face andmouth.

A similar story may hold for speech perception.There is no question that comprehension can bedisrupted by lesions to the temporal lobe in a matureadult. However, the nature and locus of this disruptionare not at all clear. Of all the symptoms that affectspeech perception, the most severe is a syndrome called“pure word deafness.” Individuals with this afflictionare completely unable to recognize spoken words, eventhough they do respond to sound and can (in manycases) correctly classify meaningful environmentalsounds (e.g., matching the sound of a dog barking tothe picture of a dog). This is not a deficit to lexicalsemantics (see below), because some individuals withthis affliction can understand the same words in awritten form. Because such individuals are not deaf, itis tempting to speculate that pure word deafnessrepresents the loss of a localized brain structure thatexists only for speech. However, there are two reasonsto be cautious before we accept such a conclusion.First, the lesions responsible for word deafness arebilateral (i.e., wide ranges of auditory cortex must bedamaged on both sides). Hence they do not follow theusual left/right asymmetry observed in language-relatedsyndromes. Second, it is difficult on logical grounds todistinguish between a speech/nonspeech contrast (adomain-specific distinction) and a complex/simple

contrast (a domain-general distinction). As we notedearlier, speech is an exceedingly complex auditoryevent. There are no other meaningful environmentalsounds that achieve anything close to this level ofcomplexity (dogs barking, bells ringing, etc.). Hence itis quite possible that the lesions responsible for worddeafness have their effect by creating a global,nonspecific degradation in auditory processing, one thatis severe enough to preclude speech but not severeenough to block recognition of other, simpler auditoryevents.

This brings us to a different but related point.There is a developmental syndrome called “congenitaldysphasia” or “specific language impairment”, a deficitin which children are markedly delayed in languagedevelopment in the absence of any other syndrome thatcould account for this delay (e.g., no evidence of mentalretardation, deafness, frank neurological impairmentslike cerebral palsy, or severe socio-emotional deficitslike those that occur in autism). Some theorists believethat this is a domain-specific syndrome, one thatprovides evidence for the independence of language fromother cognitive abilities (see Grammar, below). How-ever, other theorists have proposed that this form oflanguage delay is the by-product of subtle deficits inauditory processing that are not specific to language,but impair language more than any other aspect ofbehavior. This claim is still quite controversial, butevidence is mounting in its favor (Bishop, 1997;Leonard, 1997). If this argument is correct, then weneed to rethink the neat division between componentsthat we have followed here so far, in favor of a theory inwhich auditory deficits lead to deficits in the perceptionof speech, which lead in turn to deficits in languagelearning.

Functional Brain-Imaging Studies ofSpeech. With the arrival of new tools like positronemission tomography (PET) and functional magneticresonance imaging (fMRI), we are able at last toobserve the normal brain at work. If the phrenologicalapproach to brain organization for language werecorrect, then it should be just a matter of time before welocate the areas dedicated to each and every componentof language. However, the results that have beenobtained to date are very discouraging for thephrenological view.

Starting with speech perception, Poeppel (1996)has reviewed six pioneering studies of phonologicalprocessing using PET. Because phonology is muchcloser to the physics of speech than abstract domainslike semantics and grammar, we might expect the firstbreakthroughs in brain-language mapping to occur inthis domain. However, Poeppel notes that there isvirtually no overlap across these six studies in theregions that appear to be most active during theprocessing of speech sounds! To be sure, these studies(and many others) generally find greater activation in theleft hemisphere than the right, although many studies oflanguage activation do find evidence for some right-hemisphere involvement. In addition, the frontal and

12

temporal lobes are generally more active than otherregions, especially around the Sylvian fissure(“perisylvian cortex”), which includes Broca’s andWernicke’s areas. Although lesion studies and brain-imaging studies of normals both implicate perisylviancortex, many other areas show up as well, with markedvariations from one study to another. Most importantlyfor our purposes here, there is no evidence from thesestudies for a single “phonological processing center”that is activated by all phonological processing tasks.

Related evidence comes from studies of speechproduction (including covert speech, without literalmovements of the mouth). Here too, left-hemisphereactivation invariably exceeds activation on the right, andperisylvian areas are typical regions of high activation(Toga, Frackowiak, & Mazziotta, 1996). Interestingly,the left insula is the one region that emerges most oftenin fMRI and PET studies of speech production, acomplement to Dronkers’ findings for aphasic patientswith speech production deficits. The insula is an area ofcortex buried deep in the folds between temporal andfrontal cortex. Although its role is still not fullyunderstood, the insula appears to be particularlyimportant in the mediation of kinaesthetic feedbackfrom the various articulators (i.e., moving parts of thebody), and the area implicated in speech output deficitsis one that is believed to play a role in the mediation offeedback from the face and mouth. Aside from this onerelatively low-level candidate, no single region hasemerged to be crowned as “the speech productioncenter”.

One particularly interesting study in this regardfocused on the various subregions that comprise Broca’sarea and adjacent cortex (Erhard, Kato, Strick, &Ugurbil, 1996). Functional magnetic resonanceimaging (fMRI) was used to compare activation withinand across the Broca complex, in subjects who wereasked to produce covert speech movements, simple andcomplex nonspeech movements of the mouth, andfinger movements at varying levels of complexity.Although many of the subcomponents of Broca’scomplex were active for speech, all of these componentsparticipate to a similar extent in at least one nonspeechtask. In other words, there is no area in the frontalregion that is active only for speech.

These findings for speech illustrate an emergingtheme in functional brain imagery research, revolvingaround task specificity, rather than domain specificity.That is, patterns of localization or activation seem tovary depending on such factors as the amount and kindof memory required for a task, its relative level ofdifficulty and familiarity to the subject, its demands onattention, the presence or absence of a need to suppressa competing response, whether covert motor activity isrequired, and so forth. These domain-general but task-specific factors show up in study after study, with bothlinguistic and nonlinguistic materials (e.g., an area infrontal cortex called the anterior cingulate shows up instudy after study when a task is very new, and veryhard). Does this mean that domain-specific functions

“move” from one area to another? Perhaps, but it ismore likely that “movement” and “location” are boththe wrong metaphors. We may need to revise ourthinking about brain organization for speech and otherfunctions along entirely different lines. My hand takesvery different configurations depending on the task that Iset out to accomplish: to pick up a pin, pick up a heavybook, or push a heavy box against the wall. A “muscleactivation” study of my hand within each task wouldyield a markedly different distribution of activity foreach of these tasks. And yet it does not add very muchto the discussion to refer to the first configuration as a“pin processor”, the second as a “book processor”, andso on. In much the same way, we may use thedistributed resources of our brains in very different waysdepending on the task that we are trying to accomplish.Some low-level components probably are hard-wired andtask-independent (e.g., the areas of cortex that are feddirectly by the auditory nerve, or that portion of theinsula that handles kinaesthetic feedback from themouth and face). Once we move above this level,however, we should perhaps expect to find highlyvariable and distributed patterns of activity inconjunction with linguistic tasks. We will return tothis theme later, when we consider the brain bases ofother language levels.

III. WORDS AND GRAMMARHow Words and Sentences are Processed by

Normal AdultsThe major issue of concern in the study of word and

sentence processing is similar to the issue that divideslinguists. On one side, we find investigators who viewlexical and grammatical processing as independentmental activities, handled by separate mental/neuralmechanisms (e.g., Fodor, 1983). To be sure, these twomodules have to be integrated at some point inprocessing, but their interaction can only take placeafter each module has completed its work. On the otherside, we find investigators who view word recognitionand grammatical analysis as two sides of a singlecomplex process: Word recognition is “penetrated” bysentence-level information (e.g., Elman & McClelland,1986), and sentence processing is profoundly influencedby the nature of the words contained within eachsentence (MacDonald, Pearlmutter, & Seidenberg,1994). This split in psycholinguistics betweenmodularists and interactionists mirrors the split intheoretical linguistics between proponents of syntacticautonomy (e.g., Chomsky, 1957) and theorists whoemphasize the semantic and conceptual nature ofgrammar (e.g., Langacker, 1987).

In the 1960’s-1970’s, when the autonomy viewprevailed, efforts were made to develop real-time proces-sing models of language comprehension and production(i.e., performance) that implemented the same modularstructure proposed in various formulations of Chom-sky’s generative grammar (i.e., competence). (For areview, see Fodor, Bever, & Garrett, 1974). Thecomprehension variants had a kind of “assembly line”

13

structure, with linguistic inputs passed in a serialfashion from one module to another (phonetic -->phonological --> grammatical --> semantic).Production models looked very similar, with arrowsrunning in the opposite direction (semantic -->grammatical --> phonological --> phonetic). Accordingto this "assembly line" approach, each of theseprocesses is unidirectional. Hence it should not bepossible for higher-level information in the sentence toinfluence the process by which we recognize individualwords during comprehension, and it should not bepossible for information about the sounds in thesentence to influence the process by which we chooseindividual words during production.

The assumption of unidirectionality underwent aserious challenge during the late 1970’s to the early1980’s, especially in the study of comprehension. Averitable cottage industry of studies appeared showing“top-down” context effects on the early stages of wordrecognition, raising serious doubts about this fixedserial architecture. For example, Samuel (1981)presented subjects with auditory sentences that led up toauditory word targets like “meal” or “wheel”. In thatstudy, the initial phoneme that disambiguates betweentwo possible words was replaced with a brief burst ofnoise (like a quick cough), so that the words “meal” and“wheel” were both replaced by “(NOISE)-eel”. Underthese conditions, subjects readily perceived the "-eel"sound as "meal" in a dinner context and "wheel" in atransportation context, often without noticing the coughat all.

In response to all these demonstrations of contexteffects, proponents of the modular view countered withstudies demonstrating temporal constraints on the use oftop-down information during the word recognitionprocess (Onifer & Swinney, 1981), suggesting that theprocess really is modular and unidirectional, but onlyfor a very brief moment in time. An influentialexample comes from experiments in which semanticallyambiguous words like “bug” are presented within anauditory sentence context favoring only one of its twomeanings, e.g.,

Insect context (bug = insect): “Because they had found a

number of roaches and spiders in theroom, experts were called in to checkthe room for bugs ....”

Espionage context (bug = hidden microphone) :“Because they were concerned

about electronic surveillance, theexperts were called in to check theroom for bugs ....”

Shortly after the ambiguous word is presented,subjects see a either real word or a nonsense wordpresented visually on the computer screen, and are askedto decide as quickly as possible if the target is a realword or not (i.e., a lexical decision task). The real-wordtargets included words that are related to the "primed" orcontextually appropriate meaning of "bug" (e.g., SPY

in an espionage context), words that are related to thecontextually inappropriate meaning of “bug” (e.g., ANTin an espionage context), and control words that are notrelated to either meaning (e.g., MOP in an espionagecontext). Evidence of semantic activation or "priming"is obtained if subjects react faster to a word related to"bug" than they react to the unrelated control. If thelexicon is modular, and uninfluenced by higher-levelcontext, then there should be a short period of time inwhich SPY and ANT are both faster than MOP. On theother hand, if the lexicon is penetrated by context in theearly stages, then SPY should be faster than ANT, andANT should be no faster than the unrelated word MOP.

The first round of results using this techniqueseemed to support the modular view. If the prime andtarget are separated by at least 750 msec, priming isobserved only for the contextually appropriate meaning(i.e., selective access); however, if the prime and targetare very close together in time (250 msec or less),priming is observed for both meanings of the ambi-guous word (i.e., exhaustive access). These results wereinterpreted as support for a two-stage model of wordrecognition: a “bottom-up” stage that is unaffected bycontext, and a later “top-down” stage when contextualconstraints can apply.

Although the exhaustive-access finding has beenreplicated in many different laboratories, its inter-pretation is still controversial. For example, someinvestigators have shown that exhaustive access fails toappear on the second presentation of an ambiguousword, or in very strong contexts favoring the dominantmeaning of the word. An especially serious challengecomes from a study by Van Petten and Kutas (1991),who used similar materials to study the event-relatedscalp potentials (i.e., ERPs, or "brain waves") asso-ciated with contextually appropriate, contextuallyinappropriate and control words at long and short timeintervals (700 vs. 200 msec between prime and target).Their results provide a very different story than the oneobtained in simple reaction time studies, suggestingthat there are actually three stages involved in theprocessing of ambiguous words, instead of just two.Figure 6 illustrates the Van Petten and Kutas resultswhen the prime and target are separated by only 200msec (the window in which lexical processing issupposed to be independent of context), compared withtwo hypothetical outcomes. Once again, we are usingan example in which the ambiguous word BUG appearsin an espionage context. If the selective-access view iscorrect, and context does penetrate the lexicon, thenbrain waves to the contextually relevant word (e.g.,SPY) ought to show a positive wave (where positive isplotted downward, according to the conventions of thisfield); brain waves to the contextually irrelevant word(e.g., ANT) ought to look no different from an unrelatedand unexpected control word (e.g., MOP), with botheliciting a negative wave called the N400 (plottedupward). If the modular, exhaustive-access account iscorrect, and context cannot penetrate the lexicon, thenany word that is related lexically to the ambiguous

14

prime (e.g., either SPY or ANT) ought to show apositive (“primed”) wave, compared with an unexpected(“unprimed”) control word. The observed outcome wasmore compatible with a selective-access view, but withan interesting variation, also plotted in Figure 6. In thevery first moments of word recognition, thecontextually inappropriate word (e.g., ANT) behavesjust like an unrelated control (e.g., MOP), moving in anegative direction (plotted upward). However, later onin the sequence (around 400 msec), the contextuallyirrelevant word starts to move in a positive direction, asthough the subject had just noticed, after the fact, thatthere was some kind of additional relationship (e.g.,BUG ... ANT? ... Oh yeah, ANT!!). None of thisoccurs with the longer 700-millisecond window, whereresults are fit perfectly by the context-driven selective-access model. These complex findings suggest that wemay need a three-stage model to account for processingof ambiguous words, paraphrased as

SELECTIVE PRIMING --> EXHAUSTIVE PRIMING -->

CONTEXTUAL SELECTION

The fact that context effects actually precedeexhaustive priming proves that context effects canpenetrate the earliest stages of lexical processing, strongevidence against the classic modular view. However,the fact that exhaustive priming does appear for a veryshort time, within the shortest time window, suggeststhat the lexicon does have "a mind of its own", i.e., astubborn tendency to activate irrelevant material at alocal level, even though relevance wins out in the longrun.

To summarize, evidence in favor of context effectson word recognition has continued to mount in the lastfew years, with both reaction time and electrophysio-logical measures. However, the lexicon does behaverather stupidly now and then, activating irrelevantmeanings of words as they come in, as if it had no ideawhat was going on outside. Does this prove thatlexical processing takes place in an independent module? Perhaps not. Kawamoto (1988) has conducted simula-tions of lexical access in artificial neural networks inwhich there is no modular border between sentence- andword-level processing. He has shown that exhaustiveaccess can and does occur under some circumstanceseven in a fully interactive model, depending ondifferences in the rise time and course of activation fordifferent items under different timing conditions. Inother words, irrelevant meanings can “shoot off” fromtime to time even in a fully interconnected system. Ithas been suggested that this kind of "local stupidity” isuseful to the language-processing system, because itprovides a kind of back-up activation, just in case themost probable meaning turns out to be wrong at somepoint further downstream. After all, people dooccasionally say very strange things, and we have to beprepared to hear them, even within a "top-down,"contextually guided system.

This evidence for an early interaction betweensentence-level and word-level information is only

indirectly related to the relationship between lexicalprocessing and grammar per se. That is, because asentence contains both meaning and grammar, asentential effect on word recognition could be caused bythe semantic content of the sentence (both propositionaland lexical semantics), leaving the Great Border betweengrammar and the lexicon intact. Are the processes ofword recognition and/or word retrieval directly affectedby grammatical context alone?

A number of early studies looking at grammaticalpriming in English have obtained weak effects or noeffects at all on measures of lexical access. In asummary of the literature on priming in spoken-wordrecognition, Tanenhaus and Lucas (1987) conclude that“On the basis of the evidence reviewed ...it seems likelythat syntactic context does not influence prelexicalprocessing” (p. 223). However, more recent studies inlanguages with rich morphological marking haveobtained robust evidence for grammatical priming (Bates& Goodman, in press). This includes effects of genderpriming on lexical decision and gating in French, onword repetition and gender classification in Italian, andon picture naming in Spanish and German. Studies oflexical decision in Serbo-Croatian provide evidence forboth gender and case priming, with real word andnonword primes that carry morphological markings thatare either congruent or incongruent with the target word.These and other studies show that grammatical contextcan have a significant effect on lexical access within thevery short temporal windows that are usually associatedwith early and automatic priming effects that interestproponents of modularity. In other words, grammaticaland lexical processes interact very early, in intricatepatterns of the sort that we would expect if they aretaking place within a single, unified system governedby common laws.

This conclusion marks the beginning rather thanthe end of interesting research in word and sentenceprocessing, because it opens the way for detailed cross-linguistic studies of the processes by which words andgrammar interact during real-time language processing.The modular account can be viewed as an accidental by-product of the fact that language processing research hasbeen dominated by English-speaking researchers for thelast 30 years. English is unusual among the world’slanguages in its paucity of inflectional morphology, andin the degree to which word order is rigidly preserved.In a language of this kind, it does seem feasible toentertain a model in which words are selected indepen-dently of sentence frames, and then put together by thegrammar like beads on a string, with just a few minoradjustments in the surface form of the words to assuremorphological agreement (e.g., “The dogs walk” vs.“The dog walks”). In richly inflected languages likeRussian, Italian, Hebrew or Greenlandic Eskimo, it isdifficult to see how such a modular account couldpossibly work. Grammatical facts that occur early inthe sentence place heavy constraints on the words thatmust be recognized or produced later on, and the wordsthat we recognize or produce at the beginning of an

15

utterance influence detailed aspects of word selection andgrammatical agreement across the rest of the sentence.A model in which words and grammar interact intimate-ly at every stage in processing would be moreparsimonious for a language of this kind. As themodularity/interactionism debate begins to ebb in thefield of psycholinguistics, rich and detailed comparativestudies of language processing are starting to appearacross dramatically different language families, markingthe beginning of an exciting new era in this field.

How Words and Sentences Develop inChildren

The modularity/interactionism debate has also beena dominant theme in the study of lexical andgrammatical development, interacting with theoverarching debate among empiricists, nativists andconstructivists.

From one point of view, the course of earlylanguage development seems to provide a prima faciecase for linguistic modularity, with sounds, words andgrammar each coming in on separate developmentalschedules (Table 1). Children begin their linguisticcareers with babble, starting with vowels (somewherearound 3-4 months, on average) and ending withcombinations of vowels and consonants of increasingcomplexity (usually between 6-8 months).Understanding of words typically begins between 8-10months, but production of meaningful speech emergessome time around 12 months, on average. After this,most children spend many weeks or months producingsingle-word utterances. At first their rate of vocabularygrowth is very slow, but one typically sees a "burst" oracceleration in the rate of vocabulary growth somewherebetween 16-20 months. First word combinationsusually appear between 18-20 months, although theytend to be rather spare and telegraphic (at least inEnglish — see Table 2 for examples). Somewherebetween 24-30 months, most children show a kind of"second burst", a flowering of morphosyntax that RogerBrown has characterized as "the ivy coming in betweenthe bricks." Between 3-4 years of age, most normalchildren have mastered the basic morphological andsyntactic structures of their language, using themcorrectly and productively in novel contexts. From thispoint on, lexical and grammatical development consistprimarily in the tuning and amplification of thelanguage system: adding more words, becoming morefluent and efficient in the process by which words andgrammatical constructions are accessed in real time, andlearning how to use the grammar to create largerdiscourse units (e.g., writing essays, telling stories,participating in a long and complex conversation).

This picture of language development in Englishhas been documented extensively (for reviews see Aslin,Jusczyk & Pisoni, in press; Fletcher & MacWhinney,1995). Of course the textbook story is not exactly thesame in every language (Slobin, 1985-1997), andperfectly healthy children can vary markedly in rate andstyle of development through these milestones (Bates,

Bretherton, & Snyder, 1988). At a global level, how-ever, the passage from sounds to words to grammarappears to be a universal of child language development.

A quick look at the relative timing and shape ofgrowth within word comprehension, word productionand grammar can be seen in Figures 7, 8 and 9 (fromFenson et al., 1994). The median (50th percentile) ineach of these figures confirms that textbook summaryof average onset times that we have just recited:Comprehension gets off the ground (on average)between 8-10 months, production generally starts offbetween 12-13 months (with a sharp accelerationbetween 16-20 months), and grammar shows its peakgrowth between 24-30 months. At the same time,however, these figures show that there is massivevariation around the group average, even amongperfectly normal, healthy middle-class children. Similarresults have now been obtained for more than a dozenlanguages, including American Sign Language (alanguage that develops with the eyes and hands, insteadof the ear and mouth). In every language that has beenstudied to date, investigators report the same averageonset times, the same patterns of growth, and the samerange of individual variation illustrated in Figures 7-9.

But what about the relationship between thesemodalities? Are these separate systems, or differentwindows on a unified developmental process? A moredirect comparison of the onset and growth of words andgrammar can be found in Figure 10 (from Bates &Goodman, in press). In this figure, we have expresseddevelopment for the average (median) child in terms ofthe percent of available items that have been mastered ateach available time point from 8-30 months.Assuming for a moment that we have a right tocompare the proportional growth of apples and oranges,it shows that word comprehension, word production andgrammar each follow a similar nonlinear pattern ofgrowth across this age range. However, the respective“zones of acceleration” for each domain are separated bymany weeks or months.

Is this a discontinuous passage, as modular/nativisttheories would predict? Of course no one has everproposed that grammar can begin in the absence ofwords! Any grammatical device is going to have tohave a certain amount of lexical material to work on.The real question is: Just how tight are the correlationsbetween lexical and grammatical development in thesecond and third year of life? Are these componentsdissociable, and if so, to what extent? How muchlexical material is needed to build a grammaticalsystem? Can grammar get off the ground and go itsseparate way once a minimum number of words isreached (e.g., 50-100 words, the modal vocabulary sizewhen first word combinations appear)? Or will weobserve a constant and lawful interchange betweenlexical and grammatical development, of the sort thatone would expect if words and grammar are two sides ofthe same system?

Our reading of the evidence suggests that the latterview is correct. In fact, the function that governs the

TABLE 1: MAJOR MILESTONES IN LANGUAGE DEVELOPMENT

0-3 months INITIAL STATE OF THE SYSTEM

— prefers to listen to sounds in native language— can hear all the phonetic contrasts used

in the world’s languages— produces only “vegetative” sounds

3-6 months VOWELS IN PERCEPTION AND PRODUCTION

— cooing, imitation of vowel sounds only— perception of vowels organized along

language-specific lines

6-8 months BABBLING IN CONSONANT-VOWEL SEGMENTS

8-10 months WORD COMPREHENSION

— starts to lose sensitivity to consonants outsidenative language

12-13 months WORD PRODUCTION (NAMING)

16-20 months WORD COMBINATIONS

— vocabulary acceleration— appearance of relational words

(e.g., verbs and adjectives)

24-36 months GRAMMATICIZATION

— grammatical function words— inflectional morphology— increased complexity of sentence structure

3 years —> adulthood LATE DEVELOPMENTS

— continued vocabulary growth— increased accessibility of

rare and/or complex forms— reorganization of sentence-level

grammar for discourse purposes

TABLE 2:SEMANTIC RELATIONS UNDERLYING CHILDREN’S FIRST WORD COMBINATIONS

ACROSS MANY DIFFERENT LANGUAGES(adapted from Braine, 1976)

Semantic Function English examples

Attention to X “See doggie!” “Dat airplane”

Properties of X “Mommy pretty” “Big doggie”

Possession “Mommy sock” “My truck”

Plurality or iteration “Two shoe” “Round and round”

Recurrence (including requests) “More truck”“Other cookie”

Disappearance “Allgone airplane” “Daddy bye-bye”

Negation or Refusal “No bath” “No bye-bye”

Actor-Action “Baby cry” “Mommy do it”

Location “Baby car” “Man outside”

Request “Wanna play it” “Have dat”

16

relation between lexical and grammatical growth in thisage range is so powerful and so consistent that it seemsto reflect some kind of developmental law. Thesuccessive “bursts" that characterize vocabulary growthand the emergence of grammar can be viewed as differentphases of an immense nonlinear wave that starts in thesingle-word stage and crashes on the shores of grammara year or so later. An illustration of this powerfulrelationship is offered in a comparison of Figure 11,which plots the growth of grammar as a function ofvocabulary size. It should be clear from this figure thatgrammatical growth is tightly related to lexical growth,a lawful pattern that is far more regular (and farstronger) than the relationship between grammaticaldevelopment and chronological age. Of course this kindof correlational finding does not force us to concludethat grammar and vocabulary growth are mediated by thesame developmental mechanism. Correlation is notcause. At the very least, however, this powerfulcorrelation suggests that the two have somethingimportant in common.

Although there are strong similarities acrosslanguages in the lawful growth and interrelation ofvocabulary and grammar, there are massive differencesbetween languages in the specific structures that mustbe acquired. Chinese children are presented with alanguage that has no grammatical inflections of anykind, compared with Eskimo children who have to learna language in which an entire sentence may consist of asingle word with more than a dozen inflections. Lestwe think that life for the Chinese child is easy, thatchild has to master a system in which the same syllablecan take on many different meanings depending on itstone (i.e., its pitch contour). There are four of thesetones in Mandarin Chinese, and seven in the Taiwanesedialect, presenting Chinese children with a word-learning challenge quite unlike the ones that facechildren acquiring Indo-European languages like Englishor Italian.

Nor should we underestimate the differences thatcan be observed for children learning different Indo-European language types. Consider the Englishsentence

Wolves eat sheep

which contains three words and four distinct morphemes(in standard morpheme-counting systems, “wolf” + “-s”constitute two morphemes, but the third person pluralverb “eat” and the plural noun “sheep” only count asone morpheme each). The Italian translation of thissentence would be

I lupi mangiano le pecore

The Italian version of this sentence contains fivewords (articles are obligatory in Italian in this context;they are not obligatory for English). Depending on themeasure of morphemic complexity that we choose touse, it also contains somewhere between ten andfourteen morphemes (ten if we count each explicitplural marking on each article, noun and verb, butexclude gender decisions from the count; fourteen if each

gender decision also counts as a separate morphologicalcontrast, on each article and noun). What this means isthat, in essence, an Italian child has roughly three timesas much grammar to learn as her English agemates!There are basically two ways that this quantitativedifference might influence the learning process:

(1) Italian children might acquire morphemes atthe same absolute rate. If this is true, then it shouldtake approximately three times longer to learn Italianthan it does to learn English.

(2) Italian and English children might acquire theirrespective languages at the same proportional rate. I fthis is true, then Italian and English children should“know” the same proportion of their target grammar ateach point in development (e.g., 10% at 20 months,30% at 24 months, and so forth), and Italian childrenshould display approximately three times moremorphology than their English counterparts at everypoint.

Comparative studies of lexical and grammaticaldevelopment in English and Italian suggest that thedifference between languages is proportional rather thanabsolute during the phase in which most grammaticalcontrasts are acquired. As a result of this difference,Italian two-year-olds often sound terribly precocious tonative speakers of English, and English two-year-oldsmay sound mildly retarded to the Italian ear!

My point is that every language presents the childwith a different set of problems, and what is "hard" inone language may be "easy" in another. Table 2presents a summary of the kinds of meanings thatchildren tend to produce in their first word combinationsin many different language communities (from Braine,1976). These are the concerns (e.g., possession,refusal, agent-action, object-location) that preoccupy 2-year-olds in every culture. By contrast, Table 3 showshow very different the first sentences of 2-year-oldchildren can look just a few weeks or months later, asthey struggle to master the markedly different structuraloptions available in their language (adapted fromSlobin, 1985, 1985-1997). The content of theseutterances is universal, but the forms that they mustmaster vary markedly.

How do children handle all these options? Chom-sky's answer to the problem of cross-linguistic variationis to propose that all the options in Universal Grammarare innate, so that the child's task is simplified to one oflistening carefully for the right "triggers," setting theappropriate "parameters". Learning has little or nothingto do with this process. Other investigators haveproposed instead that language development really is aform of learning, although it is a much more powerfulform of learning than Skinner foresaw in his work onschedules of reinforcement in rats. Recent neuralnetwork simulations of the language-learning processprovide, at the very least, a kind of "existence proof",demonstrating that aspects of grammar can be learned bya system of this kind and thus (perhaps) by the humanchild.

TABLE 3:EXAMPLES OF SPEECH BY TWO-YEAR-OLDS IN DIFFERENT LANGUAGES

(underlining = content words)

English (30 months):

I wanna help wash car 1st pers. modal infinitive infinitivesingular indicative

Translation: I wanna help wash car.

Italian (24 months):

Lavo mani , sporche , apri acqua .Wash hands dirty open water 1st pers. 3rd pers. feminine 2nd pers. 3rd pers.singular feminine plural singular singular

indicative plural imperative

Translation: I wash hands, dirty, turn on water.

Western Greenlandic (26 months):

anner - punga............. anni - ler- punga hurt - 1st singular hurt - about-to 1st singular

indicative indicative

Translation: I’ve hurt myself ... I’m about to hurt myself ...

Mandarin (28 months):

Bu yao ba ta cai - diao zhege ounot want object- it tear - down this warning-

marker marker

Translation: Don’t tear apart it!

TABLE 3: (continued)EXAMPLES OF SPEECH BY TWO-YEAR-OLDS

IN DIFFERENT LANGUAGES

Sesotho (32 months):

o- tla- hlaj - uw- a ke tshehlo class 2 future stab - passive mood by thorn singular marker class 9subject-marker

Translation: You’ll get stabbed by a thorn.

J apanese (25 months):

Okashi tabe - ru tte yut- ta Sweets eat non- quote- say past

past marker

Translation: She said that she’ll eat sweets.

17

For example, it has long been known that childrenlearning English tend to produce correct irregular pasttense forms (e.g., "went" and "came") for many weeksor months before the appearance of regular marking(e.g., "walked" and "kissed"). Once the regularmarkings appear, peculiar errors start to appear as well,e.g., terms like "goed" and "comed" that are notavailable anywhere in the child's input. It has beenargued that this kind of U-shaped learning is beyond thecapacity of a single learning system, and must be takenas evidence for two separate mechanisms: a rotememorization mechanism (used to acquire words, and toacquire irregular forms) and an independent rule-basedmechanism (used to acquire regular morphemes).However, there are now several demonstrations of thesame U-shaped developmental patterns in neuralnetworks that use a single mechanism to acquire wordsand all their inflected forms (Elman et al., 1996). Theseand other demonstrations of grammatical learning inneural networks suggest that rote learning and creativegeneralizations can both be accomplished by a singlesystem, if it has the requisite properties. It seems thatclaims about the “unlearnability” of grammar have to bereconsidered.

Brain Bases of Words and SentencesLesion studies . When the basic aphasic syn-

dromes were first outlined by Broca, Wernicke and theircolleagues, differences among forms of linguisticbreakdown were explained along sensorimotor lines,rooted in rudimentary principles of neuroanatomy. Forexample, the symptoms associated with damage to aregion called Broca’s area were referred to collectively asmotor aphasia: slow and effortful speech, with areduction in grammatical complexity, despite theapparent preservation of speech comprehension at aclinical level. This definition made sense when weconsider the fact that Broca’s area lies near the motorstrip. Conversely, the symptoms associated withdamage to Wernicke’s area were defined collectively as asensory aphasia: fluent but empty speech, marked bymoderate to severe word-finding problems, in patientswith serious problems in speech comprehension. Thischaracterization also made good neuroanatomical sense,because Wernicke’s area lies at the interface betweenauditory cortex and the various association areas thatwere presumed to mediate or contain word meaning.Isolated problems with repetition were further ascribedto fibers that link Broca’s and Wernicke’s area; othersyndromes involving the selective sparing orimpairment of reading or writing were proposed, withspeculations about the fibers that connect visual cortexwith the classical language areas (for an influential andhighly critical historical review, see Head, 1926).

In the period between 1960 and 1980, a revision ofthis sensorimotor account was proposed (summarized inKean, 1985). Psychologists and linguists who werestrongly influenced by generative grammar sought anaccount of language breakdown in aphasia that followedthe componential analysis of the human language

faculty proposed by Chomsky and his colleagues. Thiseffort was fueled by the discovery that Broca’s aphasicsdo indeed suffer from comprehension deficits: Spe-cifically, these patients display problems in theinterpretation of sentences when they are forced to relyentirely on grammatical rather than semantic orpragmatic cues (e.g., they successfully interpret asentence like “The apple was eaten by the girl”, wheresemantic information is available in the knowledge thatgirls, but not apples, are capable of eating, but fail on asentence like “The boy was pushed by the girl”, whereeither noun can perform the action). Because thoseaspects of grammar that appear to be impaired inBroca’s aphasia are precisely the same aspects that areimpaired in the patients’ expressive speech, the idea wasput forth that Broca’s aphasia may represent a selectiveimpairment of grammar (in all modalities), in patientswho still have spared comprehension and production oflexical and propositional semantics. From this point ofview, it also seemed possible to reinterpret theproblems associated with Wernicke’s aphasia as aselective impairment of semantics (resulting in compre-hension breakdown and in word-finding deficits inexpressive speech), accompanied by a selective sparingof grammar (evidenced by the patients’ fluent but emptyspeech).

If grammar and lexical semantics can be doublydissociated by forms of focal brain injury, then it seemsfair to conclude that these two components of languageare mediated by separate neural systems. It was neverentirely obvious how or why the brain ought to beorganized in just this way (e.g., why Broca's area, thesupposed seat of grammar, ought to be located near themotor strip), but the lack of a compelling link betweenneurology and neurolinguistics was more thancompensated for by the apparent isomorphism betweenaphasic syndromes and the components predicted bylinguistic theory. It looked for a while as if Nature hadprovided a cunning fit between the componentsdescribed by linguists and the spatial representation oflanguage in the brain. Indeed, this linguistic approachto aphasia was so successful in its initial stages that itcaptured the imagination of many neuroscientists, and itworked its way into basic textbook accounts oflanguage breakdown in aphasia.

Although this linguistic partitioning of the brain isvery appealing, evidence against it has accumulated inthe last 15 years, leaving aphasiologists in search of athird alternative to both the original modality-basedaccount (i.e., motor vs. sensory aphasia) and to thelinguistic account (i.e., grammatical vs. lexicaldeficits). Here is a brief summary of arguments againstthe neural separation of words and sentences (for moreextensive reviews, see Bates & Goodman, in press).

(1) Deficits in word finding (called “anomia”)are observed in all forms of aphasia, includingBroca’s aphasia. This means that there can neverbe a full-fledged double dissociation betweengrammar and the lexicon, weakening claims that

18

the two domains are mediated by separate brainsystems.

(2) Deficits in expressive grammar are notunique to agrammatic Broca’s aphasia, or to anyother clinical group. English-speaking Wernicke’saphasics produce relatively few grammatical errors,compared with English-speaking Broca’s aphasics.However, this fact turns out to be an artifact ofEnglish! Nonfluent Broca’s aphasics tend to err byomission (i.e., leaving out grammatical functionwords and dropping inflections), while Wernicke’serr by substitution (producing the wrong inflec-tion). Because English has so little grammaticalmorphology, it provides few opportunities forerrors of substitution, but it does provideopportunities for function word omission. As aresult, Broca’s seem to have more severe problemsin grammar. However, the grammatical problemsof fluent aphasia are easy to detect, and verystriking, in richly inflected languages like Italian,German or Hungarian. This is not a newdiscovery; it was pointed out long ago by ArnoldPick, the first investigator to use the term“agrammatism” (Pick, 1913/1973).

(3) Deficits in receptive grammar are evenmore pervasive, showing up in Broca’s aphasia,Wernicke’s aphasia, and in many patient groupswho show no signs of grammatical impairment intheir speech output. In fact, it is possible todemonstrate profiles of receptive impairment verysimilar to those observed in aphasia in normalcollege students who are forced to process sentencesunder various kinds of stress (e.g., perceptualdegradation, time-compressed speech, or cognitiveoverload). Under such conditions, listeners find itespecially difficult to process inflections andgrammatical function words, and they also tend tomake errors on complex sentence structures like thepassive (e.g., “The girl was pushed by the boy”) orthe object relative (e.g., “It was the girl who theboy pushed”). These aspects of grammar turn outto be the weakest links in the chain of languageprocessing, and for that reason, they are the first tosuffer when anything goes wrong.

(4) One might argue that Broca’s aphasia is theonly “true” form of agrammatism, because thesepatients show such clear deficits in both expressiveand receptive grammar. However, numerousstudies have shown that these patients retainknowledge of their grammar, even though theycannot use it efficiently for comprehension orproduction. For example, Broca’s aphasics performwell above chance when they are asked to detectsubtle errors of grammar in someone else’s speech,and they also show strong language-specific biasesin their own comprehension and production. Tooffer just one example, the article before the nounis marked for case in German, carrying crucialinformation about who did what to whom. Perhapsfor this reason, German Broca’s struggle to produce

the article 90% of the time (and they usually get itright), compared with only 30% in EnglishBroca’s. This kind of detailed difference can onlybe explained if we assume that Broca’s aphasicsstill “know” their grammar.

Taken together, these lines of evidence haveconvinced us that grammar is not selectively lost inadult aphasia, leaving vocabulary intact. Instead,grammar and vocabulary tend to break down together,although they can break down in a number ofinteresting ways.

Neural Imaging Studies . To date, there arevery few neural imaging studies of normal adultscomparing lexical and grammatical processing, but thefew that have been conducted also provide little supportfor a modular view. Many different parts of the brainare active when language is processed, including areas inthe right hemisphere (although these tend to show lowerlevels of activation in most people). New “languagezones” are appearing at a surprising rate, including areasthat do not result in aphasia if they are lesioned, e.g.,the cerebellum, parts of frontal cortex far from Broca’sarea, and basal temporal regions on the underside of thecortex. Furthermore, the number and location of theregions implicated in word and sentence processingdiffer from one individual to another, and change as afunction of the task that subjects are asked to perform.

Although most studies indicate that word andsentence processing elicit comparable patterns ofactivation (with greater activation over left frontal andtemporal cortex), several studies using ERP or fMRIhave shown subtle differences in the patterns elicited bysemantic violations (e.g., “I take my coffee with milkand dog”) and grammatical errors (e.g., “I take my coffeewith and milk sugar.”). Subtle differences have alsoemerged between nouns and verbs, content words andfunction words, and between regular inflections (e.g.,“walked”) and irregular inflections (e.g., “gave”). Suchdifferences constitute the only evidence to date in favorof some kind of separation in the neural mechanismsresponsible for grammar vs. semantics, but they canalso be explained in terms of mechanisms that are notspecific to language at all. For example, content wordslike “milk,” function words like “and,” and long-distance patterns of subject-verb agreement like “Thegirls......are talking” differ substantially in their length,phonetic salience, frequency, degree of semanticimagery, and the demands that they make on attentionand working memory—dimensions that also affectpatterns of activation in nonlinguistic tasks.

To summarize, lesion studies and neural imagingstudies of normal speakers both lead to the conclusionthat many different parts of the brain participate in wordand sentence processing, in patterns that shift indynamic ways as a function of task complexity anddemands on information processing. There is noevidence for a unitary grammar module or a localizedneural dictionary. Instead, word and sentence processingappear to be widely distributed and highly variable overtasks and individuals. However, some areas do seem to

19

be more important than others, especially those areas inthe frontal and temporal regions of the left hemispherethat are implicated most often in patients with aphasia.

IV. PRAGMATICS AND DISCOURSEWe defined pragmatics earlier as the study of

language in context, as a form of communication usedto accomplish certain social ends. Because of itsheterogeneity and uncertain borders, pragmatics isdifficult to study and resistant to the neat division intoprocessing, development and brain bases that we haveused so far to review sounds, words and grammar.

Processing. Within the modular camp, a numberof different approaches have emerged. Someinvestigators have tried to treat pragmatics as a singlelinguistic module, fed by the output of lower-levelphonetic, lexical and grammatical systems, responsiblefor subtle inferences about the meaning of outputs in agiven social context. Others view pragmatics as acollection of separate modules, including specialsystems for the recognition of emotion (in face andvoice), processing of metaphor and irony, discoursecoherence, and social reasoning (including a “theory ofmind” module that draws conclusions about how otherpeople think and feel). Still others relegate pragmaticsto a place outside of the language module altogether,handled by a General Message Processor or ExecutiveFunction that also deals with nonlinguistic facts. Eachof these approaches has completely different con-sequences for predictions about processing, developmentand/or neural mediation. Within the interactive camp,pragmatics is not viewed as a single domain at all.Instead, it can be viewed as the cause of linguisticstructure, the set of communicative pressures underwhich all the other linguistic levels have evolved.

Perhaps because the boundaries of pragmatics are sohard to define, much of the existing work on discourseprocessing is descriptive, concentrating on the way thatstories are told and coherence is established andmaintained in comprehension and production, withoutinvoking grand theories of the architecture of mind(Gernsbacher, 1994). However, a few studies haveaddressed the issue of modularity in discourseprocessing, with special reference to metaphor.Consider a familiar metaphor like “kicked the bucket.”In American English, this is a crude metaphor for death,and it is so familiar that we rarely think of it in literalterms (i.e., no bucket comes to mind). However, it hasbeen suggested that we actually do compute the literalmeaning in the first stages of processing, because thatmeaning is the obligatory product of “bottom-up”lexical and grammatical modules that have no access tothe special knowledge base that handles metaphors. Byanalogy to the contextually irrelevant meanings ofambiguous words like “bug,” these literal inter-pretations rise up but are quickly eliminated in favor ofthe more familiar and more appropriate metaphoricinterpretation. Although there is some evidence forthis kind of “local stupidity,” other studies havesuggested instead that the metaphoric interpretation

actually appears earlier than the literal one, evidenceagainst an assembly line view in which each moduleapplies at a separate stage.

Development. There are no clear milestonesthat define the acquisition of pragmatics. The socialuses of language to share information or request helpappear in the first year of life, in gesture (e.g., ingiving, showing and pointing) and sound (e.g., cries,grunts, sounds of interest or surprise), well before wordsand sentences emerge to execute the same functions.Social knowledge is at work through the period inwhich words are acquired. For example, when an adultpoints at a novel animal and says “giraffe”, childrenseem to know that this word refers to the whole object,and not to some interesting feature (e.g., its neck) or tothe general class to which that object belongs (e.g.,animals). Predictably, some investigators believe thatthese constraints on meaning are innate, while othersinsist that they emerge through social interaction andlearning across the first months of life. Socialknowledge is also at work in the acquisition ofgrammar, defining (for example) the shifting set ofpossible referents for pronouns like “I” and “you,” andthe morphological processes by which verbs are made toagree with the speaker, the listener or someone else inthe room. When we decide to say “John is here” vs.“He is here,” we have to make decisions about thelistener’s knowledge of the situation (does she know towhom “he” refers?), a decision that requires the abilityto distinguish our own perspective from someone else’spoint of view. In a language like Italian, the child hasto figure out why some people are addressed with theformal pronoun “Lei” while others are addressed with“tu,” a complex social problem that often eludessophisticated adults trying to acquire Italian as a secondlanguage. The point is that the acquisition of pragma-tics is a continuous process, representing the interfacebetween social and linguistic knowledge at every stageof development (Bates, 1976).

Brain Bases. Given the diffuse and hetero-geneous nature of pragmatics, and the failure of thephrenological approach even at simpler levels of soundand meaning, we should not expect to find evidence fora single “pragmatic processor” anywhere in the humanbrain. However, there is some evidence to suggest thatthe right hemisphere plays an special role in someaspects of pragmatics (Joanette & Brownell, 1990),including the perception and expression of emotionalcontent in language, the ability to understand jokes,irony and metaphor, and the ability to produce andcomprehend coherent discourse. These are the domainsthat prove most difficult for adults with right-hemisphere injury, and there is some evidence (howeverslim) that the right hemisphere is especially active innormal adults on language tasks with emotionalcontent, and/or on the processing of lengthy discoursepassages.

This brings us back to a familiar theme: Does thecontribution of the right hemisphere constitute evidencefor a domain-specific adaptation, or is it the result of

20

much more general differences between the left and theright hemisphere? For example, the right hemispherealso plays a greater role in the mediation of emotion innonverbal contexts, and it is implicated in the dis-tribution of attention and the integration of informationin nonlinguistic domains. Many of the functions thatwe group together under the term “pragmatics” have justthese attributes: Metaphor and humor involve emotionalcontent, and discourse coherence above the level of thesentence requires sustained attention and informationintegration. Hence specific patterns of localization forpragmatic functions could be the by-product of moregeneral information-processing differences between thetwo hemispheres.

One approach to the debate about innateness anddomain specificity comes from the study of languagedevelopment in children with congenital brain injuries,involving sites that lead to specific forms of aphasiawhen they occur in adults. If it is the case that thehuman brain contains well-specified, localized proces-sors for separate language functions, then we wouldexpect children with early focal brain injury to displaydevelopmental variants of the various aphasic syn-dromes. This is not the case. In fact, children withearly unilateral brain lesions typically go on to acquirelanguage abilities that are well within the normal range(although they do tend to perform lower on a host oftasks than children who are neurologically intact).

An illustration of this point can be seen in Figure12, which ties together several of the points that wehave made throughout this chapter (Kempler, vanLancker, Marchman, & Bates, 1996). Kempler et al.compared children and adults with left- vs. right-hemisphere damage on a measure called the FamiliarPhrases Task, in which subjects are asked to point tothe picture that matches either a familiar phrase (e.g.,“She took a turn for the worse”) or a novel phrasematched for lexical and grammatical complexity (e.g.,“She put the book on the table”). Performance on thesetwo kinds of language is expressed in z-scores, whichindicate how far individual children and adults deviatefrom performance by normal age-matched controls.Adult patients show a strong double dissociation on thistask, providing evidence for our earlier conclusion aboutright-hemisphere specialization for metaphors andcliches: Left-hemisphere patients are more impairedacross the board (i.e., they are aphasic), but they doespecially badly on the novel phrases; right-hemispherepatients are close to normal on novel phrases, but theyperform even worse than aphasics on familiar phrases.A strikingly different pattern occurs for brain-injuredchildren. Although these children all sustained theirinjuries before six months of age, they were 6-12 yearsold at time of testing. Figure 12 shows that thechildren are normal for their age on the familiar phrases;they do lag behind their agemates on novel expressions,but there is no evidence for a left-right difference on anyaspect of the task. In fact, findings like these aretypical for the focal lesion population. The adult brainis highly differentiated, and lesions can result in

irreversible injuries. The infant brain is far moreplastic, and it appears to be capable of significantreorganization when analogous injuries occur. Aboveall, there is no evidence in this population for an innate,well-localized language faculty.

CONCLUSIONTo conclude, we are the only species on the planet

capable of a full-blown, fully grammaticized language.This is a significant accomplishment, but it appears tobe one that emerges over time, from simplerbeginnings. The construction of language is accom-plished with a wide range of tools, and it is possiblethat none of the cognitive, perceptual and socialmechanisms that we use in the process have evolved forlanguage alone. Language is a new machine that Naturebuilt out of old parts.

How could this possibly work? If language is nota “mental organ”, based on innate and domain-specificmachinery, then how has it come about that we are theonly language-learning species? There must beadaptations of some kind that lead us to thisextraordinary outcome.

To help us think about the kind of adaptation thatmay be responsible, consider the giraffe’s neck.Giraffes have the same 24 neckbones that you and Ihave, but they are elongated to solve the peculiarproblems that giraffes are specialized for (i.e., eatingleaves high up in the tree). As a result of this particularadaptation, other adaptations were necessary as well,including cardiovascular changes (to pump blood all theway up to the giraffe’s brain), shortening of thehindlegs relative to the forelegs (to ensure that thegiraffe does not topple over), and so on. Should weconclude that the giraffe's neck is a "high-leaf-eatingorgan"? Not exactly. The giraffe's neck is still a neck,built out of the same basic blueprint that is used overand over in vertebrates, but with some quantitativeadjustments. It still does other kinds of “neck work”,just like the work that necks do in less specializedspecies, but it has some extra potential for reaching uphigh in the tree that other necks do not provide. If weinsist that the neck is a leaf-reaching organ, then wehave to include the rest of the giraffe in that category,including the cardiovascular changes, adjustments in leglength, and so on. I believe that we will ultimatelycome to see our "language organ" as the result ofquantitative adjustments in neural mechanisms thatexist in other mammals, permitting us to walk into aproblem space that other animals cannot perceive muchless solve. However, once it finally appeared on theplanet, it is quite likely that language itself began toapply adaptive pressure to the organization of thehuman brain, just as the leaf-reaching adaptation of thegiraffe's neck applied adaptive pressure to other parts ofthe giraffe. All of the neural mechanisms thatparticipate in language still do other kinds of work, butthey have also grown to meet the language task.Candidates for this category of “language-facilitatingmechanisms” might include our social organization, our

21

extraordinary ability to imitate the things that otherpeople do, our excellence in the segmentation of rapidauditory stimuli, our fascination with joint attention(looking at the same events together, sharing newobjects just for the fun of it). These abilities arepresent in human infants within the first year, and theyare clearly involved in the process by which language isacquired. We are smarter than other animals, to be sure,but we also have a special love of symboliccommunication that makes language possible.

REFERENCESAslin, R.N., Jusczyk, P.W., & Pisoni, D.B. (in press).

Speech and auditory processing during infancy:Constraints on and precursors to language. In W.Damon (Series Ed.) & D. Kuhn & R. Siegler (Vol.Eds.), Handbook of child psychology: Vol. 5.Cognition, perception & language (5th ed.) NewYork: Wiley.

Bates, E. (1976). Language and context: Studies in theacquisition of pragmatics. New York: AcademicPress.

Bates, E., Bretherton, I., & Snyder, L. (1988). Fromfirst words to grammar: Individual differences anddissociable mechanisms. New York: CambridgeUniversity Press.

Bates, E., & Goodman, J. (in press). On theinseparability of grammar and the lexicon: Evidencefrom acquisition, aphasia and real-time processingIn G. Altmann (Ed.), Special issue on the lexicon,Language and Cognitive Processes.

Bishop, D.V.M. (1997). Uncommon understanding:Development and disorders of comprehension inchildren. Hove, UK: Psychology Press/Erlbaum.

Braine, M.D.S. (1976). Children’s first wordcombinations. With commentary by MelissaBowerman. Monographs of the Society forResearch in Child Development, 41, Serial # 164.

Chomsky, N. (1957). Syntactic structures. TheHague: Mouton.

Chomsky, N. (1988). Language and problems ofknowledge. Cambridge, MA: MIT Press.

Dronkers, N.F. (1996). A new brain region forcoordinating speech articulation. Nature, 384, 159-161.

Eimas, P.D., Siqueland, E., Jusczyk, P., & Vigorito, J.(1971). Speech perception in infants. Science,171, 305-6.

Elman, J., Bates, E., Johnson, M., Karmiloff-Smith,A., Parisi, D., & Plunkett, K. (1996). Rethinkinginnateness: A connectionist perspective ondevelopment. Cambridge, MA: MIT Press/BradfordBooks.

Elman, J. & McClelland, J. (1986). Interactiveprocesses in speech perception: The TRACEmodel. In D. Rumelhart & J.L. McClelland (Eds.),Parallel distributed processing: Explorations in themicrostructure of cognition. Cambridge, MA: MITPress.

Erhard, P., Kato, T., Strick, P.L., & Ugurbil, K.(1996). Functional MRI activation pattern ofmotor and language tasks in Broca’s area (Abstract).Society for Neuroscience, 22, 260.2.

Fenson, L., Dale, P.A., Reznick, J.S., Bates, E., Thal,D., & Pethick, S.J. (1994). Variability in earlycommunicative development. Monographs of theSociety for Research in Child Development, SerialNo. 242, Vol. 59, No. 5.

Fletcher, P., & MacWhinney, B. (Eds.). (1995). Thehandbook of child language. Oxford: BasilBlackwell.

Fodor, J. (1983). The modularity of mind. Cambridge,MA: MIT Press.

Fodor, J. A., Bever, T.G., & Garrett, M.F. (1974).The psychology of language: An introduction topsycholinguistics and generative grammar. NewYork, McGraw-Hill.

Freud, A. (1953). On aphasia: A critical study [E.Stengel, Trans.]. New York: InternationalUniversities Press. (Original work published in1891).

Gelman, D. (1986, December 15). The mouths ofbabes: New research into the mysteries of howinfants learn to talk. Newsweek, 86-88.

Gernsbacher, M.A. (Ed.) (1994). Handbook ofpsycholinguistics. San Diego: Academic Press.

Head, H. (1926). Aphasia and kindred disorders ofspeech. Cambridge, UK: Cambridge UniversityPress.

Joanette, Y., & Brownell, H.H. (Eds.). (1990).Discourse ability and brain damage: Theoretical andempirical perspectives. New York : Springer-Verlag.

Kawamoto, A.H. (1988). Distributed representations ofambiguous words and their resolution in aconnectionist network. In S. Small, G. Cottrell,& M. Tanenhaus (Eds.), Lexical ambiguityresolution: Perspectives from psycholinguistics,neuropsychology, and artificial intelligence. SanMateo, CA: Morgan Kaufman.

Kean, M.-L. (Ed.).(1985). Agrammatism. Orlando:Academic Press.

Kempler, D., van Lancker, D., Marchman, V., & Bates,E. (1996). The effects of childhood vs. adult braindamage on literal and idiomatic languagecomprehension (Abstract). Brain and Language,55(1), 167-169

Kempler, D., van Lancker, D., Marchman, V., & Bates,E. (in press). Idiom comprehension in childrenand adults with unilateral brain damage.Developmental Neuropsychology.

Kuhl, P.K., & Miller, J. D. (1975). Speech perceptionby the chinchilla: Voiced-voiceless distinction inalveolar plosive consonants. Science, 190(4209),69-72.

Langacker, R. (1987). Foundations of cognitivegrammar. Stanford: Stanford University Press.

Leonard, L.B. (1997). Specific language impairment.Cambridge, MA: MIT Press.

22

MacDonald, M.C., Pearlmutter, N.J., & Seidenberg,M.S. (1994). Lexical nature of syntacticambiguity resolution. Psychological Review,101(4), 676-703.

McCawley, J.D. (1993). Everything that linguistsalways wanted to know about logic. Chicago:University of Chicago Press.

Menn, L., & Stoel-Gammon, C. (1995). Phonologicaldevelopment. In P. Fletcher & B. MacWhinney(Eds.), Handbook of child language (pp. 335-359).Oxford: Basil Blackwell.

Onifer, W., & Swinney, D.A. (1981). Accessinglexical ambiguities during sentence comprehension:Effects of frequency of meaning and contextualbias. Memory & Cognition, 9(3), 225-236.

Poeppel, D. (1996). A critical review of PET studies ofphonological processing. Brain and Language,55(3), 352-379.

Pick, A. (1973). Aphasia. (J. Brown, Trans., & Ed.)Springfield, IL: Charles C. Thomas. (Originalwork published 1913).

Rumelhart D., & McClelland J.L. (Eds.).(1986).Parallel distributed processing: Explorations in themicrostructure of cognition. Cambridge, MA: MITPress.

Samuel, A.G. (1981). Phonemic restoration: Insightsfrom a new methodology. Journal of ExperimentalPsychology: General, 110, 474-494.

Slobin, D. (Ed.). (1985-1997). The crosslinguisticstudy of language acquisition (Vols. 1-5). Hills-dale, NJ: Erlbaum.

Sperber, D. & Wilson, D. (1986). Relevance:Communication and cognition. Cambridge, MA:Harvard University Press.

Tanenhaus, M.K., & Lucas, M.M. (1987). Contexteffects in lexical processing. In U. Frauenfelder &L.K. Tyler (Eds.), Spoken word recognition(Cognition special issue) pp. 213-234. Cambridge,MA: MIT Press.

Thelen, E., & Smith, L.B. (1994). A dynamic systemsapproach to the development of cognition andaction. Cambridge, MA: MIT Press.

Toga, A.W., Frackowiak, R.S.J., & Mazziotta, J.C.(Eds.). (1996). Neuroimage, A Journal of BrainFunction, Second International Conference onFunctional Mapping of the Human Brain, 3(3) Part2. San Diego: Academic Press.

Tomasello, M., & Call, J. (1997). Primate cognition.Oxford University Press.

Van Petten, C., & Kutas, M. (1991). Electro-physiological evidence for the flexibility of lexicalprocessing. In G.B. Simpson, Ed., Understandingword and sentence. Amsterdam: Elsevier.