plasticid..

8/4/2019 plasticid..

1/31

PLASTICITY, LOCALIZATION AND LANGUAGE DEVELOPMENT

Elizabeth Bates

University of California, San Diego

Support for the work described here has been provided by NIH-NIDCD P50

DC1289-9351 (Origins of communication disorders) and NIH/NINDS P50

NS22343 (Center for the Study of the Neural Bases of Language and Learning).

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 [email protected].

In S.H. Broman and J.M. Fletcher (Eds.), The changing nervous system:

Neurobehavioral consequences of early brain disorders (pp 214-253).

New York: Oxford University Press.


2/31

2

PLASTICITY, LOCALIZATION AND LANGUAGE DEVELOPMENT

Elizabeth Bates

Universi ty of Cal ifornia, San Diego

The term aphasia refers to acute or chronicimpairment of language, an acquired condition that ismost often associated with damage to the left side of the

brain, usually due to trauma or stroke. We have knownabout the link between left-hemisphere damage andlanguage loss for more than a century (Goodglass,1993). For almost as long, we have also known thatthe lesion/symptom correlations observed in adults donot appear to hold for very young children (Basser,1962; Lenneberg, 1967). In fact, in the absence ofother complications, infants with congenital damage toone side of the brain (left or right) usually go on toacquire language abilities that are well within thenormal range (Eisele & Aram, 1995; Feldman, Holland,& Janosky, 1992; Vargha-Khadem, Isaacs, & Muter,1994). To be sure, children with a history of earlybrain injury typically perform below neurologically

intact age-matched controls on a host of language andnonlanguage measures, including an average full-scaleIQ difference somewhere between 4-8 points from onestudy to another (especially in children with persistentseizures -- Vargha-Khadem et al., 1994). Brain damageis not a good thing to have, and some price must bepaid for wholesale reorganization of the brain tocompensate for early injuries. But the critical point forpresent purposes is that these children are not aphasic,despite early damage of a sort that often leads toirreversible aphasia when it occurs in an adult.

In addition to the reviews by other authors citedabove, my colleagues and I have also published severaldetailed reviews of language, cognition and

communicative development in children with focal braininjury, from various points of view (e.g., Bates et al.,1997; Bates, Vicari, & Trauner, in press; Elman et al.,1996; Reilly, Bates & Marchman, 1998; Stiles, 1995;Stiles, Bates, Thal, Trauner, & Reilly, 1998; Stiles &Thal, 1993; Thal et al., 1991). As these reviews attest,a consensus has emerged that stands midway betweenthe historical extremes of equipotentiality (Lenne-berg, 1967) and innate predetermination of theadult pattern of brain organization for language (e.g.,Curtiss, 1988; Stromswold, 1995). The twohemispheres are certainly not equipotential for languageat birth; indeed, if they were it would be impossible toexplain why left-hemisphere dominance for language

emerges 95%-98% of the time in neurologically intactindividuals. However, the evidence for recovery fromearly left-hemisphere damage is now so strong that it isno longer possible to entertain the hypothesis thatlanguage per se is innately and irreversibly localized toperisylvian regions of the left hemisphere.

The compromise view is one in which brainorganization for language emerges gradually across thecourse of development (Elman et al., 1996; Karmiloff-Smith, 1992), based on soft constraints that are onlyindirectly related to language itself. Hence the familiarpattern of language localization in adults is the product

rather than the cause of development, an end-productthat emerges out of initial variations in the way thatinformation is processed from one region to another.

Crucially, these variations are not specific to language,although they do have important implications for howand where language is acquired and processed. In theabsence of early brain injury, these soft constraints inthe initial architecture and information-processingproclivities of the left hemisphere will ultimately leadto the familiar pattern of left-hemisphere dominance.However, other brain plans for language are possible,and will emerge when the default situation does nothold.

In the pages that follow, I do not intend to provideanother detailed review of the outcomes associated withearly brain injury; the reader is referred elsewhere for amore complete catalogue of such findings. What I

would like to do instead is to go beyond these findingsto their implications for the nature and origins oflanguage localization in the adult, providing an accountof how this neural system might emerge across thecourse of development. With this goal in mind, thechapter is organized as follows: (1) a very brief reviewof findings from developmental neurobiology that serveas animal models for the kind of plasticity that we seein human children; (2) an equally brief illustration ofresults from retrospective studies of languagedevelopment in the focal lesion population; (3) thedistinction between prospective and retrospectivestudies, including a discussion of putative criticalperiods for language development; (4) an overview of

prospective findings on language development inchildren with congenital lesions to one side of the brain;(5) a new view of brain organization for language in theadult, an alternative to the static phrenological view thathas dominated our thinking for two centuries, one thattakes into account the role of experience in specifyingthe functional architecture of the brain.

(1) DEVELOPMENTAL PLASTICITY:ANIMAL MODELS

Evidence for the plasticity of language in thehuman brain should not be surprising in light of all thathas been learned in the last few decades aboutdevelopmental plasticity of isocortex in other species(Bates, Thal, & Janowsky, 1992; Deacon, 1997; Elmanet al., 1996, Chapter 5; Janowsky & Finlay, 1986;Johnson, 1997; Killackey, 1990; Mueller, 1996; Quartz& Sejnowski, 1997; Shatz, 1992; PAPERS IN THISVOLUME). Without attempting an exhaustive or evena representative review, here are just a few of myfavorite examples of research on developmental plas-ticity in other species, studies that provide animalmodels for the kind of plasticity that we have observedin the human case.


3/31

3

Isacson and Deacon (1996) have transplanted plugsof cortex from the fetal pig into the brain of the adultrat. These foreigners (called xenotransplants)develop appropriate connections, including functioningaxonal links down the spinal column that stop inappropriate places. Although we know very little aboutthe mental life of the resulting rat, no signs of pig-appropriate behaviors have been observed.

Stanfield and OLeary (1985) have transplantedplugs of fetal cortex from one region to another (e.g.,from visual to motor or somatosensory cortex).Although these cortical plugs are not entirely normalcompared with native tissue, they set up functionalconnections with regions inside and outside the cortex.More importantly still, the transplants developrepresentations (i.e., cortical maps) that are appropriatefor the region in which they now live, and not for theregion where they were born (When in Rome, do as theRomans do....).

Sur and his colleagues (Pallas & Sur, 1993; Sur,Pallas, & Roe, 1990) have rerouted visual informationfrom visual cortex to auditory cortex in the infant ferret.

Although (again) the representations that develop inauditory cortex are not entirely normal, theseexperiments show that auditory tissue can developretinotopic maps. It seems that auditory cortexbecomes auditory cortex under normal conditionsprimarily because (in unoperated animals) it receivesinformation from the ear; but if it has to, it can alsoprocess visual information in roughly appropriate ways.

Killackey and his colleagues have modified thebody surface of an infant rat, by removing whiskers thatserve as critical perceptual organs in this species(Killackey et al., 1994). Under normal conditions, thesomatosensory cortex of the rat develops representations(barrel cells) that are isomorphic with input from the

whisker region. In contrast, the altered animals developsomatosensory maps reflecting changes in theperiphery, with expanded representations for theremaining whiskers; regions that would normallysubserve the missing whiskers are reduced or absent(Killackey, 1990). In other words, the rat ends up withthe brain that it needs, rather than the brain that Natureintended.

Finally, in an example that may be closer to theexperience of children with early focal brain injury, arecent study by Webster, Bachevalier and Ungerleider(1995) shows that the where is it system (mediated indorsal regions, especially parietal cortex, including areaMT) can take over the functions of the what is it

system (mediated in ventral regions, especially inferiortemporal cortex, including area TE). When area TE isbilaterally removed in an adult monkey, that animaldisplays severe and irreversible amnesia for new objects,suggesting that this area plays a crucial role inmediating object memory and detection (i.e., the so-called what is it system). However, as Webster et al.have shown, bilateral removal of area TE in infantmonkeys leads to performance only slightly below age-matched unoperated controls (at both 10 months and 4years of age). If area TE is no longer available, wherehas the what is it system gone? By lesioning

additional areas of visual cortex, Webster et al. showedthat the object detection function in TE-lesioned infantmonkeys is mediated by dorsal regions of extrastriatecortex that usually respond to motion rather than form(i.e., the where is it system). In other words, a majorhigher cognitive function can develop far away from itsintended site, in areas that would ordinarily play little orno role in the mediation of that function.

These examples and many others like them have ledmost developmental neurobiologists to conclude thatcortical differentiation and functional specialization arelargely the product of input to the cortex, albeit withincertain broad architectural and computational constraints(Johnson, 1997). Such findings provide a seriouschallenge to the old notion that the brain is organizedinto largely predetermined, domain-specific faculties,i.e., the phrenological approach. An alternativeproposal that is more compatible with these findingswill be offered later on.

(2) LANGUAGE OUTCOMES INCHILDREN WITH EARLY FOCAL BRAIN

INJURY: RETROSPECTIVE FINDINGSAs noted earlier, retrospective studies of language

outcomes in children with unilateral brain injury haverepeatedly found that these children are not aphasic; theyusually perform within the normal range, although theyoften do perform slightly below neurologically intactage-matched controls (cf. Webster, Bachevalier, &Ungerleider, 1995). More importantly for our purposeshere, there is no consistent evidence in theseretrospective studies to suggest that language outcomesare worse in children with left-hemisphere damage,compared with children whose injuries are restricted tothe right hemisphere. Without attempting an exhaus-tive review, three examples will suffice to illustrate

these points.Figures 1a and 1b (adapted from Bates, Vicari, &

Trauner, in press) present idealized vs. observed resultsfor verbal vs. nonverbal IQ scores in a cross-sectionalsample of children with congenital injuries who weretested at various ages between 3 and 10 years. Figure1a illustrates what we might expect if the left/rightdifferences observed in adults were consistently observedin children: higher verbal than nonverbal IQ scores inchildren with right-hemisphere damage (RHD), whichmeans that these children should line up on the upperdiagonal; higher nonverbal than verbal IQ scores inchildren with left-hemisphere damage (LHD), whichmeans that these children ought to fall on the lower

diagonal. These idealized scores were obtained bytaking actual pairs of scores for individual children inour focal lesion sample, and reversing any scores thatwere not in the predicted direction. In contrast with thisidealized outcome, Figure 1b illustrates the actual verbaland performance IQ scores for 28 LHD and 15 RHDcases (note that there are no differences between thesetwo groups in gender or chronological age, and no meandifferences in full scale IQ). The actual data in Figure1b illustrate several points. First, in line with otherstudies of this population, the mean full-scale IQ for the


4/31

15014013012011010090807060500

Left -Hemisphere Injur

Right -Hemisphere Inju

Figure 1 a: Idealized Relat ion bet ween Verbal & Perf orm ance IQ

in Children wit h Left - vs. Right -Hemisphere Injury

(adapt ed f rom Bat es, Vicari & Trauner, in press)

Perfo rmance IQ


5/31

15014013012011010090807060500

Left -Hemisphere Inju

Right -Hemisphere Inju

Figure 1b: Observed Relat ion bet ween Verbal & Perf orm ance IQ

in Children wit h Left - vs. Right -Hemisphere Injury

(adapt ed fr om Bat es, Vicari & Trauner, in press)

Perf ormance IQ


6/31

4

sample as a whole is 93.2, within the normal range butbelow the mean of 100 that we would expect if we weredrawing randomly from the normal population. Second,the range of outcomes observed in the focal lesionpopulation as a whole is extraordinarily broad, includingsome children who can be classified as mentally retarded(i.e., 16.3% of this sample have full scale IQs at orbelow 80), and some with IQs over 120. Third, the

correlation between the verbal and nonverbal subscalesis relatively strong (+.65, p < .0001), which means thatverbal and nonverbal IQ do not dissociate markedly inthis group. In fact, as we can clearly see from thedifference between Figure 1a (predicted outcomes) and1b (the outcomes actually observed in these children),there is absolutely no evidence in these data for a doubledissociation between verbal and nonverbal IQ as afunction of left- vs. right-hemisphere injury.

Figure 2 (adapted from Reilly et al., 1998) presentsresults from a more focused study of grammaticaldevelopment, illustrating the number of differentcomplex syntactic forms produced in a narrativediscourse task by LHD, RHD and neurologically intact

controls who were tested between 6 and 12 years of ageat testing. This figure demonstrates (once again) thatchildren with focal brain injury perform within thenormal range in production of complex syntax, eventhough they do (as a group) score significantly belowneurologically intact controls. In this respect, theReilly et al. result for grammatical development inhuman children is remarkably similar to the findingsreported by Webster et al., on the relative preservationof memory for novel objects in infant monkeys withbilateral TE lesions (i.e., performance roughly 10%below that of normal controls). In addition, Figure 2shows that there is no evidence in this age range for adifference in syntactic production as a function of lesion

side or site.Finally, Figures 3a and 3b (from Kempler, van

Lancker, Marchman, & Bates, in press) compare resultsfor adults and 6-12-year-old children with LHD vs. RHDon the same sentence comprehension task. The data inboth figures are all based on z-scores, with patients ateach age level compared with the performance of age-matched normal controls (hence the difference inperformance between normal adults and normal 6-12-year-old children is factored out of the results). In thisparticular procedure, subjects are asked to match eachstimulus sentence to one of four pictured alternatives.Half the items are familiar phrases (well-knownmetaphors and figures of speech like She took a turn

for the worse), and the other half are novel phrasesmatched to the familiar phrases in length and com-plexity. As Figure 3a shows, there is a powerfuldouble dissociation between novel and familiar phrasesin adult victims of unilateral brain injury: adults withLHD score markedly better on the familiar phrases,while adults with RHD score better on the novelphrases. This is one example of a growing body ofevidence challenging the old assumption that the lefthemisphere is the language hemisphere, even inadults. The right hemisphere does make an importantcontribution to language processing, but its

contribution is qualitatively different from that of theleft hemisphere, involving a number of functionsincluding emotionality, intonation contours and (as thisexample illustrates) figurative, metaphorical and/orformulaic speech (all forms of speech in which themeaning of the sentence as a whole goes beyond themeaning one would obtain by computing across theseparate elements in the sentence). A comparison

between Figures 3a and 3b helps to clarify threeimportant points. First, children with focal injuries farefar better than adul ts with comparable damage, whenthey are compared with age-matched controls. Second,the powerful double dissociation observed in adults isnot observed in children. Third, novel sentences aremore susceptible to the effects of brain injury thanfamiliar phrases in the child group, but RHD childrenactually perform below the LHD group in compre-hension of novel sentences (significant by a one-tailed t-test), the opposite of what we might expect if the adultpattern held for children with focal brain injury.

In short, whether we are talking about globalmeasures like IQ or more subtle measures of sentence

production and comprehension, children with LHD vs.RHD do not display the profiles of impairment that wewould expect based on the adult aphasia literature -- atleast not in these and other retrospective studies, withoutcome measures at or above six years of age (i.e.,beyond the point at which fundamental aspects ofgrammar and phonology are usually in place -- Bates,Dale, & Thal, 1995).

(3) AGE OF LESION ONSET AND THE

PROBLEM OF CRITICAL PERIODS

The distinction between retrospective and prospec-tive studies is related to the controversial problem ofcritical periods for language, with special focus on the

age at which a lesion is acquired. By definition,prospective studies focus on children whose lesions areacquired very early, preferably before the point at whichlanguage learning normally begins. In contrast, manyretrospective studies collapse across children whoacquired their lesions at different points across thecourse of language learning. Our own prospectivestudies are based exclusively on children with congenitalinjuries, defined to include pre- or perinatal injuries thatare known to have occurred before six months of age,restricted to one side of the brain (left or right),confirmed through one or more forms of neural imaging(CT or MRI). Hence our results may differ from studiesof children with injuries acquired at a later point in

childhood.What might those differences be? Unfortunately,

there is very little empirical evidence regarding the effectof age of lesion onset on subsequent language out-comes. Only one fact is clear: that the outcomesassociated with left-hemisphere injury are much betterin infants than they are in adults. This means, ofcourse, that plasticity for language must decreasemarkedly at some point between birth and adulthood(Lenneberg, 1967). But when does this occur, and howdoes it happen?


7/31

LHD RHD NORMALS

Figure 2 : Number of Diff erent Complex Synt act ic Forms Produced by

Children wit h Lef t vs. Right Hemisphere Damage

in a Story-Telling Task (age = 6 to 12 years)

(adapt ed f rom Reilly, Marchman & Bat es, in press)

Lesion Group


8/31

Familiar Novel

Adult LHD

Adult RHD

Figure 3a: Perf orm ance on Familiar vs. Novel Sent ences

in Adult s wit h Left - v s. Right -Hemisphere Injury

(adapted f rom Kempler, Van Lancker, Marchman & Bat es, in press)


9/31

Familiar Novel

C

C

Figure 3b: Perf ormance on Familiar vs. Novel Sentences

in Children wit h Left - vs. Right -Hemisphere Damage

(adapt ed from Kempler, Van Lancker, Marchman & Bat es, in press)


10/31

5

Many investigators have argued that this decrease inplasticity takes place at the end of a critical period forlanguage, a window of opportunity that is alsopresumed to govern the childs ability to achieve native-speaker status in a second language (for discussions, seeBialystok & Hakuta, 1994; Curtiss, 1988; Elman et al.,1996, Chapter 5; Johnson & Newport, 1989;Marchman, 1993; Oyama, 1993; Weber-Fox & Neville,

1996). So much has been said about this presumedcritical period that a newcomer to the field (and manyconsumers within it) would be justified in assumingthat we know a great deal about its borders (i.e., when itbegins and when it comes to an end), and about theshape of the learning function in between these points.The very term critical period suggests that the abilityto acquire a native language and/or the ability to recoverfrom brain injury both come to a halt abruptly, perhapsat the same time, as the window of opportunity slamsshut. The fact is, however, that we know almostnothing about the shape of this function. In fact, weare not even justified in assuming that the function ismonotonic (i.e., that it gets progressively harder to

learn a native language, and progressively harder torecover from injuries to the left hemisphere).With regard to the presumed critical period for

recovery from brain injury, we are aware of only twolarge cross-sectional studies that have comparedlanguage and cognitive outcomes in children whoacquired their lesions at different ages, from congenitalinjuries (at or before birth) through early adolescence(Vargha-Khadem et al., unpublished results, cited withpermission in Bates, Vicari, & Trauner, in press;Goodman & Yude, 1996). Figure 4 compares resultsfrom both these studies for verbal IQ. As this figureindicates, the effect of age of injury is nonmonotonic inboth studies: The worst outcomes are observed in

children who suffered their injuries between approxi-mately 1-4 years of age. In support of the criticalperiod hypothesis, better outcomes are observed fol-lowing congenital injuries. However, in direct contra-diction to the critical period hypothesis, better outcomesare also observed in children whose injuries occurredbetween approximately 5-12 years of age, which meansthat there is no monotonic drop in plasticity. To someextent, these unpleasant wrinkles in the expectedfunction could be due to uncontrolled differences inetiology (e.g., the factors leading to injury may differ atbirth, 1-5 years, and later childhood). At the very least,however, these results ought to make us skeptical ofclaims about a straightforward critical period for

recovery from brain injury.Similar nonmonotonic findings have been reported

in at least one study of second-language acquisition andfirst-language loss (Liu, Bates, & Li, 1992). Toillustrate, compare the results in Figure 5 (adapted froma famous study of second-language acquisition byJohnson and Newport, 1989) and Figure 6 (adapted fromLiu et al., 1992). Figure 5 illustrates results from agrammaticality judgment task administered to first- andsecond-language learners of English, comparing per-formance of individuals who arrived in the U.S. atdifferent points spanning the period from birth to early

adulthood. This well-known figure suggests that thereis no single point at which the window of opportunityfor second-language learning slams shut. However, itdoes provide evidence for a monotonic drop in languagelearning ability from birth to adolescence.

Consider, however, the results in Figure 6, basedon a sentence interpretation task administered toChinese-English bilinguals in both Chinese and

English. In this task, subjects were able to use eithersemantic or word order information to interpret oddsentences like The rock chased the dog. Nativespeakers of English invariably choose the first noun,using word order to make their interpretation. Nativespeakers of Chinese invariably choose the second noun,ignoring word order in favor of semantic information.Both these strategies make perfect sense in terms of theinformation value of standard word order in these twolanguages (Chinese permits so much word ordervariation that a persistent word order strategy like theone used in English would not be very useful). Hencethis litt le task serves as a useful litmus test forretention of the first language (L1) as well as acquisi-

tion of the second (L2). The interesting point for ourpurposes is that Chinese-English bilinguals oftenperform somewhere in between these two extremes, inone or both of their two languages, and these differentweightings of word order and semantic informationvary as a function of age of acquisition. Notice thatresults for English (L2) are generally in agreement withJohnson and Newports results for a very different task:although our results asymptote at an earlier point thanthose of Johnson and Newport, they do provide evidencefor a monotonic shift from English-like interpreta-tions of English sentences in those who learned theirEnglish very early, to Chinese-like interpretation ofEnglish sentences in those who learned their English

relatively late. However, results for Chinese (L1) showa very different function, a nonmonotonic curve inwhich the best results (movement toward the secondlanguage without loss of the first language) are observedin those who are exposed to a second languagesomewhere between 4-7 years of age.

Although this is a complex result, the point of thiscomparison for our purposes here is a simple one: Thereis no single critical period for language learning;results depend on many different factors, and theprobabi lity of a positive outcome can rise or fall atdifferent points in development, in L2 learning and inrecovery from brain injury. This is where prospectivestudies can be particularly illuminating: By studying

children during their first encounters with language andother forms of higher cognition, we can learn moreabout effects associated with the initial state of thebrain, together with the processes of development and(re)organization that lead these children to a normal ornear-normal outcome.


11/31

Congenit al 1 - 5 years 5 - 12 years

Vargha-Khadem et a

Goodman & Yude (19

Figure 4: Relat ionship bet ween Age of Lesion Onset and IQ Scores

in Two Samples of Children w it h Focal Brain Injury

Age of Lesion Onset

Ment al

Cut -Of


12/31

4035302520151050

Figure 5 : Performance on a Grammat icalit y Judgment Task in

Non-Nat ive Speakers of English as a

Funct ion of Age of Exposure t o English

(adapted from Newport & Johnson, 1992)

AGE OF EXPOSURE TO ENGLISH


13/31

0 - 4 6 - 10 12 - 16 > 20

Figure 6: " English-like" vs. " Chinese-like" Grammat ical Compr ehension

as a Funct ion of Age of Exposure t o English

(adapted from Liu, Bates & Li, 1993)

AGE OF EXPOSURE TO ENGLISH

Bilinguals on

Bilinguals on

English Monolin

Chinese Mono


14/31

6

(4) LANGUAGE OUTCOMES INCHILDREN WITH EARLY FOCAL BRAIN

INJURY: PROSPECTIVE FINDINGS

All theories that take some form of plasticity intoaccount (including theories that assume a critical period)would lead us to expect relatively good outcomes inchildren with congenital injuries, i.e., the group that wehave studied in our laboratory. Evidence for the

developmental plasticity of language in this group hasmounted in the last few years, due in part to improvedtechniques for identifying children with early braininjury, including precise localization of the site andextent of damage through neuroradiology. In somecases, we have been able to identify such children in thefirst weeks of life, prior to the time when languageacquisition would normally begin, permitting us tochart the course of language, cognition and communica-tive development from the very beginning (Bates et al.,1997; Reilly, Stiles, Larsen, & Trauner, 1995; Stiles etal., 1998; Stiles & Thal, 1993), before the point atwhich alternative forms of brain organization haveemerged.

In fact, the prospective studies that we have carriedout so far provide compelling evidence for initialdeficits and subsequent processes of recovery --phenomena that are not visible later on, when mostretrospective studies take place. For example, prospec-tive studies of nonverbal cognitive development by ourcolleague Joan Stiles have revealed subtle but consistentpatterns of deficit in visual-spatial cognition. Forexample, children with RHD appear to have difficultyperceiving and/or producing the global or configuralaspects of a complex visual array; children with LHDare generally spared at the global level, but they havedifficulty with the perception and/or production of localdetails (Note: I will return to this example later on,

relating it to our findings for language). These visual-spatial deficits are qualitatively similar to those ob-served in LHD vs. RHD adults, although they areusually more subtle in children, and they resolve overtime as the children acquire compensatory strategies tosolve the same problems (Stiles et al., 1998; Stiles &Thal, 1993).

If a similar result could be found in the domain oflanguage, then we might expect (by analogy to theliterature on adult aphasia) to find the following resultsin the first stages of language development:

-- Left-hemisphere advantage for lan-guage: Children with LHD will perform below thelevels observed in children with RHD on virtually all

measures of phonological, lexical and grammaticaldevelopment, as well as measures of symbolic andcommunicative gesture.

-- The Broca pattern: By analogy to Brocasaphasia in adults, children with damage to the frontalregions of the left hemisphere will be particularlydelayed in expressive but not receptive language, andmay (on some accounts) be particularly delayed in thedevelopment of grammar and phonology;

-- The Wernicke pattern: By analogy toWernickes aphasia in adults, children with damage to

the posterior regions of the left temporal lobe will beparticularly delayed in receptive language, perhaps (onsome accounts) with sparing of grammar and phonologybut selective delays in measures of semantic develop-ment.

Our group set out to test these three hypotheses ina series of prospective studies of early languagedevelopment. In every case, we have uncovered

evidence for early deficits, and these deficits do appear tobe associated with specific lesion sites. However, incontrast with Stiles findings for visual-spatial cogni-tion, results for language provide very little evidence forhypotheses based on the adult aphasia literature.

The first study (Marchman, Miller, & Bates, 1991)focused on the emergence of babbling and first words ina small sample of five children with congenital braininjury, two with RH damage, three with LH damage,including one LH case with injuries restricted to the leftfrontal region. All the children were markedly delayedin phonological development (babbling in consonant-vowel segments weeks or months behind a group ofneurologically intact controls), and in the emergence of

first words. However, three of the children moved upinto the normal range across the course of the study.The two who remained behind had injuries to theposterior regions of the left hemisphere, results that fitwith the first hypothesis (LH advantage for language)but stand in direct contradiction to both the Broca andthe Wernicke hypotheses.

The second study (Thal et al., 1991) focused oncomprehension and production of words from 12-35months in a sample of 27 infants with focal braininjury, based on a parental report instrument that wasthe predecessor of the MacArthur CommunicativeDevelopment Inventories or MCDI (Fenson et al.,1993, 1994). In complete contradiction to Hypothesis

1 (LH mediation of language) and Hypothesis 3 (theWernicke Hypothesis), delays in word comprehensionwere actually more likely in the RH group. In linewith Hypothesis 1, but against Hypothesis 2 (the BrocaHypothesis), delays in word production were morelikely in children with injuries involving the leftposterior quadrant of the brain.

A more recent study built on the findings of Thal etal. with a larger sample of 53 children, 36 with LHDand 17 with RHD (Bates et al., 1997), using acombination of parent report (the MCDI) and analysesof free speech. This report is broken into threesubstudies, with partially overlapping samples. Study1 used the MCDI to investigate aspects of word

comprehension, word production and gesture at the dawnof language development, in 26 children between 10-17months of age. Study 2 used the MCDI to look atproduction of both words and grammar in 29 childrenbetween 19-31 months. Study 3 used transcripts ofspontaneous speech in 30 children from 20 to 44months, focusing on Mean Length of Utterance inmorphemes (MLU). In all these studies, comparisonsbetween the LHD and RHD groups were followed bycomparisons looking at the effects associated withlesions involving the frontal lobe (comparing childrenwith left frontal involvement to all RHD cases as well


15/31

7

as LHD cases with left frontal sparing) and the temporallobe (comparing children whose lesions include the lefttemporal lobe with all RHD cases and all LHD cases inwhich that region is spared). Results were compatiblewith Marchman et al. and Thal et al., but quitesurprising from the point of view of lesion/symptommappings in adult aphasia, as follows.

First, in a further disconfirmation of Hypotheses 1

and 3, Bates et al. report that delays in wordcomprehension and gesture were both more likely inchildren with unilateral damage to the right hemisphere,at least in the 10-17-month window examined here.Further studies of gestural development in our labora-tory have confirmed that the gestural disadvantage forRH children is still present between 20-24 months(Stiles et al., 1998).

Second, in a partial confirmation of Hypothesis 2(the Broca Hypothesis), frontal involvement wasassociated with greater delays in word production and theemergence of expressive grammar between 19 and 31months. However, in a surprising partial disconfirma-tion of Hypothesis 2, this frontal disadvantage was

equally severe with either left frontal or right frontalinvolvement. In other words, the frontal lobes areimportant during this crucial period of development(which includes the famous vocabulary burst and theflowering of grammar), but there is no evidence for aleft-right asymmetry in the frontal regions, and henceno evidence in support of the idea that Brocas area has aprivileged status from the very beginning of languagedevelopment.

Third, in line with Hypothesis 1 (LH mediation oflanguage) but in direct contradiction to Hypotheses 2and 3 (analogies to Brocas and Wernickes aphasia),delays in word production and the emergence ofgrammar were both more pronounced in children with

injuries involving the left temporal lobe. In contrastwith the above two findings (which only reachedsignificance within a restricted period of development),this left temporal disadvantage was reliable across allthree substudies in Bates et al., from the very firstwords (between 10-17 months of age) through crucialdevelopments in grammar (between 20-44 months ofage). Hence we do have evidence for the asymmetricalimportance of Wernickes area, but that evidencepertains equally to grammar and vocabulary (with noevidence of any kind for a dissociation between thetwo), and seems to be restricted to expressive language.

Reilly et al. (1998) conducted similar comparisonsby lesion side and lesion site in a cross-sectional sample

of 30 children with focal brain injury (15 LH and 15RH) between 3 and 12 years of age; these results werealso compared with performance by a group of 30 age-matched controls with no history of neurologicalimpairment. Analyses were based on lexical, gram-matical and discourse measures from a well-knownstory-telling task. For children between 3-6 years ofage, Reilly et al. replicated the specific disadvantage inexpressive language for children with lesions involvingthe temporal region of the left hemisphere. However,this effect was not detectable in children between 6-12years of age -- even though all children in this study had

the same congenital etiology. In fact, data for the olderchildren provided no evidence of any kind for an effect oflesion side (left vs. right) or lesion site (specific lobeswithin either hemisphere). The only effect that reachedsignificance in older children was a small but reliabledisadvantage in the brain-injured children as a group,compared with neurologically intact age-matchedcontrols. Figure 7 compares results for younger vs.

older children on one grammatical index (mean numberof errors in grammatical morphology per proposition),divided into children with left temporal involvement(+LTemp), focal lesion cases without left temporalinvolving (Ltemp, combining all RHD cases and allLHD cases with temporal sparing), and neurologicallyintact normal controls. Although we must rememberthat these are cross-sectional findings, they suggest thata substantial degree of recovery takes place in the LHgroup during the first few years of life. In subsequentlongitudinal studies, Reilly and her colleagues havefollowed a smaller group of children across this periodof development. These longitudinal findings arecompatible with the cross-sectional evidence in Figure

7, suggesting that the crucial period of recovery takesplace before the age range covered by most of theretrospective studies in the literature on cognitive andlinguistic outcomes in children with focal brain injury.

To summarize, our prospective studies of languagedevelopment in children with early focal brain injuryhave provided evidence for specific delays, correlatedwith specific lesion sites. However, the nature of theselesion/symptom correlations differs markedly fromthose that we would expect based on the adult aphasialiterature. Furthermore, these correlations are onlyobserved within specific windows of development,followed by evidence for recovery and (by implication)reorganization. None of these results are evident in

retrospective studies (including our own), where childrenare tested beyond the point at which this presumedreorganization has taken place.

We are occasionally asked why our results appear tobe incompatible with an earlier literature on the effect ofhemispherectomy (e.g., Dennis & Whitaker, 1976; butsee Bishop, 1983) and/or effects of early stroke (e.g.,Aram, 1988; Aram, Ekelman, & Whitaker, 1985;Aram, Ekelman, Rose, & Whitaker, 1985; Woods &Teuber, 1978). Our first answer is that our results arenot incompatible with the vast majority of studies.However, they do appear to be incompatible with ahandful of studies that were cited (usually in secondarysources) as evidence in favor of an innate and

irreversible role for the left hemisphere in some aspectsof language processing. As we have noted elsewhere(Bates, Vicari, & Trauner, in press; see also Vargha-Khadem et al., 1994), apparent inconsistencies betweenthe earlier studies and our more recent work disappearwhen one looks carefully at the fine print.

First, many of the earlier studies combined data forchildren whose injuries occurred at different points indevelopment, and they also combined results (usuallyon rather global measures) for children at widelydifferent ages at time of testing. As we saw in theprevious section, there may not be a monotonic relation


16/31

Under 5 Years Over 5 Years

0.00

0.05

0.10

0.15

0.20

0.25

0.30

0.35

0.40

0.45

0.50

0.55

0.60

0.65

0.70

0.75

0.80

- Left Temporal

+ Left Temporal

Normal Cont rols

Figure 7: Morpho logical Erro rs as a Funct ion of Age

and Presence/ Absence of Left Temporal Damage

(adapt ed from Reilly, Marchman & Bat es, in press)

AGE GROUP


17/31

8

between age of injury and language outcomes, and thenature of the lesion/symptom mappings that we observemay be quite different depending on the age at whichchildren are tested and the developmental events that aremost prominent at that time.

Second, some of the earlier studies had method-ological limitations that we have been able to overcomein the studies described above. In particular, a number

of well-known studies could not perform directcomparisons of children with LHD vs. RHD, because ofuncontrolled differences in age, education and/oretiology. Instead, the RHD and LHD groups were eachcompared with a separate group of matched controls.For example, Dennis & Whitaker report that their left-hemispherectomized children performed below normalcontrols on subtle and specific aspects of grammaticalprocessing; no such difference was observed betweenright-hemispherectomized children and their controls.These results were interpreted as though they constituteda significant difference between the LHD and RHDgroups, even though the latter two groups were nevercompared directly. As Bishop has pointed out in her

well-known critique (Bishop, 1983), a careful examina-tion of results for the two lesion groups suggests thatthis interpretation is not warranted. The generalproblem that one encounters with the separate controlgroup approach is illustrated in Figure 8, whichcompares hypothetical data for an LHD group, an RHDgroup, and their respective controls. As we can seefrom this figure, performance by the LHD group doesfall reliably below performance by their controls (albeitjust barely); performance by the RHD group does notfall outside the confidence intervals for their controlgroup. And yet, in this hypothetical example,performance is actually better in the LHD cases! Thekey to this conundrum lies in the standard deviations for

each control group: The standard deviation is larger forthe RHD controls, which means that a larger differencebetween RHD and controls is required to reach statisticalsignificance. Clearly, it would be unwise to drawstrong conclusions about left/right differences from adata set of this kind.

Finally, some of the better-known claims in favorof an early and irreversible effect of LH damage havebeen based on single-case studies or very small samples(including the hemispherectomy studies cited above).This fact limits the generalizability of results, and thesame result is often contradicted by other individual-caseor small-group studies.

For example, Stark and McGregor (1997) have

recently described an interesting contrast between onechild with a left hemispherectomy (seizure onset at 1;6,surgery at 4;0), and another with a right hemi-spherectomy (seizure onset at 2;0, surgery at 5;8). Bothchildren were followed longitudinally with testing at 1-2-year intervals through 9;0 and 9;6 years of age,respectively. Although both children did showsubstantial development in language and cognitionacross the course of the study, they fell behind age-matched normal controls at every point. At the end ofthe study, the LHD case had a full-scale IQ of 71 andthe RHD case had a full-scale IQ of 81, well behind the

norms for development in children who areneurologically intact. For Stark & McGregor, the mostinteresting findings lie in the contrasting patternsobserved for each child for performance IQ, verbal IQ,and series of more specific language tests. For the LHDcase, verbal and performance IQ were both quite low(separated by only four points). However, performanceon the specific language tasks followed a profile typical

of the pattern observed in children with SpecificLanguage Impairment, i.e., greater impairment inlanguage measures (especially morphosyntax) than wewould expect for her mental age. By contrast, the RHDcase displayed a sharp dissociation at the end of thestudy between verbal IQ (95) and performance IQ (70),with scores on most of the specific language measuresthat were appropriate for her mental age.

This is an interesting and provocative result, and itmight indeed reflect evidence for the emergence of somekind of left-hemisphere specialization for language priorto the age at which the surgery occurred. However, ourown experience with a relatively large focal lesionsample has made us wary of basing strong results on

case studies. Individual differences in language andcognitive ability are immense, even in perfectly normalchildren with no history of brain injury (Bates et al.,1995; Fenson et al., 1994). A similar degree ofvariation is observed even within the small cadre ofcases that have undergone hemispherectomy.

Evidence for such variation comes from the case ofAlex, recently reported by Vargha-Khadem et al. (1997).Alex was nearly mute prior to his surgery between 8-9years of age, and (to the extent that he could be tested atall) demonstrated levels of language comprehensionsimilar to those of a normal 3-year-old. Soon after hissurgery, he demonstrated remarkable recovery in bothreceptive and expressive language, and continued to

make progress into adolescence. Although Alex didsuffer some degree of mental retardation (as anadolescent, he has the mental age of a 10-12-year-oldchild on most measures), his language abilities areentirely commensurate with his mental age. In fact, hislevel of performance on language measures is superiorto both of the cases reported by Stark and McGregor,even though his surgery took place several years later.The contrast between this study and that of Stark et al.underscores two important points. First, it providesfurther evidence against the assumption that plasticitydrops monotonically across a supposed critical periodfor language. Second, it reminds us that the effects ofbrain injury are superimposed upon the vast landscape

of individual variation observed in normally developingchildren (for an elaboration of this point, see Bates etal., 1995). Because there is so much variation in thenormal population, it is difficult to know in a single-case or small-sample study whether or not the cognitiveprofiles we observe are statistically reliable. Indeed,they may be no different from the patterns that would beobserved if brain damage were imposed randomly oncases selected from the population at large (Bates,Appelbaum & Allard, 1991; Bishop, 1997; see alsoBasser, 1962, for evidence that the vast majority ofcases in a large sample of hemispherectomized children


18/31

LEFT LESION RIGHT LESION

Lesion C

Mat ched

igure 8 : A Hypot het ical Example of Comparisons bet ween Left -

and Right -Hemisphere Groups wit h t heir Respect ive Cont rols

( LHD-Cont rol Diff erence is Reliable; RHD-Cont rol Diff erence is not )

SIDE OF LESION


19/31

9

show no evidence at all of a speech/language impair-ment, regardless of side of surgery).

Despite these concerns, our results for olderchildren are largely compatible with the retrospectiveliterature on language development in the focal lesionpopulation: Children with early injuries to one side ofthe brain usually acquire language abilities within thenormal or low-normal range, with little evidence for

effects of lesion side or lesion site (as reviewed in Bates,Vicari, & Trauner, in press; Eisele & Aram, 1995;Vargha-Khadem et al., 1994). Our prospective findingsfor children under five years of age are qualitativelydifferent, but they are also so new that there is little orno comparable information in the literature, aside froma few single-case or small-sample studies with verydifferent goals (e.g., Dall Oglio, Bates, Volterra, DiCapua, & Pezzini, 1994; Feldman et al., 1992). Ofcourse it will be important to replicate all theseprospective findings with other samples of children, andin other laboratories. In the meantime, we can takesome comfort in the fact that these results are based onthe largest and most homogeneous sample of children

with focal brain injury that has ever been studied in aprospective framework. Although in some cases thesame children participate in more than one prospectivestudy, the full sample across our two largest studies(Bates et al., 1997; Reilly et al., 1998) includes 72cases of children with focal brain injury, from threedifferent laboratories. With sample sizes of 26 or morefrom one substudy to another, we have been able to useexperimental designs and inferential statistics that wouldnot be appropriate in a single-case or small-samplestudy, revealing new information about the changingnature of lesion/symptom correlations. In short, thefindings are solid enough to justify some speculationabout the development of brain organization for

language under normal and pathological conditions.

(5) HOW BRAIN ORGANIZATION FORLANGUAGE EMERGES ACROSS THE

COURSE OF DEVELOPMENT

The literature on language outcomes in humanchildren with early unilateral brain injury is quitecompatible with the burgeoning literature on neuralplasticity in other species. Many of the human resultsare new, but the information from developmentalneurobiology is now well established. Although fewneurobiologists would argue in favor of equipoten-

t ial i ty , i.e., the idea that all areas of cortex are createdequal (Lenneberg, 1967), there is now overwhelming

evidence in favor ofpluripotentiality , i.e., the ideathat cortical tissue is capable of taking on a wide arrayof representations, with varying degrees of success,depending on the timing, nature and extent of the inputto which that tissue is exposed (Elman et al., 1996;Johnson, 1997).

This conclusion is well attested in thedevelopmental neurobiology literature, but it has hadsurprisingly little impact in linguistics, cognitivescience and cognitive neuroscience. In fact, the oldphrenological approach to brain organization has found

new life in the last two decades in various proposalsthat language is an instinct (Pinker, 1994), a mentalorgan (Chomsky, 1980a,b; 1995) or an innatemodule (Fodor, 1983; Pinker, 1997a), with its ownneural architecture and its own highly specific geneticbase (see also Gopnik, 1990; Pinker, 1991; Rice, 1996;van der Lely, 1994). Indeed, Fodors 1983 monographcelebrates the contributions of Franz Gall, the original

phrenologist, and proudly bears a classical drawing ofGalls subdivided and numbered brain on its cover. Theonly real surprise is how little the claims have changedacross the last two hundred years.

Phrenology in all its reincarnations can becharacterized as the belief that the brain is organized intospatially and functionally distinct faculties, eachdedicated to and defined by a different kind of intel-lectual, emotional or moral content. In some of theproposals put forward by Gall, Spurzheim and others inthe 18th century, these included areas for hope,combativeness, conjugal love, veneration, cautiousness,calculation, tune, memory, and of course, language. Amodern variant of phrenology is represented in cartoon

form in Figure 9, which differs from the old version inat least two respects. First, the content of the proposedmodules has changed a great deal in the last twocenturies: with some exceptions, most of the ethicalcontent is gone (but see Ramachandran, 1997, for aproposed religiosity module), replaced by a smaller setof species-specific cognitive and linguistic domains(e.g., music, faces, mathematics, grammar, the lexi-con). To be sure, the particular entries and placementsin Figure 9 are of my own making, but each onerepresents explicit claims that have been made in thelast 5-10 years in the New York Times and other publicoutlets. Second, and most important for our purposeshere, the modern version of phrenology has a strong

nativist component. In contrast with the 19th-centuryphrenologists (some of whom underscored the role ofexperience in setting up the functional organiza-tion ofthe brain -- see especially Wernicke, 1874/ 1977), 20th-century champions like Fodor and Pinker have weddedtheir theory of modular localization to the doctrine ofinnateness. In this variant of phrenology, the adultbrain is organized along modular lines because the braincame packaged that way, in its fetal form, with specificfunctions assigned to specific regions by a geneticprogram (see also Gopnik & Crago, 1991; Rice, 1996;van der Lely, 1994).

In part, the phrenological approach may persistbecause alternative accounts are difficult to understand.

The adult brain is a highly differentiated organ, and theinfant brain (though underspecified in comparison to theadult brain) is certainly not a tabula rasa. And yetefforts to reintroduce experiential effects on this brainorganization (e.g., Bates & Elman, 1996; Elman &Bates, 1997) have been met with great suspicion bythose who fear a reintroduction of old behavioristaccounts (Clark, Gleitman, & Kroch, 1997; Jenkins &Maxam, 1997; Pesetsky, Wexler, & Fromkin, 1997;Pinker , 1997b). Some of the heat in this exchangecomes from the fact that several logically andempirically distinct issues are conflated in the argument


20/31


21/31

10

about mental organs for language. As a result,anyone who opposes the modern doctrine of phrenologyin its full-blown form is accused of (gasp!)behaviorism. In order to clarify the difference betweenold-fashioned tabula rasa behaviorism and theemergentist perspective that I am espousing here, weneed to break the mental organ doctrine down into aseries of separate and separable assumptions about (1)

innate representa-tions (i.e., synaptic connectionsare determined by a genetic program), (2) domain- specific processing (each region of the brain isdesigned to handle a specific kind of content), and threecorollaries about localization, (3) compact locat ion ,(3) f ixed location, and (4) universal location .Table 1 summarizes the five claims of modernphrenology, together with a character-ization of theemergentist alternative on each of these five counts.

Consider first the assumption of innaterepresentations. As my colleagues and I have acknow-ledged repeatedly, throughout this chapter and elsewhere(Bates, Elman et al. , in press; Elman et al ., 1996),cortex is not equipotential. There are powerful endo-

genous constraints in the infant brain that bias the waythat brain organization will proceed under normalcircumstances. However, claims about the nature ofthese innate constraints can be made on several differentlevels: innate representations (where representa-tions are operationally defined as the patterns ofcortical connectivity that comprise knowledge), innate

architecture (defined in terms of the global input-output architecture of the brain, and local variations indensity, speed and style of information processing), andinnate timing (including variations in length ofneurogenesis, and the onset and offset of neurotrophicfactors). The mental organ doctrine is deeply committedto the existence of innate representations. The

emergentist alternative is committed to the idea thatknowledge itself is not innate, but emerges across thecourse of development, through the interaction of innatearchitecture, innate timing, and input to the cortex.

In fact, the case for innate representations looksvery bad right now. Thirty years ago, representationalnativism was a perfectly plausible hypothesis. That is,it was reasonable to suppose that knowledge is builtinto the infant cortex in the form of detailed and well-specified synaptic connections, independent of and priorto the effects of input to the cortex (what Pinker(1997a) refers to as an innate wiring diagram). Indeed,such an assumption is critical for strong forms oflinguistic nativism (i.e., the idea that children are born

with Universal Grammar -- Chomsky,1980a,b;Pinker,1994; Rice, 1996), because synaptic connecti-vity is the only level of brain organization with thenecessary coding power for complex and domain-specificrepresentations of the sort that would be required tosupport an innate grammar. However, this particularform of innateness is difficult to defend in the face ofmounting information on the activity-dependent natureof synaptic connectivity at the cortical level. Of coursethe infant brain is certainly not a tabula rasa. At otherlevels of organization, we have ample evidence forendogenous effects that bias the learning game in

significant ways. These include constraints on theglobal input-output architecture of the brain (e.g., thefact that information from the eye usually does end upin visual cortex, in the absence of wicked interventionsby Sur and his colleagues), local variations inarchitecture and style of computation (e.g., primaryvisual cortex starts out with roughly twice as manyneurons as any other area), and variations in timing

(e.g., variations from one region to another in thelength of neurogenesis, and in the availability of NGF(nerve growth factor). It now seems that the differencebetween the human brain and that of other primatesmust be determined primarily by nonrepresentationalvariations of this kind, controlled by a genetic programsmall enough to fit into the mere 1-2% differencebetween the human genome and the genome of achimpanzee (King & Wilson, 1975; Wilson, 1985).

The second assumption in Table 1, domain-specificprocessing, is a key component of the mental organdoctrine, i.e., that distinct regions of the brain haveevolved to deal with particular kinds of content ofcompelling interest to our species (Barkow, Cosmides,

& Tooby, 1992; Pinker, 1997a). In addition tolanguage (and perhaps to distinct subcomponents oflanguage, e.g., a distinction between grammar and thelexicon), proposed modules or mental organs include aface detector, a theory-of-mind module (that containsalgorithms for detecting dishonest behavior by othermembers of the species), a mathematics module, amusic module, and so forth. These systems havepresumably evolved to deal optimally with theirassigned content, and only with that content. Indeed,Pinker (1997a) has proposed that diverse and specificforms of psychopathology may result if a module isapplied to the wrong domain (although it is not entirelyclear how this might occur, given the perceptual biases

that define a mental organ).The emergentist alternative to domain-specific pro-

cessing is that domain-specific knowledge can beacquired and processed by domain-general mechanisms,i.e., by mechanisms of attention, perception, memory,emotion and motor planning that are involved in manydifferent aspects of learning, thought and behavior. Inother words, the cognitive machinery that makes ushuman can be viewed as a new machine constructed outof old parts (Bates, Benigni, Bretherton, Camaioni, &Volterra, 1979). All of the component parts thatparticipate in language are based on phylogeneticallyancient mechanisms, with homologues up and down thevertebrate line. The specific functions that make

humans different from other species are superimposedon this Basic Vertebrate Brain Plan. Of course it islikely that some and perhaps all of the neuralcomponents that participate in human activity haveundergone quantitative changes that permit new be-haviors like language to emerge, but these componentsstill continue to carry out older and more generalfunctions of object detection, shifting attention,formation of new memories, motor planning, and soforth (i.e., they have kept their day jobs....).

To help us think about the kind of adaptation thatwould permit the construction of a new machine from


22/31

11

old parts, consider the metaphor of the giraffes neck.Giraffes have the same number of neckbones that youand I have, but these bones are elongated to solve thepeculiar problems that giraffes are specialized for (i.e.,eating leaves high up in the tree). As a result of thisparticular adaptation, other adaptations were necessary aswell, including cardiovascular changes (to pump bloodall the way up to the giraffes brain), shortening of the

hindlegs 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-reachingorgan"? 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.

In the same vein, our "language organ" can beviewed as the result of quantitative adjustments inneural mechanisms that exist in other mammals,permitting us to walk into a problem space that otheranimals cannot perceive much less solve. Of course,once language finally appeared on the planet, it is quitelikely that it began to apply its own adaptive pressuresto the organization of the human brain, just as the leaf-reaching adaptation of the giraffe's neck applied adaptivepressure to other parts of the giraffe. Hence the neuralmechanisms that participate in language still do otherkinds of work, but they have also grown to meet thelanguage task. In fact, it seems increasingly unlikelythat we will ever be in a position to explain human

language in terms of clear and well-bounded differencesbetween our brain and that of other primates. Consider,for example, the infamous case of the planum temporale(i.e. the superior gyrus of the temporal lobe reachingback to the temporal-parietal-occipital juncture). It wasnoted many years ago that the planum temporale islonger on the left side of the brain in the majority ofnormal, right-handed human adults. Because thetemporal lobe clearly does play a special role inlanguage processing, it was argued that the asymmetryof the planum may play a key role in brain organizationfor language. However, surprising new evidence hasjust emerged showing that the same asymmetry is alsoobserved in chimpanzees (Hollaway, Broadfield, Kheck,

& Braun, 1998). In fact, the asymmetry is actuallylarger and more consistent in chimpanzees than it is inhumans! I dont doubt for a moment that humans usethis stretch of tissue in a quantitatively and qualitativelydifferent way, but simple differences in size and shapemay not be sufficient or even relevant to the criticaldifference between us and our nearest relatives in theprimate line. In response to findings of this sort,Pinker (1997a) has insisted that the answer lies in thecortical microcircuitry within relevant areas. And yet,as we have seen over and over, developmentalneurobiologists have abandoned the idea that detailed

aspects of synaptic connectivity are under direct geneticcontrol, in favor of an activity-dependent account.There has to be something special about the humanbrain that makes language possible, but that some-thing may involve highly distributed mechanisms thatserve many other functions.

My own favorite candidates for this category oflanguage-facilitating mechanisms are capacities that

predate language phylogenetically, and undoubtedlyinvolve many different aspects of the brain. Theyinclude our rich social organization and capacity forsocial reasoning, our extraordinary ability to imitate thethings that other people do, our excellence in thesegmentation of rapid auditory and visual stimuli, andour fascination with joint attention (looking at the sameevents together, sharing new objects just for the fun ofit -- for an extended discussion, see Bates, Thal, &Marchman, 1991). These abilities are all present inhuman infants within the first year, and they are allimplicated in the process by which language is acquired.None of them are specific to language, but they makelanguage possible, just as quantitative adjustments in

the giraffes neck make it possible for the giraffe toaccomplish something that no other ungulate can do.Is there any evidence in favor of this domain-

general borrowed system view? I would put thematter somewhat differently: Despite myriad predic-tions that such evidence will be found, there is still nounambiguous evidence in favor of the idea that specificparts of the brain are dedicated to specific kinds ofobjects, and only those objects. For example, there arecells in the brain of the adult primate that respondpreferentially to a particular class of stimuli (e.g. faces).However, recent studies have shown that the same cellscan also respond to other kinds of content, spon-taneously and/or after an extended period of training

(Das & Gilbert, 1995; De Weerd, Gattass, Desimone, &Ungerleider, 1995; Fregnac, Bringuier, & Chavane,1996; Pettet & Gilbert, 1992; Ramachandran &Gregory, 1991; Tovee, Rolls, & Ramachandran (1996).Similarly, certain cortical regions around the sylvianfissure are invariably active in neural imaging studies oflanguage processing, including some of the same areasthat are implicated in fluent and nonfluent aphasia.However, each of these regions can also be activated byone or more forms of nonlinguistic processing. Thispoint was made eloquently clear in a recent study byErhard, Kato, Strick and Ugurbil (1996), who looked atall the proposed subcomponents of Brocas area whilesubjects were asked to carry out (covertly) a series of

verbal and nonverbal actions, including complexmovements of the mouth and fingers. Every singlecomponent of the Broca complex that is active duringspeech is also active in at least one form of covertnonverbal activity. In short, even though there isample evidence for stretches of tissue that participate inlanguage, there appears to be no candidate anywhere inperisylvian cortex for a pure language organ.

This brings us to three key assumptions about thenature of localization, the final three of the fivecontrasting issues listed in Table 1. On the phreno-logical account, precisely because of the assumptions


23/31

12

about (1) innate representations and (2) dedicatedarchitecture, it is further assumed that brain organizationfor language involves (3) a fixed architecture that cannotbe replaced and cannot be modified significantly byexperience, (4) a universal architecture that admits tovery little individual variability, and (5) a compact andspatially contiguous architecture that operates as acoherent and autonomous unit in neural imaging

studies, and creates distinct deficits in or dissociationsbetween cognitive functions when it is lesioned(disconnection syndromes -- Caramazza, 1986;Caramazza & Berndt, 1985; Geschwind, 1965; Shallice,1988). By contrast, the emergentist account is morecompatible with forms of localization that are (3)plastic and modifiable by experience, (4) variable inform as a result of variations in experience as well asindividual differences in the initial architecture, and (5)distributed across stretches of tissue that may participatein many different tasks (including spatially discon-tinuous systems that can perform separately or togetherdepending on the task). Because of these properties, theemergentist view is much more compatible with all the

mounting evidence from developmental neurobiologyfor the plasticity and activity dependence of corticalspecialization, including plasticity for language inbrain-injured children.

The emergentist view is also more compatible withthe complex and variable findings that have emerged inrecent neural imaging studies of normal adults(Courtney & Ungerleider, 1997; Poeppel, 1996).Indeed, new areas for language are multiplying at analarming rate in language activation studies, includingstudies using positron emission tomography (PET),functional magnetic resonance imaging (fMRI), mag-netoencephalography (MEG) and/or event-related brainpotentials (ERP). Although activation is usually larger

on the left than it is on the right in language activationstudies, and the familiar perisylvian regions of the lefthemisphere show up in study after study, there isincreasing evidence for participation of homologousregions in the right hemisphere (e.g., Just et al., 1996),although there is substantial variation over individuals,tasks and laboratories in the extent to which this occurs.Language activation studies that involve generation andmaintenance of codes and/or a decision betweenbehavioral options seem to result in reliable activationof several different prefrontal regions that were notimplicated in older studies of language breakdown inaphasia (e.g., Raichle et al., 1994; Thompson-Schill,DEsposi to, Aguirre, & Farah, 1997). New regions

that appear to be especially active during languageactivation have also appeared in basal temporal cortex(on the underside of the brain -- Nobre, Allison, &McCarthy, 1994), in some portions of the basalganglia, and in the cerebellum (especially on the rightside of the cerebellum). Many different aspects of bothsensory and motor cortex seem to be activated inlanguage tasks that involve imageable stimuli. Moreinteresting still for our purposes here, these patterns ofactivation vary as a function of development itself,including variations with chronological age andlanguage level in children (Hirsch et al., 1997; Mills,

Coffey-Corina, & Neville, 1997; Mueller, 1996), andvarying levels of expertise in adults (Hernandez,Martinez, Wong, Frank, & Buxton, 1997; Kim,Relkin, Lee, & Hirsch, 1997; Perani et al., 1997;Raichle et al., 1994).

The picture that has emerged is one in which mostof the brain participates in linguistic activity, invarying degrees, depending on the nature of the task and

the individuals expertise in that task. In manyrespects, this is exactly what we should expect:Language is a system for encoding meaning, and thereare now good reasons to believe that the activation ofmeaning involves activation of the same regions thatparticipate in the original experiences on whichmeanings are based. Because most of the brainparticipates in meaning, we should expect widelydistributed and dynamically shifting patterns ofparticipation in most language-based tasks. The factthat these patterns of activation change over time is alsonot surprising, reflecting changes in experience as wellas changes in the level of skill that individuals attain inactivation and maintenance of both meaning and form.

Clearly, however, there are some importantdifferences in the view of language organization thatemerges from neural imaging studies and lesion studies.Neural imaging techniques can tell us about the areas ofthe brain that participate in language. From this pointof view, we may conclude that the participation is verybroad. Lesion studies can tell us about the areas of thebrain that are necessary for normal language. The listof areas that are necessary for language (in children oradults) appears to be much smaller than the list of areasthat participate freely in a language task. Even in thiscase, however, improved techniques for structuralimaging and lesion reconstruction have yielded more andmore evidence for individual variability in

lesion/symptom mapping (Goodglass, 1993; Willmes& Poeck, 1993), and for compensatory organization inpatients who display full or partial recovery fromaphasia (Cappa et al., 1997; Cappa & Vallar, 1992).

There are of course some clear limits on thisvariability. Some areas of the brain simply cannot bereplaced, in children or adults. For example, Bache-valier and Mishkin (1993) have shown that infantmonkeys with bilateral lesions to the medial temporalregions (including the amygdala and the hippocampus)display a dense and apparently irreversible form ofamnesia that persists for the rest of the animals life, inmarked contrast to the striking recovery that followsbilateral lesions to lateral temporal cortex (Webster et

al., 1995). The key lies in the global input-outputarchitecture of those medial temporal regions, a rich andbroad form of connectivity that cannot be replacedbecause no other candidate has that kind ofcommunication with the rest of the cortex. Other partsof the brain cannot be replaced because they are thecrucial highways and offramps for information from theperiphery (e.g., the insula, which receives crucialkinaesthetic feedback from the oral articulators, or theauditory nerve, which carries irreplaceable auditory inputto the waiting cortex -- Dronkers, 1996; Dronkers,Redfern, & Ludy, in press; Dronkers, Wilkins, van


24/31

13

Valin, Redfern, & Jaeger, 1994). These irreplaceableregions form the anchor points, the universal startingpoints for brain organization in normal children, andthey are difficult if not impossible to replace once allthe exuberant axons of the fetal brain have beeneliminated.

Within this framework, learning itself also placeslimits on plasticity and reorganization in the developing

brain. For example, Marchman (1993) has shown thatartificial neural networks engaged in a language-learningtask (i.e., acquiring the past tense of English verbs) canrecover from lesions (i.e., random removal ofconnections) that are imposed early in the learningprocess. The same lesions result in a substantiallygreater language deficit when they are imposed later inthe learning process. This simulation of so-calledcritical period effects takes place in the absence of anyextraneous change in the learning potential of thenetwork (i.e., there is no equivalent of withdrawal ofneurotrophins or reduction in the learning rate).Marchman reminds us that critical period effects can beexplained in at least two ways (and these are not

mutually exclusive): exogenously imposed changes inlearning capacity (the usual interpretation of criticalperiods), or the entrenchment that results from learningitself. In other words, learning changes the nature ofthe brain, eliminates some connections and tunes othersto values that are difficult to change. Eventually thesystem may reach a point of no return, a reduction inplasticity that mimics critical period effects without anychange in the architecture other than the changes thatresult from normal processes of learning anddevelopment. Marchman does not deny the possibilityof exogenous effects on plasticity, but she arguesconvincingly that there are other ways to explain thesame result, including gradual changes in the capacity to

learn (and recover what was learned before) that are theproduct of learning itself -- change that are morecompatible with the current developmental evidencethan the notion of an abrupt and discontinuous criticalperiod (see also Bates & Carnevale, 1993; Elman et al.,1996, Chapter 4).

Finally, the emergentist view makes room for thepossibility of systematic developmental changes inlocalization, due to a shift in the processes andoperations that are required to carry out a function atdifferent points in the learning process. On the staticphrenology view, a language area is a language area,always and forever. There may be developmentalchanges that are due to maturation (i.e., an area that was

not ready before suddenly comes on line), but theprocesses involved in that content domain are alwayscarried out in the same dedicated regions. On theemergentist account, the areas responsible for learningmay be totally different from the areas involved inmaintenance and use of the same function in its matureform. In fact, there are at least three reasons why weshould expect differences in the patterns of brain activityassociated with language processing in children vs.adults.

(1) Early competition. We may assume(based on ample evidence from animal models) that the

early stages of development involve a competitionamong areas for control over tasks. This competition isopen to any region that can receive and process therelevant information, but that does not mean that everyregion has an even chance of winning. In fact, as thecompetition proceeds, those regions that are betterequipped to deal with that task (because of differences inefficiency of access and type of processing) will

gradually take more responsibility for the mediation ofthat function. In prospective studies of languagedevelopment, we are looking at this process ofcompetition as it unfolds. This leads to the predictionthat the earlier stages of development will involve morediffuse forms of processing, a prediction that is borneout by ERP studies of changes in activation across thefirst three years of language development (fromactivation to known words that is bilateral but slightlylarger in the right, towards activation that is larger onthe left and localized more focally to fronto-temporalsites -- Mills et al., 1997).

(2) Expertise. We may also expect quantitativeand qualitative change in the regions that participate in a

given task as a function of level of expertise. Thesechanges can take three different forms: expansion withinregions, retraction within regions, and a wholesale shiftin mediation from one region to another. An exampleof expansion comes from a recent fMRI study of skillacquisition in adults (Karni et al., 1995). In this study,the first stages of learning in a finger-movement tasktend to involve smaller patches of somatosensorycortex; with increased skill in this task, the areasresponsible for the motor pattern increase in size.Examples of retraction come from studies that showlarger areas of activation in the early stages of second-language learning compared with activation in nativespeakers and in more experienced second-language

learners (Hernandez et al. 1997; Perani et al., 1997).Presumably this is because the novice speaker has torecruit more neural resources to achieve a goal that wasfar easier for a more advanced speaker (equivalent to theamount of muscle a child vs. an adult must use to lift aheavy box). The third possibility may be the mostinteresting, and the one with greatest significance forour focal injury results. In the earliest stages, areasinvolved in attention, perceptual analysis and formationof new memories may be particularly important. Asthe task becomes better learned and more automatic, thebaton may pass to regions that are responsible for thereactivation of over-learned patterns, with less attentionand less perceptual analysis. A recent example of this

kind of qualitative shift is reported by Raichle et al.(1994), who observed strong fronto-cerebellar activationin the early stages of learning, replaced by activation inperisylvian cortex after the task is mastered.

(3) Matu rati on and readiness. Finally,the emergentist approach does not preclude thepossibility of maturational change. Examples mightinclude differential growth gradients for the right vs. lefthemisphere (Chiron et al., 1997), differential rates ofsynaptogenesis (synaptic sprouting) from from oneregion to another within the two hemispheres(Huttenlocher, de Courten, Garey, & van der Loos,


25/31

14

1982), changes from region to region in the overallamount of neural activity (as indexed by positronemission tomography -- Chugani, Phelps, & Mazziotta,1987), variation in rates of myelination, and so forth.As a result of changes of this kind (together with theeffects of learning itself in reshaping the brain --Marchman, 1993), we should expect to find markedshifts in the patterns of activity associated with

language processing at different points in earlychildhood.Based on these assumptions, let us return to our

findings on the early stages of language development inchildren with early focal brain injury, to see what theseresults suggest about the emergence of brainorganization for language in normal children.

(1) Right-hemisphere advantage for wordcomprehension and gesture from 10-17months. Contrary to expectations based on the adultaphasia literature, we found evidence for greater delaysin word comprehension and gesture in children withRHD. This is exactly the opposite of the patternobserved in adults, where deficits in word compre-

hension and in production of symbolic gesture are bothassociated with LHD, suggesting that some kind ofshift takes place between infancy and adulthood, withcontrol over these two skills passed from the righthemisphere to the left. This result is (as we noted)compatible with observations by Mills et al. on thepatterns of activation observed in response to familiarwords from infancy to adulthood. There are at least twopossible explanations for a developmental change, andthey are not mutually exclusive.

On the one hand, the early RH advantage could beexplained by hard maturational changes that areexogenous to the learning process itself. For example,Chiron and his colleagues have provided evidence from

positron emission tomography for a change in resting-state activation across the first two years, from bilateralactivation that is larger on the right to greater activationon the left. Based on these findings, they suggest thatthe right hemisphere may mature faster than the left inthe first year of development. As it turns out, this is theperiod in which word comprehension and gesture firstemerge in normally developing children. By contrast,word production emerges in the second year, and growsdramatically through 30-36 months, the period in which(according to Chiron et al.) the left hemisphere reachesthe dominant state that it will maintain for years tocome. Hence one might argue that the right hemispheregrabs control over comprehension and gesture in the

first year, the left hemisphere grabs control over theburgeoning capacity for production in the second year,and eventually takes over the entire linguistic-symbolicsystem (including word comprehension and meaningfulgestures).

On the other hand, it is also possible that the right-to-left shift implied by our data reflects a qualitativedifference between the learning processes required forcomprehension and the processes required for produc-tion. The first time that we figure out the meaning of aword (e.g., decoding the word dog and mapping itonto a particular class of animals), we do so by

integrating the phonetic input with information frommany different sources, including visual, tactile andauditory context (fuzzy brown thing that moves andbarks). It has been argued that the right hemisphereplays a privileged role in multi-modal integration andprocessing of large patterns (Stiles, 1995 -- more onthis below), and for this reason, we may expect theright hemisphere to play a more important role when

children are learning to comprehend words for the firsttime. Presumably, this RH advantage will disappearwhen words are fully acquired, replaced by a rapid,efficient and automatic process of mapping well-knownsounds onto well-known semantic patterns (more onthis below). If this hypothesis has merit, then wemight also expect to find evidence for greaterparticipation of the right hemisphere in the early stagesof second-language learning in adults, a testablehypothesis and one that has some (limited) support.

It is much less obvious how this shift-in-strategyhypothesis might account for the early RH advantage insymbolic gesture. Although this is admittedly aspeculative answer, this finding may be related to

results for normal children showing that comprehensionand gesture are highly correlated between approximately9-20 months of age (Fenson et al., 1994). Onepossible explanation for this correlation may lie in thefact that symbolic gestures are acquired in the context ofauditory comprehension (e.g., Wave bye-bye tograndma, Hug the baby!). Hence the two skills maycome in together in very small children because they areacquired together in real life.

(2) Deficits in expressive vocabulary andgrammar with frontal lesions to either hemi-sphere from 19-31 months. We observed specificeffects of lesions involving the frontal lobes in childrenbetween 19-31 months of age, a brief but dramatic

period of development that includes the vocabulary burstand the first flowering of grammar. Contrary toexpectations based on the adult aphasia literature, thedelays in expressive language associated with frontallesions were symmetrical, i.e., there was no differencebetween frontal lesions on the left and frontal lesions onthe right. There are a number of reasons why we wouldexpect to find specific effects of frontal involvementduring this important period in the development ofexpressive language, including contributions to theplanning and execution of motor patterns, andcontributions from working memory and/or thefashionable array of skills referred to by the termexecutive function (Pennington & Ozonoff, 1996).

However, the absence of a left-right asymmetry is moresurprising. Nor have we found any evidence for aspecific effect of left frontal injury in any of our studiesto date, at any age. This difference between infants andadults suggests to us that Brocas area is not innatelyspecialized for language. It becomes specialized acrossthe course of development, after an initial period inwhich frontal cortex makes a symmetrical contributionto language learning.


26/31

15

(3) Deficits in expressive vocabulary andgrammar with left temporal injuries from 10months to f ive years of age . This is the mostrobust and protracted finding in our prospective studies,and it is the only evidence we have for an asymmetrythat might be systematically related to a left-hemisphereadvantage for language in the adult brain. Note,however, that the effect only pertains to expressive

language (contrary to the expectation that temporalcortex is specialized for comprehension), and it appliesequally to both vocabulary and grammar (contrary to theexpectation that temporal cortex is associated withsemantics while frontal cortex handles grammar -- Zurif,1980).

We have proposed that a relatively simple bias instyle of computation may underlie this left temporaleffect, reflecting architectural differences between leftand right temporal cortex that are only indirectly relatedto the functional and representational specializationsthat are evident in adult language processing. Followinga proposal by Stiles and Thal (1993), we note that leftand right temporal cortex differ at birth in their capacity

to support perceptual detail (enhanced on the left) andperceptual integration (enhanced on the right -- seeabove). These differences are evident in nonverbalprocessing, but they may have particularly importantconsequences for language. For example, a number ofrecent studies have shown that lesions to the righthemisphere lead to problems in the integration ofelements in a perceptual array, while lesions to the lefthemisphere create problems in the analysis of perceptualdetails in the same array (e.g., Robertson & Lamb,1991). Asked to reproduce a triangle made up of manysmall squares, adult patients with left-hemispheredamage tend to reproduce the global figure (i.e., thetriangle) while ignoring information at the local level.

Adult patients with right-hemisphere damage display theopposite profile, reproducing local detail (i.e., a host ofsmall squares) but failing to integrate these features intoa coherent whole. Stiles and Thal report that childrenwith focal brain injury behave very much like thei

plasticid..

Documents