Numarul Mai 2000

nascut si dobandit - despre caracterul inascut Elizabeth Bates(1), Jeffrey Elman(1), Mark Johnson(2), Annette Karmiloff-Smith(2), Domenico Parisi(3), Kim Plunkett(4) (1) University of California, San Diego (2) MRC Cognitive Development Unit, London (3) National Council of Research, Institute of Psychology, Rome (4) Oxford University

Lucrarea de fat constituie un capitol din cartea editata de William Bechtel & George Graham, A Companion to Cognitive Science, Oxford. Lucrarea discut detaliat aspectele atribuite caracterului nnscut al functiilor psihice n general si al limbajului n mod particular, organiznd n mod coerent directiile de cercetare, asumptiile fcute n acest domeniu si delimitand clar nivelele distincte la care trebuie abordat problema. Desi mare parte din discutii sunt ilustrate prin referiri asupra problematicii inascutului in limbaj, articolul, prin organizarea transant a literaturii de specialitate n domeniu si multitiudinea de exemple ilustrative, este de real folos oricrui psiholog, avnd n vedere c aceast controvers " nnscut vs dobndit" se regseste n cercetrile asupra majorittii aspectelor psihicului uman. Traducere realizata de Andrea Demeter si Bianca Macavei Controversa nnscut/dobndit se mentine de cnd a fost pentru prima dat evidentiat de Platon si Aristotel. Astzi, majoritatea oameniilor de stiint sunt de acord c genele si mediul interactioneaz pentru a determina rezultate cognitive complexe. Oare de ce mai persist totusi controversa? n primul rnd, deoarece are implicatii practice care nu pot fi evitate (de ex. ce putem face pentru a evita unele rezultate incorecte si a asigura unele bune?), o stare de necesitate care cteodat tenteaz oamenii de stiint s fac afirmatii pe care nu le pot argumenta. n al doilea rnd, controversa persist deoarece ne lipseste o teorie precis si testabil a procesului prin care genele si mediul interactioneaz. n absenta unei teorii mai bune, nnscutul e adeseori confundat cu: (1) specificitatea domeniului (Rezultatul X e att de ciudat nct trebuie s fie nnscut), (2) specificitatea speciei (suntem singura specie care facem X, deci X trebuie s fie n genomul uman),. (3) localizarea (Rezultatul X e mediat de o zona anume din creier, deci X trebuie s fie nnscut) si (4) nvtarea (nu ne putem da seama cum ar putea s fie nvtat X, deci X trebuie s fie nnscut). Credem c o teorie explicit si plauzibil a interactiunii este pe cale s apar si multe dintre manevrele clasice de a apra sau ataca nnscutul vor disprea n curnd. ntre timp, dac separm aceste chestiuni adesea confundate, cteva erori serioase pot fi evitate. Acesta este scopul principal al acestei lucrari si nu atacarea nnscutului, ci clarificarea aspectelor la care se refer (si la care nu se refer) afirmatiile despre nnscut. Cum va arta o teorie bun a nnscutului atunci cnd va aprea? Este util s distingem aici ntre dou tipuri de interactionism: interactionism simplu (din alb cu negru rezult gri) si o form emergent (cnd se combin alb cu negru rezult ceva cu totul nou si diferit). n conceptia unei teorii emergentiste, rezultatele pot aprea din motive care nu pot fi prezise din nici unul dintre inputurile problemei. Baloanele de spun sunt sferice deoarece o sfera e singura solutie posibil pentru a atinge volumul maxim cu o suprafat minim (forma lor sferic nu e explicat nici de spun, nici de ap, nici de bietelul care face baloanele). Fagurii de miere iau form hexagonal pentru ca aceasta este solutia stabi la problema suprapunerii cercurilor (adic hexagonul nu este predictibil din ceara, mierea pe care o contine, nici din comportamentul de construire a unei albine indivi-duale). D'Arcy Thompson (1917/1968) a oferit sute de exemple de

acest gen pentru a explica emergenta diferitelor forme corporale, de-a lungul scalei filogene-tice. Jean Piaget sustinea c logica si cunostintele emerg tocmai n acest mod, din interactiuni succesive ntre activitatea sensoriomotoare si realitatea structurat. n acelasi sens, s-a sustinut c gramaticile reprezint clasa solutiilor posibile la problema "cartografierii" semnificatiilor hiperdimensionale ntr-un flux de dimensiuni limitate, puternic constrns de capacitatea omului de a procesa informatia (MacWhinney & Bates, 1989). Logica, cunostintele si gramaticile nu sunt disponibile de-a gata n mediu, dar nici nu sunt disponibile n gene. Solutii emergentiste de acest gen au fost propuse de nenumrate ori n literatura de specialitate, ca o solutie la cotroversa mediu-nnscut ('ceea ce este inevitabil nu este n mod necesar nnscut'). Din pcate, metaforele invocate de propuntorii emergentismului nu constituie o teorie convingtoare a cognitiei complexe, iar descrierile detaliate ale modificrilor comportamentale oferite de cercettorii piagetieni nu au conturat niciodat teoria formal a dezvoltrii pe care Piaget a cutat-o mai mult de sase decenii. n consecint, nativistii ferventi l-au vzut pe Piaget ca pe un empiricist radical, nediferentiindu-l de Skinner privind ncrederea sa n mediu ca si cauz fundamental a dezvoltrii (Chomsky, 1980). O soart similar au avut cei care au propus o viziune emergentist asupra limbajului (Gibson, 1992). Considerm c actualmente este posibil o perspectiv emergentist mai convingtoare asupra dezvoltrii si aceasta din trei motive. n primul rnd, developmentalistii au nceput s utilizeze principii din dinamica nonliniar (Elman & co.,1996; Thelen & Smith, 1994). Aceasta este cea mai actual si probabil ultima frontier a fizicii teoretice, aruncnd lumin asupra proceselor prin care rezultate complexe, surprinztoare si aparent discontinue pot emerge din modificri cantitative mici, de-a lungul unei singure dimensiuni. Metafora 'fagurelui de albin' a dat astfel curs unei perspective explicite, formale asupra emergentei. n al doilea rnd, acum este posibil simularea schimbrilor comportamentale n retele neuromimetice multinivelare, sisteme care integreaz principiile dinamicii nonlineare necesare explicrii emergentei solutiilor complexe din inputuri mai simple (Elman & co., 1996; Rumelhart & McClelland, 1986). n al treilea rnd, cercetatorii din psihologia dezvoltarii devin constienti de unele rezultate inovatoare n neurobiologia dezvoltrii. Dup cum vom vedea, rezultatele actuale din neurobiologie aduc vesti proaste pentru nativistii de ieri, deoarece ei subapreciaser natura extraordinar de plastic si dependent de activitate a specializrii corticale si sustin cazul abordrii emergentiste a dezvoltrii functiilor cognitive superioare. Chiar si n cazul unei perspective interactioniste de acest gen, trebuie nceput de undeva. Constrngerile impuse de gene si de mediu asupra formei emergente trebuie specificate. Ce se ntelege prin faptul c un anumit rezultat este constrns de caracterul nnscut? ntr-o prim aproximare, putem defini nnscutul drept o supozitie privind cantitatea de informatie dintr-o finalitate (psihic) complex care a fost determinat de gene (reamintindu-ne, desigur, c genele nu actioneaz independent si c ele pot fi activate sau inhibate de semnale din mediu, de-a lungul ntregii vieti a organismului). Elman & co. au propus o taxonomie trinivelar a abordrilor "nnscutului", ordonndu-le de la asumptii 'tari' la asumptii 'slabe', cu referire la cantitatea de informatie la care trebuie s contribuie genele pentru ca abordarea respectiv s fie viabil. Fiecare nivel e definit operational n termeni care corespund creierului uman si respectiv retelelor neuromimetice, dup cum urmeaz: I. Constrngeri reprezentationale - se refer la structurarea a priori a reprezentrilor mentale/neuromimetice care constituie si stau la baza cunostintelor. Conectivitatea sinaptic la nivel cortical este cel mai probabil candidat la implementarea cunostintelor detaliate n creier, deoarece aceste este singurul nivel care are capacitatea de codare necesar unor finalitti cognitive superioare. n retelele neuromimetice, acest nivel este operationalizat prin conexiunile ponderate dintre unittile procesoare. II. Constrngeri arhitecturale - se refer la structurarea a priori a sistemului de procesare a informatiei care trebuie s dobndeasc si/sau s contin aceste reprezentri. Desi reprezentrile si arhitectura nu sunt totuna, nu este nici o ndoial c gama de reprezentri pe care o poate suporta un sistem este puternic constrns de arhitectura acestuia. n computerele traditionale se-rial digitale, unele programe pot rula doar pe dispozitive de capacitatea, viteza si puterea corespunztoare. n retelele neuromimetice, unele forme de cunostinte pot fi reprezentate doar ntr-un sistem cu o structur adecvat (cu un numr optim de unitti, nivele, tipuri de conexiuni ntre nivele, etc). De fapt, exist acum un ntreg subdomeniu n aria cercetrilor asupra retelelor neuromimetice, n care algoritmi genetici sunt aplicati pentru a

gsi clasa de arhitecturi optime pentru un set dat de probleme de nvtare. Pentru a operationaliza constrngerile arhitecturale n creier si n retelele neuromimetice, Elman & co. le mpart n trei subnivele: A. Unitti computationale de baz. n creier, acest subnivel se refer la tipurile de neuroni, pragurile de descrcare a impulsurilor, neurotransmittori, pro-prietti excitatorii sau inhibitorii, etc. n retelele neuromimetice, acest nivel se refer la unittile computationale, cu functia lor de activare, algoritmul de nvtare, temperatur, momentum si rat de nvtare. B.Arhitectura local. n creier, acest subnivel se refer la factori regionali, ca numrul si grosimea straturilor, densitatea diferitelor tipuri de celule n cadrul straturilor respective, tipul circuitelor neuronale (cu sau fr recurent). n retelele neuromimetice, se refer la la factori ca numrul de nivele, densitatea unittilor n cadrul nivelelor respective, prezenta sau absenta unittilor cu caracter recurent, etc. C. Arhitectura global. n creier, acest subnivel include caracteristici arhitecturale globale, ca sursele caracteristice de input (ci aferente) si patternuri de output (ci eferente), care conecteaz regiunile cerebrale cu lumea exterioar si unele cu altele. n numeroase mo-dele de retele neuronale, dimensiunea sistemului e att de redus nct distinctia dintre arhitectura local si cea global nu este util. Totusi, n cadrul asa numitelor retele modulare sau tip "expert", este adesea util s se discute despre subretele distincte si interconexiunile lor. III. Constrngeri cronotopice - se refer la constrngerile nnscute ce se vor manifesta in dezvoltarea individului, incluznd interactiunile spatio-temporale. n creier, acestea ar include constrngerile privind numrul de diviziuni celulare care au loc n geneza neuronal, undele spatio-temporale de crestere si scurtare sinaptic si diferentele relative privind sincronizarea subsistemelor (de ex. diferente ntre vz, auz, etc. n ceea ce priveste ritmul stimulrii cortexului de ctre talamus). Acelasi nivel este reprezentat n retelele neuromimetice prin prezentatea gradual a datelor, programe de diviziune celular n retelele cu posibilitti de crestere, rate de nvtare autoadaptabile si modificri intrinsece n nvtare care apar datorit saturrii nodurilor. Cititorul este rugat s consulte Elman & co. pentru exemple detaliate ale tuturor acestor nivele. Pentru scopul lucrrii de fat, important este c asumptiile nativiste 'tari' asupra limbajului (Fodor, 1985; Pinker,1994), fizicii (Spelke, 1991) sau rationamentului social (Horgan, 1995, Leslie, 1994) trebuie s ia n considerare n mod implicit sau explicit nativismul reprezentational, deoarece acesta este singurul nivel care are capacitatea de codare necesar implementrii informatiei independente de experient. De exemplu, Noam Chomsky (1975) a propus c teoria lingvistic, teoria gramaticii universale ... "este o proprietate nnscut a mintii umane" (p. 34) si c ar trebui s concepem "dezvoltarea limbajului ca fiind analoag dezvoltrii unui organ corporal" (p. 11). Metafora 'organului mental' las putin spatiu nvtrii. ntr-adevr, Chomsky a sustinut c "o teorie general a nvtrii ... mi se pare dubioas, neargumentat si lipsit de orice suport empiric" (1980, p. 110). Piatelli-Palmarini (1980, p. 2) reia tema, afirmnd c "nu vedem nici un avantaj n pstrarea termenului de nvtare. Suntem de acord cu aceia care sustin c am cstiga n claritate dac utilizarea stiintific a termenului ar fi abandonat". n ce ar putea consta o astfel de bogat structur nnscut? Pinker sugereaz c aceste cunostinte nnscute rezid n microcircuitele creierului. Considerm c are absolut dreptate: dac notiunea de "instinct lingvistic" nseamn ceva, atunci se refer la microcircuitele corticale, deoarece acestea sunt (dup cunostintele noastre) singurul mod prin care informatia detaliat poate fi stocat n creier. Acest gen de nativism reprezentational este teoretic plauzibil si atractiv, dar s-a dovedit a fi dificil de sustinut att din punct de vedere matematic, ct si empiric. Din punct de vedere matematic, este dificil de nteles cum pot fi controlate 1014 conexiuni sinaptice din creierul uman de un genom de aproximativ 106 gene, mai ales c (a) maxim 20-30% din aceste gene se implic n constructia sistemului nervos (Wills, 1991) si (b) oamenii au n comun cca 98% din genele lor cu cele mai apropiate primate (King & Wilson, 1975). Dar problema e si mai grav dect att. Paul Churchland (1995) ne aminteste c fiecare conexiune sinaptic poate lua valori multiple. Dac presupunem, n mod conservator, c fiecare conexiune poate lua 10 valori, Churchland calculeaz c puterea de codare sinaptic a creierului uman contine mai multe stri potentiale de conectivitate dect numrul particulelor din univers! Genele ar avea nevoie de foarte mult informatie pentru a orchestra un sistem de acest calibru. Desigur c o corespndent detaliat ntre ntre gene si cortex ar fi totusi posibil dac genele s-ar comporta ca literele dintr-un alfabet, dnd nastere unui numr indefinit de combinatii. Dar nu este cazul, dimpotriv, genele opereaz ntr-o matrice spatio-temporal si chimic puternic constrns (Edelman, 1987;
3

Wills, 1991), folosind si refolosind principii topologice ce s-au conservat timp de milioane de ani, de-a lungul a mii de specii. Se poate argumenta c aceasta componenta nnscut a cunostintelor ocup doar o fractiune din marea diversitate de stri, n ciuda bogtiei sale. n orice caz, ultimele dou decenii de cercetri asupra dezvoltrii creierului la vertebrate sugereaz c paternurile minuscule de conexiuni corticale sunt n mare parte determinate de inputuri corticale (vezi Elman & co., cap. 5; Johnson, 1997). De exemplu, stim c cortexul auditiv poate prelua imagini retinotopice dac inputul de la nivelul ochiului este redirijat de la tinta sa vizual normal (Sur, Pallas & Roe, 1990); stim c portiuni din cortex prelevate dintr-o arie cortical si transplantate ntr-alta vor prelua reprezentrile corespunztoare inputului pe care-l primesc n noua locatie (Stanfield & O'Leary, 1985); stim c modificri ale suprafetei corporale ale unui pui de sobolan duc la modificri corespunztoare ale zonelor corticale (Killackey, 1990), stim c sistemul "unde ?" (de localizare a informatiei spatiale) poate prelua functia sistemului "ce este?" (de indentificare) la puii de maimut cu leziuni bilaterale ale cortexului temporal inferior (Webster, Bachevalier & Ungerleider, 1995) si c la copiii suferind de leziuni ale emisferei stngi care ar duce la afazie ireversibil n cazul unui adult, limbajul continu s se dezvolte n limite normale (Bates & co.,1997; Eisele & Aram, 1995). Pe scurt, actualmente exist foarte putine dovezi care s sprijine ideea c genele determin direct patternul de conexiuni corticale. Dimpotriv, dezvoltarea creierului la vertebratele superioare pare s implice o supraproductie de elemente (neuroni, axoni si sinapse) la nceputul vietii, urmat de un proces competitiv, prin care elementele viabile sunt mentinute, iar cele care dau gres sunt eliminate (Edeman, 1987). Pasco Rakic se refer la aceast competitie ca la procesul prin care experienta "sculpteaz" efectiv creierul. Pe lng aceast sculptare prin regresia numrului de elemente, experienta poate de asemenea s adauge structuri de-a lungul vietii, inducnd cresterea numrului de sinapse tocmai n ariile care se confrunt cu sarcini nounoute (Greenough, Black & Wallace, 1993; Meyernich, 1995; Pons & co.,1991). Desi sunt surprinztor de putine dovezi n favoarea existentei unor reprezentri nnscute la nivel cortical (Balaban, Teilet & Le Douarin, 1988), exist dovezi substantiale n favoarea unor arhitecturi nnscute si a unor variatii nnscute n sincronizare. Aceasta include dovezi conform crora neuronii "stiu" unde trebuie s ajung n timpul migratiei celulare (Rakic, 1988), precum si dovezi cum c axonii prefer tinte specifice n timpul lungii lor cltorii de la o regiune la alta (Neider, Maimon & Finlay, 1995; dar vezi Molnar & Blakemore, 1991). Ar putea acest gen de "nnscut" s ofere baz unei "gramatici universale" ? Probabil c nu, deoarece (a) aceste biasri arhitecturale grosiere nu au capacitatea de codare necesar pentru ceva att de detaliat si specific cum sunt cunostintele gramaticale si (b) principiile de crestere la acest nivel par s opereze ntr-o msur remarcabil la mai multe specii. De exemplu, Deacon (1997) prezint dovezi despre cresterea legitim a axonilor n creierul unui sobolan adult, din transplantul cortical prelevat de la porci fetali! Exist, de asemenea, si variatii regionale n substratul neurochimic (de exemplu, cortexul somatosenzorial transplantat ntr-o arie vizual va prelua imagini retinotopice, mentinndu-si totusi neurochimicalele caracteristice unei arii somatosenzitive - Cohen-Tannoudji, Babinet & Wassef, 1994), ct si variatii regionale n densitatea celular (de exemplu, cortexul vizual primar e deosebit de dens, o caracteristic ce pare s fie determinat n timpul genezei neuronale, nainte ca orice informasie s fie primit - Kennedy, Dehaz, Horsburg, 1990). Astfel, regiunile corticale tind s difere nc de la bun nceput prin stilul de computatie, ceea ce nseamn c vor diferi si n privinta tipului de sarcini pe care le execut cel mai bine. Cu alte cuvinte, competitia care caracterizeaz dezvoltarea creierului nu are loc pe un teren cu sanse egale de cstig. Competitia este de la bun nceput orientat astfel nct s privilegieze anumite ?planuri? generale ale creierului, n defavoarea altora. Totusi, este de asemenea clar c multe "planuri" alternative ale creierului sunt disponibile dac cel optim este mpiedicat s se desfsoare dintr-un motiv sau altul. Bates & co. (1997) au argumentat c, la om, specializarea emisferei stngi n limbaj depinde de acest gen de constrngeri indirecte, arhitecturale. Regiunile temporale si frontale ale emisferei stngi joac un rol important n medierea productiei lingvistice la peste 95% din adultii normal, ducnd la afazie ireversibil n cazul lezrii unor locatii specifice din aceast emisfer. Si totusi, dup cum s-a mentionat anterior, copii cu leziuni similare nu devin afazici.

Cum este posibil? Dac cortexul perisylvian stng nu este necesar dezvoltrii limbajului normal, atunci de unde provine "harta" cerebral tipic a adultului? Studiile efectuate asupra copiilor prezentnd leziuni cerebrale focale, arat c regiunea temporal (dar nu si cea frontal) a emisferei stngi este ntr-adevr specializat din nastere, deoarece copii cu leziuni n regiunile temporale stngi prezint ntrzieri semnificative n dezvoltarea gramaticii si a vocabularului expresiv (dar nu si a celui receptiv). Totusi, aceast diferent regional nu mai este detectabil atunci cnd copiii cu aceleasi leziuni timpurii ajung la vrsa de 7 ani, ceea ce nseamn c un mare numr de reorganizri trebuie s fi avut loc n primii ani de viat. Evident c alte regiuni ale creierului sunt capabile s preia reprezentrile necesare limbajului normal. Bates & co. sugereaz c cortexul temporal stng este initial specializat nu pentru limbaj n sine, ci pentru extragerea detaliilor perceptuale (de exemplu, leziuni ale acestor regiuni determin efecte specifice asupra extragerii de detalii dintr-un patern vizuo-spatial). n conditii normale, aceast biasare indirect a stilului computational duce la specializarea emisferei stngi n limbaj. Dar reprezentrile necesare limbajului nu sunt (si aparent nici nu e necesar s fie) prezente de la bun nceput. Deoarece dovezile nu sunt suficiente pentru o form "tare", reprezentational a nativismului, diferentele care pot fi observate de la o specie la alta probabil c sunt capturate n principal de factori arhitecturali si cronotopici. Rezultatul final emerge din interactiunea dintre aceste constrngeri si problemele specifice cu care se confrunt aceste structuri n lumea real. n aceast ordine de idei, s reconsiderm cteva din argumentele clasice n favoarea cunostintelor nnscute, argumente care, credem, confund nivele de analiz care ar trebui separate. Caracterul nnscut si specificitatea de domeniu . S-a argumentat c limbajul este att de caracteristic, att de specific domeniului n chestiune, nct nu este posibil s fi fost nvtat sau procesat de un sistem domeniu-general. Asumptii similare s-au fcut si despre perceptia fetelor, despre muzic, matematic si rationamentul social. Elman & co. argumenteaz c asumptiile asupra specificittii de domeniu trebuie (la fel ca n cazul caracterului nnscut) s fie submprtite n diferite nivele nainte de a aborda aceast problematic din punct de vedere empiric. Utiliznd limbajul ca domeniu de testare, iat o expunere succint a problematicii. Specificitatea comportamental. Limbajurile reprezint o clas de solutii la o problem care este n mod indubitabil unic prin scopul si natura sa: problema codrii unui spatiu de semnificatii hiperdimensionale ntr-un canal de dimensiui limitate (MacWinney & Bates, 1989). S-ar putea face analogii ntmpltoare cu domenii ca si cntecul psrilor (nvtarea n cadrul sistemului vocal), sahul (un set de solutii complexe la un joc pe care l practic doar oamenii), sau muzica (tran-zitii de la un sunet la altul pe baza unor reguli), dar aceste similaritti sunt n mare parte superficiale. Limbile au foarte putine n comun cu alte sisteme cognitive, ns au foarte multe n comun unele cu altele. De unde provin aceste asemnri? Spatiul de semnificatii implicat n problema de encodare a limbajului include experiente mprtsite de toti membrii normali ai speciei, iar canalele utilizate de limbajul uman exercit constrngeri universale asupra procesrii de informatie (de ex. perceptie, memorie, planificarea articulrii). n aceste conditii, nu ar trebui s fim surprinsi s descoperim c clasa de solutii la problema expus este limitat, constituind un set de alternative la care ne putem referi ca fiind Gramatica Universal. Vom stipula c comportamentele domeniu-specifice au aprut ca si rspuns la la aceast problem de encodare si c limbajele naturale provin dintr-un set comun de solutii domeniu-specifice. Dar asemenea factori nu constituie prin ei nsisi dovezi "pro" nnscut, deoarece aceleasi solutii ar fi putut s apar si conform unui scenariu emergentist. Specificitatea reprezentational. Dac un individ produce n mod constant comportamente necesare pentru a rezolva o problem domeniu specific, nseamn c acesta are n mod necesar un set de reprezentri mentale domeniu specifice, ce sustin acele comportamente. Altfel spus, fiecare reprezentare trebuie s fie implementat ntr-o form ce poate fi diferentiat de alte aspecte ale cunoasterii (vezi mai jos localizarea). Aceast generalizare se mentine, fie c reprezentrile n discutie sunt nnscute, fie c sunt dobndite; asadar, specificitatea unei reprezentri nu este relevant pentru disputa asupra

caracterului nnscut. Specificitatea proceselor mentale. Acesta este nivelul la care caracterul nnscut si specificitatea de domeniu se intersecteaz n cele din urm: este oare posibil ca o arhitectur domeniu general s dobndeasc si/sau s proceseze reprezentri domeniu specifice? De remarcat c sintagma "domeniu-general" nu nseamn neaprat "mecanism ce poate s nvete si s fac orice". Am evidentiat deja necesitatea unei bune concordante ntre problem si arhitectur n cadrul cercetrilor asupra retelelor neuromimetice. Nu exist dispozitiv care s poat nvta si face orice. Discutia este mai specific: poate un domeniu ca limbajul s fie nvtat si/sau procesat de orice sistem ce nu este n mod special proiectat si menit s sustin productii lingvistice? Aceasta este o ntrebare empiric la care nc nu s-a gsit rspunsul. Cu toate acestea, exist dovezi n favoarea abordrii domeniu generale, dovezi ce vin dinspre experimentele de simulare a nvtrii domeniu- specifice, n retelele adem, Watkins, Alcock, Fletcher & Passingham, 1995). Treizeci de ani de cercetri asupra altor copii si adulti cu SLI conduc la concluzii similare (Leonard,1996): SLI coreleaz cu o gam de deficite re-lativ discrete dincolo de limitele corectitudinii limbajului, inclusiv aspecte ale atentiei, jocului simbolic, imageriei mentale si detectiei unor secvente rapide de sunete. Pe scurt, SLI este o tulburare cu transmitere genetic, dar nu mai este att de sigur (n ciuda denumirii) c este specific limbajului, si cu att mai putin unei particularitti a gramaticii. Reversul este de asemenea adevrat: s-a demonstrat c deficitele specifice limbajului nu sunt nnscute, cel putin nu ntr-o acceptiune ce ar putea trezi interes. De exremplu, stim acum c morfologia gramatical (inclusiv toate acele "cuvintele" si terminatii) este un domeniu extrem de vulnerabil; fie c este afectat n mod izolat sau nu, morfologia este un teren problematic major n SLI, sindromul Down si alte situatii n care se cunoaste sau se suspecteaz existenta unei baze genetice. Totusi, probleme specifice legate de morfologia gramatical au fost evidentiate si n variate forme de le-ziuni cerebrale dobndite (indiferent de locatia leziunii), n cazul indivizilor normali din punct de vedere neurologic dar cu tulburari de auz, precum si n cazul studentilor obligati s proceseze propozitii n conditii dificile (de exemplu, cu stimuli deteriorati din punct de vedere perceptiv sau cu resurse atentionale si mnezice diminuate datorit unor sarcini concurente). Morfemele gramaticale tind s nu se impun perceptiei si sunt dificil d e imaginat si probabil de aceea ele constituie o "veriga slab n lantul procesrii". Faptul c ele sunt afectate preferential n sindroamele cu substrat genetic nu constituie cu necesitate o dovad c acestea ar fi nnscute n modalitate domeniu-specific. O ran la cot are efecte extrem de specifice asupra jocului de tenis, dar nu se poate concluziona c aceast parte anatomic ar procesa informatiile despre tenis, nici c genele ce particip la formarea cotului ndeplinesc aceast functie cu scopul de a sustine tenisul. S-a stipulat deja c limbajul nu e similar jocului de tenis, ns metafora prezentat sustine argumentul n cauz. n concluzie, caracterul nnscut si specificitatea de domeniu nu nseamn acelasi lucru, iar argumentele pro "nnscut" nu se pot limita la enumerarea unor fenomene neobisnuite (strategia Madame Tussaud). Vom lua n continuare n discutie specificitatea de specie, localizarea si caracterul dobndit (nvtarea), domenii preferentiale n efortul de a demonstra cara-cterul nnscut doar artnd c un anumit domeniu e "special". Caracterul nnscut si specificitatea speciei. n aceast variant a abordrii specificittii de domeniu, se sustine conceptia conform creia un anumit domeniu trebuie s fie nnscut, deoarece acesta apare doar la specia umana sau cel putin doar oamenii l au foarte bine reprezentat. Astfel de domenii ar fi: limbajul, muzica, politica, religia, finantele si hockey-ul pe gheat. Pentru rigurozitate, trebuie subliniat faptul c exist variante rudimentare de deprinderi umane si la alte specii, inclusiv forme de limbaj. Identificarea unor astfel de precursori infraumani este util, deoarece poate furniza multe informatii n legatur cu evolutia limbajului si a altor functii specific umane. Dar nu se poate argumenta existenta unui sistem domeniu-specific nnscut doar artand c nici o alt specie nu are ce avem noi. Desi suntem singura specie care joac sah, nimeni nu s-ar aventura s sustin c am avea o facultate psihic destinat exclusiv jocului de sah. Desigur, multi dintre noi nu jucm sah, ns toate fiintele umane dezvoltate normal utilizeaz limbajul. Constituie specificitatea de specie pe de o parte si universalitatea pe de alta dovezi pentru existenta unei facultti psihice nnscute? Este posibil, ns ambele aspecte amintite pot fi explicate apelnd la factori care sunt legati doar indirect de domeniul n discutie. Pn n prezent, nu s-a putut identifica nici o structur neural care s apar doar la

fiintele umane, de exemplu un tip de neuron specific uman, un neurotransmittor, un pattern de stratificare cortical sau (depinde de cum definim notiunea de "arie") o arie cerebral unic omului (Deacon, 1997; Finlay & Darlington, 1995). Se pare c gama variat de deprinderi 100% umane pe care specia noastr la posed se ntemeiaz pe variatii cantitative n planul cerebral al primatelor, e.g. expansiunea cortexului frontal n raport cu alte regiuni, mrirea proportio-nal a ariilor secundare n cortexul vizual, cresterea controlului cortical direct asupra gurii si degetelor. Cea din urm inovatie pare a fi aprut cu scopul de a sustine limbajul (sau, mai general, cultura), si s-ar putea ca ntr-o anumit masur acest lucru s fie adevrat. Oricum, este interesant de retinut faptul c aceleasi conexiuni directe de la cortex spre periferie sunt prezente la embrionul de sobolan, ns ele sunt eliminate nainte de a deveni functionale (Deacon, 1997). Desi nu apartinem scolii evolutioniste care explic orice fenomen prin mrimea creierului, abilittile specifice unei specii ar putea fi un produs secundar neintentionat al unei schimbri mult mai generale n forta computational (Wills, 1991). Specificitatea de specie luat separat nu constituie dovada existentei unei structuri mentale specifice. Caracterul nnscut si localizarea. Acesta este, de asemenea, un tip de argument n favoarea specificittii: structurile mentale sunt speciale pentru c apar n anumite prti ale creierului, ceea ce Fodor numeste "arhitectur neural specializat" (termen ce combin nivelele reprezentational si arhitectural stabilite de Elman & co.). Dac am putea demonstra, spre exemplu, c structurile cerebrale se comport diferit atunci cnd e vorba de morfeme gramaticale regulate, respectiv nere-gulate, nu ar constitui acest lucru o dovad a existentei unor procesori domeniu-specifici a priori specializati? Nu n mod necesar. n primul rnd, tot ceea ce stim noi este mediat de creier. Dac experientiem doi stimuli exact n acelasi fel, atunci (prin definitie) noi nu "stim" c ei sunt diferiti. Dac i experientiem diferit, atunci respectiva diferent trebuie s fie reflectat n creier. Orice nou fapt de cunoastere modific structura creierului, indiferent ct de minor este aceast schimbare. S lum n discutie, spre exemplu, o recent demonstratie ce evidentia faptul c, n cazul sahistilor experti, apar patternuri diferite de activare c ortical n diferite etape ale jocului (Nochell i & co. 1994). Aceasta nu nseamn c avem o structur mental responsabil de finalul jocului, nici mcar n stadiul de adult. Si cu sigurant nu se poate spune c ne-am nscut cu o astfel de structur. Toate cunostintele presupun localizare intr-o forma sau alta (compact sau local, sau distribuit pe o arie mai larga), asadar demonstrand existenta localizarii nu furnizam o proba pentru asertiunea de (caracter) innascut. Acest lucru este valabil fie ca localizarea este universala (toti oamenii prezinta acelasi pattern), fie ca este variabila (diferiti oameni prelucreaza acelasi continut in locatii diferite - Caplan, 1981). Cu toate acestea reciproca nu este valabila: daca o abilitate cognitiva este innascuta, ea trebuie sa fie realizata intr-o forma specificabila topografic. Aceasta este modalitatea de exprimare a genelor, i.e. codand proteinele intr-o matrice definita spatial, temporal si chimic (Edelman, 1987, Wills, 1991). Acesta este motivul pentru care dovezile in favoarea plasticitatii corticale se constituie in contraargumente pentru nativismul reprezentational. Pentru a evita aceste dovezi s-ar putea imagina un scenariu in care genele responsabile de formearea sistemului nervos ar calatori libere prin fluxul sangvin, cautand un mediu favorabil construirii unui anumit "organ mental".La urma urmei, fiecare celula din organism contine intregul genom. Poate ca genele responsabile de aparitia limbajului se plimba de colo-colo asteptand un semnal de genul "apucati-va si construiti o structura anatomica pentru limbaj aici". Exista desigur in literatura de specialitate exemple de structuri potrivite aparute in locuri nepotrivite, sau cel putin atipice (e.g. gena dominanta pentru ochi, ce se poate multiplica in diferite locatii). Dar se pare ca aceste cazuri constituie exceptii. Si cel putin in cazul acesta (spre deosebire de functiile cognitive superioare ca limbajul), exista un set specificabil de constante fizice implicatet asupra gurii si degetelor. Cea din urm inovatie pare a fi aprut cu scopul de a sustine limbajul (sau, mai general, cultura), si s-ar putea ca ntr-o anumit masur acest lucru s fie adevrat. Oricum, este interesant de retinut faptul c aceleasi conexiuni directe de la cortex spre periferie sunt prezente la embrionul de sobolan, ns ele sunt eliminate nainte de a deveni functionale (Deacon, 1997). Desi nu apartinem scolii evolutioniste care explic orice fenomen prin mrimea creierului, abilittile specifice unei specii ar putea fi un produs secundar neintentionat al unei schimbri mult mai generale n forta computational (Wills, 1991). Specificitatea de specie luat separat nu constituie dovada existentei unei structuri mentale specifice. Caracterul nnscut si localizarea. Acesta este, de asemenea, un tip de argument n favoarea specificittii: structurile mentale sunt speciale pentru c apar n anumite prti ale creierului, ceea ce Fodor numeste " arhitectur neural specializat" (termen ce combin

nivelele reprezentational si arhitectural stabilite de Elman & co.). Dac am putea demonstra, spre exemplu, c structurile cerebrale se comport diferit atunci cnd e vorba de morfeme gramaticale regulate, respectiv nere-gulate, nu ar constitui acest lucru o dovad a existentei unor procesori domeniu-specifici a priori specializati? Nu n mod necesar. n primul rnd, tot ceea ce stim noi este mediat de creier. Dac experientiem doi stimuli exact n acelasi fel, atunci (prin definitie) noi nu "stim" c ei sunt diferiti. Dac i experientiem diferit, atunci respectiva diferent trebuie s fie reflectat n creier. Orice nou fapt de cunoastere modific structura creierului, indiferent ct de minor este aceast schimbare. S lum n discutie, spre exemplu, o recent demonstratie ce evidentia faptul c, n cazul sahistilor experti, apar patternuri diferite de activare cortical n diferite etape ale jocului (Nochelli & co. 1994). Aceasta nu nseamn c avem o structur mental responsabil de finalul jocului, nici mcar n stadiul de adult. Si cu sigurant nu se poate spune c ne-am nscut cu o astfel de structur. Toate cunostintele presupun localizare intr-o forma sau alta (compact sau local, sau distribuit pe o arie mai larga), asadar demonstrand existenta localizarii nu furnizam o proba pentru asertiunea de (caracter) innascut. Acest lucru este valabil fie ca localizarea este universala (toti oamenii prezinta acelasi pattern), fie ca este variabila (diferiti oameni prelucreaza acelasi continut in locatii diferite - Caplan, 1981). Cu toate acestea reciproca nu este valabila: daca o abilitate cognitiva este innascuta, ea trebuie sa fie realizata intr-o forma specificabila topografic. Aceasta este modalitatea de exprimare a genelor, i.e. codand proteinele intr-o matrice definita spatial, temporal si chimic (Edelman, 1987, Wills, 1991). Acesta este motivul pentru care dovezile in favoarea plasticitatii corticale se constituie in contraargumente pentru nativismul reprezentational. Pentru a evita aceste dovezi s-ar putea imagina un scenariu in care genele responsabile de formearea sistemului nervos ar calatori libere prin fluxul sangvin, cautand un mediu favorabil construirii unui anumit "organ mental".La urma urmei, fiecare celula din organism contine intregul genom. Poate ca genele responsabile de aparitia limbajului se plimba de colo-colo asteptand un semnal de genul "apucati-va si construiti o structura anatomica pentru limbaj aici". Exista desigur in literatura de specialitate exemple de structuri potrivite aparute in locuri nepotrivite, sau cel putin atipice (e.g. gena dominanta pentru ochi, ce se poate multiplica in diferite locatii). Dar se pare ca aceste cazuri constituie exceptii. Si cel putin in cazul acesta (spre deosebire de functiile cognitive superioare ca limbajul), exista un set specificabil de constante fizice implicate ce determina forma structurii ce se constituie. Pentru acelasi motiv pentru care nu se poate recrea un dinozaur din ADNul gasit intr-o boaba de chihlimbar (vezi scenariul din Jurassic Park), genele ce determina limbajul sau muzica sau perceptia fetelor nu voiajaza intr-o barcuta de salvare cautand un loc sa acosteze. Localizarea nu presupune caracterul innascut, dar asertarea interventiei ereditatii presupune o baza de ordin fizic. De aceea nativistii cauta pe buna dreptate corelatele neuronale ale sistemului vizat la un moment dat. Caracterul innascut si caracterul dobandit. In cadrul lingvisticii au fost facute afirmatii ce eludeaza toate dovezile empirice mai sus mentionate.Argumentul cel mai excentric legat de subiectul in discutie suna in felul urmator: X (de obicei, limbajul) este atat de neobisnuit, atat de unic in felul sau, incat nu poate fi dobandit pe baza unei game variate de mecanisme de invatare (e.g. Crain, 1991). Copiii (dupa cum se afirma) trebuie sa dobandeasca o gramtica mai puternica decat ar sustine inputul sara si deteriorat de care ei dispun. Ei pot sa mearga dincolo de datele disponibile si sa selecteze tinta gramaticala adecvata deoarece ei au deja cunostinte detaliate referitoare la clasa de gramatici posibile. Varianta formalizata a acestui argument se bazeaza pe o dovada formala a posibilitatii de achizitie, din informatica numita Teorema lui Gold (Gold, 1967), conform careia gramatica unei clase particulare nu poate fi indusa sau "ghicita" dintr-o baza finita de date pozitive (i.e., exemple de propozitii dintr-o anumita limba) in absenta datelor negative (i.e., exemple de propozitii ce nu apartin respectivei limbi). O discutie de detaliu sau de suprafata legata de aceasta problema depaseste intentiile acestui articol; dorim doar sa subliniem ca dovezile in favoarea caracterului dobandit se bazeaza pe cel putin patru tipuri de asumptii: o definitie a gramaticii ce va fi achizitionata (e.g., gramaticile sunt definite ca siruri de simboluri generate de una sau mai multe reguli recursive), o caracterizare a datelor de care dispune cel ce invata, specificitatea mecanismului de invatare ce va opera cu aceste date si un criteriu ce defineste succesul invatarii. Daca gramatica ce trebuie invatata este foarte abstracta, daca criteriul de succes este foarte dificil de atins si/sau daca mecanismul de invatare este slab iar datele deteriorate, atunci incontestabil ca gramatica respectiva nu poate fi

invatata in lipsa unui important bagaj de cunostinte innascue ce vor compensa imperfectiunile amintite. Se aplica oare aceste reguli limbajului uman? Se pare ca Teorema lui Gold se aplica doar daca emitem asumptii despre mecanismul de invatare ce sunt foarte improbabile atunci cand vorbim de orice tip,de sistem nervos cunoscut. Si trebuie subliniat faptul ca aceasta este singura dovada formalizata disponibila la aceasta dat. Nu se cunosc alte studii care sa investigheze posibilitatea invatarii unor alte tpuri de gramatici (e.g., aplicarea probabilistica a sensului la sunete), sau posibilitatea ca un mecanism de invatare cu proprietati foarte diferite sa achizitioneze o astfel de gramtica (e.g. o retea neuromimetica multinivelara). Intre timp, exista simulari de invatare a unor gramatici n retele neuromimetice ce ar putea fi intepretate ca dovezi in favoarea ideii de "dobandit"(pt.detalii, vezi Elman et. al.). De exemplu Elman (1993) a aratat ca o gramatica artificiala cu elemente subordonate si dependente la distanta (i.e., acordul gramatical), poate fi invatata de o simpla retea recurenta ce opereaza in timp si incearca sa ghiceasca ce urmeaza, cate un cuvant pe rand. Sistemul realizeaza aceasta performanta apeland doar la o baza de date pozitive (i.e., feeback-ul vine sub forma supozitiilor ce sunt sau nu reconfirmate la aparitia urmatorului cuvant). De asemenea, face erori pe care apoi le corecteaza, in absenta datelor negative, oferind o dovada prima facie impotriva generalitatii Teoremei lui Gold vizand invatarea limbajului intr-un alt tip de sistem. Oricum nu orice retea neuromimetica poate raliza aceasta sarcina. Elman a descoperit ca reteaua sa poate invata gramatica doar daca a fost expusa in prealabil la propozitii simple, propozitiile complexe fiind introduse ulterior. Dar aceasta situatie nu apare si in cazul copiilor, ce sunt expusi cel putin la citeva propozitii subordonate inca de la inceput. Elman a constatat ca poate obtine acelasi rezultat printr-o metoda simpla: se initializeaza sistemul cu o memorie foarte instabila (cantonata in unitatile ce copiaza starea interna a sistemului dintr-o incercare anterioara), crescand gradat memoria (independent de invatare in sine) pana la forma adulta. In consecinta, reteaua poate invata doar siruri scurte in primele etape de achizitie - cu toate ca inputul contine inca de la inceput atat siruri simple, cat si complexe. Acest exemplu ilustreaza conceptia noastra expusa anterior in legatura cu diferitele nivele la care se manifesta caracterul innascut (ereditatea); o gramatica ce nu a putut fi invatata in conditiile existente la un anumit moment in timp poate fi achizitionata intr-o retea recurenta cand aceste conditii se modifica - toate acestea realizandu-se fara a ingloba in arhitectura reprezentari innascute, inainte de debutul invatarii. Pe scurt, nu putem deduce din prezenta unor structuri excentrice ca aceste structuri sunt innascute - nici macar daca acestea apar exclusiv la specia umana, sunt caracteristice tuturor membrilor normali ai acestei specii, localizate in anumite parti specifice ale sistemmului si achizitionabile doar in conditii determinate. Aceleasi date pot fi explicate inlocuind cunostintele innascute (i.e. reprezentarile) cu constrangeri structurale si temporale ce necesita mult mai putin informatie specificata genetic. Acest tip de solutie energentista la controversa innascut - dobandit exista de mai multi ani, dar a devenit o alternativa viabila din punct de vedere stiintific doar pe parcursul ultimului deceniu. In consecinta, mult asteptata reconciliere intre Platon si Aristotel este pe cale sa se produca.

BIBLIOGRAFIE SELECTATA Bates, E., Thal, D., Aram, D., Eisele, J., Nass, R., & Trauner, D. (1997). From first words to grammar in children with focal brain injury. In D. Thal & J. Reilly, (Eds.), Special issue on Origins of Communication Disorders, Developmental Neuropsychology, 275-343. Caplan, D. (1981). On the cerebral organization of linguistic functions: Logical and empirical issues surrounding deficit analysis and functional localization. Brain and Language, 14, 120-137. Chomsky, N. (1980). On cognitive structures and their development: A reply to Piaget. In M. Piatelli-Palmarini (Ed.), Language and learning: the debate between Jean Piaget and Noam Chomsky. Cambridge, MA: Harvard University
9

Press. Crain, S. (1991). Language acquisition in the absence of experience. Behavioral and Brain Sciences, 14, 597-611. Deacon, T.W. (1997). The symbolic species: The coevolution of language and the brain. New York: Norton. Edelman, G.M. (1987). Neural Darwinism: The theory of neuronal group selection. New York: Basic Books. Eisele, J., & Aram, D. (1995). Lexical and grammatical development in children with early hemisphere damage: A cross-sectional view from birth to adolescence. In Paul Fletcher & Brian MacWhinney (Eds.), The handbook of child language (pp. 664-689). Oxford: Basil Blackwell. Elman, J. L. (1993). Learning and development in neural networks: The importance of starting small. Cognition, 48, 71-99. Elman, J., Bates, E., Johnson, M., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking innateness: A connectionist perspective on development. Cambridge, MA: MIT Press/Bradford Books. Fodor, J.A. (1983). The modularity of mind: An essay on faculty psychology. Cambridge, MIT Press. Gold, E.M. (1967). Language identification in the limit. Information and Control, 16, 447-474. Hubel, D.H., & Wiesel, T.N. (1963).Receptive fields of cells in striate cortex of very young, visually unexperienced kittens. Journal of Neurophysiology, 26, 944-1002. Johnson, M.H. (1997). Developmental cognitive neuroscience: An introduction. Oxford: Oxford University Press. MacWhinney, B. & Bates, E. (Eds.) (1989).The cross-linguistic study of sentence processing. New York: Cambridge University Press. Piatelli-Palmarini, M. (1989). Evolution, selection and cognition: From "learning" to parameter setting in biology and the study of language. Cognition, 31(1), 1-44. Pinker, S. (1994). The language instinct: How the mind creates language. New York: William Morrow. Rumelhart, D. & McClelland, J. (1986).Parallel distributed processing: Explorations in the microstructure of cognition. Cambridge, Mass.: MIT Press. Spelke, E.S. (1991). Physical knowledge in infancy: Reflections on Piaget's theory. In S. Carey & R. Gelman, (Eds.), Epigenesis of the mind: Essays in biology and knowledge. Hillsdale, NJ: Erlbaum. Thelen, E., & Smith, L.B. (1994). A dynamic systems approach to the development of cognition and action. Cambridge, MA: MIT Press. Webster, M.J., Bachevalier, J., & Ungerleider, L.G. (1995). Development and plasticity of visual memory circuits. In B. Julesz & I. Kovacs (Eds.), Maturational windows and adult cortical plasticity in human development: Is there reason for an optimistic view? Reading, MA: Addison-Wesley. Wills, C. (1991). Exons, introns, and talking genes: The science behind the Human Genome Project. New York: Basic Books.

10

Functional Brain Development in Infants: Elements of an Interactive Specialization Framework Mark H.. Johnson Child Development, 71, 75-81, 2000 One future direction for cognitive development research involves a closer integration with our knowledge about the developing brain. I present a framework for analyzing and interpreting postnatal functional brain development in human infants. Three specific hypotheses contribute to this framework, within which a variety of phenomena associated with the neural basis of perception and cognition in normal and abnormal development can be characterized. INTRODUCTION Over the past decade there has been increasing interest in relating behavioral changes during infancy to underlying changes in the brain, and in particular to the postnatal development of the cerebral neocor tex (e.g., Diamond, 1991; Johnson, 1997a, 1997b; Nelson & Bloom, 1997; Richards, in press). Aside from the potential clinical benefits, increased knowledge of the relation between the developing brain and behavior may change ideas and theories about the mechanisms underlying perceptual, motor, and cognitive development. In addition, the increasing availability of methods for imaging the brain at work will allow us to study the neural basis of cognitive ability during childhood much more readily. It will therefore be necessary to develop theories about the functional development of the brain that allow us to integrate information about the structural (neuroanatomical) development of the brain with data about perceptual, cognitive, and behavioral changes. Many of the efforts so far to relate brain to behavioral development have been based on a maturational framework in which it is assumed that as particular regions of the brain mature they allow or enable new sensory, motor, and cognitive functions to appear. A related assumption has been that there is a progres sion of maturation from posterior to anterior regions, with the primary visual cortex being functional from shortly after birth and the frontal lobes, and prefrontal cortex, being the last parts to become functional. This posterior-to-anterior assumption is largely based on evidence from structural neuroanatomical studies (see Johnson, 1997a, for review), but it is assumed to have direct functional consequences 1. I suggest that in the forthcoming decades we have to move beyond the maturational framework. In this paper I present elements of an alternative framework (way of thinking) for postnatal functional cortical development. This framework includes three specific hypotheses about postnatal functional brain development. Finally, I discuss the implications of this proposed view for developmental disorders and early brain damage. A maturational framework underlies many current models that attempt to relate the physical growth of the brain to changes in behavior. For example, Johnson (1990) advanced a model of how the differen tial maturation of cortical pathways could explain changes in visual orienting and attention in human infants over the first six months of life. This model assumed that frontal regions (such as frontal eye fields) were on the last pathway to mature and in general corresponded to a posterior-to-anterior maturational progression. While neural and behavioral evidence available at the time corresponded well with predictions of the model, more recent studies suggest that frontal structures may be playing an active role earlier than more posterior structures. For example, in studies of saccade-related potentials in 6-month-olds we
1

By "functional" in this paper, I refer to the information processing properties of the neural circuits in question, and not to basic physiological functions associated with maintaining cells.

11

have observed eye-movement related potentials over frontal sites but not over the more posterior sites where they are observed in adults (Csibra, Tucker, & Johnson, 1998, in press). Converging results are obtained when eye-movement tasks are studied in infants with perinatal focal damage to cortex: it is infants with damage to the frontal quadrants of the brain who show deficits, and not those with more pos terior damage which produces such deficits in adults (Johnson, Tucker, Stiles, & Trauner, 1998). To briefly preview the framework to be presented, I argue that functional brain development can be analyzed in terms of the differential specialization of multiple coactive pathways. In addition to differential specialization between brain pathways, I hypothesize that differential patterns of specialization within a pathway can occur during development, sometimes resulting in patterns and temporal dynamics of the neural basis of behavior in infants that differ from those observed in adults. These differences in the neural control of behavior in infants can be empirically observed and may help make sense of some currently puzzling data. While the hypotheses presented can be tested independently, together they form the basis of a framework for a new way of thinking about functional brain development in infancy within which more specific models dedicated to particular domains can be developed. What is presented here is far from a complete theory of functional brain development in infants, but it is intended to set a direction and agenda for future research. ISSUES AND ASSUMPTIONS Background issues. Gottlieb (1992) distinguished between two approaches to the study of development. The first of these, predetermined epigenesis assumes that there is a unidirectional causal path from genes to structural brain changes to psychological function. In contrast, the second approach, probabilistic epigenesis, views interactions between genes, structural brain changes, and psychological function as bidirectional. Integral to the latter approach is the importance of activity-dependent development. One manifestation of this in postnatal life is that infants themselves select appropriate inputs for the subse quent further specialization of their brains Johnson and Morton, 1991). As discussed above, much current theorizing on the neural basis of sensory, motor, and cognitive change, is based on the view that the mat uration of particular neocortical regions or pathways allows or enables new functions to appear (e.g., Diamond, 1991; Johnson, 1990). This clearly follows a predetermined epigenesis viewpoint in which the primary cause of a cognitive change is neural maturation. A number of recent reviews of pre- and postnatal brain development have concluded that probabilistic epigenesis is a more appropriate way to view postnatal brain development (and also prenatal development, though that is not the topic of this paper) (e.g., Johnson, 1997a; Nelson & Bloom, 1997). Explaining developmental change when there are bidirectional interactions between brain structure and (psychological) function is, however, more challenging. Possibly the recent advent of connectionist modeling will offer a formal way to analyze interactions between neural structure and function (Elman et al., 1996). When adopting a probabilistic epigenesis viewpoint, the aim is still to unite developmental neuroanatomical observations with functional development. Until we have more detailed knowledge of the bidirectional relations between structure and function, however, it would be unwise to use structural neuroanatomical data to constrain theories of functional brain development too closely. In a related manner, a probabilistic epigenesis approach emphasizes the need for notions of partial functioning of neural pathways because in order for bidirectional interactions between brain structure and function to work, there needs to be early partial functioning which then shapes subsequent structural developments. From this viewpoint, then, structural and functional changes in regions of the cortex codevelop. It does not appear to be the case that cortical regions are functionally silent before they become mature in terms of their neuroanatomical structure . In contrast to some other regions of the brain, it is clear that the cerebral cortex shows structural and functional changes well into postnatal life in humans (see Johnson, 1997a, for review). It is also the part of the brain most commonly associated with higher cognitive functions, and therefore likely to have central importance for cognitive and perceptual development. The general neuroanatomy of the neocortex is remarkably similar across both regions and species. It remains controversial to what extent the detailed differences in neuroanatomy sometimes detectable between and within regions of the neocortex are the

12

result of prenatal genetic and molecular factors, or the result of pre- and postnatal neuronal activitydependent processes (see Johnson, 1997a, for review). Although it is not feasible to use functional brain imaging methods such as positron emission tomography (PET) or functional magnetic resonance imaging (FMRI) with young infants, it is possible to use scalp recorded event-related electrical potentials (ERP). Recent advances in this technology allow recording with high-density arrays for improved spatial resolution (Tucker, 1993) . This method is thought to detect electrical changes, as groups of neurons fire within the cerebral cortex. It this.- provides a good method for studying the time course and spatial distribution of functional activity in cortex. Assumptions. A number of specific assumptions underlie the framework to be presented, all of which have at least some supporting evidence. The first of these is that from shortly after birth multiple path ways and structures within the cerebral cortex can be partially activated by stimulus presentation or task situations. Whether this includes the entire set of pathways and structures available to the adult for in formation processing is unknown. It is assumed that in the newborn, while major cortical pathways, such as the dorsal and ventral routes of visual processing, are relatively immature, neurons on the pathway are capable of being activated and of transmitting information to output centers of the brain. While even in the adult multiple pathways are potentially engaged in a range of tasks, particular pathways appear to dominate, in the sense of most directly controlling behavioral output in specific task contexts. An example of this comes from the study of the neural basis of eye-movement control. Schiller (1998) describes multiple pathways that are involved in eye-movement planning and control in adult primates. In specific task situations, however, such as making an anticipatory eye movement to an expected target location, a particular subset of these pathways dominate the response of the subject. The next assumptions relate to why particular pathways dominate the behavioral response in particular stimulus or task situations. It is assumed that two temporal factors are important here: the relative speed of processing for a given type of information in different pathways, and the speed of response of the subject. Behavioral responses are partly guided by the information processed when the action is initiated. A delayed response allows slower pathways to influence the response. Rapid reactions are made on the basis of faster, more automatic, brain pathways. For. example, in tasks in which adult participants are re quired to make eye movements to a series of visual targets, pressure to respond rapidly can lead to sys tematic errors generated by more reflexive eye movement pathways (Dassonville, Schlag, & SchlagRey, 1992), some of which are subcortical (Schiller, 1998). AN INTERACTIVE SPECIALIZATION FRAMEWORK Hypotheses. In this section, I present three interrelated working hypotheses that contribute to the position I wish to advance. I first summarize these hypotheses and then examine each in more detail. 1. Cortical pathways differ from each other by virtue of their particular pattern of inputs and outputs to other brain structures and biases in their information-processing properties. The latter refer to slight differences, such as those in the detailed patterns of intrinsic connectivity, the balance of neurotransmitters, or synaptic density. Such differences correspond to those that Elman et al. (1996) referred to as "architectural constraints." In the newborn, however, they are thought to be slight biases on a very similar, general, immature neocortical architecture. These initial biases are sufficient to ensure that particular types of sensory input, or input-output pairings, are more efficiently processed by a subset of the pathways. There is thus a process of "recruitment" of particular pathways and structures for certain functions. 2. During development, cortical pathways and structures go through a process of specialization. By specialization I mean the extent to which a given cortical region is selective in its response properties. Shortly after birth a cortical region may respond to a wide variety of visual stimuli, but with development the same region may only be engaged by a subset of these stimuli. This process may be somewhat akin to the tuning of response properties of single neurons. I hypothesize that in addition to narrowing response properties, the process also facilitates speed of processing for the stimuli or task situations on which it becomes specialized.

13

3. Specialization within a cortical pathway does not necessarily proceed from sensory input to output during development. In at least some cases the process starts at later stages of the cor tical processing stream and progresses to earlier (shorter latency ) stages. The first hypothesis outlined above is that initial biases in the different pathways activated, along with the patterns of connectivity of these pathways to other parts of the brain, lead to some pathways and structures being marginally more suited to processing information about some kinds of input, or input output relations. Hypothesis 1 suggests that in the newborn infant few, if any, of these cortical pathways are specialized (in the above sense) for most tasks, but by virtue of initial biases some are engaged more than others by particular tasks or stimuli. Hypothesis 2 asserts that these initial biases are then greatly am plified by a subsequent process of specialization. This notion further specifies the view initially advanced by Elman et al. (1996) that cortical structures are "recruited" for computational functions during development. Specialization refers to the extent to which a given cortical region is selective in its response properties {Johnson, 1999), and is related to the notion of "perceptual narrowing" previously described by Nelson (1993). For example, a recent high-density ERr study investigated the effects of inversion on face process ing in infants and adults. The "inversion effect" refers to the observation that inverted faces activate neural and cognitive mechanisms different from those activated by upright faces. De Haan, Oliver, and Johnson (1998) found that infants at 6 months showed the same "inversion effect" with monkey faces as they do with human faces, a finding consistent with their processing human and monkey faces in the same way. In contrast, in adults, human face stimuli produce different patterns of activation from those produced by other species (de Haan et al., 1998). These results suggest that cortical processing of faces becomes more finely tuned (specialized) for human faces during human development. While in this paper specialization is used to describe a particular type of change in development, there are several plausible neurocomputational mechanisms which could underlie it. One likelihood is that it in volves the selective pruning of inappropriate synaptic connections, and possibly also inhibition of other alternative pathways Jacobs, 1999; Shrager & Johnson, 1996). For the purposes of the present framework, the mechanisms underlying specialization need only be sufficient to cause particular cortical pathways to become more adept than other pathways (which were initially alternative and coactivated pathways) at processing a particular class of stimuli. Predictions and supporting evidence. In this section I explore some predictions and implications of the framework presented, and use it to account for a number of phenomena associated with postnatal functional brain development in human infants. The three sets of phenomena I will focus on are: 1. Changes in localization during development, that is, developmental changes in the extent of cortex activated following presentation of a stimulus or a task situation. 2. Evidence that infants often act before fully processing information relevant to the action. 3. The precocial involvement of some frontal cortical regions during early infancy. A number of authors have described developmental changes in the spatial extent of cortical activation in a given situation during postnatal life. Event-related potential experiments with infants have indicated that both for word learning (Neville, 1991) and face processing (de Haan et al., 1998) there is increasing localization of processing of a stimulus class with age/ experience. That is, a larger area of the scalp shows a stimulus-specific effect in younger infants than in older ones. Within the present framework, such changes may be attributed to more pathways being partially activated before experience with a class of stimulus leads to the specialization of one or more of those pathways. In the example of face processing, both the left and the right ventral visual pathways are differentially activated by faces in early infancy, but in many (but not all) adults this localizes further to the right ventral pathway only Johnson & de Haan, in press). In the example of word recognition, differences are initially found over widespread cortical areas, but be come localized to left temporal leads after further experience with this class of stimulus (Neville, 1991). Thus, changes in localization can be viewed as a direct consequence of specialization. Fewer pathways

14

become activated by a given stimulus because most of them become tuned to other functions and therefore are no longer engaged by the broad range of stimuli they responded to earlier in development. Addition ally, it is possible that there is inhibition from pathways that are becoming increasingly specialized for that function. In this sense, then, there is competition between pathways to recruit functions. The second class of phenomena that can be interpreted within the framework outlined above concerns the relative temporal dynamics of alternative cortical routes of information processing. Given that more cortical pathways may be activated in infants, how is it that their behavior shows evidence of deficits in processing relative to that of adults? I have I already suggested that one part of the answer to this question lies in the lack of specialization of these pathways for computational functions suited to the information they process. I now suggest that another (and related} part of the answer lies in the speed with which these pathways act to influence output. Many ERP studies with infants have shown that differences between trial types (such as words and nonwords) are found at much longer latencies in infants than in adults. For example, face-sensitive effects in the ERP occur around 170 ms after stimulus onset in adults, but are not observed until around 350 ms in 6-month-old infants (de Haan et al., 1998). It is widely assumed that such differences are due to slower neural conduction rates in the infant brain, but recent spatiotemporal analyses of high-density ERP recordings suggest that this may be only one part of the story. Specifically, infants may show longer latency effects because specialization occurs later in their cortical processing stream than it does in adults. One line of evidence consistent with this account comes from the fact that the early visually evoked po tentials (VEPs) to foveal stimuli are only slightly delayed in 6-month-olds relative to adults, suggesting that at least for these early components of cortical visual processing neural processing speed is only one contributory factor. A second line of evidence comes from the spatiotemporal analyses of the ERP pro duced when an infant views faces which suggest that the face-specific response occurs at a later stage of cortical processing in infants than in adults. Both infants and adults show three distinct temporal phases in their cortical response to complex visual stimuli. While adults show face-sensitive responses at the sec ond of the three stages, it is not until the third phase that infant ERPs to faces differ from those to other stimuli. Thus, in at least this case, the face-specific response occurs much later in time in the infant than in the adult partly because it takes further stages of cortical processing to achieve a selective response. If we extend this observation from face processing to other domains, it raises the further hypothesis that Specialization (in the sense defined above) begins at later real-time stages of cortical processing ( commonly those regions more associated with output), and, with experience, moves forward in real time to early stages of the processing (commonly associated with analyzing sensory input). Since parts of the frontal cortex have major outputs to subcortical regions to control behavior, and since the frontal cortex appears to be the terminal cortical structure for many pathways (Fuster, 1989), I speculate that this region may show evidence of specialization (stimulus selectivity ) before many other cortical regions. This prediction stands in contrast to the currently prevailing view that frontal, and especially prefrontal, cortical regions are the last to show functional activity in human postnatal brain development. It is important to clarify that these claims do not call into question specific hypotheses about the rela tion between prefrontal cortex development and performance in several detour and occluded object reaching tasks toward the end of the first year of life (e.g., Diamond, 1991). They merely suggest that other associations between frontal and prefrontal development and behavioral performance will be found at younger ages. Some preliminary indications of this come from recent high-density ERP studies of saccade-related potentials in 6-month-olds. As mentioned earlier, Csibra and colleagues (1998, in press) examined the cortical activity associated with the planning of eye movements in 6-month-olds and observed eye movement related potentials over frontal sites, but not over the more posterior sites where they are observed in adults (see also Richards, in press). Converging resuits are obtained when eye movement tasks are studied in infants with perinatal focal damage to cortex. Infants with damage to the frontal quadrants of the brain show deficits in such tasks, whereas those with more posterior damage, which causes such effects in adults, do not Johnson et al., 1998; see also Craft, White, Park, & Figiel, 1994). To summarize the related hypotheses presented so far, in the first few weeks after birth many brain path ways are partially activated by task situations in infants, but most or all of the cortical pathways are un-

15

specialized. Action in young infants is therefore often based on the output from the first available pathway, which may often result in an impoverished or incorrect response when the task situation is more complex or demanding. In very young infants, subcortical pathways will often be faster than any of the cortical routes, and may thus dominate behavioral responses. With further age/experience, pathways become more specialized and therefore fewer of them are initially activated in a given stimulus/ task context (resulting in increased localization). Those that remain activated, however, may compete more vigorously to control action. At this later stage, the child has more options available for guiding action appropriately. For example, by delaying action, the child can bring more complex representations to bear on behavior. By adulthood, more specialized pathways mean that fewer pathways are engaged by a task situation (although it is still likely to be more than one), and response time (imposed by task demands) may determine which pathway guides the action. Disorders and deviations. We can use the framework outlined above to consider (1) the effects of early brain damage, and (2) developmental disorders of known or suspected genetic etiology. Starting with the former, it should be evident that the ;framework presented allows for the possibility that, prior to the specialization of the relevant pathways, other pathways that are also initially activated by the same stimulus/ task can compensate for the damaged pathway or region. In other words, pathways have the capacity to specialize for a variety of functions. It is important to note that the same process of specialization that would occur normally is involved, albeit with a different allocation of functions to pathways. It is likely, however, that in at least some domains there will be some cost to the reallocation of pathway specialization given that functions will not always be allocated to pathways with the optimal initial biases and/or connectivity to other regions. Although the effects of early focal lesions on cognition are often variable, complex, and domain dependent, at least in the area of effects on language acquisition some general conclusions can be reached (Stiles & Thal, 1993). 1. Regions of the left temporal lobe may be best suited to language processing, but they are not critical since language can develop in close to normal ways with this region damaged. 2. Focal pre- and perinatal lesions often cause de lay regardless of the site of the lesion. 3. Different regions of the cortex may be involved in the acquisition of language in infants from those that are important for language in adults. Specifically, there may be changes in the extent and pattern of lateralization as well as changes along the anterior-posterior axis. These general conclusions are largely consistent with the framework outlined above in that prior to the specialization of relevant pathways, other pathways can specialize in their place in the event of damage. For example, right ventral visual pathways (such as those that normally specialize for face processing) could be replaced by left ventral visual pathways in the event of early damage. Such rearrangements of the normal patterns of specialization are likely to have some general cost even on other domains, however, since there are fewer pathways available for specialization and the rearrangement of specialization may not be optimal. The observation that the cortical regions critical for a function may change during acquisition is also broadly consistent with hypothesis 3, which states that specialization within a pathway sometimes progresses from cortical output regions toward those closer to input structures. If the patterns of specialization within a pathway are different for infants from those seen in adults, we would expect differences between infants and adults in the effects and locations of lesions that cause effects. With regard to developmental disorders of genetic etiology, it would be premature to offer any explana tions of specific disorders. With a framework such as that presented here, however, we can reverse the form of question normally asked about such disorders. Instead of seeking specific explanations for particular disorders, we can inquire what the characteristics of "theoretical" disorders would be following different types of deviation within the framework (see also Oliver, Johnson, Karmiloff-Smith, & Pennington, 2000). At a later stage it would then be valuable to assess the extent to which certain real disorders might share these characteristics. For example, one type of deficit could involve defects in the molecular mechanisms underlying the specialization process itself. In terms of the neural substrates of

16

behavior such disorders would manifest as II arrested development" with many relatively unspecialized pathways still competing for influence over behavior, a situation which would presumably result in variable and poor performance over a range of tasks. Other kinds of disorders could possibly affect particular pathways, and compensation by other pathways could come into effect in a way similar to that following early focal damage. Behaviorally, patterns of mild deficit could result from this. More cognitive-specific kinds of disorder could result from subtle molecular and cellular distortions that change some of the initial biases in some pathways. Thus, the normal patterns of biases that eventually result in the characteristic adult allocation of functions to pathways would be altered, resulting in a different neural basis for behavior even when near normal behavioral competence is achieved (see also Karmiloff-Smith, 1998). A more detailed account of how differing initial biases can alter the nature of the functions that subsequently emerge has been presented elsewhere (Oliver et al., 2000). CONCLUSIONS In this paper I have presented a framework for the analysis of postnatal functional brain development. The framework has a number of implications for views on cognitive and behavioral development in infancy. As opposed to stage-like transitions between levels of cognitive ability, the framework encourages cognitive models with multiple competing representations that are engaged by particular stimuli or task demands, and which compete to influence behavioral output. Such models are already being generated. For example, Mareschal, Plunkett, and Harris (1999) have modeled behavioral data concerning infant responses to occluded objects in a connectionist network model with two pathways which are differentially activated by tracking either an object's spatiotemporal properties or its surface features. During the next decades, I anticipate great advances as we come closer to incorporating evidence from neuroscience into our understanding of mental development in infants and children. ACKNOWLEDGMENTS This research was financially supported by MRC Programme grant G97 15587, EU Biomed grant BMH4CT97-2032, and by Birkbeck College. I wish to thank Ron Barr, Annette Karmiloff-Smith, and Denis Mareschal for their comments on earlier versions of this paper. ADDRESS AND AFFILIATION Corresponding author: Mark H. Johnson, Center for Brain and Cognitive Development, School of Psychology, Birkbeck College, University of London, London WCIE 7HX, United Kingdom; e-mail: mark.johnson@psyc.bbk.ac.uk. REFERENCES Craft, S., White, D. A., Park, T. S., & Figiel, G. (1994). Visual attention in children with perinatal brain injury: Asymmetric effects of bilateral lesions. Journal of Cognitive Neuroscience, 6, 165-173. Csibra, G., Tucker, L. A., & Johnson, M. H. (1998). Neural correlates of saccade planning in infants: A high-density ERP study. International Journal of Psychophysiology, 29, 201-215. Csibra, G., Tucker, L. A., & Johnson, M. H. (in press). Anticipatory and reactive saccades in infants: A high-density ERP study. Infancy. Dassonville, P., Schlag, J., & Schlag-Rey, M. (1992). The frontal eye field provides the goal of saccadic eye movement. Experimental Brain Research, 89,300-310. de Haan, M., Oliver, A., & Johnson, M. H. (1998). Electrophysiological correlates of face processing by adults and 6-month-old infants. Journal of Cognitive Neuroscience (Annual Meeting Supplement), 36. Diamond, A. (1991). Neuropsychological insights into the meaning of object concept development. In S. Carey & R. Gelman (Eds.), The epigenesis of mind: Essays on biology and cognition (pp. 67-110). Hillsdale, NJ: Erlbaum. Elman, J ., Bates, E., Johnson, M. H., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking Innateness: A connectiorist perspective on development. Cambridge, MA: MITPress.

17

Fuster, J. M. (1989). The prefrontal cortex (2nd ed.). New York: Raven Press. Gottlieb, G. (1992). Individual development and evolution. New York: Oxford University Press. Jacobs, R. A. (1999). Computational studies of the development of functionally specialized neural modules. Trends in Cognitive Sciences, 3,31-38. Johnson, M. H. (1990). Cortical maturation and the development of visual attention in early infancy. Journal of Cognitive Neuroscience, 2, 81-95. Johnson, M. H. (1997a). Developmental cognitive neuroscience: An introduction. Oxford: Blackwell. Johnson, M. H. (1997b). The neural basis of cognitive development. In D. Kuhn & R. S. Siegler (Eds.), w. Damon (Series Ed.), Handbook of child psychology: Vol. 2. Cognition, perception and language (pp. 149). New York: Wiley. Johnson, M. H. (1999). Cortical plasticity in normal and abnormal cognitive development: Evidence and working hypotheses. Development and Psychopathology, 11,419-437. Johnson, M. H., & de Haan, M. (in press). Developing cortical specialization for visual-cognitive function: The case of face recognition. In J. L. McClelland & R. S. Siegler (Eds.), Mechanisms of Cognitive Development: Behavioral and Neural Perspectives. Mahwah, NJ: Erlbaum. Johnson, M. H., & Morton, J. (1991). Biology and cognitive development: The case of face recognition. Oxford: Blackwell. Johnson, M. H., Tucker, L.A., Stiles, J., & Trauner, D. (1998). Visual attention in infants with perinatal brain damage: Evidence of the importance of anterior lesions. Developmental Science, 1, 53-58. Karmiloff-Smith, A. (1998). Development itself is the key to understanding developmental disorders. Trends in Cognitive Sciences, 2, 389-398. Mareschal, D., Plunkett, K., & Harris, P. (1999). A computational and neuropsychological account of object-oriented behaviours in infancy. Developmental Science, 2,306-317. Nelson, C. A. (1993). The recognition of facial expressions in infancy: Behavioral and electrophysiological evidence. In B. de Boysson-Bardies, S. de Schonen, P. Jusczyk, P. McNeilage, &J. Morton (Eds.), Developmental neurocognition: Speech and face processing in the first year of life (pp. 187 -198). London: Kluwer Academic Press. Nelson, C. A., & Bloom, F. E. (1997). Child development and neuroscience. Child Development, 68,970987. Neville, H. J. (1991). Neurobiology of cognitive and language processing: Effects of early experience. In K. R. Gibson & A. C. Petersen (Eds.), Brain maturation and cognitive development: Comparative and cross-cultural perspectives (pp. 355-380). Adaline de Gruyter Press. Oliver, A., Johnson, M., Karmiloff-Smith, A., & Pennington, B. (2000). Deviations in the emergence of representations: A neuro-constructivist framework for analysing developmental disorders. Developmental Science, 3,1-23. Richards, J. E. (in press). Cortical indices of saccade planning in infants. Infancy. Schiller, P. H. (1998). The neural control of visually guided eye movements. In J. E. Richards (Ed.), Cognitive neuroscience of attention: A developmental perspective (pp. 3-50). Mahwah, NJ: Erlbaum. Shrager, J., & Johnson, M. H. (1996). Dynamic plasticity influences the emergence of function in a simple cortical array. Neural Networks, 9,1119-1129. Stiles, J., & ThaI, D. (1993). Linguistic and spatial cognitive development following early focal brain injury: Patterns of deficit and recovery. In M. H. Johnson (Ed.), Brain development and cognition: A reader (pp. 643-664). Oxford: Basil Blackwell. Tucker, D. (1993). Spatial sampling of head electrical fields: the geodesic sensor net. Electroencephalography and Clinical Neurophysiology, 87, 154-163.

18