Scribd Logo
Upload

Welcome to Scribd!

  • Pinker PDF

    Uploaded by

    S
    14 views7 pages

    Original Title

    pinker.pdf

    Copyright

    © © All Rights Reserved

    Available Formats

    PDF or read online from Scribd

    Did you find this document useful?

    Is this content inappropriate?

    Report this Document

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    14 views7 pages

    Pinker PDF

    Uploaded by

    S

    Copyright:

    © All Rights Reserved
    Download as PDF or read online from Scribd
    Flag for inappropriate content
    of 7
    Rules of Language Steven Pinker Science, New Series, Vol. 253, No, 5019 (Aug. 2, 1991), 530-535. Stable URL hitp:/flinks.jstor-org/sicisici=0036-8075% 28199 10802%293%3A253%3A5019%3C530%3 AROL%3E2.0,CO%3B2-Q Science is eurrently published by American Association for the Advancement of Science, Your use of the ISTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at hup:/www,jstororglabout/terms.hml. ISTOR’s Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at hutp:/wwwjstor.org/joumals/aaas hum. ch copy of any part of'a JSTOR transmission must contain the same copyright notice that appears on the sereen or printed page of such transmission, ISTOR is an independent not-for-profit organization dedicated to creating and preserving a digital archive of scholarly journals. For more information regarding JSTOR, please contact support @ jstor.org. hupulwww jstor.org/ Fri Jan 6 11:07:11 2006 sigalatons ae dicused by M. Le Norman [Ball Am Aton So. 19, 65L (0987)} and ©: Token, 0. Burs, DUA. Ck, and ML. Noman it 2. hae fee in hie 22. Tien th ld Cadet pnd yo sete Sty (Clean, Our, 1951; sce abo ade [crop | 280, {19 (98iyfand A-Fera, E-Ten, and Zanes [Anon pine 79, 190 (1979). 28, MUL. Noman, K-H. A. Winkler, L. Soar in Phys of Ene Trop in gli Ro Source, Nand Rd Asronomy Ohno Worhap Ni 9 (GNRAO, Green Baa, WV, 1988), p 180. 24, DeA- inte, M. Le Nonman, J 6. Bars, Aaophys, J. 810, U63 (1986) i. $342, 700 (1989) KR Lind, D.G. Payne, DL: Me, RD. Blaoed, hi 344, 39 1989) 25, ARO BaL, Mon Not R, Aton, Soc 182, 147 (1978); R Band and. Eicher, Ps. fp, 184, 1 1987), 26, The mage of 36273 (Fg, 3D) fom R.A. Pee, unpublisied images of {800"F 808 (Pg. 30) ean be ound nN Jakion, LW" Browne DL Sve, KOR Lind Mow. Now Aston Soe 244, 750 (1990), 27, DLA. Carkey AH Bide, JO" Bor, RUA. Py, M. L Noeman, in Irequrton. 28, RP Mathews and PA. Scheer, Mon Nb. Aiton Sa, 242, 616 (199) ‘hp 628, 29, D.A'Chike and J. 0. Bus, Asp. J. 369, 308 (1991); he sample of ACR vores with ets icin hs paper, 10% hve lobes with partes. 230, RCA Tain, in Late Noten Phys No 27, Hot Spi Erg Rao ‘Seu (Springer Veg, New York, 1989), . 84. 81, P'D Baral Gc Miley, RT, Sehila,C) Lonsl, Aston. Asp. Sapp. Ser 73,518 (1988), PA. G. Scheer in somata Aural Union Sypran 97, agate a Sour (Heel, Dore, 1982) p16. 38, A°G. Willams sn 8. F- Gall Nate 315, 34 (2985. HAD. arn teay Unies of in at Urb 90, 35. D-E Hardee and M.L: Nowan, sto J. 342, 60 (1989) hid. 365,134 (1990), 36, J-H. Zia, thess, University of New Meso (199). 2 1.0, bum, M.L Noman ME Sahin prepion 34 Engle cane newer ro wnt MH in, ENO A ith sips 347,718 1989), Fores Aig BB Fermoy Roe 1M sun Hc AD. Boru 46, LAF {i5n) Hels Ala} hr at pets Rue 308,18 29,10 Bis an Py 4,578 GN) KA Doge FN One] A Ek evi Ser 72.75 099, 40. FA STA Wl dt Aa, Aso J 80018) 3082 lise SE San 1.0 is M. Sane, Na 35, 46 (198 10am 9,18 390, BSR GnIH MI Novan Ball Am Ac, Sa 2: 0 (190). Reason BS e org ante 97109 (9p fay pre ening ay He eS an a ac by fs eguin, wt gency wl be eng sd en ot _ at ws he i of aa fe pt dno he mn ‘eundng he ee py vet a Ra pom gh Crean pane ec Hotere, mmc eis ght ‘iid yh ng uel ony etn Be Young coe 45, RE Wikams als PCa, Nowe 10 Nee, 46, CF Operant N.Over Amn 301 81 0980) {5:5 at Rann) Bon Balin. to: 2” 2 (199, {8 Tarra on ane ow a prac ah nang oe nets Asma Sy i'n Alpe Nex Me, Ts ‘Sesh wa uppored by Sin Scce eanlon pe AT 61711 ak ‘SSFon bea a Na emma re Abt ge ACR a7 ee tc cobras ng Hada FA ta Shane lnc 'hites A Wola Cote Ser ‘Sensors work. Wea ha, Sn LR Babe bBo oumg. A Oompa 1 Eid F Ove, aod Rr Foy hr ang ito alte ore pape Rules of Language STEVEN PINKER Language and cognition have been explained as the prod- ucts of a homogencous associative memory structure oF alternatively, of a set of genetically determined computa- tional modules in which rules manipulate symbolic rep- resentations. Intensive study of one phenomenon of En- lish grammar and how it is processed and acquired suggests that both theories are partly right. Regular verbs (walcwaled) are computed by 2 suhiaaton ole in 2 neural system for grammatical processing; irregular verbs, (run-ran) are retrieved from an associative memory. VERY NORMAL HUMAN CAN CONVEY AND RECEIVE AN’ unlimited number of discrete messages through a highly structured stream of sound or, in the case of signed lan- ‘guages, manual gestures, This remarkable piece of natural engineer- ing depends upon a complex code or grammar implemented in the brain that is deployed without conscious effort and that develops, without explicit training, by the age of four. Explaining this talent is an important goal of the human sciences. ‘Theories of language and other cognitive processes generally fall Tie autor bi de Deparment of Bri and Copan Scca, Mach Insite of Fiology, Cambridge, MA 02139 into owo clases, Associtionism describes the brain asa homogs- neous network of imerconnecred units modified by a laming tmechanism that records coratons among, frequently co-occurring input patterns (J). Ruleand-representtion theories deseribe the thin ara computational device in which ules and principles operate fon symbolic data strctres (2, 3). Some rule theories further propor that the brain is divided inm modular computational ystems that have an organization that is largely specified genetically, done ofthe systems being language (3,4) Daring thc last 35 years there hasbeen an unprecedented empirical ‘nody of human langage tructre, acuston, se, and breakdown, allowing these centuries old proposals to be refined and teed. I wll ikstrate how intensive muldsiplinary sty of one linwstic phe- nomenon shows that both akscaionam and role theories ae pry Comet, bt about diferent components of the language system Modules of Language A grammar defines a mapping between sounds and meanings, but the mapping is not done in a single step but through a chain of intermediate data structures, each governed by a subsystem. Mor. phology is the subsystem that computes the forms of words, I focus ‘on a single process of morphology: English past tense inflection, in which the physical shape of the verb varies to encode the relative time of occurrence of the referent event and the speech act. Regular SCIENCE, VOL, 253 past tenses marking (for example, walk-walked) i arulelike process resulting in addition ofthe suffi -d. In addition there are about 180 ‘regula verbs that mark the past tense in other ways (for example, itt, come-came, fel-fl) ast tense inflection is an isolable subsystem in which grammatical ‘mechanisms can be studied in detail, without complex interactions ‘with the rest of language. Ic is computed independently of syntax, the subsystem that defines the form of phrases and sentences: The syntax of English forces its speakers to matk tense in every sentence, but no aspect of syntax works differenly with regula and ieregulat ‘verbs. Past tense marking is also insensitive to lexical semantics (5, 6): the regulsrirregular distinction does not correlate with any feature of verb meaning, For example, hit-hi, srie-stuck, and’ slap-slapped have similar meanings, but three different past tense forms; stand-stood, stand me up-stood me up, and understand-snderstood, hhave unrelated meanings but identical past tense forms, Past mark. ing is also independent of phonology, which determines the possible sound sequences in a language: the three pronunciations of the regular suffix (in ripped, ribbed, and ridéed) represent not three independent processes but a single sulix - modified to conform with general laws of English sound patterning (5). Rulelike Processes in Language English inflection can illustrate the major kinds of theories used to cxphin linguistic procescs. Traditional grammar offers the fllow- ing fst approximation: Regular infection, being fally predictable, is computed by a rule that concatenaes the affix -d tothe verb stm, “This allows a speaker to inflect an unlimited number of new verbs, an ability seen both in aduls, who easily create past forms for ncologisms like féxed, and in preschoolers, who, given a novel verb lke rik in experiments, freely produced ricked (7) Tn contrat, icreguar verb forms ate unpredictable: compare se-sut and hihi, singsong and string-srang, feel and tell-old. Therefore they must te individually memorized, Retrieval ofan ieregular form from memory ordinary blocks application ofthe regular rule, alehough in children retrieval occasionally fils, yelling “overregularization” errors ike bredked (8, 8,10) The rule-ote theory, although appealingly straightforward, is inadequate, Rote memory, i thought ofa list of shoe, designed for the very rare verbs with unrelated past tense forms, like hewus and go-went. But forall other inregular verbs, the phonological Content of the stem is largely preserved in the past form, 38 in swing-swung (5, 11). Moreover, a given ireegular pattern such as a vowel change is typically scen in a family of phoncically similar items, such a8 singsang ring-rng, spring-sprang shrinkshnonk, and swin-swan, or grow-grew, low-blew, drow-threw, and lyflew (5,9, 11. The rote theory cannot explain why verbs with ireguar pat forms come in similarity families, rather than belonging to arbitrary lists. Finally, regula pairs are peychologically not a closed lst, but their patterns can somtimes be extended to new forms on the basis of similarity to existing forms. All children occasionally use forms such a8 brng-rang and bie-ore (5, 9) A few ieegular past forms have entered che language historically under the influence of existing, forms. Qui, cost, catch are from French, and fling, sling, stick have joined imegular clusters in the last few hundred years (12); such tffecs are obvious when dialects are compared (for example, hep-olp, rierx,drag-dng, clinistome (13). Such analogizing cin bbe demonstrated in the laboratory: faced with infecting nonsense verbs like spling, many adults produce splung (6,7, 14, 13) ‘The partial sjstematicity of regular vers has been handled in ‘opposite ways by modem rule and associationist theories. One version of the theory of Generative Phonology (11) posts rules for 2aucusT 1991 ineegular verbs (for example, change {t0 4) a8 well as for regular ‘ones. The theory is designed to explain the similarity between verb stems and their past tense forms: ifthe rue ost changes a specified segment, the rest of the stem comes through in the output tne touched, by default, just asin the fully regula case. But the ale theory does not address the similarity among diferent verbs in the input setand people's tendency to generalize irregular pattems. fan irregular rule is restricted to apply toa list of words, the similarity among the words inthe list is unexplained, Bur if common pattern shared by the words is identified and the rule is restricted to apply to all and only the verbs displaying that pater (for example, change i to. when it appears afer an consonant clastcr and precedes 1g), the rue fils because the similarity to be accounted for is one of family resemblance rather than necessary 0 sificent conditions (5, 4, 14, 18: such a rule, while succesfully applying to spring, shrink, drink, would incorrectly apply to bring-roughe and fing flung and ‘would fil to apply to begin-began and swin-swam, where it should apply. ‘Associationst theories also propose that regular and irregular pattems are computed by a single mechanism, but here the mecha- nism isan associative memory. A formal implementation in neural net tems is the “connections model of Rumelhart and McClcl- land (16), which consists of an array of input units, an array of courpuc units, and a matrix of modifiable weighted links berween every input’ and every output. None of the elements or links corresponds exactly t0 2 word or rule. The stem is represented by turing on a subset of input nodes, each corresponding to a sound patter in the stem, This sends signal across each ofthe links to the ‘output nodes, which represent the sounds of the past tense form, Each output node sums ts incoming signals and tur on if the sum ‘exceeds threshold; the output form isthe word most compatible with the set of active output nodes. During the learning phase, the past tense form computed by the network is juxtaposed with the Correct version provided by a “teacher,” and the strengths of the links and thresholds are adjusted so a5 to reduce the diference. By recording and superimposing associations between stem sounds and past sounds, che model improves its performance and can generalize to new forms tothe extent that thee sounds overlap with old ones "This proces is qualitatively the same for regule and ieregular verbs stopped is produced because input ep units were linked to output, ‘pped units by previous verbs; clung is produced because ing Was, linked to ung. Asa result such models can imitate people's analo- sing of ireegular patterns to new forms. “The models, however, ae inadequate in other ways (5, 17). The precise patterns of inflectional mappings in the world’s languages are unaccounted for: the network can lea input-output mappings found in no human language, such as mirror-reversing the order OF segments, and cannot learn mappings that are common, such a8 redupliating the stem. The actual outputs are often unsystematic blends such as mal-membled and four-tourde. Lacking a representa- tion of words a lexical entries, distinct ftom their phonological or semantic content, the model cannot explain how languages can contain semantically unrelated homophones with different pas tense forms such as fesed (prevaicate) and lielay (recline, ring-ag and sering-srung, mect-met and mete-meted "These problems call for a theory of language with both a ‘computational component, containing specific kinds of rules and. representations, and an associative memory system, with certain properties of connectionist models (5, 6, 10). In such a theory, regular past tense forms are compured by a rule that concaenates 3 als with a variable standing forthe stem. Iregulars are memorized pis of words, but the linkages between the pair members restored in an asociative memory structure fostering some generalization by analogy (9, 14, 18): although string and sirmg are represented 38, ARTICLES 531 separate, linked words, the mental representation of the pair over- laps in part with similar forms like sling and bring, so that the learning of slung i easier and extensions like brane can occur asthe result of noise or decay in the parts of the representation that code the identity of the lexical entry. Because it categorically distinguishes regular ftom irregular forms, the rule-association hybrid predicts thatthe two processes should be dlissociable from virtually every point of view. With respect to the psychology of language use, iregular forms, as memorized items, should be strongly affected by properties of associative memory such as frequency and similarity, whereas regular forms should not. With respect to language structure, iregular forms, as memory-listed words, should be available as the input to other word-formation processes, whereas regular forms, being the final ourputs of such processes, should not. With respect to implementation in the brain, because regular and irregular verbs are subserved by different ‘mechanisms, it should be possible to find one system impaired while the other is spared. The predictions can be tested with methods ‘ranging from reaction time experiments to the grammatical analysis of languages to the study of child development to the investigation of brain damage and genetic deficits Language Use and Associative Laws Frequency. Ieregular verbs are memorized items, they should be better remembered the more they are encountered. Indeed, children make errors like bresked more often for verbs ther parents use in the pastrense forms less frequently (9, 10, 19). To adults, low-frequency inregular past tense forms lke smote, bade, slew, and srode sound odd of stilted and often coexist with regularized eouinterpars such as slayed and strided (5, 18, 0). As these psychological elects accumu- late over generations, they shape the language. Old English had ‘many mote irregular verbs than Modern English, such as ebide- bode, chide-hid, gil-glt; the ones used with lower frequencies have become regular over the centuries (18). Most surviving iregular verbs are used at high frequencies, and the 13 most frequent verbs in English—be, have, d, say, make, go, take, come, se, get, Know, give, find—ae al iregular (2), “Although any theory positing a frequency-sensitive memory can account for frequency effects on irregular verbs [with inverse effets ‘on their corresponding regularized versions (20)], the rule-assoia- tive-memory hybrid model predicts that regular inflection i dffe- cent If regular past tense forms can be computed on-line by ‘concatenation of symbols forthe stem and afi, they do not require Prior storage of a past tense entry and thus need not be harder or stranger for low-frequency verbs than higher ones (22) Jgments by native English speakers ofthe naturalness of word forms bear this prediction out. Unlike irregular verbs, novel oF low-frequency regular verbs, although they may sound unfamiliar in ‘themselves, do not accrue any increment of oddness or uncertainty ‘when putin the pas tense: infrced is as natural a past tense form of infarct 28 walked is of walk (8). The contrast can be seen clearly in idioms and clichés, because they can contain a verb that is not ‘unfamiliar itself but that appears in the idiom exclusively in the present or infinitive form. Irregular verbs in such idioms can sound strange when put in the past tense: Compare You'll excuse me if I forgo the pleasure of reading your paper before i's published with Last night Lfonvent the pleasure of reading student papers, or don't know how she can bear the guy with I don’t now how she bore the guy. In contrast, regular verbs in nonpas idioms do not sound worse when putin the past: compare She doesn’t sufér fools ladly with None of them ever sufered fool gladly. Similaely, some regular verbs ike aford and cope usually appear with can’, which requires the stem form, and hence sa hhave common stems but very low-frequency past tense forms (21) Bur the uncommon I don’t know how he aforded i (coped) does not sound worse than He can’t afford it (cope) "These effects can be demonstrated in quantitative studies (20) Subject? ratings of regular past tense forms of diferent verbs correlate significantly with their ratings of the corresponding stems (r= 0.62) but not with the frequency of the past form (0.14, partialing out stem rating). In contrast, ratings of irregular past tense forms correlate less strongly with their stem ratings (0.32), and significantly with past frequency (0.29, parialing ot stem rating). Experiments on how people produce and comprehend inflected forms in real time confirm this diflerence. When subjects sce verb stems ona screen and must utter the past form as quickly as possible, they take significantly less time (16- to 29msce difference) for irregular verbs with high past frequencies than irregular verbs with, low past frequencies (stem frequencies equated), but show no such difference for regular verbs (<2-msecdiference) (23). When recog- nizing words, people are aided by having seen the word previously ‘on an earlier tral in the experiment; their mental representation of the word has been “primed” by the first presentation. Presenting a regular past tense form speeds up subsequent recognition of the stem no less than presenting the stem itself (181- versus 166-msec reduction), suggesting that people store and prime only the stem and analyze a regula inflected form as a stem plus 2 suffix. In contrast, prior presentation of an irregular form is significantly less fective at priming its stem than presentation ofthe stem itself (39- versus 99-msec reduction), suggesting that the two are stored as, separate but linked items (24). ‘Sinilarity, regular verbs fall into families with similar stems and, similar past tense forms, partly because the associative nature of _memory makes it easier to memorize verbs in such families. Indeed, children make fewer overregularization error for verbs that fll nto families with more numerous and higher frequency members (5, 4-10, 25). As mentioned above, speakers occasionally extend irreg- ular patterns to verbs that are highly similar to irregular families (brane), and such extensions are seen in dialects (13). A continuous ‘effect of similarity has been measured experimentally: subjects frequently (4496) convert spling to sphune (based on string, sling, et ‘cetera, less often (249%) convert sink to shunk, and rarely (79%) ‘convert sid to sud (14). ‘The rule-associative-memory theory predicts that the ability to generate regula past tense forms should not depend on similarity to ‘existing regular verbs: The regular rule applies as a default, treating all nonieregular stems as equally valid instantiations of the mental, symbol “Verb.” Within English vocabulary, we find that a regular verb can have any sound patter, rather than falling into similarity clusters that complement the irregulars (5): for example, need-neaded coexist with bleei-bled and feedfed,bink-blinked with shrink-shranke and drink-drank. Regularieregular homophones such 4 lilayie- lied, meet.mesymete-eted, and hang-hungshang-hanged are the clearest examples. Moreover verbs with highly unusual sounds are easily provided with regular pasts, Although no English verb ends inv ot 4 neutral vowel (21), novel verbs with these patterns are readily inflecable as natural past tense forms, such as Yelsin ou-Gor- bachev’ed Gorbachev or We rhumba'd al night. Children are no more likey to overregularize an irregular verb if it resembles a family of similar regular verbs than if is dissimilar from regulars, suggesting, that regulars, unlike irregulars, do not form attracting clusters in memory (10, 25). Adults, when provided with novel verbs, do not rate regular past forms OF unusual sounds like ploomphed as any worse, relative to the stem, than familia sounds like plipped (similar to dip, flip, sip, etcetera), unlike their ratings for irregalars (15, 26). In contrast, in associationist models both irregular and regular SCIENCE, VOL. 253 ‘generalizations tend to be sensitive to similarity. For example the Rumelhare-McClelland model could not produce any output for ‘many novel regular verbs that did not resemble other regulars in the training set (5, 15, 17) Organization of Grammatical Processes Grammars divide into faily autonomous submodule in which blocks of rules produce outputs that serve (or cannot serve) as the input for other blocs of rules, Lingusic research suggests an information fow of lexicon to dervational morphology (complex vrord-formation) to regular infecon, wth regular and regular processes encapsulated within diferent subcomponents (27,28). If iegolar past tense forms are stored in memory as entries inthe mental lexicon, then like other stored words they should be the input to rules of complex word formation, If regular past tense forms are computed from words by a rule acting a a default, they should be formed from the outputs of complex word formation ‘ils. Two phenomena istrate this organization ‘Apotent demonstration ofthe earlier point tha regular processes

    You might also like