14 Spring 2001 Eye on Psi Chi

ANGUAGE COMES SO NATURALLYto us that we are apt to forget whata strange and miraculous gift it is.Over the next hour you will sit in

your chairs listening to a man makenoise as he exhales. Why would you dosuch a thing? Not because the sounds areparticularly melodious, but because thesounds convey information in the exactsequence of hisses and hums and squeaksand pops. As you recover the informa-tion, you think the thoughts that I wantyou to think. Right now I am conveyingideas about language itself, but with aslightly different sequence of hisses andpops I could be talking about anythingfrom theories of the origin of the universeto the latest plot twists in your favoritedaytime drama. The fundamental scien-tific problem raised by language is toexplain this vast expressive power. Whatis the trick behind our ability to fill eachother’s heads with so many kinds ofthoughts?

The point of this piece is that there isnot one trick but two. Each was identi-fied in the 19th century by continentallinguists. The first is the principle of thememorized word, which Ferdinand deSaussure called the arbitrary sign. Theword duck doesn’t look like a duck, walklike a duck, or quack like a duck, but Ican use it to cause you to think thethought of a duck because all of us atsome point in our lives have memorizedan association between that sound andthat meaning.

Though a memorized link between asound and a meaning is rather simple, it

can be effective. Since human memory isvast, we can convey a large number ofconcepts, simply by memorizing soundsthat are paired with them. A typical high-school graduate knows around 60,000words, which works out to a rate oflearning a new sound–meaning associa-tion approximately every 90 wakingminutes starting at the age of one. Also,these entries require little in the way ofcomputation. Given the sound, you canlook up the meaning (in comprehension);given a meaning, you look up the sound(in production).

But of course we don’t just blurt outindividual words. We combine them intophrases and sentences, and that brings upthe second trick behind language, combi-natorial grammar—what Wilhelm vonHumboldt called “the infinitive use offinite media.” Everyone who speaks agiven language has a recipe or algorithmfor combining words in such a way thatthe meaning of the combination can bededuced from the meanings of the wordsand the way they are arranged. Forexample, one English rule says that asentence is composed of a subject—anoun phrase—followed by a predicate—a verb phrase. The verb phrase in turncan consist of a verb, followed by a nounphrase—the object—followed by a sen-tence—the complement.

The advantage of combinatorial gram-mar is that by allowing us to combinesymbols we can express new combina-tions of ideas. Journalists say that when adog bites a man, that isn’t news, butwhen a man bites a dog, that is news.

Grammar allows us to convey news, byreshuffling words in particular orders.Moreover, because our knowledge oflanguage is couched in abstract symbols—noun, verb, subject, object—the samerules allow us to talk about a big dogbiting a man and a big bang creating auniverse.

Another advantage of grammar is thatthe number of combinations it can gener-ate grows exponentially with the lengthof the string. If there are, say, 10,000nouns with which to begin a sentence,and then 4,000 words one can use tocontinue it, there are 10,000 × 4,000 =40 million two-word beginnings to a sen-tence, and the number of possible sen-tences explodes as you continue to addwords to the tail of the growing sentence.A final advantage is that human gram-mars are recursive: A sentence contains apredicate, which can in turn contain asentence, which can contain a predicate,and so on. That provides an ability togenerate structures of arbitrary size,hence an unlimited number of differentsentences.

I suggest that the basic design ofhuman language combines the advan-

Words and RulesSteven Pinker, PhD

Department of Brain and Cognitive SciencesMassachusetts Institute of Technology

This article is based on Dr. Pinker’s Psi Chi/Frederick Howell Lewis Distinguished Lecturepresented on August 4, 2000, during the Psi ChiNational Convention, held in conjunction withthe annual meeting of the American Psychologi-cal Association in Washington, D.C.

This material is presented at greater lengthin Words and Rules: The Ingredients of Language,which was published in paperback in October2000 by Basic Books.

PSI CHI/FREDERICK HOWELL LEWIS DISTINGUISHED LECTURE

L...............................................................................................................................................

...............................................................................................................................................

Eye on Psi Chi Spring 2001 15

tages of these two principles. We have alexicon of words for both common andidiosyncratic entities like ducks and dogsand men, and which depends on thepsychological mechanism called memory.And we have a set of grammatical rulesfor novel combinations of entities, fordogs biting men and men biting dogs,which depends on a mental mechanismof symbol combination.

To test this idea, we need a case inwhich words and rules can express thesame idea, but are psychologically, andultimately neurologically, distinguish-able. I believe we do have such a case:regular and irregular inflection.

Verbs in English and in many otherlanguages come in two flavors. Regularverbs such as walk–walked, jog–jogged,and kiss–kissed are monotonously pre-dictable: All form the past tense bysuffixing the stem with -ed. The regularverbs are open ended. English hasthousands of existing regular verbs, andnew ones are being added all the time.When to fax entered common parlanceabout 15 years ago, no one had to run tothe dictionary to look up its past-tenseform; everyone knew it was faxed. Simi-larly, flamed, dissed, moshed, andspammed all can be deduced withouthaving to hear them in the past-tenseform.

This productivity is visible even inchildren. In 1958, Jean Berko Gleasonbrought some 4-year-olds into the laband said, “Here is a man who knows howto wug. He did the same thing yesterday.He . . . .” The children filled in the blankwith wugged, a word they had neverheard before, so they could not havememorized it beforehand; they must havegenerated it on the fly. And in one senseall children are participants in an experi-ment like that, because they pass througha stage in which they produce otherforms they could not have heard fromtheir parents, forms like comed, goed,bringed, taked, and holded.

And that brings us to the second flavorof word in English, the irregular verbs,such as bring–brought, hit–hit, go–went,sleep–slept, make–made, ring–rang, andfly–flew. In contrast to the regulars, theirregulars are unpredictable. The pasttense of sink is sank or sunk, but the pasttense of cling is clung but not clang. Thepast tense of think is neither thank northunk but thought, and the past tense ofblink is neither blank nor blunk norblought, but is regular, blinked.

Also unlike the regulars, the irregularsform a closed class. About 180 verbs areirregular in standard English, and therehave not been any recent new ones.

All this leads to a simple theory.Irregular verbs are simply pairs of words.Just as we memorize duck, we memorizebring and we memorize brought, andthen we link the two in memory. Regularverbs are generated by a rule, akin to therule generating sentences out of subjectsand predicates. This rule says that a verbin the past tense may be composed of theverb stem plus the suffix -ed. If a verbdoes not supply a past-tense form frommemory, the regular rule applies by de-fault; that is how children and adults cansay things like wugged and faxed andspammed, which cannot have been storedin memory beforehand.

Alas, there is a complication for thisneat theory: The irregular verbs displaypatterns. We find families ofirregular verbs such as:keep–kept, sleep–slept,feel–felt, and dream–dreamt;wear–wore, bear–bore,tear–tore, and swear–swore;string–strung, swing–swung,sting–stung, and fling–flung.This is not what we wouldexpect if the irregular verbswere memorized individu-ally by rote, in which casethey could just as easily allbe idiosyncratic.

Moreover, these aren’tjust redundancies in mem-ory; they are occasionallygeneralized. Occasionallychildren make errors likebring–brang, bite–bote, andwipe–wope. In the history oflanguage, every once in awhile a new irregular verbappears. Quit and knelt areonly about 200 years old(Jane Austen, for example,used quitted), and snuck,which is now standardamong Americans and Canadians under50, has been in the language only for 100years. This is especially obvious if youcompare nonstandard dialects, whichcontain forms like help–holp, drag–drug,and bring–brung. Experimental psy-chologists can even catch people in theact of generalizing an irregular pattern:When Joan Bybee and Carol Moderasked students, “What is the past tense ofto spling?” many said splang or splung.

How do we account for these patternsand generalizations, which are neitherclearly word-like nor clearly rule-like?Two alternatives to the words-and-rulesdichotomy have been proposed. Eachtries to stretch one of the components tocover the territory ordinarily allotted tothe other.

According to the theory of generativephonology from Noam Chomsky andMorris Halle, there are rules all the waydown. Just as we have a rule adding -edto form the regular past tense, we have asuite of rules that generate irregular past-tense forms by substituting vowels orconsonants. For example, one rulechanges i to u in verbs like cling–clung.

A problem for this theory is the familyresemblance among the verbs undergoingthe rule, such as string–strung, sting–stung, fling–flung, cling–clung. How doyou get the rule to apply to them? If yousimply link the rule by stipulation toeach of the words, you have no explana-tion for why the words are so similar.Why string–strung, sting–stung, and

fling–flung, which share the consonantsbefore and after the i, and not fib–fub,wish–wush, and trip–trup? The obviousmove at this point is to distill some com-mon pattern out of the set of words thatundergo a rule and append it to the ruleas a condition. But that does not workeither. Say the rule is restricted to applyonly to verbs that begin with two conso-nants and end with ng. Such a rule wouldfalsely include verbs like bring andspring, which fit the pattern but whosepast-tense forms are brought and sprang(not brung and sprung). At the sametime, the rule falsely excludes words likestick–stuck and spin–spun, which missthe condition by a whisker.

The problem is that the words show-ing an irregular pattern are family-resem-

..............................................................................................................................................................................................................................................

......................................................................

..................................................................

16 Spring 2001 Eye on Psi Chi

blance categories in the sense of LudwigWittgenstein. No set of properties runsthrough the entire class; rather, patternsof overlapping similarities probabilis-tically link various subsets.

This led to something completelydifferent: the theory of parallel distrib-uted processing or artificial neural net-works from David Rumelhart and JamesMcClelland and their followers. Ratherthan having words all the way down, thistheory has memory associations all theway up. Rumelhart and McClellanddevised a neural network model called apattern-associated memory, which linksnot an item to an item, but the featuresof an item—the sounds composing it—to the features of an item. A word ispresented to the network by turning onunits corresponding to the word’s sounds.The model is trained with examples of averb and its past-tense form: sing–sang,walk–walked, and so on. It records corre-lations between features of the stem andfeatures of the past-tense form, and thatallows it to generalize a pattern to a newverb if it is similar to verbs it has beentrained on: Once trained on ring and singand spring and cling, it automaticallygeneralizes to spling because some of thepieces of spling occupy the same repre-sentational real estate as the pieces ofring and sing. In the same way, it gener-alizes from walked and talked to balkedand stalked. It generalizes reasonablywell despite not having any distinctionbetween words and rules; a single mecha-nism handles regular and irregular forms.

Despite the ingenuity of these alter-natives, I will present evidence that thetraditional words-and-rules model isright after all, though with a twist.Irregulars really are words stored inmemory, but memory is not just a list ofslots, but is partly associative, a bit likeRumelhart and McClelland’s pattern-associator network: Features are linkedto features, as well as words being linked

to words. As a result, similar words rein-force each other and are easier to memo-rize, and they create a temptation togeneralize to new similar words.

But we cannot do without a rule forthe regulars. Irregular forms can getaway with a pattern-associator memorybecause people’s use of irregular patternsreally is limited by memory: Peopleapply the patterns only to forms thathave been memorized or to forms similarto them. But people generalize far morepowerfully when it comes to regularforms. The regular inflection can beapplied to any word, regardless of itsstatus in memory. As we shall see, thatshows that regular inflection is computedby a mental operation that does not needaccess to the contents of memory,namely a symbol-combination rule thatapplies to any instance of the symbolverb. The evidence consists of unrelatedcircumstances in which memorizedforms are not accessed for one reason oranother, but people can still apply theregular pattern.

One example is what happens whenthe memory entry for a word is weakbecause the word is rare. We know thatmemory benefits from repetition: Themore often you hear something, the bet-ter you remember it. If a word is not usedvery often, its memory trace will beweak. The prediction of the words-and-rules theory is that this should hurt theirregulars, but not the regulars.

The first test of the prediction comesfrom the statistical structure of theEnglish language. Here is a list of thetop 10 verbs in English in order of fre-quency of occurrence in a million-wordcorpus: be, have, do, say, make, go, take,come, see, get. Notice that all 10 areirregular: be–was, have–had, do–did, andso on. Now, there cannot be a bottom-10list, because in the million-word corpusabout 800 words are tied for last place,namely, one-in-a-million, the lowest

frequency you can measure in a million-word corpus. But the first 10 of thatlist, in alphabetical order, are abate,abbreviate, abhor, ablate, abridge,abrogate, acclimatize, acculturate,admix, and absorb. Notice that all 10are regular, as are the vast majority ofthe uncommon verbs in English.

The explanation is simple. If a worddeclines in frequency, there may be ageneration of children that hasn’t prop-erly memorized its irregular past-tenseform. Unable to retrieve an irregularform, they default to the regular, and theverb will change to regular for them andfor all subsequent generations. If Chaucerwere here today, he would say that thepast tense of cleave is clove, the pasttense of crow is crew, and the past ofchide is chid. Old and Middle Englishhad about twice as many irregular verbsas modern English does. Joan Bybee hasshown that it is the rarer verbs that havebecome regular; the ones that are com-mon remain irregular to this day.

You can feel this force of historyyourself. Many low-frequency irregularssound strange to us, and they are slippingout of the language before our ears.Complete this sequence: I stride, Istrode, I have ———. Stridden doesn’tsound quite right to most people, pre-sumably because it is not a form youhear every day. Similarly, smite–smote,slay–slew, bid–bade, and forsake–forsook have a quaint or stilted soundto them, and some are used in regularalternatives such as slayed. In contrast,low-frequency regulars always soundfine. If I asked you to complete this se-quence—I abrogate, I abrogated, I have———, there is nothing particularlystrange about abrogated, presumablybecause rarity doesn’t hurt a word if it isnot dependent on memory to begin with.

A nice illustration of this effect comesfrom idioms and clichés, where peoplemay be familiar with a verb in only one

Steven Pinker, PhD,a native of Montreal,received his BA fromMcGill University in1976 and his PhD inpsychology fromHarvard in 1979. Afterserving on the facul-ties of Harvard andStanford Universitiesfor a year each, he

moved to MIT in 1982, where he is currentlyProfessor of Psychology in the Department ofBrain and Cognitive Sciences and a MacVicarFaculty Fellow. His research on visual cognitionand on the psychology of language has receivedthe Troland Award from the National Academyof Sciences and two prizes from the AmericanPsychological Association. He has also received

awards for his graduate teaching at MIT and forhis undergraduate teaching at MIT, two prizesfor general achievement, an honorary doctorate,and five awards for his popular science booksThe Language Instinct and How the Mind Works.Professor Pinker is a fellow of several scholarlysocieties, including the American Academy ofArts and Sciences and the American Associationfor the Advancement of Science. He is an associ-ate editor of Cognition and serves on manyprofessional panels, including the Usage Panelof the American Heritage Dictionary and theScientific Advisory Panel of an 8-hour NOVAtelevision series on evolution. Professor Pinkeralso writes frequently in the popular press,including the New York Times, Time, Slate, andThe New Yorker. His fifth book, Words and Rules:The Ingredients of Language, was published inOctober.

................................................................................................................................................................................

................................................................................................................................................................................

Eye on Psi Chi Spring 2001 17

tense. For example, to forgo is not terri-bly common, but does have a certainliveliness in the sarcastic expression toforgo the pleasure of, as in “You’ll ex-cuse me if I forego the pleasure ofwatching the video of your wife givingbirth.” But if you force the cliché into thepast tense, you get something strange:“Last night I forewent the pleasure ofwatching Herb’s vacation slides.” Like-wise, you can say, “I don’t know how shecan bear that guy,” but it’s odd to say, “Idon’t know how she bore that guy.” Youmight say, “I dig the Doors, man,” but“In the ’60s, your mother and I dug theDoors” is peculiar.

This contrast is never found withregular past-tense forms, which alwayssound as good or as bad as their stems.“We can’t afford it” comes out as “Idon’t know how he afforded it”; “Shedoesn’t suffer fools gladly” transformsinto “None of them ever suffered foolsgladly,” both unexceptionable. The rarityof their past-tense forms doesn’t hurtthem because their past-tense formsneedn’t be stored to begin with. They canbe computed on the fly by a rule, andinherit whatever sense of naturalness orunnaturalness inheres in the verb stem.The irregulars, in contrast, consist of twoentries in memory, which can part com-pany, one familiar, the other unfamiliar.

A second instance in which memoryis unhelpful occurs when a new word isdifficult to analogize to words in memorybecause it is unusual in sound. Recallthat Bybee and Moder asked volunteersfor the past tense of spling and othernovel words. With spling, which is simi-lar to existing irregulars such as cling,fling, string, and sling, most peopleoffered splang or splung. With krink,which is less similar, only about halfproduced krank or krunk. And with vin,which shares only the vowel, hardly

anyone suggested van or vun. SandeepPrasada and I replicated the experiment,but also varied the similarity of novelwords to regular forms. To plip is similarto many regular verbs such as clip, flip,strip, nip, slip, sip, tip, and trip. To smeejdoesn’t rhyme with any English verbroot, and to ploamf doesn’t sound likeanything. We presented the words tohuman participants and to the trainedpattern-associator model. With theirregular-sounding verbs, the model dida reasonably good impersonation of thehuman being: They both produced manysplangs and splungs, fewer kranks andkrunks, and very few vans and vuns. Butwith the regular-sounding words, peopleand the model diverged: People offeredplipped, smeejed, and ploamfed, whereasthe model could get only plipped. Wherepeople deduced smairfed, the modelcame up with sprurice. Smeejed cameout as leefloag, and frilged came out asfreezled.

This failure highlights an infirmity ofpattern-associator network models: Theydon’t have the computational devicecalled a variable, a symbol that can standfor an entire class regardless of its con-

tent, such as the symbol verb. A pattern-associator can only associate bits andpieces with bits and pieces. The modelhas nothing to fall back on if a new itemdoesn’t overlap similar, previouslytrained items, and can only cough up ahair ball of the bits and pieces that areclosest to the ones that it has beentrained on. People, in contrast, reasonthat a verb is a verb, and, no matter howstrange the verb sounds, they can hangan -ed on the end of it.

Yet another circumstance showcasingthe power of a rule arises when an ir-regular form is trapped in memory be-cause of the word’s grammatical struc-ture. Some irregular verbs mysteriouslyshow up in regular garb in certain con-texts. For example, you might say, “Allmy daughter’s friends are lowlifes”; youwouldn’t say, “All my daughter’s friendsare lowlives,” even though the ordinaryirregular plural of life is lives. Manypeople refer to more than one Walkmanas Walkmans, not Walkmen. People say,“Powell ringed the city with artillery,”not rang, and that a politician grand-standed, not grandstood. This immedi-ately shows that sound alone cannotbe the input to the inflection system,because a given input, say life, can comeout the other end of the device either aslifes or as lives, depending on somethingelse.

What is that something else? Manylanguage writers have suggested that itis meaning: A semantic stretching of aword dilutes the associations to itsirregular past-tense form, causing peopleto switch to the regular. But this is sim-ply not true. In example after example, aword’s meaning can change in large orsmall ways, and the irregular form sticksto it like glue. For example, if you prefixa word, its irregular form survives: eat–ate, overeat–overate (not overeated);similarly, we find overshot, overdid,preshrank, and so on. If we form a new

........................................................................................................................................................................

You’ll excuse me if I forego the pleasure of watchingthe video of your wife giving birth.

???—Last night I forewent the pleasure of watchingHerb’s vacation slides.

I don’t know how she can bear that guy.???—I don’t know how she bore that guy.

I dig the Doors, man.???—In the ’60s, your mother and I dug the Doors.

................................................................................................................................................................................

18 Spring 2001 Eye on Psi Chi

noun by compounding, any irregularform comes along with it—workmen (notworkmans), stepchildren, milk teeth,musk oxen. If you use a noun metaphori-cally, that too leaves the irregular un-touched: straw men, snowmen, sawteeth,all God’s children. And English hashundreds of idioms in which a verb takeson a wildly different meaning, but in allcases it keeps its irregular past-tenseform: cut a deal (not cutted), took a leak,caught a cold, hit the fan, blew them off,put them down, came off well, went nuts,and hundreds of others. So it is notenough simply to add a few units formeaning to an associative memory andhope that any stretch of meaning willcut loose an irregular form and therebyexplain why people say lowlifes andgrandstanded.

A better theory, from linguistic theory,says that headless words become regular.Let me explain. The point of rules ofgrammar is to assemble words in such away that one can predict the propertiesof the new combination from the proper-ties of the parts and the way they arearranged. That is true not just when westring words into sentences, but when westring bits of words into complex words.

Start with the verb to eat. We combineit with the prefix over- to form a newword, overeat. The scheme for deducingthe properties of the new word from itsparts is called the right-hand head rule:Take the properties of the rightmostelement and copy them up to apply to thewhole word. What kind of word is over-eat? It’s a verb, because eat is a verb—the verbhood of eat gets copied. Whatdoes overeat mean? It’s a kind of eating,eating too much. That’s because themeaning of eat is to eat, and that getscopied to apply to the whole combina-tion. Finally, what’s the past tense ofovereat? It’s overate, because the pasttense of eat is ate, and that informationgets copied up to the new combination,too.

Another example: Start with the nounman. Combine it with work to produce anew word, workman. What kind of wordis workman? It’s a noun, because man isa noun; the nounhood gets copied. Whatdoes workman mean? It’s a kind of man,a man who does work: The meaning ofman is passed upstairs. And what is theplural of workman? It’s workmen, be-cause the plural of man is men, and thatinformation, too, gets copied.

But there is a family of exceptions:headless words, which don’t get theirfeatures from the rightmost morpheme.In some compound words, for example,the meaning pertains to something that

the rightmost noun has rather than some-thing the rightmost noun is. For example,what is a lowlife? A kind of life? No, it isa kind of person, namely, a person whohas (or leads) a low life. In forming theword, you have to turn off the right-handhead rule—that is, plug the informationpipeline from the root in memory to thewhole word—in order to prevent theword from meaning “a kind of life.” Andwhen the pipeline is plugged, there is nolonger any way for the irregular plural oflife, lives, to percolate up. That informa-tion is sealed in memory, and the regular“add s” rule steps in as the default. Otherexamples include still lifes (not stilllives), which is not a kind of life but akind of painting. We say saber-tooths,not saber-teeth, because the word refersnot to a kind of tooth but to a kind of cat.Flatfoots is American slang for police-men; the plural is not flatfeet because aflatfoot is not a kind of foot. This alsosolves the mystery of Walkmans: AWalkman is not a kind of man.

children frequently make speech errorslike “We holded the baby rabbits” and“The alligator goed kerplunk.” Thewords-and-rules theory offers a simpleexplanation: Children’s memory retrievalis less reliable than adults’. It is based onan uncontroversial fact about the differ-ence between children and adults: Chil-dren haven’t lived as long. (That is whatbeing a child means.) Now, among theexperiences we accumulate through theyears, is hearing the past-tense forms ofirregular verbs. Since children haven’theard held and came and went very often,they have a weak memory trace for thoseforms. Retrieval will be less reliable, andas long as the child has acquired theregular rule, he or she will fill thevacuum by applying the rule, resulting inan error like comed or holded.

Evidence? First, we can show thatweak memory is a factor: Gary Marcusand I found that the more often a child’sparent uses an irregular in casual speech,the less often the child makes an erroron it. Second, the theory explains whychildren, for many months, produce noerrors with these forms—at first they sayheld and came and went, never holdedand comed and goed. Why does a childwake up one morning and start to sayholded? Perhaps because that is the pointat which the child has just acquired the“-ed” rule. How can we tell? By lookingat what children do with regular verbs.Very young children say things like“yesterday we walk,” leaving out the pasttense altogether. Then they pass from astage of leaving out the -ed more oftenthan supplying it to a stage of supplyingit more often than leaving it out. And thattransition is precisely at the point inwhich the first error like holded occurs.This is exactly what we would expect ifthe child has just figured out that thepast-tense rule in English is “add -ed.”Before that, if the child failed to come upwith an irregular form, he had no choicebut to use it in the infinitive: “Yesterday,he bring . . .”; once he has the rule, hecan now fill the gap by overapplying theregular rule, resulting in bringed.

The final kind of evidence comesfrom cases in which the memory systemis directly compromised by neurologicaldamage or disease. Michael Ullman andI have asked a variety of neurologicalpatients to fill in the blank in items suchas “Everyday I like to [verb]; yesterday,I . . . .” We tested patients with anomia,an impairment in word finding, typicallyassociated with damage to the posteriorperisylvian region of the left hemisphereof the brain (see Figure 1), which leavesthe patient in a constant tip-of-the-tongue

“Since children haven’t heardheld and came and went very

often, they have a weak memorytrace for those forms. Retrieval

will be less reliable, and as long asthe child has acquired the regularrule, he or she will fill the vacuumby applying the rule, resulting inan error like comed or holded.”

Another example showing off thismental machinery comes from verbs thatare based on nouns. We say that the artil-lery “ringed the city,” not rang, becausethe verb comes from a noun: To ring inthis sense means “to form a ring around.”To get a noun to turn into a verb, theusual percolation pipeline has to beblocked, because ordinarily the pipelineallows part-of-speech information to becopied from the root to the newly formedword. And that blocked pipeline preventsany irregularity associated with thesound of the verb from applying to thenewly formed word. For similar reasons,we say that a politician grandstanded,not grandstood, because the verb comesfrom the noun grandstand, as in “play tothe grandstand.”

Let’s switch to a very different kind ofcircumstance in which memorized formsare not accessed but regular inflection isapplied: childhood. As I mentioned,

..............................................................................................................................................................................................................................................

..............................................................................................................................................................................................................................................

Eye on Psi Chi Spring 2001 19

state. The patient can’t get words intospeech quickly enough, and resorts togeneric fillers like stuff, thing, guy, andthis and that. Patients with anomia canoften produce fluent and mostly gram-matical speech, suggesting that theirmental dictionaries are more impairedthan their mental grammars. With suchpatients, we found that irregular verbsare harder than regulars, which fits thetheory that irregulars depend on memorywhereas regulars depend on grammar.We also predicted and observed regular-ization errors like swimmed, which occurfor the same reason that children (whoalso have weaker memory traces) pro-duce such errors: They cannot retrieveswam from memory in time. And thepatients are relatively unimpaired indoing a wug test (Today I wug, yesterdayI wugged), because that depends ongrammar, which is relatively intact.

By symmetrical logic, brain-injuredpatients with agrammatism—a deficit instringing words together into grammati-cal sequences, typically caused by dam-age to anterior perisylvian regions of theleft hemisphere (see Figure 1)—shouldshow the opposite pattern. They shouldhave more trouble with regulars, whichdepend on grammatical combination,than with irregulars, which depend onmemory. They should produce few errorslike swimmed, and they should havetrouble doing the wug test. And that isexactly what happens.

TO SUM UP, DESPITE THE IDENTICALfunction of regular and irregular inflec-tion, irregular are avoided, but the regu-lar suffix is applied freely in a variety ofcircumstances, from chided to ploamfedto lowlifes to anomia—that have nothingin common except a failure of access to

information in memory. These circum-stances are heterogeneous and exotic;obviously we don’t have separate neuralmechanisms that generate regular formsin each of these cases. Rather, they fallout of the simple theory that the rulesteps in whenever memory fails, regard-less of the reason that memory fails.

These seemingly disconnectedphenomena thus suggest that regularinflection is computed by a mentaloperation that does not need access to thecontents of memory, namely a symbol-combination rule. I believe that thisreveals distinct mental mechanismsthat implement the two principles oflanguage responsible for its breathtakingexpressive power: memory, for the arbi-trary sign that underlies words, and sym-bolic computation, for the infinite use offinite media underlying grammar.N

Selected BibliographyBerent, I., Pinker, S., & Shimron, J. (1999). De-

fault nominal inflection in Hebrew: Evidencefor mental variables. Cognition, 72, 127–138.

Clahsen, H., Marcus, G. F., Bartke, S., & Wiese, R.(1996). Compounding and inflection in Germanchild language. In G. Booij and J. van Marle(Eds.), Yearbook of Morphology 1995 (pp. 115–142). Dordrecht, The Netherlands: Kluwer.

Clahsen, H., Rothweiler, M., Woest, A., & Marcus,G. F. (1992). Regular and irregular inflection inthe acquisition of German noun plurals. Cogni-tion, 45, 225–255.

Kim, J. J., Marcus, G. F., Pinker, S., Hollander, M.,& Coppola, M. (1994). Sensitivity of children’sinflection to grammatical structure. Journal ofChild Language, 21, 173–209.

Kim, J. J., Pinker, S., Prince, A., & Prasada, S.(1991). Why no mere mortal has ever flown outto center field. Cognitive Science, 15, 173–218.

Marcus, G. F. (1995). Children’s overregulariza-tion of English plurals: A quantitative analysis.Journal of Child Language, 22, 447–459.

Marcus, G. F. (1995). The acquisition of theEnglish past tense in children and multilayeredconnectionist networks. Cognition, 56, 271–279.

Area damagedin patient withagrammatism

Area damagedin patient withanomia

FIGURE 1

Location of damaged areas of the brain related to anomia and agrammatism.

Marcus, G. F. (1996). Why do children say“breaked”? Current Directions in PsychologicalScience, 5, 81–85.

Marcus, G. F. (2000). Children’s overregularizationand its implications for cognition. In P. Broeder& J. Murre (Eds), Models of language acquisi-tion: Inductive and deductive approaches (pp.154–176). Oxford: Oxford University Press.

Marcus, G. F., Brinkmann, U., Clahsen, H., Wiese,R., & Pinker, S. (1995). German inflection: Theexception that proves the rule. Cognitive Psy-chology, 29, 189–256.

Pinker, S. (1991). Rules of language. Science, 253,530–535.

Pinker, S. (1993). The central problem for thepsycholinguist. In G. Harman (Ed.), Concep-tions of the human mind: Essays in honor ofGeorge A. Miller (pp. 59–84). Hillsdale, NJ:Erlbaum.

Pinker, S. (1994). The language instinct. NewYork: Morrow.

Pinker, S. (1995). The irregular verbs. In L. B.Frumkes, The logophile’s orgy (pp. 148–152).New York: Delacorte.

Pinker, S. (1995). Why the child holded the babyrabbits: A case study in language acquisition. InL. R. Gleitman & M. Liberman (Eds.), Aninvitation to cognitive science: Vol. 1. Language(2nd ed., pp. 107–133). Cambridge, MA: MITPress.

Pinker, S. (1997). Words and rules in the humanbrain. Nature, 387, 547–548.

Pinker, S., & Prince, A. (1988). On language andconnectionism: Analysis of a parallel distributedprocessing model of language acquisition.Cognition, 28, 73–193.

Pinker, S., & Prince, A. (1994). Regular andirregular morphology and the psychologicalstatus of rules of grammar. In S. D. Lima, R. L.Corrigan, & G. K. Iverson (Eds.), The reality oflinguistic rules (pp. 321–351). Philadelphia:Benjamins.

Pinker, S., & Prince, A. (1996). The nature ofhuman concepts: Evidence from an unusualsource. Communication and Cognition, 29,307–362. Reprinted (1999) in P. Van Loocke(Ed.), The nature of concepts: Evolution, struc-ture and representation (pp. 8–51). London:Routledge.

Prasada, S., & Pinker, S. (1993). Generalization ofregular and irregular morphological patterns.Language and Cognitive Processes, 8, 1–56.

Prince, A., & Pinker, S. (1988). Rules and connec-tions in human language. Trends in Neuro-sciences, 11, 195–202. Reprinted (1989) in R.G. M. Morris (Ed.), Parallel distributed pro-cessing: Implications for psychology and neuro-biology (pp. 182–199). New York: OxfordUniversity Press.

Prince, A., & Pinker, S. (1988). Wickelphoneambiguity. Cognition, 30, 189–190.

Ullman, M. (in press). Evidence that lexicalmemory is part of the temporal lobe declarativememory, and that grammatical rules are pro-cessed by the frontal/basal-ganglia proceduralsystem. Brain and Language.

Ullman, M., Corkin, S., Coppola, M., Hickok, G.,Growdon, J. H., Koroshetz, W. J., & Pinker, S.(1997). A neural dissociation within language:Evidence that the mental dictionary is part ofdeclarative memory, and that grammatical rulesare processed by the procedural system. Journalof Cognitive Neuroscience, 9, 266–276.

Ullman, M., & Gopnik, M. (1999). Inflectionalmorphology in a family with inherited specificlanguage impairment. Applied Psycholinguis-tics, 20, 51–117.

Xu, F., & Pinker, S. (1995). Weird past tenseforms. Journal of Child Language, 22, 531–556.

..............................................................................................................................................................................................................................................

.................................................................................................................................................................