In the latest issue:

The Politics of Like and Dislike

William Davies

The Shrine

Alan Bennett

After the Shock

Adam Tooze

Punishment by Pressing

Hazel V. Carby

The Suitcase

Frances Stonor Saunders

Short Cuts: Thanington Without

Patrick Cockburn

The Lessons of Reconstruction

Randall Kennedy


Linda Colley

Eva Hesse

Anne Wagner

Parachuted into France

Neal Ascherson

The Age of Sail

N.A.M. Rodger

Poem: ‘Near Gleann nam Fiadh’

Robin Robertson

‘You People’

Clare Bucknell

What Didn’t Happen

Michael Wood

Forster in Cambridge

Richard Shone

Diary: In Ashgabat

James Lomax

Dig-dug, think-thunkCharles Yang
Vol. 22 No. 16 · 24 August 2000

Dig-dug, think-thunk

Charles Yang

2040 words
Words and Rules: the Ingredients of Language 
by Steven Pinker.
Phoenix, 176 pp., £7.99, October 2000, 0 7538 1025 5
Show More
Show More

The mastery of the English past tense isn’t a glamorous topic, not these days. That, however, is what Words and Rules is about and, as if to sweep any doubts away, Steven Pinker assures us of the subject’s importance: the study of the past tense is ‘the only case I know in which two great systems of Western thought may be tested and compared . . . like ordinary scientific hypotheses’. As for his own theory of the tense, it is ‘an opening statement in the latest round of a debate on how the mind works that has raged for centuries’ – this book never runs low on hubris or hyperbole.

The ‘two great systems’ are the rationalist and the empiricist. On the rationalist side, so far as linguistics is concerned, we have Chomsky, who believes that much of what we are able to do with language cannot be derived from experience, but must be innate, hidden somehow in our genes. In the enterprise known as generative linguistics, he and his colleagues have studied the elaborate system of rules and structures that make an ‘infinite use of finite means’ – Wilhelm von Humboldt’s description of the unbounded combinatorial power of language. According to Pinker, the dominance of rationalism in linguistics is now being challenged by the model of ‘connectionist networks’, the modern embodiment of empiricism. This model understands the mind as a massive collection of elementary, but intricately connected computing devices called artificial neurons, by analogy with the brain, a massive collection of real neurons. Our linguistic capabilities are acquired from what the network is exposed to: when a furry feline is seen at the same time as the sound ‘cat’ is heard, a link between the two is established. It is the intrusion into Chomsky’s generative linguistics of this new breed of empiricists that brings irregular English verbs into the limelight.

In generative linguistics, the past tense of verbs has traditionally been accounted for by means of rules. Irregular verbs are grouped into classes, where special rules apply to modify the sound of the stem: for example, the past tense of deal, keep and lose results from adding a -t suffix and changing the quality of the vowel, giving usdel-t, kep-t and los-t. Other irregular verbs – sing-sang, sit-sat – change the stem vowel, but add no suffix; still others – hit, put – neither change the stem vowel nor add a suffix. Learning an irregular verb, therefore, involves learning what class it belongs to. If a verb doesn’t form its past tense by any of these means, it is picked up by the default rule that attaches an -ed suffix to the stem. The vast majority of English verbs are regular, on the lines of walk-walked.

The connectionist approach is different. In 1986, the psychologists James McClelland and David Rumelhart proposed a network that learns verbs by building associative links between stems and their past tense forms. The network is ‘trained’ by repeated exposure to such pairs, without any appeal to rules or innate knowledge. Pinker published a critique of this model at the time and has remained a visible opponent of connectionism. He emphasises that humans and networks perform differently. Both adults and children generally treat nonce verbs as though they are regular: given smeej and frilge, they supply smeejed and frilged. In contrast, a connectionist network comes up (hopelessly) with leefloag and freezled.

Pinker is attempting to adopt a middle way between rationalism and connection-ism. He follows generative linguistics in assigning regular verbs to the default -ed rule, and then, embracing connectionism, he treats irregular verbs and their past tense as pairs memorised by rote. Finally, he adds a twist of his own, by breaking both the stem and the past tense into sequences of consonants and vowels, and pairing them by means of links between these bits and pieces.

Take the verb form fly out, as used when a baseball player hits the ball high into the air and it’s caught. The past tense isn’t flew out, as non-American English users might suppose, but flied out, as every baseball fan knows. This, Pinker explains, has to do with a simple notion of heads in linguistic structure. Not everything is equal when stems and suffixes are built up into words, and the words into phrases: one particular part, the head, always determines the character of the whole. Thus the compound verb overshoot is headed by shoot and therefore inherits its irregular status, making the past tense overshot. To fly out derives from fly, the noun, meaning a high catchable hit, which in turn is derived from fly, the verb. In this verb-to-noun-back-to-verb detour, the irregular tag of the first verb is blocked by the interfering noun, and the final verb is then dealt with by the default -ed rule.

Or take another example of what Pinker calls the ‘darnedest things’ in child English: the teacher holded the rabbit. It’s clear that children didn’t learn holded from their mothers, who would never say such a thing. Pinker points out that children must know how to extract the -ed suffix from words, must know it signifies a past event and must know that the default rule can in principle apply to all words, including the irregular ones they are still trying to master. We can deduce this despite not knowing exactly how in neurological terms the brain goes about synthesising these strands of information. In language use, the behavioural level is still the one most likely to reveal the secrets of the mind.

Pinker insists that his compromise theory is not ‘a sappy attempt to get everyone to make nice and play together’. The end result, however, may well not satisfy either side. The middle ground is the right place to be only if the two extremes are demonstrably wrong; otherwise it begins to look like no man’s land. Consider his main complaint about linguists’ rules for irregular verbs: that since an irregular verb forms its past tense by fiat, there is no explanation of why verbs like sting, string, sling,stink, sink, swing, and spring all change i to u in the past participle and all sound so similar. Pinker’s explanation is based on Wittgenstein’s notion of a ‘family resemblance’, the sort of fuzzy association connectionist networks excel at. In Pinker’s representation of verbs as sound segments, the bits they share become like a gravitational attraction for word families, some of them central, such as string-strung and sling-slung, others on the fringe, such as dig-dug and win-won. But this reasoning seems circular: why do these particular verbs gravitate into families? So far as one can tell, because they sound similar. Their stem similarity is only partial, moreover: the i-u family doesn’t include think or blink, both of which might seem closer to the family ‘centre’ than dig or win. Nowhere does Pinker tell us how his fuzzy family resemblance works to prevent the formation of thunk and blunk. More important, the ultimate explanation for the stem similarities Pinker identifies seems to be historical. Irregular verbs were both more abundant and more regular in Old English. There was once a rule, for example, that produced ew whenever ow occurred: mow, sow and row behaved liked know-knew, grow-grew and blow-blew in modern English. Since verbs are presumed to be regular unless proven otherwise, when our ancestors saw a rare mew, sew and rew, they assumed these to be regular. What we are left with today is only a partial similarity among irregular verbs, vestiges of once systematic rules. To demand an explanation that goes beyond the historical contingencies misses the point.

The goal of linguistics, as of all sciences, is successful generalisation. But if it’s already hard to write a grammar for English, it’s far harder to write a grammar that captures the similarities that all languages share but which still leaves room for their differences. Once we look beyond modern English, notorious for its verbal simplicity, the problems with Pinker’s verbs become apparent. To take one of his own examples (from a chapter entitled ‘The Horrors of the German Language’), noun plurals in German employ five suffixes: Kind-er (‘children’), Wind-e (‘winds’), Ochs-en (‘oxen’), Daumen-ø (‘thumbs’ – ø indicates an ‘empty’ suffix, as in the English plural sheep-ø and past tense hit-ø), and Auto-s (‘cars’). Pinker argues convincingly that, despite its low frequency, the -s is the default suffix. However, it’s hard to believe that German speakers memorise all four classes of irregular plural, i.e. the majority of nouns in the language, on a word-by-word basis, as if each were entirely different from the others. The partial similarity among English irregular verbs, which looks like nothing more than a historical accident, has misled Pinker into looking for family resemblances: a quick glance at German shows that the four irregular classes of plural show no systematic similarity whatever. The horrors of German are real: one must sort each irregular noun into its proper class, as in the traditional rule-based view.

Pinker’s theory fares even worse when we turn to African, Pacific, American and even Indo-European languages, which typically have very long ‘words’ built out of prefixes and suffixes, each of them expressing an individual meaning and all of them glued together by the grammar and sound system of the language: ‘he will kiss her’ may literally be a concatenation of ‘kiss’, followed by suffixes for future-tense, third person, singular, male, nominative, and third-person, singular, female, accusative. Here, too, there are default forms: not only past tense for verbs, but also present tense and future, and not only plurals for nouns, but also gender and number. It is inconceivable that such ‘words’ are all individually memorised, for there are tens of billions of such combinations.

Despite the book’s ambitious subtitle, almost all the ingredients of language are missing from it. The main reason linguists have failed to get worked up about the connectionists’ incursion is, as Pinker says, that the English irregular verb is ‘famous among linguists for being so boring’. Indeed, these verbs constitute only a trivial part of language and its underlying cognitive processes. Children make plenty of interesting mistakes before they start telling us that ‘the teacher holded the rabbit.’ Why do they say ‘me going’ instead of ‘I am going’? Why do they selectively drop some parts of words but not others (mato instead of tomato, but nevertoma)? How do they figure out puppy is a noun and waggle is a verb, but that bark is both? Or, for that matter, how do they pick out Mummy’s voice against the noise from the radio in the background? As children grow up, they present linguists with a host of new challenges: how do we understand sentences on the fly, what tricks does the mind play on us when we make a slip of the tongue, what happens to our language when we get old? Linguistics must, in principle, study all of these questions in every one of the planet’s several thousand languages.

My final grumble arises from Pinker’s insistence that his irregular verbs provide the ‘only case’ by which to assess the state of the empiricist-rationalist divide. Let’s not forget, however, that that disagreement was revived in recent times by Chomsky’s demonstration of innate linguistic knowledge. All Pinker (and the connectionists) are doing is turning over the rocks at the base of the intellectual landslide caused by the Chomskian revolution. The testing-ground for the debate extends not only far beyond irregular verbs but even beyond language. It is now known that we are born with considerable knowledge of the world around us: very young infants can reason about motion and spatial relationships among objects. We have learned a great deal about the unique abilities of individual species, discoveries which point to innate traits and instincts, endowed through millions of years of evolution. At the same time, we have learned that human infants and other higher primates can organise patterns, linguistic and otherwise, in apparently inductive, hence learned and empirical, ways. The debate is still on and hyping up irregular verbs à la Pinker does no justice to the scientific efforts of the past fifty years.

Send Letters To:

The Editor
London Review of Books,
28 Little Russell Street
London, WC1A 2HN

Please include name, address, and a telephone number.

Read anywhere with the London Review of Books app, available now from the App Store for Apple devices, Google Play for Android devices and Amazon for your Kindle Fire.

Read More

Sign up to our newsletter

For highlights from the latest issue, our archive and the blog, as well as news, events and exclusive promotions.

Newsletter Preferences