Words and Rules: The Ingredients of Language (review)

Words and rules (W&R) can be read on several levels. On one level, it is a popularization of research in morphology and psycholinguistics. On another level, it is a compilation of arguments on one side of a long-standing debate over the viability of connectionist models of language acquisition. And on yet another level, it is an essay on the nature of cognition and the structure of the human mind. At all these levels, its contributions deserve serious attention.

The fundamental thesis of W&R is stated succinctly on page 278: ‘Regular and irregular forms coexist but require different computational mechanisms: symbol combination for regular forms, associative memory for irregular forms. The same may be true for classical and family resemblance categories’. In support of this claim, Pinker draws on an impressive variety of types of evidence, including the distribution of forms in inflectional paradigms, historical change, corpus studies, child language, various psycholinguistic experimental paradigms, and neurophysiological data. No single argument is knock-down, but their combined weight is imposing. Nevertheless, I am not persuaded that P’s dichotomy is as sharp as he makes it out to be.

The language instinct (TLI) established P as the preeminent interpreter of our discipline for general audiences. Many of the features that made that book so successful are in evidence in W&R. The most obvious is the use of amusing examples from popular culture. More important, however, is P’s facility at presenting ideas from scholarly works in clear, nontechnical ways, accessible to intelligent lay people. Like his previous popularizations, W&R is fun and easy to read.

By the same token, those aspects of TLI that rubbed some people the wrong way recur in W&R. It advocates a position, rather than attempting to give a balanced overview. Consequently, a lay audience is likely to take as established fact some claims that are actually quite controversial. Moreover, the literature citations are disproportionately to work by P and his students, a fact that may lead naїve readers to believe that P is the dominant figure in the field.¹

On the other hand, W&R attempts something more difficult than TLI. Instead of surveying a variety of topics loosely tied together by the idea of a ‘language instinct’, this book ‘tries to illuminate the nature of language and mind by choosing a single phenomenon and examining it from every angle imaginable’ (ix). Whether a general audience will grow weary of that phenomenon (inflectional morphology, primarily the English past tense) remains to be seen.

W&R’s central claims are the following:

1. 1. Productive regularities like the English past tense -ed suffix are fundamentally different from subregularities like ring/rang, sink/sank, etc.

2. 2. Productivity is not always correlated with frequency.

3. 3. Therefore, purely inductive methods (like connectionist networks) are incapable of acquiring productive rules with idiosyncratic exceptions.

4. 4. However, these inductive methods provide a good model for the acquisition of nonproductive subregularities.

5. 5. Human conceptual structure may exhibit a parallel dichotomy between rule-based categories (which facilitate general reasoning) and a messy collection of family resemblances based on experience.

P presents numerous and diverse arguments in support of claim 1. They include evidence like the following:

• • Newly-coined words invariably take the regular inflection.

• • Speakers apply the regular inflection to unfamiliar words in wug-type tests.

• • Children acquiring a language overgeneralize, giving irregular words regular inflections.

• • Rarely used words with irregular inflections tend to adopt the regular inflections over time.

• • Studies of aphasics, as well as various neuroimaging techniques, suggest that regular and irregular inflections may be processed in different regions of the brain.

The principal argument for claim 2 concerns German noun plurals. P claims that the plural used on novel nouns is the suffix -s although the vast majority of German nouns are pluralized in other ways, showing that regularity must not be conflated with frequency. According to P, precisely this conflation is built into connectionism. His case for claim 2 is...