Words and Rules: The Ingredients of Language

Pinker argues that language comprises two fundamentally distinct mental mechanisms: a mental dictionary of memorized words and a mental grammar of creative rules . Irregular past-tense forms like "drank" or "sang" are stored as whole lexical items and retrieved from associative memory, while regular forms like "walked" are computed online by the combinatorial rule that adds -ed to a verb stem . The book rejects the connectionist claim, advanced by Rumelhart and McClelland in 1986, that a single associative network can account for both regular and irregular inflection without appealing to rules . Pinker contends that the past tense is a uniquely rich test case for two ancient competing views of mind: rationalism's combinatorial grammar versus empiricism's associative memory .

Pinker argues the past tense is "the only case I know in which two great systems of Western thought (rationalism and empiricism) may be tested and compared on a single rich set of data" . The regular-irregular distinction provides a natural experiment: irregulars are a closed, memorized set of roughly 165 verbs, while regulars are an open-ended set generated by a productive rule . Connectionist models trained on verb lists appeared to reproduce patterns of child language acquisition and neurological impairment without positing separate rule and memory systems . Pinker's laboratory spent over a dozen years testing predictions that would distinguish the hybrid model from pure connectionism, including frequency effects, novel verb generalization, and dissociations in brain-damaged patients .

Pinker cites converging evidence from multiple methodologies. Studies by Ullman and by Prasada and Snyder found that high-frequency irregular verbs are retrieved faster than low-frequency irregulars—a frequency effect predicted by memory storage—while regular verbs show no such frequency effect, consistent with rule-based computation . Neuropsychological dissociations show that patients with Alzheimer's disease, Parkinson's disease, and Huntington's disease exhibit selective impairments in one system but not the other . Brain imaging studies associate irregular retrieval with temporal-lobe networks implicated in memory, while regular computation is associated with frontal-lobe networks . Pinker's team also demonstrated that speakers intuitively generalize novel verbs to irregular patterns based on family resemblance—e.g., "spling/splang/splung" by analogy to "spring/sprang/sprung" .

Chomsky and Halle's 1968 analysis in The Sound Pattern of English showed that the entire repertoire of English irregular verbs could be accounted for by merely three phonological rules, implying that memory plays a minimal role in language . Pinker rejects this as implausible, arguing that irregular forms are stored as whole words rather than generated by rules. His hybrid model gives memory a far greater role than the generative phonology tradition allowed. However, Pinker's account also differs from pure associationism: regular forms are computed by a combinatorial grammar, not retrieved from memory. The debate between Pinker and the Chomskyan tradition concerns whether irregularity reflects deep abstract rules or stored lexical items, while the debate with connectionism concerns whether rules exist at all .

Connectionists, including Rumelhart and McClelland, argue that a single pattern-associator network can learn both regular and irregular past-tense forms from exposure to verb lists, without positing separate rule and memory systems . Pinker acknowledges the model's successes but identifies three flaws: the pattern associator can only produce forms, not recognize verbs as verbs; it depends too heavily on phonology and cannot distinguish homonyms; and it required manipulation to replicate children's actual errors . A UCLA review noted that Pinker's own criticisms of Chomsky and Halle's assumption that "memory is compressed to a minimum" were not fully convincing, since the frequency effects he cited were small and the generative model could also accommodate them . The debate remains unresolved, with each side claiming the empirical record supports its architecture.

Words and Rules shares the wit and accessible style of The Language Instinct (1994) but explores language in a fundamentally different way. Where The Language Instinct argued broadly for an innate language faculty, Words and Rules zooms in on a single phenomenon—the regular-irregular verb distinction—and examines it from every angle: psychology, biology, history, philosophy, linguistics, and child development . A Nature review noted that for a broad audience, The Language Instinct was Pinker's more amusing book, while Words and Rules is "an academic work" that enables the reader to reason through each step of the argument . The two books are complementary: one makes the case for language as an instinct, the other provides the detailed psycholinguistic evidence from a specific domain.