Key Takeaways
Language runs on memorized words and productive rules
Two different tools. A word links a sound with a meaning by convention: nothing about the sound rose naturally resembles a flower. Speakers memorize such links in a mental lexicon. Rules do a different job, combining words and word parts into forms whose meanings can be worked out from their pieces and arrangement. The past tense makes the contrast visible: walk becomes walked by a regular pattern, while sing becomes sang through a stored, irregular form.
A useful division. The rule can supply a form for a new verb such as fax, while memory preserves special cases such as went. Pinker’s central claim is that language needs both systems. Neither is an inferior version of the other: memory stores particular knowledge, and rules generate combinations beyond what has been memorized.
The distinction is powerful because it explains how language can be both conventional and creative. A dictionary alone cannot account for a speaker producing a sentence never heard before; a rulebook alone cannot explain why go becomes went. The model also resists a familiar false choice between learning and innate structure: speakers learn words and patterns, while the architecture of the mind determines how those inputs can be combined. Its main caution is that words and rules are explanatory categories, not necessarily two neatly isolated brain compartments.
A finite grammar can generate thoughts nobody has said before
Combinations multiply. Grammar applies to categories such as noun and verb, not just to memorized strings. That lets speakers substitute new words into familiar structures and build novel sentences. Recursion, in which a sentence can contain another sentence, lets the process repeat without a fixed upper limit. A finite inventory of words and rules can therefore support an effectively unbounded range of expressions.
Why not encode everything? John Wilkins tried to build a language whose words described objects through a logical taxonomy. His elaborate system made users mentally calculate a word’s meaning, and it struggled with unfamiliar concepts, pronunciation, and small errors in speaking. Ordinary languages instead combine productive grammar with short, arbitrary words. The balance is more useful than making every word a miniature definition.
The contrast between grammar and Wilkins’s designed vocabulary makes a general engineering point: a system that compresses everything into one elegant code can become costly to use. Human language trades perfect transparency for quick retrieval, then relies on grammar to recombine what memory supplies. The result resembles a toolkit with both stored components and reusable procedures. The claim of boundless expression describes the formal capacity of recursion, not the practical ability to produce or understand infinitely long sentences, which remains constrained by attention, memory, and time.
Treat children's errors as evidence of rule discovery, not carelessness
The wug test. Psychologist Jean Berko Gleason showed children a made-up creature and asked them to name two of it. About three-quarters of preschoolers supplied the regular plural, and 99 percent of first-graders did. Because the children could not have memorized that invented word’s plural, they had applied a pattern. Children likewise produce forms such as breaked and goed, showing that they actively build language rather than merely repeat adult speech.
Why errors fade. Children often use correct irregular forms before overgeneralizing, then gradually improve, a pattern called U-shaped development. Pinker argues that an emerging rule supplies a past tense when the child cannot retrieve the irregular from memory. The irregular form blocks the rule when it is recalled. Parents rarely correct these errors consistently, and a child unable to speak still judged many forms correctly, suggesting correction is not the mechanism that teaches the distinction.
A child’s error can reveal a hidden mental operation more clearly than a correct answer. Breaked exposes the rule that ordinary adult speech conceals behind the memorized form broke. The account also makes learning compatible with limited feedback: children can infer structure from patterns in speech without receiving a correction for every mistake. This does not mean every developmental error proves an innate rule. The evidence is strongest when spontaneous speech, invented-word experiments, and judgments of correctness converge, rather than when a single amusing utterance is treated as decisive.
Irregular verbs are historical fossils, not random exceptions
Rules leave traces. Many irregular forms began as predictable products of older rules. English feet arose when a plural suffix changed the vowel in foot; after the suffix disappeared, the altered vowel remained. Other irregular verbs preserve fragments of ancient vowel-changing patterns, as in sing, sang, sung. Over centuries, pronunciation shifts, dialect contact, and changing word frequency blur the original system, leaving families of similar but imperfectly matched forms.
History keeps editing. The metaphor of Broken Telephone captures how each generation infers a language from what it hears, sometimes reanalyzing or reshaping it. Pinker counts about 164 irregular verbs in modern English, compared with 325 strong verbs in Old English. Common verbs are more likely to retain irregular forms because speakers hear and remember them often; less familiar forms tend to regularize.
The fossil metaphor prevents two mistakes: treating irregularities as meaningless noise, and assuming every surviving pattern is an active rule in a modern speaker’s mind. Historical origin and present-day mental representation are different questions. A form can preserve the outline of an old system even after speakers no longer know that system. This is a useful parallel to biological vestiges, where inherited structure records past processes without necessarily serving its original function. The comparison should remain an analogy, since linguistic change is transmitted through learning and social interaction, not biological inheritance.
Let memory handle familiar exceptions and rules cover the unknown
Two theories compete. One approach makes rules responsible for regular and irregular forms alike. Another tries to explain both through a network of learned associations among similar sounds. Each captures part of the evidence. A network can explain why similar irregular verbs, such as sing and ring, encourage similar patterns. But it struggles to produce regular forms for unfamiliar sound combinations. A pure rule account, meanwhile, can describe patterns among irregulars but may require implausibly elaborate representations.
The hybrid explanation. Pinker’s account assigns regular forms to a rule and irregular forms to memory, while allowing memory to connect similar words. The rule can attach -ed to an unfamiliar verb without relying on a similar example. Associative memory helps speakers learn and sometimes extend irregular families. This division explains both the productivity of regular forms and the fuzzy, similarity-based behavior of irregular ones.
The hybrid proposal is more than a compromise: it assigns different problems to mechanisms suited to solve them. Symbolic rules support generalization to unfamiliar cases, while associative memory captures graded similarities and the influence of repetition. That division is familiar in other accounts of cognition, where rapid pattern matching coexists with explicit procedures. A scientific caution remains: showing that a hybrid model fits the evidence does not establish its exact neural implementation. The theoretical distinction is clearest at the level of behavior, where generalization and retrieval leave different signatures.
Word structure, not stretched meaning, determines which irregulars survive
Roots carry irregulars. A canonical root is a basic word stored with a sound, meaning, and grammatical category. Irregular forms are linked to such roots. A complex word can inherit properties from its head, the element that determines what kind of thing the whole word refers to. Thus workman is a kind of man and takes the plural workmen.
Structure can block inheritance. A lowlife is a person with a low life, not a kind of life, so the compound cannot inherit the plural lives and instead becomes lowlifes. Baseball’s fly out is formed through a noun meaning a high hit, so the past is flied out rather than flew out. By contrast, metaphor alone does not erase irregularity: people still say threw up and took a leak. What matters is how the word is built, not simply how far its meaning has shifted.
These examples show why meaning cannot be the only guide to grammar. Similar meanings can have different forms, while an irregular form can persist through an opaque metaphor. Word structure offers a more precise explanation by tracking which part supplies the whole word’s grammatical identity. It also explains why speakers can make a judgment without knowing a word’s etymology: they respond to how they currently analyze it. The account is not a license to explain every odd form after the fact; its value depends on predictions that can be tested with new words and speaker judgments.
A default rule can dominate even when it is a minority
Frequency is not the definition. Textbooks often call the most common pattern regular. Pinker uses a psychological definition instead: a regular pattern is a default that applies when memory supplies no special form. German weak verbs take the suffix -t, and speakers use it for invented, rare, unusual, borrowed, and structurally unusual verbs, much as English speakers use -ed.
The minority case. Among the thousand most common German verbs, only 45 percent are regular, compared with 86 percent in English. Yet German -t behaves as a default. German noun plural -s is an even clearer case: it applies to a small minority of nouns, but speakers use it for names, foreign borrowings, and invented words. The book argues that the rule’s power comes from its mental operation, not from a majority of examples.
This distinction separates a rule’s psychological role from its statistical prevalence. That matters because exposure counts can be misleading: a few highly frequent irregulars may dominate conversation even when regular forms cover many more dictionary entries. The German evidence is especially useful because it breaks the usual correlation between being common and acting as a default. Still, the conclusion depends on how researchers count words and define a pattern. Pinker notes that the regularity gap narrows under different counting choices, so the strongest evidence comes from speakers’ behavior with novel and structurally unusual forms.
Brain damage separates word retrieval from grammatical rule use
Opposite impairments. People with agrammatism often struggle to apply grammatical endings but can retain many memorized forms. People with anomia have difficulty retrieving words yet can produce regular forms and sometimes overapply the rule, saying forms such as digged. Alzheimer’s patients in the studies showed a similar tendency, while Parkinson’s patients showed greater trouble with grammatical processing. These contrasting patterns suggest that stored word knowledge and rule-based processing rely on partly different systems.
Evidence has limits. Electrical brain responses also differed: irregular-form errors produced a response associated with an unexpected word, while regular-form errors produced one associated with a grammatical violation. Brain-imaging studies found differences between regular and irregular processing, but disagreed about the precise locations. Pinker’s claim is therefore strongest at the level of distinct processes, not a simple map with one word center and one rule center.
The double dissociations matter because a single deficit could reflect general difficulty, while opposite patterns suggest different underlying operations. The comparison also connects language to broader distinctions between remembering facts and carrying out learned procedures. Yet brain evidence requires restraint: a task such as producing a past tense also involves attention, speech planning, and memory, so activation alone does not identify a grammar module. The book handles this tension well by treating neural findings as converging clues rather than a neat anatomical verdict.
Keep fuzzy family resemblance separate from rule-defined categories
Two kinds of categories. Irregular verb families resemble everyday categories such as games or birds: their members share overlapping traits, but no single feature cleanly defines the group. Some verbs are central examples, while others sit at the edge. Regular verbs work differently. A verb qualifies for the regular pattern by being a verb, and the rule can apply regardless of its sound or familiarity.
Different jobs in thought. Pinker links these patterns to two kinds of reasoning. Associative memory supports judgments based on similarity among particular cases. Rules support crisp, reusable deductions, such as those in arithmetic or kinship. Both are useful: similarity helps people navigate a varied world, while formal rules let them reason beyond familiar examples. The analogy suggests that language offers a small-scale view of a broader, mixed system of thought.
The closing analogy is valuable because it avoids forcing all cognition into either fuzzy association or formal logic. A person can recognize a typical bird by resemblance and also know that a dolphin is not a fish by using a taxonomic rule. Likewise, speakers can generalize irregular patterns by similarity while applying regular morphology through a category rule. The parallel is suggestive, not proof that linguistic and conceptual categories share one mechanism. Its strongest contribution is to make room for both modes of thought without treating either as a defect.
Analysis
Words and Rules makes a compact linguistic puzzle carry an unusually large theoretical burden. Its central move is to compare regular and irregular forms that express the same grammatical meaning, then ask why speakers produce them differently. The contrast yields a layered account: memorized forms connect to particular lexical entries, similar irregulars reinforce one another through associative memory, and productive rules operate over grammatical categories. That model explains why children invent forms such as breaked, why rare irregulars erode, and why a German default can remain productive even when it is statistically uncommon.
The book’s strongest methodological lesson is that a theory should explain behavior across settings, not merely summarize a list. The wug test, speech errors, reaction times, word judgments, language disorders, and cross-language comparisons each probe different consequences of the same proposed architecture. The argument is especially persuasive when independent measures converge, as in children’s novel-word generalizations and the contrast between agrammatic and anomic patients.
The evidence is not equally decisive everywhere. Neural imaging in the book yields inconsistent locations, and the proposed distinction between rules and memory is clearer as a functional account than as a detailed map of circuitry. Nor does the model imply that every irregular form is literally stored as a complete, isolated item. The account allows associations, partial structure, and parallel processing, which makes its most defensible claim a division of labor rather than two sealed mental boxes.
The final comparison to categories extends the argument beyond language: similarity-based judgments and rule-based deductions solve different problems. That is an illuminating synthesis, though the analogy should not be mistaken for identity. The book’s lasting achievement is to show how tiny verbal quirks can expose deep design choices in a mind that must remember particular things and still create the unprecedented.
Review Summary
Words and Rules by Steven Pinker examines how the mind processes regular and irregular verbs and plurals, exploring the dual cognitive systems of memorization and rule application. Reviewers praise Pinker's accessible writing and fascinating linguistic insights, particularly regarding children's language acquisition and brain function. Common criticisms include excessive repetition, exhaustive example lists, and presenting his controversial position without adequately addressing opposing views. Some found it too specialized or dry, while others appreciated the deep dive into linguistics. Most agree it overlaps significantly with The Language Instinct and works best for readers genuinely interested in linguistic theory.
People Also Read
Glossary
words-and-rules theory
memory plus combinationAn account of language in which memorized words and irregular forms are stored in memory, while productive grammatical rules assemble forms and sentences. In Pinker’s version, irregular patterns can also be linked by similarity in memory. The theory explains how language preserves arbitrary exceptions while generating forms speakers have never heard before.
listeme
memorized language itemA language element whose form or meaning must be learned and stored rather than fully generated by a general rule. A listeme can be smaller than a word, such as a suffix, or larger, such as an idiom. The term distinguishes memorized units from combinations assembled productively.
blocking
memory suppresses a ruleThe principle that a stored form can prevent a general rule from producing a competing form with the same grammatical function. If the irregular past tense sang is available for sing, it blocks the regular alternative singed. If the stored form is not retrieved, the rule may apply instead.
default
fallback ruleA pattern that applies when no more specific form or instruction is available. In the book, regular inflection is a default because it can supply a past tense or plural for unfamiliar, rare, or structurally unusual words. A default can be psychologically productive even when it applies to fewer words than another pattern.
wug-test
test of rule generalizationAn experiment that presents a person with an invented word and prompts an inflected form. Because the word is new, the response cannot simply be recalled as a previously learned form. Jean Berko Gleason used the test with children, who supplied regular plurals and past tenses for made-up words.
canonical root
basic stored wordA basic word represented in memory with a sound, a meaning, and a grammatical category, in a sound pattern typical of simple words in its language. Pinker argues that irregular forms attach to roots of this kind. Words created through quotation, sound imitation, names, borrowing, or category conversion may lack such a root.
head
element defining the wholeThe word or part of a word that determines important properties of a larger construction, including its grammatical category and central meaning. Workman is a kind of man, so man is its head, and the plural follows man. A head can also provide stored irregular forms to a complex word.
bahuvrihi
headless possessive compoundA compound that refers to a person or thing by something it has or does, rather than by the category named by its final element. A lowlife is a person associated with a low life, not a kind of life. Because the final element does not define the whole, irregular forms stored with it may not transfer to the compound.
pattern associator memory
similarity-based networkA connectionist model that learns associations between patterns of input and output features. Similar inputs overlap in their learned connections, allowing a pattern learned for one item to influence another. Pinker argues that such models capture generalization among similar irregular verbs, but have difficulty generating regular forms for novel, unfamiliar sound patterns.
family resemblance category
category of overlapping similaritiesA category whose members share overlapping traits without every member possessing one defining feature. Games, birds, and families of irregular verbs can have central examples and less typical edge cases. Similarity can guide judgments about new members, but it does not provide a crisp rule that separates every member from every nonmember.
Download PDF
Download EPUB
.epub digital book format is ideal for reading ebooks on phones, tablets, and e-readers.