Words and Rules

1. Language's Power: Words and Rules as Core Ingredients

The premise of this book is that there are two tricks, words and rules.

Boundless expression. Human language, a miraculous gift, allows us to share an unfathomable vastness of ideas by fashioning breath into sounds. This boundless expressive power stems from two fundamental "tricks": words and rules. Words are memorized arbitrary pairings of sound and meaning, like "rose" conveying the concept of a flower.

Grammar's combinatorial magic. Rules, or grammar, combine these words into phrases and sentences, where meaning is inferred from arrangement. This combinatorial system is productive, allowing for novel sentences; abstract, using symbols like "Noun Phrase" (NP) for any noun; and recursive, enabling sentences of infinite length. For example:

• NP + VP = Sentence
• Determiner + Noun = NP
• Verb + NP = VP

Efficiency and creativity. This combinatorial power is astonishingly efficient. A small set of elements and rules can generate an immense number of distinct thoughts, far exceeding the number of seconds since the universe's birth. This dual system of memorized words and generative rules is the engine of human communication, allowing us to articulate thoughts never before conceived.

2. Regular vs. Irregular: A Fundamental Mental Divide

I believe that languages do provide us with such specimens. They are called regular and irregular words.

Two flavors of verbs. English verbs come in two distinct types: regular and irregular, serving as perfect specimens to illuminate the words-and-rules dichotomy. Regular verbs, like "jog-jogged," are predictable, formed by adding "-ed," and are an open-ended class, readily accepting new words like "fax-faxed."

Chaotic exceptions. Irregular verbs, such as "buy-bought" or "sing-sang," are chaotic and idiosyncratic, forming a closed list of about 150-180 verbs with no new additions. This stark contrast suggests a simple theory: regular forms are generated by a rule ("add -ed"), while irregular forms are individually memorized.

• Regular: jog → jogged (rule-based)
• Irregular: buy → bought (memorized)

Blocking principle. A crucial principle, "blocking," explains their interaction: if a word's past tense is found in memory (e.g., "held"), the regular rule is blocked. Otherwise, the rule applies by default (e.g., "Bork-Borked"). This dual mechanism is not an evolutionary quirk but an inevitable outcome of two mental subsystems—memory for words and rules for combinations—performing the same task.

3. Historical Echoes: Irregulars as Fossils of Old Rules

Most of the forms were originally created by rules, but a later generation never grasped the rules and instead memorized the forms as words.

Broken Telephone. Language evolves like a game of "Broken Telephone," with each generation reanalyzing and subtly altering the language they hear. Irregular forms are not random but "fossils" of long-dead rules from ancestral languages like Proto-Indo-European and Proto-Germanic. For example, the vowel changes in "foot-feet" originated from an Old English phonological rule (umlaut) that later vanished, leaving the altered vowel as a memorized irregularity.

Patterns persist. Despite millennia of change, these ancient patterns persist in clusters of similar-sounding irregular verbs. Examples include:

• sing-sang-sung
• bind-bound-bound
• blow-blew-blown

Frequency's role. The survival of irregular verbs is strongly linked to their frequency. The most common verbs in English (e.g., "be," "have," "do," "go") are almost all irregular. If an irregular verb's usage declines, children are less likely to memorize its unique form, defaulting to the regular "-ed" rule, causing the irregular form to eventually disappear from the language. This ongoing process explains why English has fewer irregular verbs today than in Old English.

4. The Mind's Default: Regularity Beyond Frequency

The omnipotence of a regular rule does not seem to depend on a person’s previously having been swamped with regular forms.

Productivity's engine. The mind's ability to generate new word forms is called productivity. For regular forms, this is driven by a rule that applies universally, regardless of a word's familiarity. Studies show that rare regular past tenses (e.g., "maimed") are processed as easily as common ones (e.g., "walked"), indicating they are computed on the fly, not retrieved from memory.

German's counter-example. This is powerfully demonstrated in German, where the regular participle suffix "-t" applies to a minority of verbs (45% of common verbs), yet functions as the default. It attaches to:

• Rare and novel verbs (e.g., "gefabt" for a made-up verb)
• Unusual-sounding verbs
• Verbs derived from other categories (e.g., "gefrühstückt" from "Frühstück" - breakfast)

Challenging connectionism. This "emergency plural ending" in German, despite its statistical rarity, behaves exactly like English's regular "-ed." This directly refutes connectionist theories that claim generalization is solely driven by the statistical prevalence of a pattern. Instead, it confirms that regularity stems from a mental rule that applies to any item categorized as a "verb" or "noun," regardless of its sound or frequency.

5. Word Structure Matters: Headless Words Force Regularization

A word that violates the standards has to give up its irregular form; these are the words, such as flied out and lowlifes, that arouse the curiosity of language-lovers.

Rootless words. Words are stored in memory as "roots," canonical forms with arbitrary sound-meaning pairings. Irregular inflections are tied to these roots. However, words that are "rootless"—not perceived as canonical roots—cannot access irregular forms and default to regular inflection. Examples include:

• Onomatopoeia: ping-pinged (not pang)
• Quotations: three "man"s (not men)
• Names: the Childs (not Children)
• Foreign borrowings: mongooses (not mongeese)
• Acronyms/Truncations: PCs, sysmans
• Converted words: to high-stick → high-sticked (not high-stuck)

Headless compounds. Similarly, "headless" compounds, where the meaning of the whole doesn't derive from its rightmost part (the "head"), also force regularization. A "lowlife" is a person, not a type of "life," so its plural is "lowlifes" (not "lowlives"). This "word structure theory" explains why "flied out" is used in baseball (the verb "to fly out" is derived from the noun "a fly ball," not the verb "to fly").

Grammar's deep logic. These systematic regularizations demonstrate that speakers instinctively analyze a word's grammatical structure. If a word lacks a proper root or a clear head to inherit irregularity from, the default rule steps in. This highlights that rules operate on abstract categories and structural relationships, not just sound or meaning.

6. Children's Errors: A Window into Rule Acquisition

Overgeneralization errors are a symptom of the open-ended productivity of language, which children indulge in as soon as they begin to put words together.

U-shaped development. Children's acquisition of irregular verbs often follows a U-shaped curve: they initially use correct irregulars (e.g., "sang"), then overgeneralize the regular rule (e.g., "singed"), and finally return to correct irregulars. This "getting worse before getting better" phenomenon is a hallmark of rule acquisition, as children extract a pattern and apply it broadly.

Rule discovery. Children, around ages two to three, discover the "-ed" rule by noticing patterns in regular verbs. Once acquired, this rule is applied to all verbs, including irregular ones, leading to errors like "goed" or "bleeded." These errors are not due to a lack of knowledge of the correct irregular form, but rather a temporary failure to retrieve it from memory, allowing the newly acquired rule to apply.

• Early stage: sing (bare stem) or sang (memorized)
• Rule acquisition: singed (overgeneralization)
• Adult stage: sang (memorized, blocking singed)

No negative feedback. Crucially, children do not learn to correct these errors through parental feedback, which is often ineffective or absent. Instead, the "blocking" principle, likely innate, helps them. As children hear irregular forms more frequently, their memory traces strengthen, allowing the correct irregular form to block the regular rule, gradually eliminating overgeneralizations.

7. Cross-Linguistic Universals: Rules Across Diverse Tongues

To see these deep parallels in the languages of the French and the Germans, the Arabs and the Israelis, the East and the West, people living in the Age of the Internet and people living in the Stone Age, is to catch a glimpse of the psychic unity of humankind.

Universal default rules. The phenomenon of a "default" or "emergency" rule, which applies when memory fails, is not unique to English or even Germanic languages. It is a universal feature found across diverse language families, suggesting a fundamental aspect of human language faculty.

• Dutch: Regular suffix -den for verbs from nouns.
• French: Regular plural -als for names, foreign words, etc., despite irregular -aux for native words.
• Hungarian: Regular declension for names, despite irregular patterns for native nouns.
• Arabic: "Sound plural" suffixes (-uun, -aat) for names, diminutives, borrowings, despite "broken plural" for native words.
• Hebrew: Regular plural -im for foreign words and names, despite irregular patterns for native nouns.
• Mandarin Chinese: Classifier ge for objects without a memorized classifier, or for abstract nouns.
• Arapesh (New Guinea): A "default gender" for nouns that don't fit other categories.

Psychic unity. These widespread parallels, observed in languages with vastly different histories and structures, underscore the "psychic unity of humankind." They demonstrate that the human mind is equipped with a universal toolkit for language acquisition, comprising both a frequency- and similarity-loving associative memory and a promiscuous combinatorial grammar, which interact to shape the unique characteristics of each language.

8. Brain's Blueprint: Distinct Neural Systems for Words and Rules

If words and rules are the ingredients of language, we should be able to tell them apart in the brain.

Neurological evidence. Modern cognitive neuroscience, using techniques like fMRI, PET, and EEG, provides compelling evidence that words and rules are processed by distinct neural systems. This is supported by "double dissociations" in brain-damaged patients and genetic disorders.

• Agrammatism (Broca's area damage): Patients struggle with grammatical rules (regular verbs, complex sentences) but are better at retrieving memorized words (irregular verbs).
• Anomia (temporal/parietal lobe damage): Patients struggle with word retrieval (irregular verbs) but maintain grammatical fluency (better at regular verbs, make childlike overgeneralizations).

Neurodegenerative diseases. This pattern is mirrored in neurodegenerative diseases:

• Alzheimer's disease (memory loss): Impaired irregular verb inflection, preserved regular verb inflection.
• Parkinson's disease (procedural memory impairment): Impaired regular verb inflection, preserved irregular verb inflection.

Real-time brain activity. Electroencephalography (EEG) and Magnetoencephalography (MEG) show distinct brain activity patterns. Incorrect regular suffixes elicit a "Left Anterior Negativity" (LAN), indicating a grammatical violation, while incorrect irregular suffixes elicit an "N400," suggesting a semantic or lexical anomaly. MEG studies even show a temporal shift: initial activity in memory areas for word recognition, then a shift to frontal areas for regular suffixing.

9. Beyond Language: Rules and Associations Shape All Concepts

I believe there is something beneath the similarities, and that the facts of regularity and irregularity offer glimmers of insight into the nature of our conceptual categories.

Two ways of knowing. The distinction between regular and irregular inflection parallels two fundamental ways the human mind forms conceptual categories: classical (Aristotelian) and family resemblance (Wittgensteinian).

• Irregular verbs are like family resemblance categories (e.g., "games," "furniture"): no single defining trait, fuzzy boundaries, prototypes, and crisscrossing features. They are relics of history, memorized associatively.
• Regular verbs are like classical categories (e.g., "odd number," "grandmother"): defined by crisp rules, clear membership, and no prototypes. They are products of a rule system, a "virtual category."

Digital mind in an analog world. This duality reflects how our "digital minds" (rule-based, symbolic computation) interact with an "analog world" (complex, historically shaped phenomena). Family resemblance categories help us make predictions about objects shaped by myriad historical contingencies (like biological species or chairs). Classical categories, part of rule systems like arithmetic or kinship, allow for abstract, deductive reasoning about an unlimited range of cases.

Consilience of knowledge. The study of regular and irregular verbs, bridging linguistics, psychology, and neuroscience, offers a glimpse into this deeper cognitive architecture. It suggests that our capacity for abstract, combinatorial thought, manifest in grammar and classical categories, is a key to human intelligence, enabling us to understand and manipulate the world beyond mere associative learning.

Last updated: February 2, 2026