[IA Series 12/n] The Grammar of Natural Language — A Term Sheet

This is a term sheet for the grammar of natural language, in the same spirit as the Intelligent Agents Term Sheet and the Grammar of Logic. The core terms come from the standard linguistics literature and Russell and Norvig’s ‘Artificial Intelligence: A Modern Approach’ (Chapter 23).

Introduction

My interest in meaning goes back to reading the Thesaurus as a kid — grammar didn’t interest me much, but meaning did. The agents in this series are natural-language interfaces: they read prompts and write answers in this language. The grammar of logic (Series 11) is how an agent reasons; the grammar of natural language (Series 12) is how it communicates. The LLM agents I build sit exactly on that bridge, so the language they process deserves the same term-sheet treatment as the logic they reason in. This post is deliberately lean — a shared vocabulary, not a textbook.

The Grammar — Core Terminology

Morphology — the shape of words

  • Morpheme: The smallest unit of meaning. A word is built from morphemes.
    • Free morpheme: Stands alone (run, book).
    • Bound morpheme: Attaches to another morpheme (-s, -ed, un-).
  • Inflection: Modifying a word to fit grammar without changing its core meaning — walk → walked (tense), cat → cats (number). It does not create a new word.
  • Derivation: Building a new word — run → runner, happy → unhappy. It can change meaning and category.

Lexical categories (parts of speech)

  • Noun (N): Names an entity — agent, pipeline.
  • Verb (V): Names an action or state — query, deploy.
  • Adjective (Adj): Modifies a noun — observable, stochastic.
  • Adverb (Adv): Modifies a verb, adjective, or sentence — quickly, partially.
  • Determiner (Det): Marks a noun phrase — the, a, this.
  • Preposition (P): Relates a noun phrase to the rest of the sentence — in, through, at.
  • Pronoun: Stands in for a noun phrase — it, they.
  • Conjunction: Joins constituents — and, or, if.

Phrases and sentences

  • Phrase: A group of words functioning as a unit. Named by its head — noun phrase (NP), verb phrase (VP), prepositional phrase (PP).
  • Constituent: A word or group of words that behaves as a single unit in the grammar.
  • Clause: A phrase built around a verb. Independent clauses stand alone as sentences; dependent clauses do not.
  • Sentence (S): An independent clause — a complete unit of the grammar.
  • Parsing: Recovering a sentence’s grammatical structure from its word string.

Grammatical functions

  • Subject: The noun phrase a sentence is about; typically the agent of the action.
  • Object: The noun phrase the action is directed at.
  • Predicate: What the sentence says about the subject — the verb and its dependents.
  • Agreement: Grammatical matching between constituents — subject and verb agree in number (the gate is vs the gates are).
  • Case: The grammatical role of a noun phrase — I (subject) vs me (object).
  • Tense: When the event occurs — past, present, future.
  • Aspect: How the event unfolds in time — progressive, perfective.
  • Mood: The speaker’s attitude to the event — indicative, imperative, subjunctive.
  • Voice: How the action relates to its participants — active (the agent queries), passive (is queried).
  • Word order: The ordering of subject, verb, and object — English is SVO.

The Grammar, stated

The standard first approximation is a context-free grammar of phrase-structure rules:

S  → NP VP
NP → Det N | Det Adj N | Pronoun | ProperNoun
VP → V NP | V NP PP | V AdvP
PP → P NP

A caveat worth naming: natural language is not fully context-free — agreement, case, and dependencies leak across the boundaries this grammar draws. The CFG is the workable approximation, not the whole truth.

Semantics — Meaning

  • Reference: What an expression points at in the world — the merge gate refers to a particular object.
  • Sense: The meaning of an expression independent of what it points at — the morning star and the evening star differ in sense but share a reference.
  • Compositionality: The meaning of a whole is a function of the meanings of its parts (Frege’s principle).
  • Semantic roles: The parts participants play in an event — agent (does it), theme (undergoes it), instrument (does it with), recipient (receives it).
  • Lexical semantics: Word meaning, organized into relations — synonymy (same meaning), antonymy (opposite), hyponymy (kind-of: deployment is a hyponym of action), meronymy (part-of), polysemy (one word, related senses: gate = the physical object or the decision), homonymy (one word, unrelated senses: bank).
  • Distributional semantics: The view that meaning is determined by distribution — a word’s meaning is fixed by the contexts in which it occurs. The distributional hypothesis is due to Harris (1954); Firth (1957) gave its slogan: “a word is characterized by the company it keeps." Meaning is quantified as vectors over co-occurrence (LSA, word2vec) — the basis of modern LLM semantics, and the natural-language counterpart of the logic sheet’s Determination: where logic’s truth is attributed by a human-chosen model, natural-language meaning is determined by the distribution of the corpus — learned from usage, not attributed.
  • Compositional distributional semantics: Merging the grammar’s compositionality with distributional vectors, so the meaning of a phrase is computed from the meanings of its parts (Clark, Coecke & Sadrzadeh 2008). The same family as distributed representations (Rieger 1991), where meaning lives spread across many dimensions rather than in one place.
  • Ambiguity: A phrase with more than one meaning.
    • Lexical: a word has multiple senses.
    • Structural: the grammar permits two structures — “the agent sees the man with the telescope” attaches the PP to the object or the verb.
  • Truth conditions: The conditions under which a sentence is true. This is where natural-language meaning meets the logic term sheet: a sentence’s truth conditions are determined by the grammar and attributed by the reader, exactly as the logic sheet’s Determination defines.
  • Anaphora: A word that refers back to an earlier one — “the agent finished its run; it then merged."

Pragmatics — Using the Grammar

  • Pragmatics: What a speaker means by an utterance, beyond what the sentence literally says.
  • Utterance: A particular use of a sentence in a context.
  • Context: The situation of the utterance — who speaks, to whom, when, and against what shared background.
  • Speech acts: What an utterance does — asserting (the pipeline is green), asking (is it green?), requesting (make it green), promising.
  • Implicature: What is communicated beyond what is said — “the gate is open” can imply “you may go,” without stating it.
  • Presupposition: What an utterance takes for granted — “the merge failed” presupposes a merge existed.
  • Deixis: Meaning that depends on the context of utterance — I, you, here, now, this.
  • Discourse: A sequence of utterances — the level above the sentence, where anaphora, coherence, and topic live.

Closing

Logic is how the agent reasons; natural language is how it communicates; the ubiquitous language of the ontology work is the shared vocabulary expressed in this grammar. The LLM sits on that bridge — parsing the natural language into meaning, and generating natural language that carries it back. This term sheet names the pieces of that interface so the series can describe it precisely.

ia-series natural-language