Derivation Versus Generation in Formal Grammar

Core Idea

Generative approaches show how rules progressively build structures from base elements (derivation). Constraint-based approaches specify well-formedness conditions that structures must satisfy (generation). Both can model the same phenomena but differ in how they organize formal knowledge.

Explainer

From your work with constituent trees and phrase structure rules, you know that a grammar can be written as a set of rules like S → NP VP, NP → Det N, and so on. These rules work by expansion: start with the symbol S, pick a rule that rewrites it, apply more rules to the resulting symbols, and continue until you have a string of words. The sequence of steps — each rule application transforming one symbolic expression into another — is a derivation. The final tree records the history of those steps. Derivational thinking asks: what is the *procedure* that built this structure?

Constraint-based frameworks invert the question. Instead of asking how a structure is built step by step, they ask: what conditions must any well-formed structure satisfy? A constraint-based grammar specifies a set of constraints — requirements about word order, case, agreement, argument structure — and a structure is grammatical if and only if it satisfies all of them simultaneously. There is no ordered sequence of rule applications; all constraints evaluate the candidate structure at once. Lexical-Functional Grammar (LFG), Head-Driven Phrase Structure Grammar (HPSG), and Optimality Theory (OT) all work in this mode.

The conceptual difference matters because it changes what "explanation" looks like. In a derivational account, explaining why a structure is ill-formed means showing that no valid derivation produces it — some rule was violated or failed to apply. In a constraint-based account, an ill-formed structure violates one or more constraints, and explaining it means identifying which constraints are at stake. In practice, both frameworks can capture the same empirical facts; the debate is partly about which organization better reflects the mental grammar humans actually use and which scales more gracefully to complex phenomena like long-distance dependencies, ellipsis, or cross-linguistic variation.

The distinction also maps onto a broader division in formal systems: procedural versus declarative representations. A derivational grammar is procedural — it describes a process. A constraint-based grammar is declarative — it describes a state of affairs that must hold. This same split appears in programming languages (imperative vs. declarative), logic (proof-theoretic vs. model-theoretic), and machine learning (generative models vs. discriminative models). Recognizing this isomorphism helps you transfer intuitions across fields: whenever you see a system that generates outputs via sequential rule application versus one that filters candidates against constraints, you are encountering the derivation/constraint distinction in another domain.