Construction Grammar (Goldberg)

How the framework works

The starting move of Construction Grammar is to take seriously the things generative grammar treated as marginal — idioms, partially productive patterns, conventionalized expressions. Kick the bucket means "die" with no compositional path; the more X the more Y ("the bigger the better") obeys a partial template; Time is money instantiates conceptual metaphor. Generative grammar handles these as exceptions in the lexicon. Construction Grammar makes them the model.

The core claim: grammar is a structured inventory of constructions. A construction is a learned pairing of form (phonological, syntactic, morphological) with meaning (semantic, pragmatic, discourse). Constructions span scale:

• Morphemes. The English suffix -er in baker, runner, teacher pairs a phonological form with the meaning "agentive nominalizer."
• Words. Avocado pairs a phonological string with a fruit-meaning. Words are constructions.
• Idioms. Kick the bucket pairs a fixed phrase with a non-compositional meaning ("die").
• Partially filled idioms. What's X doing Y? ("What's that fly doing in my soup?") fixes parts and leaves slots.
• Abstract schemas. The ditransitive Subj V Obj1 Obj2 fixes only the syntactic configuration and contributes its own meaning (transfer).
• Discourse patterns. Topic-comment, list intonation, presentational structures.

Knowing a language is knowing its construction inventory. There is no clean separation between lexicon (irregular) and grammar (regular) — they form a continuum of items varying in size, abstraction, and productivity. Generalizations across constructions are stored as schemas; specific items are stored as instances. The architecture is non-derivational: there is no underlying form that surfaces by movement.

Worked example: the ditransitive construction

Goldberg's central argument concerns the English ditransitive: Subj V Obj1 Obj2. Examples include "Pat baked Chris a cake," "Sue gave Tom a book," "She handed me the keys."

The traditional generative analysis: each verb has lexical entries specifying its argument structure. Give can select two objects; bake can select two objects under certain conditions. Argument structure is verb-projected.

Goldberg's 1995 argument: the ditransitive schema itself contributes meaning, namely "agent intends to cause recipient to receive theme." The verb fills the V slot, but the transfer meaning belongs to the construction. Three pieces of evidence:

1. Novel verbs slot in. "She faxed Tom a memo." The verb fax is not lexically a transfer verb but the ditransitive imposes transfer reading. Children and adults extend this productively to invented verbs.
2. Verbs gain transfer meaning in the construction. "He sneezed the napkin off the table" (caused-motion construction). The verb sneeze is intransitive lexically but takes the caused-motion schema. The motion meaning is contributed by the construction, not by sneeze.
3. The construction has constraints. Only intentional transfer fits the prototypical ditransitive. "*She baked the cake the oven" is bad because the oven is not a sentient recipient. The constraint belongs to the schema, not lexical baking.

Goldberg (2006, Constructions at Work) elaborates with quantitative corpus and acquisition evidence. Bencini and Goldberg (Journal of Memory and Language, 2000) showed adults use construction-level cues to interpret novel verbs. Children produce ditransitive innovations from age four. The case for constructional meaning is strong; whether it eliminates lexical argument structure entirely is debated.

Worked example: caused-motion and resultative constructions

Two further argument-structure schemas illustrate the program:

• Caused-motion construction. Subj V Obj Path, with meaning "X causes Y to move along PATH." "She kicked the ball into the goal," "The wind blew the leaves off the tree," "He sneezed the napkin off the table." The path-PP is licensed by the construction; verbs that are not lexically motion verbs (sneeze, blow) acquire motion meaning when they fill the V slot.
• Resultative construction. Subj V Obj Resultative-XP, with meaning "X causes Y to become STATE." "She hammered the metal flat," "He drank the teapot dry," "They painted the house red." Again, the verb need not entail the result. Hammer does not entail flatness lexically; the construction adds it.

These schemas are productive but constrained. "*She drove the soles off her shoes" in the literal sense is degraded because driving does not normally cause soles to detach. Goldberg argues each schema has a semantic frame and licensing conditions. The frame inventory is rich and partially language-specific — French and German have different constructional inventories than English, generating typological variation without parameters.

Construction Grammar vs other syntactic frameworks

	Construction Grammar	Principles & Parameters	Minimalism	LFG	HPSG	Systemic Functional
Founders	Goldberg 1995; Fillmore & Kay 1980s	Chomsky 1981	Chomsky 1995	Bresnan & Kaplan 1982	Pollard & Sag 1994	Halliday 1985
Basic unit	Construction	Phrase, head, parameter	Merge, feature	F-structure, c-structure	Sign, type	Function (Theme, Subject, Actor)
Lexicon-grammar split	Continuum, no split	Sharp split	Sharp split	Lexicalist split	Lexicalist, types	No split, lexicogrammar
Movement / transformation	None	Move-α	Internal Merge	None (functional)	None (feature percolation)	None
Argument structure	Constructional schema	Verbal projection	Lexical features	Lexical mapping	Lexical valence	Process-Participant frames
Cross-linguistic strategy	Construction inventory varies	Parameter setting	Lexical features	Lexical entries	Type hierarchy	Functional registers
Acquisition	Item-based, frequency-driven	Switch-setting	Feature acquisition	Lexical learning	Schema abstraction	Functional learning

Acquisition in Construction Grammar

Goldberg's Constructions at Work (2006) and Tomasello's Constructing a Language (2003) developed the acquisition story. Children learn specific items first — Tomasello's "verb island hypothesis" holds that toddlers' early productions are organized verb by verb. "Cut paper, cut bread, cut hair" generalize from individual cut-instances. Only later do children abstract over multiple verbs to form schemas like SVO or the ditransitive.

Joan Bybee's frequency-based approach (Frequency of Use and the Organization of Language, 2007; Language, Usage and Cognition, 2010) provides the quantitative engine. Token frequency entrenches specific items; type frequency drives schema abstraction. Past tense -ed is highly productive because it applies to many types; past tense ablaut (sing-sang) is unproductive because it applies to few.

Statistical preemption (Goldberg 2006) explains some learnability gaps. Why is "*She explained me the theorem" bad when "She told me the theorem" is fine? Children hear "She explained the theorem to me" often. The repeated alternative preempts the ditransitive analogy. The principle: when a less-similar construction is repeatedly heard for the same meaning, it preempts the more-similar generalization. This solves part of the no-negative-evidence puzzle without invoking innate UG.

Generative counterarguments

Stefan Müller (Grammatical Theory, 2018) gives a thorough comparative analysis. The generative case against Construction Grammar:

• Information duplication. If each construction stores its own form-meaning pair, regularities across constructions are missed. Generative theories factor common information into principles; Construction Grammar must postulate explicit inheritance hierarchies, which formalize but do not eliminate the duplication.
• Long-distance dependencies. Wh-movement, parasitic gaps, and across-the-board extraction involve dependencies spanning indefinite distance. Construction Grammar handles them through unification and feature percolation (especially in Sign-Based Construction Grammar), but generative critics argue movement-based accounts are simpler.
• Predictive precision. Generative theory makes detailed predictions about subtle phenomena (binding scope, reconstruction effects). Construction Grammar tends to be descriptively rich but less predictively constrained.
• Universal patterns. Cross-linguistic universals are easier to express via Universal Grammar; Construction Grammar must derive them from shared cognition and communicative pressure, which is a heavier explanatory load.

Construction grammarians reply that the costs are worth the benefits: explicit modeling of idiom and partial productivity; fit with usage data and acquisition timelines; integration with frame semantics, conceptual metaphor, and discourse pragmatics. Boas and Sag's Sign-Based Construction Grammar (2012) addresses several formal worries by adopting HPSG-style typed feature structures, putting Construction Grammar on a comparably formal footing.

Variants of Construction Grammar

• Berkeley Construction Grammar (Fillmore, Kay 1980s–90s). The original. Formal, unification-based, focused on argument structure and lexicogrammatical interface. Influenced HPSG.
• Cognitive Construction Grammar (Goldberg, Lakoff). Closer to cognitive linguistics. Emphasizes constructional meaning, frame semantics, conceptual metaphor. Goldberg 1995 is the foundational text; Constructions at Work (2006) the elaboration.
• Radical Construction Grammar (Croft 2001). Cross-linguistic. Argues syntactic categories (noun, verb) are construction-specific, not universal primitives. Each language has its own construction inventory; cross-linguistic universals are typological tendencies, not innate.
• Sign-Based Construction Grammar (Boas & Sag 2012). Merger with HPSG. Uses typed feature structures, inheritance hierarchies, formal unification. Addresses long-distance dependencies through the same mechanisms HPSG uses.
• Embodied Construction Grammar (Bergen & Chang, 2005). Links constructions to simulation semantics — comprehension as mental simulation in sensorimotor systems. Computational implementations exist.
• Fluid Construction Grammar (Steels 2011). A computational implementation focused on language emergence in artificial agents.

Common pitfalls

• Treating constructions as just templates. They are form-meaning pairings; meaning is integral. A pure syntactic template without semantics is not a construction in Goldberg's sense.
• Assuming Construction Grammar is just "stored idioms." The framework treats abstract schemas as constructions too. The ditransitive is as much a construction as kick the bucket; only the level of abstraction differs.
• Confusing Construction Grammar with Cognitive Grammar. Cognitive Grammar (Langacker 1987, 2008) is a related but distinct framework focused on conceptual structure as the basis of meaning. The two are compatible and often discussed together but have different formalisms and emphases.
• Reading "no UG" as "no innate cognition." Construction grammarians assume domain-general cognitive abilities (categorization, analogy, statistical learning) — they reject specifically grammatical UG, not all biological contribution.
• Mistaking the framework for purely surface-oriented. Constructions encode rich semantic and pragmatic information. The case for constructional meaning is what distinguishes the program from purely structural syntactic theories.
• Treating Construction Grammar as synonymous with Goldberg. Goldberg is the most-cited proponent but the family is wider. Croft's typological program, Sign-Based variants, and computational approaches diverge significantly from Goldberg's cognitive emphasis.

Legacy and current status

Construction Grammar has reshaped argument-structure research, frame semantics, and usage-based acquisition. Goldberg's 1995 book is among the most-cited works in linguistics; her 2006 Constructions at Work consolidated the program into a textbook framework. The constructional approach has spread to corpus linguistics, second-language acquisition, language change (Hilpert, Constructional Change in English, 2013), and computational NLP.

The framework remains a serious alternative to generative theory, especially for researchers prioritizing usage data, idiomatic and partially productive patterns, and cross-linguistic descriptive coverage. The dialogue with Sign-Based Construction Grammar and HPSG has narrowed the formal gap with mainstream syntactic theory; the dialogue with cognitive linguistics keeps the program embedded in broader theories of meaning and conceptualization.