collostructions

Collostructions are statistical associations between lexical items and syntactic constructions in corpus linguistics and construction grammar. The term describes how often a given word tends to occur in a particular syntactic construction compared with what would be expected if words were distributed independently of constructions. Researchers extract instances of a construction from a corpus, tally the occurrence of each lexical item in the construction's slot(s), and compare observed counts to expected counts under a null model, using statistics such as log-likelihood ratio, Pearson's chi-square, or Fisher's exact test. The result is a collostruction strength, often expressed as a measure of excitatory (more frequent than expected) or inhibitory (less frequent than expected) association. A positive strength indicates that the item preferentially co-occurs with the construction; a negative strength suggests avoidance.

Collostruction analysis supports construction grammar by providing empirical distributional evidence for the association between words and constructions, enabling studies of lexical choice, constructional productivity, and diachronic change. It is applied to a wide range of constructions, including ditransitives, causatives, and alternations, and can be used cross-linguistically. Layering with distributional semantics can further reveal semantic aspects of the associations.

Limitations include sensitivity to corpus size, sampling bias, polysemy, and the alignment of a word form to a construction; results depend on construction inventory and statistical method. Despite these caveats, collostructions offer a systematic framework for testing predictions about how lexical items participate in constructions and how constructional patterns evolve over time.