A computational model of S-selection

Abstract

We develop a probabilistic model of S(emantic)-selection that encodes both the notion of systematic mappings from semantic type signature to syntactic distribution—i.e., projection rules—and the notion of selectional noise—e.g., C(ategory)-selection, L(exical)-selection, and/or other independent syntactic processes. We train this model on data from a large-scale judgment study assessing the acceptability of 1,000 English clause-taking verbs in 50 distinct syntactic frames, finding that this model infers coherent semantic type signatures. We focus in on type signatures relevant to interrogative and declarative selection, arguing that our results suggest a principled split between cognitive verbs, which select distinct proposition and question types, and communicative verbs, which select a single hybrid type.