Distributional Models of Preposition Semantics

Prepositions are often considered to have too little semantic content or be too polysemous to warrant a proper semantic description. We first illustrate the suitability of distributional similarity methods for analysing preposition semantics by way of an inter-preposition similarity task, and make the claim that any semantic account of preposition semantics must be partially conditioned on valence. We further apply distributional similarity methods to a particle compositionality task.