Description |
While quantifiers have traditionally been studied in the context of formal logic, recent work in psycholinguistics has shown that their use in grounded situations is highly dependent on pragmatic factors. In this talk, I explore non-grounded quantification -- that is, quantification over concepts without directly available referents (e.g. \'some airplanes are used for freight\', \'all cats are mammals\'). I present the results of a study involving the quantification of feature norms, which illustrate that some substantial disagreements can occur between speakers with regard to the quantification of common concept-feature pairs. I then propose a model of quantification based on distributional semantics. The model is derived from observed language use and therefore lends itself to pragmatic interpretation and (theoretically) to speaker-dependent variations. Tested on simple subject-predicate statements, it shows promising performance in generating quantifiers which match human majority opinion.
|