Remodelling complement coercion interpretation

Existing (experimental and computational) linguistic work uses participant paraphrases as a stand-in for event interpretation in complement coercion sentences (e.g. she finished the coffee \u3e she finished drinking the coffee). We present crowdsourcing data and modelling that supports broadening this conception. In particular, our results suggest that sentences where many participants do not give a paraphrase, or where many different paraphrases are given, are informative about to how complement coercion is interpreted in naturalistic contexts