Idiomatic Expression Identification using Semantic Compatibility

Abstract Idiomatic expressions are an integral part of natural language and constantly being added to a language. Owing their non-compositionality ability take on figurative or literal meaning depending the sentential context, they have been classical challenge for NLP systems. To address this challenge, we study task detecting whether sentence has idiomatic expression localizing it when occurs in sense. Prior research studied specific classes offering limited views generalizability new idioms. We propose multi-stage neural architecture with attention flow as solution. The network effectively fuses contextual lexical information at different levels using word sub-word representations. Empirical evaluations three largest benchmark datasets varied syntactic patterns degrees show that our proposed model achieves state-of-the-art results. A salient feature is its identify idioms unseen during training gains from 1.4% 30.8% over competitive baselines dataset.