Classification of Idiomatic Sentences Using AWD-LSTM

Abstract

Idioms are a mixture of terms with a figurative meaning distinct from the literal meanings of each word or expression. Automatic meaning detection of idioms represents a serious challenge in understanding the language. Because their meaning cannot be directly retrieved from the words, the development of computational models for human processing human languages is concerned with natural language processing (NLP). Idiomatic phrase identification is utmost important in many NLP applications like machine translation system, chatbot and information retrieval system (IR). Text classification is one of the fundamental tasks of NLP and is mostly attempted using supervised algorithms. This paper has perceived the identification of idioms as a text classification task. In this paper, we propose a classification model to classify the idioms and the literal sentences using ASGD weight-dropped LSTM (AWD-LSTM) model and universal language model fine-tuning (ULMFiT) for transfer learning to fine-tune the language model. The proposed model has been evaluated using precision, recall and F1-score metrics. The proposed model has been tested with the TroFi metaphor dataset and an in-house dataset and achieved 81.4 and 85.9% of F-Score, respectively.