Substituting phrases with idioms: A sequence-to-sequence learning approach

Abstract

In this paper, a sequence-to-sequence model is proposed for translating sentences without idiom to sentence with the idiom. The problem is challenging in two ways, predicting the correct idiom based on context and generating the correct sentence using the idiom due to complex semantic and syntactic rules of language. Sequence-to-sequence learning has gained popularity in the past few years due to their surprising results on machine translation task. This work is based on sequence-to-sequence learning of word sequences along with their part-of-speech tags to predict sentences with correct idiomatic phrases. Results have shown that models have achieved higher BLEU score on using part-of-speech tags as the input sequences. These observations show the prominence of part-of-speech tags in identifying the hidden writing patterns in the language.