Text Entailment Generation with Attention-based Sequence-to-sequence Model

Abstract

Text entailment needs semantic similarity judgment between two sentences and is a good task to measure text understanding. If we realize entailment generation, we can apply it to summarization that keeps semantics between an original text and a generated text. These days, neural networks are employed to construct modules that encode an original text and generate summarization. In natural language processing, sequence-to-sequence models, which realize sequential learning, are employed to develop machine translation. Moreover, attention mechanism is proposed to improve machine translation considering word alignment between a source language and a target language. In this paper, we applied an attention-based sequence-to-sequence model to an entailment generation task and confirmed the system realized entailment generation. The proposed method can capture important words in the input text and generate a frequent sentence, which is grammatically correct and semantically appropriate. The results mean the proposed system understands a text semantically.