Recognizing entailments in legal texts using sentence encoding-based and decomposable attention models

Abstract

This paper presents an end-to-end question answering system for legal texts. This system includes two main phases. In the first phase, our system will retrieve articles from Japanese Civil Code that are relevant with the given question using the cosine distance after the given question and articles are converted into vectors using TF-IDF weighting scheme. Then, a ranking model can be applied to re-rank these retrieved articles using a learning to rank algorithm and annotated corpus. In the second phase, we adapted two deep learning models, which has been proposed for the Natural language inference task, to check the entailment relationship between a question and its related articles including a sentence encoding-based model and a decomposable attention model. Experimental results show that our approaches can be a promising approach for information extraction/entailment in legal texts.

Siamese Networks for Inference in Malayalam Language Texts

“…Entailment recognition has been part of Competition on Legal Information Extraction / Entailment, where sentence encoding and decomposable attention models perform entailment recognition in the context of legal texts (Son et al, 2017). Automatic translation-based approaches are used in the Italian dataset, where the dataset is translated into the English language for entailment recognition (Pakray et al, 2012).…”