This paper proposes several recurrent neural network-based models for recognizing requisite and effectuation parts in Legal Texts. Firstly, we propose a modification of BiLSTM-CRF model that allows the use of external features to improve the performance of deep learning models in case large annotated corpora are not available. However, this model can only recognize RE parts which are not overlapped. Secondly, we propose two approaches for recognizing overlapping RE parts including the cascading approach which uses the sequence of BiLSTM-CRF models (...) and the unified model approach with the multilayer BiLSTM-CRF model and the multilayer BiLSTM-MLP-CRF model. Experimental results on two Japan law RRE datasets demonstrated advantages of our proposed models. For the Japanese National Pension Law dataset, our approaches obtained an \ score of 93.27% and exhibited a significant improvement compared to previous approaches. For the Japan Civil Code RRE dataset which is written in English, our approaches produced an \ score of 78.24% in recognizing RE parts that exhibited a significant improvement over strong baselines. In addition, using external features and in-domain pre-trained word embeddings also improved the performance of RRE systems. (shrink)

This paper investigates the task of reference resolution in the legal domain. This is a new interesting task in Legal Engineering research. The goal is to create a system which can automatically detect references and then extracts their referents. Previous work limits itself to detect and resolve references at the document targets. In this paper, we go a step further in trying to resolve references to sub-document targets. Referents extracted are the smallest fragments of texts in documents, rather than the (...) entire documents that contain the referenced texts. Based on analyzing the characteristics of reference phenomena in legal texts, we propose a four-step framework to deal with the task: mention detection, contextual information extraction, antecedent candidate extraction, and antecedent determination. We also show how machine learning methods can be exploited in each step. The final system achieves 80.06 % in the F1 score for detecting references, 85.61 % accuracy for resolving them, and 67.02 % in the F1 score for the end-to-end setting task on the Japanese National Pension Law corpus. (shrink)

This article addresses the problem of automatically extracting legal indices which express the important contents of legal documents. Legal indices are not limited to single-word keywords and compound-word keywords, they are also clause keywords. We approach index extraction using structural information of Japanese sentences, i.e. chunks and clauses. Based on the assumption that legal indices are composed of important tokens from the documents, extracting legal indices is treated as a problem of collecting chunks and clauses that contain as many important (...) tokens as possible. Each token is assigned a weight which is a statistical score, e.g. TF–IDF and Okapi BM25, to indicate its importance. The importance of a chunk or clause is determined based on the average weight of tokens included in that chunk or clause. Then, highly weighted chunks and clauses are recognized as the indices for legal documents. The experimental results on Japanese National Pension Act data show that our proposed method achieves better performance than TextRank, the most popular unsupervised method in extracting single-word and compound-word keywords. In addition, this approach is also applicable to extract clause keywords with high performance. (shrink)