What's New

We published the 2018 progress report about joint research for social problem-solving using LINE. (link in Japanese) (6/6)

We will present the following paper at SIGIR2019 (2019/7/21-25) (4/15)

  • Wataru Sakata, Tomohide Shibata, Ribeka Tanaka and Sadao Kurohashi:
    FAQ Retrieval using Query-Question Similarity and BERT-Based Query-Answer Relevance

We will present the following papers at NAACL2019 (2019/6/2-7)

  • Yin Jou Huang, Jing Lu, Sadao Kurohashi and Vincent Ng:
    Improving Event Coreference Resolution by Learning Argument Compatibility from Unlabeled Data
  • Arseny Tolmachev, Daisuke Kawahara and Sadao Kurohashi:
    Shrinking Japanese Morphological Analyzers With Neural Networks and Semi-supervised Learning

We will present the following paper at IJCNN2019 (2019/7/14-19)

  • Abhishek Kumar, Asif Ekbal, Daisuke Kawahara and Sadao Kurohashi:
    Emotion helps Sentiment: A Multi-task Model for Sentiment and Emotion Analysis

Pretrained BERT model for Japanese has been released. (link in Japanese) (4/3)

The following paper received the Best Paper award at the 25th Annual Meeting of The Association for Natural Language Processing (link in Japanese) (3/18)

  • Shuhei Kurita and Anders Søgaard:
    The semantic dependency parsing model using deep reinforcement learning finds out the easy-first parsing strategy

Research Overview

Language is the most reliable medium of human intellectual activities. Our objective is to establish the technology and academic discipline for handling and understanding language, in a manner that is as close as possible to that of humans, using computers. These include syntactic language analysis, semantic analysis, context analysis, text comprehension, text generation and dictionary systems to develop various application systems for machine translation and information retrieval.

Search Engine Infrastructure based on Deep Natural Language Processing


The essential purpose of information retrieval is not to retrieve just a relevant document but to acquire the information or knowledge in the document. We have been developing a next-generation infrastructure of information retrieval on the basis of the following techniques of deep natural language processing: precise processing based not on words but on predicate-argument structures, identifying the variety of linguistic expressions and providing a bird's-eye view of search results via clustering and interaction.

Machine Translation


To bring automatic translation by computers to the level of human translation, we have been studying next-generation methodology of machine translation on the basis of text understanding and a large collection of translation examples. We have already accomplished practical translation on the domain of travel conversation, and constructed a translation-aid system that can be used by experts of patent translation.

Fundamental Studies on Text Understanding

To make computers understand language, it is essential to give computers world knowledge. This was a very hard problem ten years ago, but it has become possible to acquire knowledge from a massive amount of text in virtue of the drastic progress of computing power and network. We have successfully acquired linguistic patterns of predicate-argument structures from automatic parses of 7 billion Japanese sentences crawled from the Web using grid computing machines. By utilizing such knowledge, we study text understanding, i.e., recognizing the relationships between words and phrases in text.