Our NLP research focuses on processing and analyzing human language to address real social problems.
Harmful Content Detection and Cyberbullying
Online hate speech and cyberbullying are serious social issues worldwide. We develop automatic detection systems using NLP and deep learning, covering multiple languages including Japanese, English, Polish, and Korean. We apply cross-lingual transfer learning to extend detection capabilities to low-resource languages where training data is scarce. Our tools have been deployed on commercial platforms.
Ainu Language Processing
Approximately 90% of the world’s languages are at risk of extinction. Ainu, the indigenous language of the Ainu people in Hokkaido, has no native writing system and has been transmitted entirely through oral tradition. We have digitized over 1.75 million characters of Ainu text and developed NLP tools including morphological analysis, part-of-speech tagging, and machine translation to support language preservation and revitalization.
Sentiment Analysis
We research automatic recognition of emotions and intent from text. The ML-Ask sentiment analysis system, developed in our lab, is the world’s first open-source Japanese sentiment analysis tool and has been widely adopted as a research baseline and experimental tool.
Cross-lingual Transfer Learning
We develop methods for transferring NLP knowledge from high-resource to low-resource languages. Our contributions include qWALS, a typology-based language similarity metric that predicts cross-lingual transfer performance across NLP tasks and multilingual models.
Computational Models of Metaphor
We analyze cross-linguistic differences in metaphorical expression and have developed the Murasaki knowledge acquisition system, implemented across Japanese, English, Vietnamese, Korean, and Chinese.