Institute of Information Science, Academia Sinica

Adaptation technologies are always useful in NLP when there is discrepancy between the training scenario and use scenario.  They are also effective in alleviating the data scarcity problem.  Domain adaptation is the most popular kind of adaptation technologies and is intensively researched.  In this talk we will introduce two other kinds of adaptation technologies: cross-standard adaptation and cross-lingual adaptation.  Cross-standard adaptation is used to improve the performance of an automatic annotation task by leveraging corpora with different annotation standard, while cross-lingual adaptation is used to solve an NLP problem in one language by utilizing the linguistic knowledge which is learnt from solving the same problem in another language.  We propose conditional mapping approach for cross-standard adaptation and decomposed projection approach for cross-lingual adaptation.  We investigate these technologies mainly for the tasks of Chinese word segmentation and dependency parsing, and finally we verify their effectiveness in Irish POS tagging and dependency parsing.

Qun Liu is currently a Professor in Dublin City University (DCU) and the leader of Natural Language Processing Group in ADAPT Centre (formerly CNGL), Ireland.  He received his Master's degree in Computer Science from the Institute of Computing Technology at Chinese Academy of Sciences (ICT/CAS) in 1992, and his PhD degree in Computer Science from Peking University in 2004. Since 1992 he has been a Researcher and Professor in ICT/CAS, Beijing, where he was the Leader of the Natural Language Processing Research Group.   He moved to Ireland and joined CNGL/DCU in July 2012, while still working with the NLP group at ICT/CAS as an Adjunct Professor and Group Leader. His research interests focus on machine translation and natural language processing, especially on Chinese language processing, statistical machine translation models, approaches and evaluation methods. He led the development of the broadly used ICTCLAS open source Chinese word segmentation and POS tagging systems in 2002.  In recent years he and his research group mainly work on syntactic-based statistical machine translation models, context-aware translation methods and adaptation technologies for NLP.