Page 90 - 2017 Brochure
P. 90
究員
蘇克毅 Keh-Yih Su
Research Fellow
Ph.D., Electrical Engineering, University of Washington
Tel: +886-2-2788-3799 ext. 1801 Fax: +886-2-2782-4814
Email: kysu@iis.sinica.edu.tw
http://www.iis.sinica.edu.tw/pages/kysu
• Research Fellow, Institute of Information Science, Academia Sinica (2014-present)
• President/CEO, Behavior Design Corporation (1998-2014)
• Associate Professor, Professor, Electrical Engineering, National Tsing Hua University (1984-1998)
• Ph.D., Electrical Engineering, University of Washington (1984)
• Advisor, New Series in Natural Language Processing , John Benjamins (1999-present)
• President, Asian Federation of Natural Language Processing (2011-2012)
• Executive Member, Association for Computational Linguistics (2005-2007)
• President, Association for Computational Linguistics and Chinese Language Processing (1996-1997)
Research Description meaningful problem solving interpretation.
My research interests include Machine Translation, Machine I have also worked on Chinese word segmentation with new
Reading, Chinese Language Processing, Natural Language generative models (also with additional resources later), and have
Understanding, Statistical Language Modeling and Machine obtained the best performance to be reported on the SIGHAN
Learning. I have worked on statistical modeling and machine close/open-test. Performance on this test is classified based
learning for natural language processing (NLP) since 1986. In on whether resources other than the training-set data, such as
this work, I mostly adopted principled approaches to solving NLP dictionaries, could be adopted.
problems, in contrast to the prevailing techniques of simply adding
a lot of features under the Maximum Entropy framework. Currently, I am beginning to build a platform to explore new
knowledge from multiple documents. Traditional single-document
My experience mainly lies in building a statistical semantic machine information extraction (IE) does not utilize redundancy and
translation system. I proposed and built a statistics-oriented interaction between the content that may be achieved by extraction
framework for machine translation, which automatically obtains from multiple documents. As a result, the quality of the information
associated parameters from the given pairs of Source and Target directly extracted from an individual document is relatively poor,
Semantic-Normal-Forms via semi-supervised learning. In addition, and knowledge that could be deduced from the interactions
I proposed a joint model to integrate Translation Memory into a between information extracted from various documents/domains is
phrase-based machine translation system during decoding, which not explored. This project will develop a generic cross-document
made significant improvements to the system. Recently, I have built analysis package to refine the noisy information extracted from
a goal-oriented meaning-based statistical framework for solving traditional IE, and to explore new knowledge cross domains.
English multi-step math word problems, which resembles the
human cognitive understanding of problems and produces a more Transactions on Asian Language Information Processing, volume 14,
number 3, pages 12:1-12:29, June 2015.
Publications 6. Kun Wang, Chengqing Zong, and Keh-Yih Su, “Dynamically
Integrating Cross-Domain Translation Memory into Phrase-Based
1. Chao-Chun Liang, Shih-Hong Tsai, Ting-Yun Chang, Yi-Chung Machine Translation during Decoding,” Proceedings of COLING
Lin and Keh -Yih Su, “A Meaning-bas ed Englis h Math W ord 2014, August 2014.
Problem Solver with Understanding, Reasoning and Explanation,” 7. Kun Wang, Chengqing Zong, and Keh-Yih Su, Aug. 4-9, 2013,
Proceedings of COLING 2016, Osaka, Japan, December 2016, System “Integrating Translation Memory into Phrase-Based Machine
Demonstration. Translation during Decoding”, Proceedings of ACL 2013, Sofia,
Bulgaria.
2. Chao-Chun Liang, Kuang-Yi Hsu, Chien-Tsung Huang, Chung-Min 8. Yufeng Chen, Chengqing Zong, and Keh-Yih Su, “A Joint Model
Li, Shen-Yun Miao and Keh-Yih Su, “A Tag-based Statistic English to Identify and Align Bilin gual Named Entities,” Jo urn al of
Math Word Problem Solver with Understanding, Reasoning and Computational Linguistics, Vol.39, No. 2, June 2013, pages 229-266.
Explanation,” Proceedings of IJCAI 2016, New York City, New York, 9. Xiaoqing Li, Kun Wang, Chengqing Zong, and Keh-Yih Su, Dec. 8-15,
U.S.A., July 2016, System Demonstration. 2012, “Integrating Surface and Abstract Features for Robust Cross-
Domain Chinese Word Segmentation”, Proceedings of COLING 2012,
3. Yi-Chung Lin, Chao-Chun Liang, Kuang-Yi Hsu, Chien-Tsung Mumbai, India, pages 1653-1669.
Huang, Shen-Yun Miao, Wei-Yun Ma, Lun-Wei Ku, Churn-Jung Liau, 10. Kun Wang, Chengqing Zong, and Keh-Yih Su, “Integrating Generative
Keh-Yih Su, “Designing a Tag-Based Statistical Math Word Problem and Discriminative Character-Based Models for Chinese Word
Solver with Reasoning and Explanation,” International Journal of Segmentation,” ACM Transactions on Asian Language Information
Computational Linguistics and Chinese Language Processing, volume Processing, Vol.11, No. 2, Article 7, June 2012, pages 7:1-7:41.
20, number 2, pages 1-26, December 2015.
4. Chien-Tsung Huang, Yi-Chung Lin, and Keh-Yih Su, “Explanation
Generation for a Math Word Problem Solver,” International Journal of
Computational Linguistics and Chinese Language Processing, volume
20, number 2, pages 27-44, December 2015.
5. Xiaoqing Li, Chengqing Zong, and Keh-Yih Su, “A Unified Model for
Solving the OOV Problems of Chinese Word Segmentation,” ACM
88 研究人員 Research Faculty
蘇克毅 Keh-Yih Su
Research Fellow
Ph.D., Electrical Engineering, University of Washington
Tel: +886-2-2788-3799 ext. 1801 Fax: +886-2-2782-4814
Email: kysu@iis.sinica.edu.tw
http://www.iis.sinica.edu.tw/pages/kysu
• Research Fellow, Institute of Information Science, Academia Sinica (2014-present)
• President/CEO, Behavior Design Corporation (1998-2014)
• Associate Professor, Professor, Electrical Engineering, National Tsing Hua University (1984-1998)
• Ph.D., Electrical Engineering, University of Washington (1984)
• Advisor, New Series in Natural Language Processing , John Benjamins (1999-present)
• President, Asian Federation of Natural Language Processing (2011-2012)
• Executive Member, Association for Computational Linguistics (2005-2007)
• President, Association for Computational Linguistics and Chinese Language Processing (1996-1997)
Research Description meaningful problem solving interpretation.
My research interests include Machine Translation, Machine I have also worked on Chinese word segmentation with new
Reading, Chinese Language Processing, Natural Language generative models (also with additional resources later), and have
Understanding, Statistical Language Modeling and Machine obtained the best performance to be reported on the SIGHAN
Learning. I have worked on statistical modeling and machine close/open-test. Performance on this test is classified based
learning for natural language processing (NLP) since 1986. In on whether resources other than the training-set data, such as
this work, I mostly adopted principled approaches to solving NLP dictionaries, could be adopted.
problems, in contrast to the prevailing techniques of simply adding
a lot of features under the Maximum Entropy framework. Currently, I am beginning to build a platform to explore new
knowledge from multiple documents. Traditional single-document
My experience mainly lies in building a statistical semantic machine information extraction (IE) does not utilize redundancy and
translation system. I proposed and built a statistics-oriented interaction between the content that may be achieved by extraction
framework for machine translation, which automatically obtains from multiple documents. As a result, the quality of the information
associated parameters from the given pairs of Source and Target directly extracted from an individual document is relatively poor,
Semantic-Normal-Forms via semi-supervised learning. In addition, and knowledge that could be deduced from the interactions
I proposed a joint model to integrate Translation Memory into a between information extracted from various documents/domains is
phrase-based machine translation system during decoding, which not explored. This project will develop a generic cross-document
made significant improvements to the system. Recently, I have built analysis package to refine the noisy information extracted from
a goal-oriented meaning-based statistical framework for solving traditional IE, and to explore new knowledge cross domains.
English multi-step math word problems, which resembles the
human cognitive understanding of problems and produces a more Transactions on Asian Language Information Processing, volume 14,
number 3, pages 12:1-12:29, June 2015.
Publications 6. Kun Wang, Chengqing Zong, and Keh-Yih Su, “Dynamically
Integrating Cross-Domain Translation Memory into Phrase-Based
1. Chao-Chun Liang, Shih-Hong Tsai, Ting-Yun Chang, Yi-Chung Machine Translation during Decoding,” Proceedings of COLING
Lin and Keh -Yih Su, “A Meaning-bas ed Englis h Math W ord 2014, August 2014.
Problem Solver with Understanding, Reasoning and Explanation,” 7. Kun Wang, Chengqing Zong, and Keh-Yih Su, Aug. 4-9, 2013,
Proceedings of COLING 2016, Osaka, Japan, December 2016, System “Integrating Translation Memory into Phrase-Based Machine
Demonstration. Translation during Decoding”, Proceedings of ACL 2013, Sofia,
Bulgaria.
2. Chao-Chun Liang, Kuang-Yi Hsu, Chien-Tsung Huang, Chung-Min 8. Yufeng Chen, Chengqing Zong, and Keh-Yih Su, “A Joint Model
Li, Shen-Yun Miao and Keh-Yih Su, “A Tag-based Statistic English to Identify and Align Bilin gual Named Entities,” Jo urn al of
Math Word Problem Solver with Understanding, Reasoning and Computational Linguistics, Vol.39, No. 2, June 2013, pages 229-266.
Explanation,” Proceedings of IJCAI 2016, New York City, New York, 9. Xiaoqing Li, Kun Wang, Chengqing Zong, and Keh-Yih Su, Dec. 8-15,
U.S.A., July 2016, System Demonstration. 2012, “Integrating Surface and Abstract Features for Robust Cross-
Domain Chinese Word Segmentation”, Proceedings of COLING 2012,
3. Yi-Chung Lin, Chao-Chun Liang, Kuang-Yi Hsu, Chien-Tsung Mumbai, India, pages 1653-1669.
Huang, Shen-Yun Miao, Wei-Yun Ma, Lun-Wei Ku, Churn-Jung Liau, 10. Kun Wang, Chengqing Zong, and Keh-Yih Su, “Integrating Generative
Keh-Yih Su, “Designing a Tag-Based Statistical Math Word Problem and Discriminative Character-Based Models for Chinese Word
Solver with Reasoning and Explanation,” International Journal of Segmentation,” ACM Transactions on Asian Language Information
Computational Linguistics and Chinese Language Processing, volume Processing, Vol.11, No. 2, Article 7, June 2012, pages 7:1-7:41.
20, number 2, pages 1-26, December 2015.
4. Chien-Tsung Huang, Yi-Chung Lin, and Keh-Yih Su, “Explanation
Generation for a Math Word Problem Solver,” International Journal of
Computational Linguistics and Chinese Language Processing, volume
20, number 2, pages 27-44, December 2015.
5. Xiaoqing Li, Chengqing Zong, and Keh-Yih Su, “A Unified Model for
Solving the OOV Problems of Chinese Word Segmentation,” ACM
88 研究人員 Research Faculty
