A Brief Summary of Research Interests
I am a former graduate student of Michael Collins at MIT CSAIL. In previous work, I have received a B.S., M.Eng, and Ph.D in EECS at MIT. My research interests include structured classification, discriminative methods, and semi-supervised learning.
I am an alumnus of MIT CSAIL, advised by Michael Collins. I am interested in discriminative modeling for structured classification and, more recently, in semi-supervised learning methods for NLP.
2 Dependency parsing
My research has focused on dependency parsing, a syntactic formalism whose grounded, lexicalized nature makes it an attractive target for feature-rich discriminative models. At the same time, dependency parsers are able to recover core head-modifier relationships.
3 Structured linear models
Structured linear models are a discriminative classification framework composed of three components: (1) a factorization, which specifies a decomposition of structured labels into sets of parts, allowing efficient decoding and inference; (2) a feature mapping, which represents each part as a vector of real-valued features; and (3) a parameter estimation algorithm, which learns a weighting for each feature.
Each component can be improved in a modular fashion. For example, the factorization can be improved by using a second- or third-order parser instead of a first-order parser, features can be improved by adding cluster based information, and parameter estimation can be adjusted by selecting between the structured perceptron, max-margin models, or log-linear models.
4 Future work
I am currently at Google Research under the auspices of Fernando Pereira.
T. Koo, A.M. Rush, M. Collins, T. Jaakkola, and D. Sontag. 2010. Dual Decomposition for Parsing with Non-Projective Head Automata. In Proceedings of EMNLP, pages 1288–1298. [Best paper award]. [slides]
T. Koo. Advances in Discriminative Dependency Parsing 2010. PhD thesis, MIT, June 2010. [4up] [slides]
T. Koo and M. Collins. 2010. Efficient Third-order Dependency Parsers. In Proceedings of ACL, pages 1–11. [slides] [code]
X. Carreras, M. Collins, and T. Koo. 2008. TAG, Dynamic Programming, and the Perceptron for Efficient, Feature-rich Parsing. In Proceedings of CoNLL, pages 9–16. [Best paper award].
T. Koo, X. Carreras, and M. Collins. 2008. Simple Semi-supervised Dependency Parsing. In Proceedings of ACL, pages 595–603. [clusters]
M. Collins, A. Globerson, T. Koo, X. Carreras, and P. Bartlett. Exponentiated Gradient Algorithms for Conditional Random Fields and Max-Margin Markov Networks. Journal of Machine Learning Research 9(Aug):1775–1822, 2008. [src]
T. Koo, A. Globerson, X. Carreras, and M. Collins. 2007. Structured Prediction Models via the Matrix-Tree Theorem. In Proceedings of EMNLP, pages 141–150. [src]
A. Globerson, T. Koo, X. Carreras, and M. Collins. 2007. Exponentiated Gradient Algorithms for Log-Linear Structured Prediction. In Proceedings of ICML, pages 305–312.
T. Koo and M. Collins. 2005. Hidden-Variable Models for Discriminative Reranking. In Proceedings of EMNLP, pages 507–514.
M. Collins and T. Koo. Discriminative Reranking for Natural Language Parsing. Computational Linguistics 31(1):25–69, 2005.
T. Koo and M. Collins. 2004. Parse Reranking with WordNet Using a Hidden-Variable Model. M.Eng Thesis, Massachusetts Institute of Technology, Cambrige, MA, USA.
J. Hajič, M. Čmejrek, B. Dorr, Y. Ding, J. Eisner, D. Gildea, T. Koo, K. Parton, G. Penn, D. Radev, and O. Rambow. 2002. Natural Language Generation in the Context hyphenate of Machine Translation; Section 3.3, Chapter 6 (with Jan Hajič). NLP WS-02 Final Report.