ECE Special Seminar: Young-Bum Kim (Microsoft) – “Unusually supervised NLP”
Speaker : Dr. Young-Bum Kim
In this talk, I will present my work on cross-lingual supervised learning. The key idea is that by jointly modeling a broad array of languages, apparent ambiguities can be resolved by building generic and universally plausible models of human language. I will talk about the application of this idea to several longstanding problems in NLP, including part-of-speech induction, morphological induction, grapheme-to-phoneme prediction, and computational decipherment of lost languages.
I will also present a novel method for rapid proto-typing and efficient construction of natural language understanding systems. The key insight is that even small amounts of annotated data can yield powerful results when the examples to be labelled are chosen carefully. I will talk about two novel methods to achieve this goal, one based on matrix factorizations and the other based on a notion of feature coverage.
If time permits, I will also present the new transfer learning techniques for disparate label sets. In natural language understanding (NLU), a user utterance can be labeled differently depending on the domain or application (e.g., weather vs. calendar). Standard domain adaptation techniques are not directly applicable to take advantage of the existing annotations because they assume that the label set is invariant. I will talk about a solution based on label embeddings induced from canonical correlation analysis (CCA) that reduces the problem to a standard domain adaptation task and allows use of a number of transfer learning techniques. I will also talk about a new transfer learning technique based on pre-training of hidden-unit CRFs (HUCRFs).
Dr. Young-Bum Kim is a Scientist in the Language Understanding and Dialog System group at Microsoft. He is building language understanding, dialog system and machine learning capabilities into Microsoft products such as Cortana, Xbox One and Windows 10.
Kim received a Ph.D. in computer science from the University of Wisconsin-Madison. His dissertation work, which focuses on multilingual statistical models and the optimal data set selection, received the EMNLP Plenary Award honorable mention and UW-Madison research award. He also organized an ACL shared task on named entity recognition in informal text.