Event

Semantic Role Labeling and Kernel Methods for Question Answering and Spoken Language Understanding

Alessandro Moschitti

University of Trento

Tuesday, October 28, 2008
2:00 pm

A deep approach to Natural Language Understanding has been proven to be infeasible from a computational complexity point of view. This has led to the development of shallow methods for semantic processing, where Semantic Role Labeling (SRL) systems are the most recent example of such research trend. However, the noise and errors in the data automatically produced by SRL and other natural language parsers prevent logic-based approaches from effectively carrying out semantic reasoning.

Machine learning approaches have been shown to be robust to noise but they require expertise, intuition and deep knowledge about the target problem to convert semantic structures into attribute-value representations. Kernel Methods are powerful techniques, which can simplify the data representation modeling, by encoding semantic information at a more abstract level than the usual attribute-value method.

In this talk, after briefly introducing the theory of kernel methods and Support Vector Machines, we will show the use of different kernels for complex semantic processing such as Spoken Language Understanding and Question Answering.

In particular, we will show (a) the use of kernels for improving standard spoken language understanding, (b) the design of a FrameNet system for spoken dialog data, which demonstrates that SRL annotation can be automatically generated for spoken dialog systems and (c) the effective use of SRL for Question/Answer Classification.

Speaker Bio:

Alessandro Moschitti is an Assistant Professor at the Information Engineering and Computer Science Department of the University of Trento. In 1998, he graduated from the University of Rome "La Sapienza" with a Master Degree in Computer Science, and then in 2003 he obtained his PhD in Computer Science at the University of Rome "Tor Vergata". Between 2002 and 2004, he worked as an associate researcher in the University of Texas at Dallas for two years.

His expertise concerns machine learning approaches to Natural Language Processing, Information Retrieval and Data Mining. In particular, he has designed applications of supervised and unsupervised learning for Text Categorization, Named Entity Recognition, Co-Reference Resolution, Text Summarization, Textual Entailment Recognition, Question Answering, Semantic Role Labeling and Spoken Dialog Systems. He has recently devised innovative kernels within Support Vector and other kernel-based machines for advanced syntactic/semantic processing. His work has been published in the major conferences of different research communities, e.g. ACL, ICML, CIKM and ICDM.