home · mobile · calendar · colloquia · 2010-2011 · 

Colloquium - Moschitti

Kernel Engineering: The Natural Language Processing Case
University of Trento, Italy

In recent years, machine learning (ML) has been used more and more to solve complex tasks in different disciplines, ranging from Bioinformatics to Software Systems, Information Retrieval or Natural Language Processing (NLP). These tasks often require the processing of structured input. For example, NLP applications critically deal with syntactic and semantic structures. To model the latter in terms of feature representation suitable for ML algorithms, expertise, intuition and deep knowledge about the target linguistic phenomenon are required.

Kernel Methods are powerful mathematical tools that can alleviate the data representation problem as kernel functions can be used to define similarities between structured objects at a more abstract level than feature vectors. Unfortunately, not all functions are valid kernels, consequently their appropriate modeling still requires noticeable creativity and expertise. To simplify such work, an engineering approach is needed. This consists in using basic and well-known kernel functions to build more complex object similarities to effectively carry out learning for the target task. With this aim, we identify three different engineering techniques: basic kernel combinations, object transformations and merging of kernels in new complex functions.

In this talk, after introducing kernel methods, we will show several examples of the techniques above to engineer effective kernels for Natural Language Applications, e.g., Semantic Role Labeling and Question/Answer Classification. Additionally, we will report experiments in which kernel-based machines often improve the state-of-the-art.

Alessandro Moschitti is a professor of the Computer Science and Information Engineering Department of the Trento University. In 2003, he took his PhD in Computer Science from the University of Rome "Tor Vergata" and he worked as an associate researcher for the University of Texas at Dallas for two years, between 2002 and 2004. His expertise concerns theoretical and applied machine learning in the areas of NLP, IR and Data Mining. He has devised innovative kernels within Support Vector and other kernel-based machines for advanced syntactic/semantic processing. These have been documented in more than 110 scientific articles, published in the major conferences of several research communities, e.g., ACL, ICML, ECML-PKDD, CIKM, ECIR and ICDM. He is also an active PC member for the conferences/journals of the areas above. He is currently guest editor of the Journal of Natural Language Engineering, an ML area co-chair for ACL2011 and a co-chair for TextGraphs 6. He has participated in six projects of the European Community (EC) and in three US projects: MTBF with Con-Edison, IQAS for ARDA AQUAINT PROGRAM and Deep QA (the Jeopardy! challenge) with IBM. Currently, he is the project consortium coordinator of the EC Coordinate Action, EternalS, and project coordinator of several Italian projects. He has received several awards, e.g., the IBM Faculty award.

Hosted by Martha Palmer and James Martin.
A video of the talk is now available.

Department of Computer Science
University of Colorado Boulder
Boulder, CO 80309-0430 USA
May 5, 2012 (14:13)