Linking sense to syntax in FrameNet
Charles J. Fillmore
Professor, International Computer Science Institute and Graduate School in Linguistics, UC Berkeley

Short Abstract:
In this talk I want to show, not "rules" for linking semantic notions to their syntactic realization, but patterns of valence and realization as discovered in the Corpus-based project known as FrameNet. In this computational lexicography project,sponsored by the U.S. National Science Foundation, we proceed by (1) associating words to the conceptual structures or "frames" which underlie them, (2) discovering and characterizing the semantic roles of the constituents that they govern, if they are governing words, or the semantic roles that they fulfil if they are dependent words, (3) annotating sentences containing these words, as found in a very large Corpus (the British National Corpus), according to the semantic roles found in them, and (4) devising the means of summarizing and displaying the results of our discoveries.
The paper will explain the theoretical underpinnings of the project's work, will briefly sketch the nature of the resulting database, will demonstrate some of the results, and will outline some of the NLP applications to which information from the database can make a contribution. http://linguistics.berkeley.edu/lingdept/Current/people/facpages/fillmore.html

New Chances for Deep Linguistic Processing
Hans Uszkoreit
Professor of Computational Linguistics, Saarland University

Short Abstract:
The Grammatical analysis with linguistically designed grammars has always been a central topic of investigation in theoretical computational linguistics. However, deep grammatical analysis has played a negligible role in the development of language technology applications. Deep parsers have been lacking efficiently and robustness. Building up coverage has been slow and costly. The few linguistic grammars that truly exhibit large coverage have caused a constant fight with extensive ambiguity. In my talk I will present recent developments that provide new challenges and opportunities for linguistic methods. Increased efficiency of deep parsing and the embedding of a selective deep analysis into a robust shallow regime for information extraction offer ways to employ deep parsing in an environment where it can improve results without ruining robustness. I will argue that such a selective utilization as we have realized it by an integration of HPSG parsing into a hybrid IE system comprised of statistical and FST components, also provides a promising direction for gradual, measurable controlled progress in the development of deep grammars and lexicons. http://www.coli.uni-sb.de/~hansu/

Evolutionary Linguistics and Computer Modeling
William S.-Y. Wang
Academician, Academia Sinica

Short Abstract:
In recent years, a new approach to studying the evolution of language has been initiated by Hurford, Nowak, and others, which is importantly based on computer modeling. The topics which have been investigated include the various mechanisms for language acquisition, the maximum size of the lexicon, and the transitions into segmental phonology and into syntax. I will offer a synthetic review of the major developments in this exciting new area, as well as discuss some future possibilities. http://www.ee.cityu.edu.hk/~wsyw/