Accurate Prediction of Argument Structure
Accurate Prediction of Argument Structure using Syntactic Features
29th February 2016 | Room 432 | Battelle | 16h00
Many current state-of-the-art parsers have now reached a performance plateau of ninety percent in capturing tree structures. For these systems, the next challenge lies in breaking this ceiling. It has long been thought that the bottleneck preventing more accurate syntax-to-semantic interfaces lies in the quality of the first phase of analysis: the better the parse, the better the output. In fact, most of the structures used to train current state-of-the-art parsing models today are simplified versions of a more informative data set --- the Penn Treebank --- which is often stripped of its richer set of annotations (i.e. traces and functional labels are removed), while, for reasons of efficiency and availability, projective dependency trees are often given preference over richer graph structures. This simplification has led to the emergence of surface syntax-based parsers whose output cannot directly be used to extract full-fledged predicate-argument structures. For example, control verb constructions, cleft structures, argument sharing in ellipsis coordination, are among the phenomena requiring a graph to be properly accounted for.In this talk, I will present transition-based and high-order parsing methods to predict deep syntactic graphs and show that we need to have an informative feature set to get more accurate predictions. Finally, I will present an overview of semantic representations and talk about some of the challenges they raise.
16 février 2016