CSE Distinguished Speaker Katrin Erk "Probabilistic inference over natural language semantics with weighted corpus-derived inference rules"

Abstract: The meaning of natural language sentences can be represented through logic. Traditionally, the resulting logical form has been paired with manually created taxonomies for inference. But there are many valid inferences that these taxonomies do not cover, leading to the often-raised complaint that logic-based approaches are “brittle”.
One of the greatest recent success stories of computational lexical semantics has been the large-scale extraction of lexical information from large corpora. But the resulting information is weighted and very noisy. We combine logical form representations of sentence meaning with weighted corpus-derived inference rules, and use Markov Logic Networks to perform probabilistic inference over the resulting representations. Evaluating on the task of textual entailment, we find that the corpus-derived information provides a performance boost, especially for inferences not covered by the WordNet taxonomy.

Bio: Katrin Erk is an associate professor in the Department of Linguistics at the University of Texas at Austin. Her research expertise is in the area of computational linguistics, especially semantics. In the last five years, her work has focused on distributional approaches, as well as the combination of logic and distributional representations into a probabilistic logical form.
Katrin Erk completed her dissertation on tree description languages and ellipsis at Saarland University in 2002, under the supervision of Gert Smolka and Manfred Pinkal. From 2002 to 2006, she held a researcher position at Saarland University, working on manual and automatic frame-semantic analysis.