Seminar "Corpus-Based Semantics"

Katrin Erk, Alexander Koller

Thursdays, 11-13, building C7.1, room U.15

Seminar for BSc and MSc students

Classical semantic theory focuses on the assignment of semantic representations to sentences. It assumes that there is an underlying knowledge base that represents information about word meaning and world knowledge. However, it is currently unfeasible to use methods of deep computational semantics in real-world applications: It is very labour-intensive to specify a wide-coverage syntax-semantics interface by hand, and the manual construction of a useful knowledge base is nearly impossible.

A new approach that has emerged over the past few years is to derive such resources from corpora. On the one hand, assignments of predicate-argument structures to (statistical) parses can be learned from corpora such as the PropBank or FrameNet annotations. On the other hand, semantic information ranging from synonymy or hyponymy to rather complex verb relations can be learned with a surprising degree of success even from unannotated corpora.

In this seminar, the students will get to know some of the most important methods for learning semantic knowledge from corpora. The seminar is suitable both for (advanced) B.Sc. and for M.Sc. students; students who want a Hauptseminar certificate will cover more material in more depth in their presentations and seminar papers.