CSAIL Event Calendar: Previous Series

"What's the Meaning of This?" Detecting Word Senses for General-Purpose vs. Domain-Specific WSD

Speaker: Anna Rumshisky , Postdoctoral Research Fellow, CS Dept., Brandeis Univ
Date: January 6 2011
Time: 2:00PM to 3:00PM
Location: Kiva 32-G449
Host: Peter Szolovits, Clinical Decision Making, CSAIL

Contact: Fern, fern@csail.mit.edu

Knowledge acquisition is one of the main challenges in NLP. Typical strategies include the development of annotation schemes and annotated corpora, mining unstructured text data for the relevant information, and using distributed environments to glean sophisticated information from users. In this talk, I will explore the issues of knowledge acquisition as it applies to the problem of word sense definition and detection. I will begin by looking at lexical resource development for general-purpose WSD, where word senses are often fuzzy, determined in composition, and extensible "on the fly". I will discuss the manual identification of context patterns responsible for sense detection and the way it can be crowd-sourced, while avoiding the usual pitfalls of sense inventory construction. I will then examine the problem of identifying lexical items that activate the same sense of a predicate in composition. Such lexical items are often semantically diverse, and their similarity is only relative to a given
context. I will describe a method for resolving this problem in a fully unsupervised setting, using the notion of contextualized similarity to model sense differentiation. Finally, I will touch on domain-specific WSD, where senses are typically well-distinguished and linked to well-established structured knowledge sources maintained by domain experts. I will briefly describe a methodology for leveraging
domain-specific taxonomies for WSD using information-theoretic methods.

See other events that are part of

See other events happening in January 2011


About Us Research News Resources Directory