Title: Learning for Deep Language Understanding
Speaker: Smaranda Muresan (Rutgers)
Time: 2:15pm-3:30pm, Friday, Nov 11
Place: Room 4102, CUNY Graduate Center, 365 Fifth Ave (34str&35str).
		

Abstract:

Recently, several machine learning approaches have been proposed for
mapping sentences to their formal meaning representations. However, only a
few of them integrate the semantic representation with a grammar formalism,
which is needed for linking semantics to other grammatical information, most
notably syntax. Lexicalized Well-Founded Grammars (LWFG) is a syntactic-semantic
formalism that balances expressiveness with provable learnability results.
LWFGs are suited to learning in data-poor settings. In this talk I will introduce
several tractable algorithms for LWFG learning, and I will compare them in
terms of a-priori knowledge needed by the learner, hypothesis space, algorithm
complexity and amount of annotated data needed. I will also briefly
present our ontology-based semantic interpreter that can be linked to the
grammar through grammar rule constraints, providing access to meaning during
parsing and generation. In this approach, the parser will take as input natural
language utterances and will produce ontology-based semantic representations.
I will show that even with a weak "ontological model", the semantic interpreter
at the grammar rule level can help remove erroneous parses obtained when we
do not have access to meaning.

Speaker's Bio:

Smaranda Muresan is an assistant professor in the Library and Information 
Science Department, School of Communication and Information at Rutgers 
University. She is the co-director of the Laboratory for the Study of Applied 
Language Technologies and Society, and a graduate faculty in the department of 
Computer Science. She received her PhD in Computer Science from Columbia 
University in 2006. Before coming to Rutgers she was a Postdoctoral Research 
Associate at the  Institute for Advanced Computer Studies at University of
Maryland. Her research focuses on computational models for language 
understanding and learning, as well as language in social media, and is funded 
primarily by the National Science Foundation (NSF).