We are pleased to present two talks on September 16, 2011.


Time: 215pm-3pm, Friday, Sept 16, 2011
Place: Room 4102, CUNY Graduate Center, 365 Fifth Ave (34str&35str).
Title: Cross-Domain Bootstrapping for Named Entity Recognition

Speaker: Ang Sun (New York University)

We propose a general cross-domain bootstrapping algorithm for domain
adaptation in the task of named entity recognition. We first
generalize the lexical features of the source domain model with word
clusters generated from a joint corpus. We then select target domain
instances based on multiple criteria during the bootstrapping process.
Without using annotated data from the target domain and without
explicitly encoding any target-domain-specific knowledge, we were able
to improve the source model’s F-measure by 7 points on the target

Ang Sun is a PhD candidate in Computer Science at New York University,
working with Prof. Ralph Grishman. His work at NYU focuses on
combining semi-supervised learning and un-supervised learning for
relation extraction and named entity recognition. He received his MS
degree in Computer Science from NYU in 2010, and is expected to get
his PhD degree in May 2012.


Time: 3pm-345pm, Friday, Sept 16, 2011
Place: Room 4102, CUNY Graduate Center, 365 Fifth Ave (34str&35str).
Title: Syntactic Approach vs. Semantic Approach to Temporal Slot Filling

Speaker: Qi Li (City University of New York)

The goal of temporal slot filling task at NIST TAC Knowledge Base
Population track is to extract attributes and their temporal
boundaries for entities across a large scale of documents. There are
several challenges in this new task: (1) temporal information is
scattered from different documents; (2) Temporal information usually
includes a lot of uncertainty and vagueness; (3) Temporal expressions
may play ambiguous roles in the relation between an entity and its
attributes. This requires techniques such like cross-document entity
co-reference, temporal classification, and information aggregation to
reduce temporal uncertainty. In this talk, we will present and compare
various approaches that use deep dependency parsing based kernel
methods, semantic features and surface features based classification.
We will also present detailed analysis on the remaining errors and
point out the possible research directions for this new pilot study.
This is based on joint work with Javier Artiles, Taylor Cassidy,
Suzanne Tamang and Heng Ji.

Qi Li is a second year Ph.D. student in Computer Science department of
the Graduate Center, City University of New York, working with
Professor Heng Ji. His current research interests include temporal
information extraction (IE), global inference for IE and transfer