Title: GLARF and the 2nd Stage of Parsing: Combining Parsing, SRL, NE tagging, Temporal Tagging, ...
Speaker: Adam Meyers (NYU)
Time: 2:15pm-3:30pm, Friday, March 25
Place: Room 4102, CUNY Graduate Center, 365 Fifth Ave (34str&35str).
		

Abstract:

Over the last decade, we have built a system for producing second
stage parses that combine the output of parsers, NE taggers, semantic
role labelers and other transducers to produce a single
theoretically-consistent representation. The resulting output includes
predicate argument dependencies that connect all the words in the
sentence in a single graph. This approach contrasts with that of
Semantic Role Labeling programs which link verb predicates to their
arguments, but (typically) ignore relations anchored by other parts of
speech (nouns, adjectives, etc.). Other systems that combine different
types of annotation (Ontonotes, MASC, etc.) take an approach that
links annotation at the token and/or character level, without changing
any of the input. In contrast, GLARF imposes the consistency of a
GLARF-based theory on the data, making tokenization and constituent
boundary decisions when the data sources conflict -- errors in the
data are corrected in the process. English GLARF achieves F-scores
ranging from 77% for spoken language telephone transcripts to 88% for
correspondence and news text, only slightly lower than parsing scores
for the same types of data. This talk will discuss how GLARF can be
used by the NLP community; the theoretical context in which GLARF was
created; GLARF systems that process Chinese and Japanese; as well as
new features we are adding to GLARF to support temporal and
causational analysis.

The first public release of the English version of GLARF software
occurred on January 13, 2011. The GLARF website:
	http://nlp.cs.nyu.edu/meyers/GLARF.html
includes more information on GLARF as well as instructions for
downloading and using English GLARF.

Speaker's Bio:

Adam Meyers is a research assistant professor at New York University,
specializing in several sub-fields of computational linguistics
including: the manual and automatic creation of linguistic resources
(lexicons, corpus annotation); machine translation; noun phrase
processing and knowledge-based methods.