Type of Talk: AI

Speaker Name: Katrin Erk

Speaker Affiliation: Department Of Linguistics University of Texa

s at Austin

Date: Friday November 10 2006

Start Time: 11:

00 a.m.

Location: ACES 2.402

Host: Raymond Mooney

Ta

lk Title: Automatic Meaning Analysis of Free Text: Small Steps Towards a B

ig Goal

Talk Abstract:
Viewed as a whole the problem of doing an
automatic meaning analysis of free text is huge. But maybe the problem can
be carved up in more manageable pieces: Recently several approaches to an

automatic predicate-argument structure analysis have been proposed what ha

s been called a who does what to whom analysis. This can be seen as a first
building block in a modular meaning analysis (where other very much neces

sary building blocks would include negation and modals). It is an importan

t building block which focuses on lexical semantics and on the link to sem

antic taxonomies -- and a building block that has recently become much more
accessible with the availability of manually annotated corpora.

In
this talk I first take a closer look at the data for predicate-argument st

ructure from the viewpoint of a manual annotation effort where we annotat

ed a German corpus with FrameNet-style information. I then present a system
for automatic predicate-argument structure analysis its architecture and

the statistical modeling for its subtasks. Lastly I discuss a study on cro

ss-lingual semantic analysis -- which opens up the possibility of deriving

cross-lingual paraphrases.

Speaker Bio:
Katrin Erk is an assista

nt professor in the Department of Linguistics at the University of Texas at
Austin. She completed her dissertation on tree description languages at Sa

arland University in 2002 advised by Gert Smolka. From 2002 to 2006 she h

eld a researcher position in Saarbruecken working with Manfred Pinkal. Her

current work includes research on machine learning methods for semantic ana

lysis the acquisition of lexical information from corpora manual semantic
annotation the detection of multiword expression and computational model

s for word sense.