Campus Event Calendar
Event Entry
New for: D1, D2
What and Who
Combining Linguistic and Statistical Information
Rens Bod
Univ. Amsterdam
Talk
AG 1, AG 2
AG Audience
Note: We use this to send email in the morning.
Date, Time and Location
Monday, 4 May 98
14:00
-- Not specified --
43.1 - DFKI
+1.01 (Turing)
Saarbrücken
Abstract
When computational models of language processing are not constructed in a
purely linguistic context, but aim at being relevant for practical
applications or for psychological theory, they ought to cope with
problems like ambiguity and robustness. For this reason, increasingly
many models take statistical properties of a sample corpus into account
when they process new input. In this talk, I will argue for a statistical
approach to language processing, called Data-Oriented Parsing ("DOP"),
which works with representations of concrete past language experiences,
rather than with abstract grammar rules. Models that instantiate this
approach maintain large corpora of linguistic representations of
previously occurring utterances. They operate by decomposing the given
representations into fragments and recomposing those pieces to analyze
new utterances. The occurrence frequencies of the fragments are used to
determine which is the most probable analysis of an utterance.
purely linguistic context, but aim at being relevant for practical
applications or for psychological theory, they ought to cope with
problems like ambiguity and robustness. For this reason, increasingly
many models take statistical properties of a sample corpus into account
when they process new input. In this talk, I will argue for a statistical
approach to language processing, called Data-Oriented Parsing ("DOP"),
which works with representations of concrete past language experiences,
rather than with abstract grammar rules. Models that instantiate this
approach maintain large corpora of linguistic representations of
previously occurring utterances. They operate by decomposing the given
representations into fragments and recomposing those pieces to analyze
new utterances. The occurrence frequencies of the fragments are used to
determine which is the most probable analysis of an utterance.
During the last few years, DOP models for various kinds of linguistic
representations have been developed, ranging from tree representations,
compositional semantic representations, and lexical-functional grammar
representations. In this talk, I will go into the computational aspects
of these DOP models and discuss a number of experiments which were
carried out in the context of a practical spoken dialogue system called
OVIS. I will also go into some of the conceptual issues that this
data-oriented approach brings to light.
Contact
Corinna Johanns
302-5282
--email hidden
passcode not visible
logged in users only
Uwe Brahm, 04/12/2007 12:01 -- Created document.