Natural Language Processing Group: U of U CS
Natural Language Processing Research Group
-
-
Calvin: "I like to verb words."
Hobbes: "What?"
Calvin: "I take nouns and adjectives and use them as verbs. Remember when `access' was a thing? Now it's something you do . It got verbed."
Calvin: "Verbing weirds language."
Hobbes: "Maybe we can eventually make language a complete impediment to understanding."
- Calvin & Hobbes, by Bill Watterson
NLP Group Meeting Schedule
Current Research Projects
SUNDANCE: Practical, Conceptual
Sentence Analysis
We are currently building a new conceptual sentence analyzer called
Sundance (Sentence UNDerstanding ANd Concept Extraction). Sundance is
designed as a partial parser that is robust in the face of
ungrammatical or ill-formed input, but can produce conceptual case
frames to represent the meaning of sentences in a particular domain.
One of the applications for Sundance is information extraction
, which involves recognizing and extracting specific types of
information from text.
Suggestions for Improvements to current Sundance Release can be
entered here:
Sundance Suggestion Log
The motivation for Sundance is that it will be a conceptual NLP system
that can be easily adapted for different domains. Consequently, we
have several ongoing projects in developing corpus-based methods for
knoweldge acquisition. We developed the AutoSlog system that
generates dictionaries of extraction patterns automatically using an
annotated corpus. A newer version, AutoSlog-TS, can generate
domain-specific extraction patterns using only preclassified texts as
input. We have also developed a corpus-based approach for building
semantic lexicons.
We are investigating methods for using NLP techniques to improve
performance in text classification and other information retrieval
applications. We have demonstrated that using an underlying
information extraction system can produce phrases that achieve higher
precision than individual words in some domains. We also have ongoing
projects involving other IR applications such as topic segmentation.
Click here for list of current and former NLP students at Utah!
RuSpace.ru - Russian Social Network
Mail: riloff@cs.utah.edu
for more information.
Corpus-based Knowledge Acquisition