|
Courses and Workshops
|
| Title: | Acquisition of lexical and terminological information from corpora |
| Lecturer(s): | Ulrich Heid (Universitaet Stuttgart) |
| Type: | Introductory Course |
| Section: | Language and Computation |
| Week: | First | | Time: | 14.00-15.30 (Slot 3) |
| Webpage: | http://www.ims.uni-stuttgart.de/~uli/
|
| Reader: | Course will not have a reader | | Room: | C2 |
Description We give an overview of corpus-based methods and tools for lexical
acquisition using the following applications as examples:
* the creation of dictionaries for NLP (acquisition of
subcategorization and distribution data for words, as well as of
evidence for lexical semantic classes (e.g. mass vs. measure
nouns));
* corpus-based updating of printed dictionaries (the above plus
collocation acquisition);
* terminology and specialized dictionaries (multiword terms, roles
in a frame-based term description, concept relations).
After an overview of necessary corpus properties (`representativity',
metadata, annotations, representation), emphasis is on the use of
linguistically annotated corpora for (a) the testing of linguistic
hypotheses (grammar and lexicon), (b) large-scale acquisition of
individual phenomena (methods leading to higher precision vs. higher
recall), (c) acquisition methodologies for shallow vs. for richly
annotated corpora (e.g. multilevel corpora, i.e. corpora with
annotations from different levels of linguistic description).
We then discuss query and retrieval from corpora and linguistic and
statistical analysis from different angles:
* query languages for word-based annotation, for (recursive)
chunking, for syntax trees and for multilevel corpora; (research)
questions on the combined use of linguistic search criteria and
metadata:
* automating query and extraction in embedded tools (e.g. for term
extraction and dictionary updating);
* combining linguistic search and statistical filtering (e.g. in
collocation extraction).
The following tools can be made available to participants (all un
UNIX/LINUX, partly on WINDOWS):
* the IMS Corpus Workbench (CWB) for word-based query and
chunk-based query (corpora for DE, EN, FR, IT, NL);
* the TIGERSearch treebank query tool (DE corpora);
* the NXT Search tool for multilevel corpora (developed in the EU
project NITE, likely with DE and EN corpora).
|
© 2003, 2004 LORIA, Carlos Areces. |
2003-11-20
| |
