This is to announce the availability (for research and teaching purposes)
of the µ-TBL system -- a software package that uses the search and database
capabilities of the Prolog programming language to implement a generalized
form of transformation-based learning. The µ-TBL system is designed to be:
General
The system supports four kinds of transformational operators
(four kinds of rules) by means of which not only traditional
'Brill-taggers', but also Constraint Grammar disambiguators,
are possible to train.
Easily extendible
Through its support of a compositional rule/template formalism
and 'pluggable' algorithms, the system can easily be tailored
to different learning tasks.
Logically transparent
Rules have a declarative, logical semantics which, among other
things, has proven to be of great value during the implementation
work.
Efficient
A number of benchmarks have been run which show that the system
is fairly efficient -- an order of magnitude faster than Brill's
contextual-rule learner. With another choice of algorithm,
the speed improvement is around two orders of magnitude.
Interactive
Prolog is an interactive programming environment and this is
something that the µ-TBL system inherits.
Small
Thanks to the choice of implementation language, the system's
code base can be kept quite small. Indeed, a 'light' version
of the µ-TBL system, consisting of just one page of Prolog code,
has been implemented.
The relevant URL is: http://www.ling.gu.se/~lager/mutbl.html
You may download papers and software, and there are three example applications
to experiment with:
o Training a Brill-tagger
o Training a Constraint Grammar disambiguator
o Training a NP-chunker
Send mail to Torbjorn.Lager@ling.uu.se if you want to be notified of further
developments of the software.
Best regards,
Torbjörn Lager
Department of Linguistics
Uppsala University
Sweden
Homepage: http://www.ling.gu.se/~lager/