OTKit: Tools, Tricks and Techniques for Optimality Theory. 31st TABU Dag, June 3-4, 2010, Groningen, Netherlands. See also http://www.tabudag.nl. Poster.
Lauri Karttunen demonstrates in an article, whose title is already
provocative ('The Insufficiency of Paper-and-Pencil Linguistics: The
Case of Finnish Prosody', 2005, ROA-818), that even the best linguists
are prone to make errors when solving a too complex problem in
Optimality Theory (OT). In fact, listing all the candidates is often
not feasible due to their very large or infinite number; and so
linguists might leave out some of them, which escape their expert
intuition but are crucial for the analysis. These forms can be
"absurd", but prove to be optimal (most harmonic) unless the
analysis is refined. Consequently, Karttunen concludes, one needs
software that is able to find the best candidate without
enumerating all of them.
We present the advantages and disadvantages of using
OTKit: Tools for Optimality Theory, a new, Java-based package
freely available at http://www.birot.hu/OTKit/. Unlike most
software used by the OT community, OTKit is able to handle an
infinite candidate set. You do not have to list all the candidates,
and constraints can be defined without specifying explicitly for each
form the number of violation marks assigned. However, finding the best
candidate in an infinite candidate set is not simple: without being
led or misled by the linguist's intuition, computers need smart algorithms
(finite-state technology, dynamic programming,...). Or the use of an
error-prone heuristic technique (simulated annealing), as it
happens in OTKit.
The talk will introduce the basic components of the user interface of
OTKit, ignoring the script language and the Java library. It will turn
out that intuitively simple OT notions must be decomposed in order to
formalize the computational problem. The result is, hopefully, a tool
that can benefit linguists in their research. The same linguists will
certainly benefit the further development of OTKit by providing feedback.