AUTOR:     Damir Æavar
AFILIACJA: Uniwersytet w Zadarze (Chorwacja) / Indiana University (USA) /
           Institut za hrvatski jezik i jezikoslovlje (Chorwacja)
TYTU£:     On inducing lexical properties with probabilistic methods

STRESZCZENIE:

In this talk we will discuss various computational methods and
experiments that utilize probabilistic and information theoretic
measures over single and groups of lexical items, based on corpora
from various languages (including the Croatian Language Corpus at the
Institute of Croatian Language and Linguistics).  On the one hand, we
will tackle the question to what extent lexical categories could be
reconstructed on the basis of distributional properties.  On the
other hand, we will discuss the possibilities to reconstruct or
simulate syntactic behavior and typing of lexical elements via
information theoretic concepts and methods.