An Architecture for a Federation of Highly Heterogeneous Lexical
Information Sources

				 Artola, X.
		  Dept. of Computer Languages and Systems
		      University of The Basque Country
		 649 p.k., 20080 Donostia (Basque Country)
			     jiparzux@si.ehu.es

				 Soroa, A.
	     Dept. of Comp. Science and Artificial Intelligence
		      University of The Basque Country
		 649 p.k., 20080 Donostia (Basque Country)
			     ccpsoeta@si.ehu.es


The purpose of this paper is to present an integrating architecture for a
federation of highly heterogeneous lexical information sources. The term
lexical information source is used here in its broadest sense, and so it
includes from very different lexical and dictionary databases, up to
heterogeneously structured and coded electronic dictionaries or
language-processing lexica, or even language-processing programs such as
lemmatisers or part-of-speech taggers. Our proposal is that lexical and
dictionary stores can be brought together in an integrated federation of
databases (used also in its broadest sense), so that we can query them by
using a unique language, regardless of the particular resource to which the
query will be finally targeted. As desirable as it might be, it is
unrealistic to expect that the great variety of available lexical
information resources could be converted into a single and standard
representation schema in the near future. Hence, our approach aims for a
solution that does not require any conversion of the sources, and that
participates of the emerging "wrapper" technologies that enhance external
communication with legacy systems.

We address the problem of querying very different sources of lexical
information using for this purpose a unique and common query language from
the point of view of information integration. The so-called local-as-view
paradigm is used for describing each lexical source as a view over a general
conceptual model (GCM). So, it is necessary to specify a GCM to describe
lexical knowledge, a source conceptual model (SCM) for each lexical source
expressed much in the way of the GCM, as well as the manner to describe the
content of each source in terms of the classes and relationships of the GCM
source content descriptions. The GCM and some of the SCMs have been
specified and implemented using a description logic language NeoClassic, and
an answering algorithm that translates queries from the general model into
each particular schema has also been implemented.

Currently two lexical resources have been integrated in our system. One is a
TEI-conformant general purpose Basque monolingual dictionary, locally stored
as a collection of SGML documents. The second is a pure lexical database
called EDBL, used in language-processing tasks and stored in a classic
relational DBMS.