Machine-Readable Dictionaries, Lexical Data Bases And The Lexical System

I should like to raise some issues concerning the conversion from a traditional Marhine-Readable Dictionary (MRD) on tape to a Lexical Data Base (LDB), in order to highlight some important consequences for computational linguistics which can follow from this transition. The enormous potentialities of the information implicitly stored in a standard printed dictionary or a MRD can only be evidenced and made explicit when the same data are given a new logical structure in a data base model, and exploited by appropriate software. A suitable use of DB methodology is a good starting point to discover several kinds of lexical, morphological, syntactic, and semantic relationships between lexieal entries which would otherwise have remained unexploited. Moreover, the transformation of a "very large-scale" MRD into a LDB provides the means of operating throughout the lexicon in a really extensive manner. I think in fact that an "almost exhaustive" approach to lexical facts is essential both for reliable investigations of a lexical system, and for many kinds of linguistic applications which cannot be restricted to a particular domain of discourse. The possibility of abstracting significant regularities from recurrent patterns of natural language definitions by means of suitable computational methods, and of reaching a formalization of a number of important structuring relations within the lexicon will be discussed. An overview of the "associative links" already produced in the Italian LDB, and of other allowable interconnections will be given. In a "relational" organization of a computerized dictionary with complex interlinked structures, each word acquires its meaning as a result of its position in some of the partitionings created by the formalized relations. When an entry is activated, all of its relations with other entries can be activated, too. Conversely, when a relation is activated, all of its linked concepts are made immediately available. Conceptual and linguistic information at many levels is thus interactively retrievable from the LDB following the appropriate pointers. I shall especially take into consideration those types of relations which can be of relevance not only for "Computational Lexicology" research, but also in a more general Computational Linguistics framework. An example is provided by derivational relationships which, when formalized, give rise to families of semantically and syntactically connected entries, linked to the same base-word node, and substitutable in different syntactic formulations of the same conceptual meanings. Another example concerns ease or argument relations, both (a) between lexieal items, and (b) governed by lexical …