Lexicon-grammars are used in applications such as information extraction (Guarasci et al. 2020) and opinion mining (Elia et al. 2015). They allow for a formalization of syntactic constructions that has been proposed as a framework for sentence parsing (De Bueriis, Langella, 2019).

Lexicon-grammars are lexical databases built by linguists (La Valley 2020; Mota et al. 2018) A lexicon-grammar records, in full size, the elementary syntactic structures of its entries. A lexicon-grammar of verbs and other predicative elements (adjectives, nouns and adverbs) has been constructed for several languages in the form of two-dimensional tables that store their properties (for example: an excerpt of table 38LH). These tables comprise a significant number of entries (in the thousands for each language). They result from the implementation of a common set of linguistic principles:

• Meanings are carefully distinguished (examples: miss, check, crack up) based on distributional and transformational criteria;
• The elementary sentence is the minimal unit of study, and the form of the sentence is schematically represented by a predicative form and a characteristic set of arguments;
• In particular, principles must allow for separating essential complements (subjects and objects) from non-essential complements (adverbials, circumstantial complements);
• A set of syntactic transformations is experimentally established and tested on an extended vocabulary in order to assess the reproducibility of acceptability judgments (Elia et al. 2011).

When these general principles are respected, experience has shown that different individuals or teams can arrive at identical evaluations, which ensures the cumulative nature of descriptions. This point is crucial for symbolic approaches to natural language processing: the amount of data that must be accumulated and represented in a consistent model is such that many research and development teams need to cooperate, and their results must be such that they can be merged without requiring rewriting substantial parts of the grammar and lexicon of each language. Meeting these conditions is in no way trivial: projects of constructing non-lexicalized phrase-structure or dependency grammars tend to show that their construction is not cumulative; besides, there is no example of such grammars which is of significant size and not the work of a single specialist.

The Lexicon-Grammar method starts from building data tables describing syntactic-semantic features, instead of giving priority to grammar at the expense of the lexicon. Indeed, the separation between grammar and lexicon in linguistic description is counterproductive: there are often more differences between two sentences than common traits, which is why attempts to build general rules often prove ineffective.

The theoretical framework adopted for this enterprise is that of the syntactic transformations of Z.S. Harris. The level of formalization used by this theory is satisfactory: syntactic parsers have been implemented as early as the 1960s.

The notion of Lexicon-Grammar is briefly presented in The RELEX Network, Section 4. The term “Lexicon-Grammar” was introduced in 1978. The ideas underlying Lexicon-Grammar were formulated by Maurice Gross during the 1960s (Gross 1968).

References in English on Lexicon-Grammar: 1968–2003, 2004–…
View Lexicon-Grammar tables