A semiotic metamodel for bridging lexical and formal semantics

The amount of lexical resources that are developed either as long-term repositories, or as short-term products of NLP techniques, is growing significantly, posing the problem of understanding their commonalities and their potential for reusability and interoperability. A major concern for reusability and interoperability is the ability to control, both intellectually and computationally, the semantics of the data that are contained in a lexical resource. Lexical data semantics is usually left implicit, either because there is not a shared agreement on how to represent that semantics, or because developers of lexical resources do not customarily employ formal methods. This article presents Semion, a metamodel for bridging lexical and formal semantics, thus allowing the representation of relations (e.g. annotation, reuse, mapping, transformation, etc.), which can hold between the elements of a lexicon (either established, like WordNet, or created for local purposes), and the elements of a typed logical theory, e.g. an ontology. Semion is based on a standard version of semiotics, and is applied to define metamodels for lexical resources, such as a FrameNet (Baker, Fillmore and Lowe, 1998) metamodel (OntoFrameNet), and to introduce a formal method for lexical information integration. Elsewhere (Gangemi, 2008b), Semion has been founded on an ontology of schemata (Lakoff, 1987, Johnson, 1987, Langacker, 1990, Talmy, 2003), where schemata are considered as invariances that emerge from the co-evolution of organisms and environment, and that are exemplified by neurobiological, cognitive, linguistic, and social constructs. That foundation allows to liaise lexical and formal semantics with cognitive linguistics and construction grammar. However, it is not described here for space reasons. While specific relations between individual ontologies and lexica are addressed in literature quite often, e.g. (Gangemi, Guarino, Masolo and Oltramari, 2003, Buitelaar, Choi, Gangemi, Huang and Oltramari, 2007, De Luca, Eul and Nürnberger, 2007, Scheffczyk, Baker and Narayanan, 2008, Huang, Calzolari, Gangemi, Lenci, Oltramari and Prevot, 2008), it is far less usual to propose a metamodel to formally represent the liaison between lexical and formal semantics. Metamodels have been proposed to abridge different lexical resources, starting with OLIF (McCormick, Lieske and Culum, 2004), and recently with reference to lexical semantics, as in LMF (Francopoulo, George, Calzolari, Monachini, Bel, Pet and Soria, 2006), where an attempt has been made to informally align some lexica under a same