Final Report Summary - CONTENT (Content Across Domains: From Words to Discourse)
The main question for 20th century semantics and philosophy of language was the problem of compositionality: How can human beings understand and produce a potentially infinite set of utterances? The answer, on which modern formal semantics was based is that a sentence’s meaning should be determined by the denotation of words and the ways in which they are put together. Despite its success, in its current state formal semantics has a serious shortcoming. A complete and realistic modeling of human semantic competence also means being able to model what speakers understand sentences as saying. Thus, a semantic theory should make predictions concerning what lexical predicates imply, what a sentence or a text is about, what is its main subject matter, or how it connects to previous discourse. Formal semantics has no handle on such issues.
The main question of 21st century semantics is shaping up to be: How can we model the information that sentences convey in tandem with compositionality? The Marie Curie CIG CONTENT project (project #618550) contributes to this research agenda. It combines the strengths of traditional linguistics (in particular formal semantics) and natural language processing in computer science in order to develop more sophisticated theories of semantic content at the word, sentence and discourse level. Methodologically, it aims at enlarging the set of theoretical tools both in formal semantics and in computational (distributional) semantics. It is hoped that by developing novel methods towards semantic content and by building bridges with the computer science community this research can open new theoretical perspectives as well as contribute to the development of further technical advances in the field of natural language processing. Such innovations could help various natural language tasks become much more efficient and thus might have a significant impact for a number of automatic language processing tasks as well as information retrieval.
The project resulted in 11 scientific publications by the PI and a number of conference presentations. The project has also contributed to the principal investigator obtaining a tenured academic position within the CNRS. A highly successful international workshop, organized by the principal investigator, was held in February 2015 in Toulouse. The project has contributed to the visibility of the research on formal and distributional semantics in France, and also to the strengthening of collaborative ties between the MELODI group at IRIT, Université Paul Sabatier, The Institut Jean Nicod (ENS) in Paris and other leading centers of computational and theoretical linguistics in France and Europe. Some of the achievements of the project were also disseminated towards the larger public during a European Researcher’s Night held in 2015 at the Cité de l’Espace in Toulouse.
The main question of 21st century semantics is shaping up to be: How can we model the information that sentences convey in tandem with compositionality? The Marie Curie CIG CONTENT project (project #618550) contributes to this research agenda. It combines the strengths of traditional linguistics (in particular formal semantics) and natural language processing in computer science in order to develop more sophisticated theories of semantic content at the word, sentence and discourse level. Methodologically, it aims at enlarging the set of theoretical tools both in formal semantics and in computational (distributional) semantics. It is hoped that by developing novel methods towards semantic content and by building bridges with the computer science community this research can open new theoretical perspectives as well as contribute to the development of further technical advances in the field of natural language processing. Such innovations could help various natural language tasks become much more efficient and thus might have a significant impact for a number of automatic language processing tasks as well as information retrieval.
The project resulted in 11 scientific publications by the PI and a number of conference presentations. The project has also contributed to the principal investigator obtaining a tenured academic position within the CNRS. A highly successful international workshop, organized by the principal investigator, was held in February 2015 in Toulouse. The project has contributed to the visibility of the research on formal and distributional semantics in France, and also to the strengthening of collaborative ties between the MELODI group at IRIT, Université Paul Sabatier, The Institut Jean Nicod (ENS) in Paris and other leading centers of computational and theoretical linguistics in France and Europe. Some of the achievements of the project were also disseminated towards the larger public during a European Researcher’s Night held in 2015 at the Cité de l’Espace in Toulouse.