The MELODI group gathers researchers from the previous IC3 team and former members of the LiLAC team at IRIT.

MELODI members’ research connects natural language analysis and formalization on the one hand, and knowledge modeling and representation on the other hand. The team relies on a variety of research approaches, from theoretical study of the linguistic grounding of semantics and its formal representation, of interaction and discourse structures, to more experimental work based on software developments, on building resources (such as annotated corpora, lexica or ontologies) and corpus based experiments. A part of this research takes place in a knowledge engineering perspective which aims at defining tools and developing applications using these semantic features and knowledge models.

Work on natural language carried out in MELODI includes the following topics:

• Analysis and description of linguistic semantics at the lexicon, sentence and discourse levels;
• Machine learning of structured representations for Natural Language Processing.

Work on knowledge representation addresses the following issues:

• Methods and principles of ontology engineering and ontology evolution ;
• Knowledge modeling and representation with graphs and/or logics.

At the border of natural language and knowledge representations appear the following research issues:

• Building ontologies from text analyses ;
• Linguistic grounding of semantic relationships;
• Methods and tools to design applications that require the joint reference to language and to knowledge.

Reference Theories

• Representational theories: knowledge and problem solving explicit modeling, ontology.
• Semantics of linguistic actions (formal semantics)
• Theory of interaction structure: dialog games, formal theory of discourse (SDRT)
• Textual and corpus semantics
• Activity theory and usage theory

Designed methods and tools

Examples of recent results include:

• Methods and tools for corpus annotation with discourse, temporal relationships (ANNODIS ANR project)
• Learning algorithms to identify discourse units in corpora.
• Study of discourse structures in dialogues (STAC project with ERC fund).
• Identification of linguistic marks of opinions (CASOAR project).
• Extraction of semantic relations from text by using lexico-syntactic patterns (Caméléon).
• Tools for text semantic annotation and for ontology evolution management (Textviz, Dynamo and CorpusLogicistes ANR projects).
• Expression of SPARQL queries with the help of query patterns.
• Exploitation of document structure and content to build an ontology (Géonto ANR project).
• Exploitation of discourse relations and textual architecture to identify concepts and relationships in text (BQR OntoDis).
• Contribution to the DAFOE method and workbench to build ontologies and terminologies from text (Dafoe4App ANR project).
• Web Service to query ontologies (WebContent ANR project).