Semantic Analysis through Ant Algorithms, Conceptual Vectors and Fuzzy UNL Graphs

In the context on the UNL project, we focus on the automatization of enconversion process, that is the building of UNL graphs from sentences. We present an extension of the UNL graph structure aiming at handling lexical and relational ambiguities. On this intermediate structure, we can apply ant algorithm propagation of conceptual vectors and other constraints. Graph nodes and relations have a level of excitement and when this level remains too low for too long they are deleted. This way, both acception and attachment selections can be performed.