Combining Different Knowledge Sources for Text Understanding

We describe a project under development whose objective is to build a model of natural language text understanding. This model has a form of a converter of a natural language text to its semantic structure. The project basically consists in enriching the ETAP-4 linguistic processor, developed at the Institute for Information Transmission Problems (IITP) of the Russian Academy of Sciences, with a new module – that of semantic analysis. An important feature of this module is that it uses not only linguistic knowledge incorporated in the grammar and the combinatorial dictionary, but also extralinguistic knowledge stored in the ontology and contextual information accumulated in the repository of individuals. We developed an ontology that serves as the semantic metalanguage for semantic structures. This knowledge permits to make different types of inferences, which extract implicit information. Several examples are given that show how all these knowledge sources can be used, how the meaning is represented and what inferences can be made.