INSTITUTE OF INFORMATION TECHNOLOGIES - BAS

Cybernetics and Information Technologies
Volume 5, No 2. Sofia, 2005, Bulgarian Academy of Sciences


Parallel Language and Semantic Treatment in AGN

Velina Slavova*, Alona Soschen**

* New Bulgarian University, Department of Computer Science, Sofia
E-mail: vslavova@nbu.bg
** Massachusetts Institute of Technology, MIT Department of Linguistics and Philosophy
E-mail: soschen@mit.edu


Abstract: AGN is a Generalized Net model, developed for simulating the cognitive process of natural language comprehension. It allows a more global repre-sentation, on a high level of abstraction, of overall cognitive process of message acquisition. AGN imitates the cognitive system’s functions of control and coordination of the included sub-processes, some of which run in parallel. The treatment and processing of information includes joint operations in two knowledge spaces – language and semantics. The functioning of AGN can be combined with a database to represent human memory with structured knowledge. For this purpose, language is formalised and presented as Language Information System (LIS).

This leads to the development of a database structure based on human cognitive resources, semantic primitives, semantic operators, syntactic rules and data. In LIS, the grammatical rules of the language are related to operators in the semantic space. The method is applied for modelling a specific grammatical rule (secondary predication in Russian). The results of applying LIS are consistent with the stages of treatment, modelled with AGN. The processing of examples from the linguistics domain are tracked by the transitions of AGN, which perform operations in two knowledge spaces. Analysis and the tracking of these examples suggests that the mechanisms of language comprehension are strongly assisted by a Top-down information flow, based on semantic primitives and operators. The result of this formalization supports the idea that AGN model can be combined with a database, which represents the structure of memory knowledge.

Keywords: natural language processing, generalized net, parallel treatment.