Natural Language Text Generation from Knowledge Bases of OSTIS-systems

Abstract

The article discusses an approach towards generating coherent natural language texts from fragments of ostis-system knowledge bases. The architecture of the abstract sc-agent of translating a fragment of the knowledge base into a natural language is described. The task of generating a natural language text is subdivided into three sub-tasks: structure fi ltering, knowledge base fragment decomposition, and generating an equivalent natural language text. Two ways of linearizing a knowledge base fragment are proposed: one based on a predefi ned order specifi cation and an algorithmic one. Generation of an equivalent natural language text is proposed to be done in two stages. Firstly, an intermediate rough natural language representation is generated using rule-based mappings between sc-text constructions and their corresponding natural language verbalizations. Secondly, the intermediate representation is converted into a coherent natural language text with the help of a large language model. Finally, three possible applications of the proposed approach are described.