Solution of the answer formation problem in the question-answering system in Russian

摘要

The question-answering systems were being investigated for several decades, but the majority of researches were carried out in English. The subject of this paper is the knowledge-based question-answering system. The unique mathematical model describes the process of answering when the question is presented in Russian as a natural language. The model is executed by mapping the question to the existing structure of the offered knowledge base. Question mapping in the natural language to any logical form is performed by using the semantic-syntactical analysis under the conditions of a limited annotated semantic corpus in Russian. There exist loads of approaches to execute the syntactic, semantic-syntactic, semantic analysis and approaches to create the answer, but these approaches are not fully transferrable into Russian. The paper describes features of implementation of the semantic-syntactical analysis in Russian with using SRL algorithm, and features of answers creation. The described rules are based on the ontology offered by A. Grasser and containing 18 categories to determine the category of a question and to retrieve relations from the text. The paper shows the results of experiments in forming the answers for ifferent subject domains, including technical texts from the journal “Izvestiya SPbGETU "LETI"”, books of Turgenev, and results of the user's requests to a search engine. The directions of further researches in the field, which will increase quality of the model work, and supposed expansion of the available ontology of questions' categories are also described in this paper.