Метод частиномовної розмітки на основі квазіфлексій

Abstract

The quality of the recommendations provided as a result of the application of decision support systems largely depends on the quality and reliability of the knowledge provided by experts. Solving the problem of automatic detection of ambiguity in the textual formulations of experts is a significant step towards increasing the reliability of knowledge and the adequacy of the models on the basis of which decision support is provided. Most approaches to automatic ambiguity detection rely on the use of part-of-speech tagging as the first step in detecting ambiguity. The article proposes a method of automatic part-of-speech tagging based on quasi-inflections (variable word components), the accuracy of which is commensurate with the existing implementations of the rule-based approach. The advantages of the rule-based approach include a significant reduction in the required amount of information, simple implementation of analyzer improvements and a high degree of portability of components (rules, dictionaries, quasi-inflections, exceptions).Comparing the reported accuracy of part-of-speech analyzers, Markov models and the transformation approach achieve an accuracy of up to 97 %. At the same time, the accuracy of the rule-based approach varies from 97 % to 100 %.The novelty of the proposed method is the use of quasi-inflections as the main and only method for determining the part of speech and grammatical characteristics of the word. To check the effectiveness of the proposed method, testing was conducted on the basis of textual formulations in 7 files with the goal hierarchy structures, a total of 4378 words. The proposed method showed an accuracy of 98,70 % in the part-of-speech tagging of textual formulations in the test corpus. Demonstrated high accuracy of the method can be achieved with strict following of the described method of compiling dictionaries of rules and quasi-inflections. Refs: 23 titles.