Approach to Automatic Determining of Speakers of Direct Speech Fragments in Natural Language Texts

Abstract

Natural language text consists of an author’s or narrator’s text and direct speech fragments. They have different speakers so they could use different vocabularies and syntactic structures. In order to analyze the dependency of vocabulary and sentence structure on speaker, it is necessary to attribute each text fragment to its speaker. The results of such analysis can be used in natural language text generation tasks, allowing to convey different narrative voice depending on the purpose of the generated text. The authors developed a set of rules for attributing direct speech fragments to speaking characters, created a method of direct-speech scene analysis and implemented it in a software tool. In order to evaluate the accuracy of the attribution of direct speech fragments to speakers, an experiment was carried out. The results of the experiment show the viability of the developed method and allow to improve it for further use. The potential applications of the developed method and the software tool are discussed.