The Task of Synthesizing the Kazakh Language Based on the seq2seq Approach for a Question-Answer System

Abstract

Currently, many texts are synthesized using a computer. Sentence synthesis is becoming more common in various fields. For example, it is used in intelligent systems that can explain to the user the progress of solving a specific problem, decision support systems that help the user make decisions based on the developed alternative, smart homes, voice assistants, chat bots, and so on. Machine learning is one of the most effective approaches to solving the problem of text synthesis. Machine learning algorithms can determine how to perform important tasks by analyzing examples. In the task of synthesizing sentences using machine learning, it is possible to replace a number of components of the entire system with neural networks, which allows not only to approach existing algorithms in quality, but even to significantly surpass them. The article investigates analogous sentence synthesis systems, considers the problem of sentence synthesis in the Kazakh language for the task of a question-answer system. This work consists of two subtasks. The first describes the collection and processing of text resources necessary to solve the problem. For this subtask, an electronic question-answer corpus for the Kazakh language was collected and processed. This corpus consists of 60,000 questions and answers on various topics. The second sentence synthesis problem was solved on the basis of machine learning using the seq2seq method. Based on the chosen method and the assembled corpus, a number of experiments were carried out and the results were obtained. Based on the results obtained, the quality was assessed using the BLEU metric and an expert.