Empirical Studies of Everyday Professional, Domestic and Client-Service Communication for the Development of Voice Assistants in Russian

Abstract

Voice assistants are gradually becoming an increasingly common feature of our everyday life. However, the naturalness of communication provided by them usually leaves much to be desired. It may be caused by the fact that many chat-bots are trained on artificially created linguistic data sets and on fictional dialogues modeled by linguists on the basis of common phrasebooks or communication textbooks. As a result, the necessary pragmatic result can be achieved, but the feeling of “unnatural” communication of a voice assistant remains, which often reveals itself by the use of archaic phrases or remarks that are not quite suitable for the situation. This state of affairs seems to be improved by referring to real speech data―namely, to a representative volume of sound recordings of real speech communication. The paper discusses some approaches to the analysis of speech data from the sound corpus “One Day of Speech”, which is the most representative resource of Russian everyday spoken communication. The pragmatic structure of professional and everyday conversations is considered, as well as linguistic content of standard modules, such as Greeting and Farewell. As a practical recommendation, we can suggest increasing the variability of answers not due to the lexical diversity of phrases, but due to a more diverse intonation implementation for the most typical replicas in spoken Russian.