On the issue of evaluation of speech instruments (prosodics and extralinguistics) as objectives for improving artificial intelligence
Mustafin, Airat Ravilyevich Psychiatrist, psychotherapist, Medical clinic "Academy of Health", Arkhangelsk, Russia firstname.lastname@example.org
Mustafin, Rodion Ayratovich 5th year student of the Medical Faculty of the Northern State Medical University, Arkhangelsk, Russia Nibori07@gmail.com
Abstract An artificial intelligence does a good job of understanding texts, including the most common metaphors and other means of narrative artistry. However, the written text does not always reflect the possibilities of oral reproduction. At the same time, nonverbal tools account for a significant share of the information load. This applies primarily to the emotional assessment of situations described in the texts. We are talking about what we only hear, that is, prosody and extralinguistics. In addition, in a more complex task, we can have a combined assessment of the oral text and performance information: facial expressions, pantomimics, proxemics and takesics. Probably, the developers of artificial intelligence will face the fact that without personalized standards of text execution by one or another person, we will not be able to understand a significant part of the information accurately enough. In other words, developers can follow the path of creating numerous individual standards for non-speech instruments, which is extremely difficult and cumbersome to perform, or look for alternatives. An alternative way, it seems to us, will consist in typing by significant signs. Moreover, it will be ideal if the set of features itself can be set independently, receiving evaluation options with varying degrees of accuracy of understanding. That is, with different assumptions. It is possible that the accuracy will depend on the semantic context comparison function, as well as other types of non-speech contexts – situational and behavioral ones.