Semantic cross-correlation as a measure of social interaction

Achieving human-level social and emotional intelligence is vital for the integration of future virtual agents and robots into the human society. To solve the challenge, it is necessary to have efficient metrics and criteria that would allow us to measure how close the solution is to the state of the art. In this work, one such potentially useful measure is introduced: semantic cross-correlation computed in addition to semantic trajectories characterizing a dialogue. This measure characterizes social interaction. Two different kinds of dialogues are used: (1) between physicians and schizophrenic patients and (2) between ad hoc participants assigned the roles of a user and a system who cooperatively solve problems. Similar in some aspects and at the same time different patterns of semantic characteristics are observed in the two cases. The new measure is applicable to dialogue transcripts as well as to sequences of actions with known semantic characteristics attributed to them. Therefore, it can be expected to become a useful general metric in evaluation of social-emotional interactions and social intelligence of virtual agents.