We presented a new paper on Automatic Personality perception at the ACM International Conference on Multimodal Interaction.
The work was done in collaboration with Nisreen Alshubaily, one of the best PhD students ever, and Tanaya Guha, all at the University of Glasgow.
The work addresses a very interesting problem related to personality perception, namely whether understanding what people say changes our perception of the personality traits in unacquainted speakers.
For this reason, we performed a new annotation of an established benchmark for Automatic Personality Perception (the SSPNet Speaker Personality Corpus) that includes personality raters that both understand and do not understand the language of the Corpus speakers (whose is French).
The results show that such difference has a major impact on the performance of approaches for Automatic Personality Perception. In fact, when the annotators do not understand the language of the speakers, multimodal approaches do not benefit from the use of both paralanguage and language, even if this latter, when used in isolation, produces results better than chance. The probable reason is that, when language is not accessible, it cannot contribute with independent information to a multimodal approach.
Leave a Reply