Machine Learning and Emotions

Are machine learning algorithms able to decipher emotional state from everyday speech?

iibp-admin AI & ML in Psychology Leave a Comment

The words that people use have been found to reflect stable psychological traits, but less is known about the extent to which everyday fluctuations in spoken language reflect transient psychological states.
Researchers explored within-person associations between spoken words and self-reported state emotion among 185 participants who wore the Electronically Activated Recorder (EAR; an unobtrusive audio recording device) and completed experience sampling reports of their positive and negative emotions four times per day for seven days (1,579 observations).
Following this, they examined language using the Linguistic Inquiry and Word Count program (LIWC ; theoretically created dictionaries) and open vocabulary themes (clusters of data driven semantically related words). Although some studies give the impression that LIWC’s positive and negative emotion dictionaries can be used as indicators of emotion experience, researchers found that when computed on spoken language, LIWC emotion scores were not significantly associated with self-reports of state emotion experience.
Exploration of other categories of language variables suggests a number of hypotheses about substantive everyday correlates of momentary positive and negative emotion that can be examined further. These findings
  • (1) suggest that LIWC positive and negative emotion dictionaries may not capture self-reported subjective emotion experience when applied to everyday speech, either they are not effective enough or people have an incongruence in their self-report and everyday speech,
  • (2) emphasize the importance of establishing the validity of language based measures within one’s target domain,
  • (3) demonstrate the potential for developing new hypotheses about personality processes from the open ended words that occur in everyday speech, and
  • (4)extend perspectives on intra-individual variability to the domain of spoken language
(Final accepted version, 4 Feb 2019 (in press, Journal of Personality and Social Psychology)

Leave a Reply

Your email address will not be published. Required fields are marked *