Personality is the set of characteristics that define one person. Personality
influences the friends-making process, one’s approach to problems, the
reaction to different news, the language they use, etc. It can also be considered an
invariant in the life of a person, where one can expect very little change in
it after the development age of a person. Understanding one’s own
personality and those of the surrounding people is claimed to help improve
personal and interpersonal relationships as well as one’s career. In this project, we will adopt the Big Five personality framework, a commonly used framework which aims to describe personality based on the five traits of Openness (to experience), Conscientiousness, Extraversion, Agreeableness and Emotional (in)stability.
Previous works in Human-Computer Interaction have demonstrated the positive potential benefit of designing agents which express specific personalities. In this work, we look to investigate whether current open-source language models could support more autonomous generations of such personality-expressive agent output. We will begin our work focusing on language generation, where language is defined as the choice of words,
their order and how they are used to formulate sentences in order to express a
certain or multiple ideas. For example, a very extraverted person is generally
more talkative and uses less complex sentences than an introverted one. Further, we will begin experimenting with the extraversion trait due to its higher impact on spoken interactions and will look to extend the results to the whole spectrum of personality.