Why Is MBTI Personality Detection from Texts a Difficult Task?

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 3580–3589 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 3580 Why Is MBTI Personality Detection from Texts a Difficult Task? Sanja ˇStajner, Seren Yenikent Symanto Research Nuremberg, Germany {sanja.stajner, seren.yenikent}@symanto.com Abstract Automatic detection of the four MBTI per- sonality dimensions from texts has recently attracted noticeable attention from the natu- ral language processing and computational lin- guistic communities. Despite the large collec- tions of Twitter data for training, the best sys- tems rarely even outperform the majority-class baseline. In this paper, we discuss the theoreti- cal reasons for such low results and present the insights from an annotation study that further shed the light on this issue. 1 Introduction Apart from being long and requiring to be admin- istrated by a skilled human assessor in artificial circumstances (laboratory conditions), the tradi- tional questionnaire-based personality tests often introduce social desirability bias (Krumpal, 2011) and the reference-group effect (Heine et al., 2002), which can be introduced either from the subject’s or assessor’s side. To avoid those biases, it was suggested that analysing person’s writing provides more objective assessment of one’s personality than the traditional questionnaires (Stachl et al., 2019). 1.1 The MBTI The original MBTI model was based on the com- prehensive theoretical work of Carl Jung (1921), and was further developed by Myers and Briggs by adding the fourth dimension (judgment/perception) and several decades of extensive practical use within the industrial and educational settings (Briggs-Myers and Myers, 1995). Today, the MBTI is one of the most widely used non-clinical psycho- metric assessments, regularly used in understand- ing team building processes in work environments (Kuipers et al., 2009), for career suggestions (Gar- den, 1997), and in marketing and consumer behav- ior (Gountas and Gountas, 2001). The MBTI lays out a binary classification based on four distinct functions, and draws the typology of the person according to the combination of those four values (e.g. INFP, ESTJ): • Extraversion/Introversion (EI) - preference for how people direct and receive their energy, based on the outer or inner world • Sensing/INtuition (SN) - preference for how people take information in, by five senses or by interpretation and meanings • Thinking/Feeling (TF) - preference for how people make decisions, by relying on logic or emotions towards people and special circum- stances • Judgment /Perception (JP) - how people deal with the world, by organizing it or staying open for new information While many studies investigated linguistic char- acteristics of the Big 5 personality traits (Mairesse et al., 2007; Furnham, 1990; Pennebaker and King, 1999; Gill and Oberlander, 2002; Scherer, 2003; Pennebaker and King, 1999; Gill and Oberlander, 2003), to the best of our knowledge, there have been no studies reporting on linguistic characteris- tics of different MBTI types. The most probable reason for this is a different nature of the MBTI framework. Unlike the Big 5 model that origi- nated from lexical analyses (Cattell, 1946; Tupes and Christal, 1961; …

Content truncated to 500 words