PERFECTA: Politeness-based Evaluation of Regard for Finding Efforts to Create Trust and Affiliation

SIFT will identify a set of linguistic events or strategies, occurring in on-line discourse and dialogues (such as email, tweets, blog discussions, etc.) that are expected to alter trust, affiliation, and/or regard on the recipients. The Brown and Levinson model suggests several of these such as "bullying", "buttering up", "discrediting", etc. A politeness-based linguistic analysis approach, augmented as needed with deep semantic processing and plan-recognition techniques, will be combined to provide “recognizers” for such strategies. Candidate strategies will be derived from a literature review of fields such as the psychology of team and group affiliation, marketing and communication, and from interaction with SMEs in the detection of social influence. Recognizers will be validated first through laboratory experiments establishing correlations with human rating of handcrafted dialogues and then to dataset(s) of real discourse data in the modalities of interest. Concurrently, if suitable datasets can be found or created (such as in rating data for products offered or changed after discourse), SIFT will explore the real, behavioral effects of these strategies. Because the Brown and Levinson model is culturally universal, SIFT will also illustrate transitioning our approach to recognize influence strategies across discourse domains, interaction modes and/or cultures.