[pella]

related research:

"Talking with tact: Polite language as a balance between kindness and informativity"

Keywords: Politeness; computational modeling; communicative goals; pragmatics

PDF: https://langcog.stanford.edu/papers_new/yoon-2016-cogsci.pdf

"Discussion"

"Why would a speaker ever say something that is not maximally truthful and informative? Communication is often examined from the perspective of successful information transfer from speaker to listener. In the social realm, however, communication also can serve the social function of making the listener feel good and saving her face. We proposed here that intuitively “polite” utterances arise from the desire to be kind (i.e. save face). A cooperative speaker then tries to balance the goals to be kind and to be informative, and produces utterances of varying degrees of politeness that reflect this balance. To test this proposal, we examined inferential judgments on a speaker’s utterance, which was a potentially face-threatening evaluation of the listener’s performance. As we predicted, participants’ inferences about the true state of the world differed based on what the speaker said and whether the speaker’s intended goal was to be honest, nice or mean (Expt. 2). We were also able to predict participants’ attributions of different social goals to speakers depending on how well the literal utterance meaning matched the actual rating the performance deserved (Expt. 3). The model presented here relates to other work done in game-theoretic pragmatics. Van Rooy (2003) uses a gametheoretic analysis of polite requests (“Could you possibly take me home?”) to argue the purpose of polite language is to align the preferences of interlocutors. Our notion of social utility Usocial is similar in that it motivates speakers to signal worlds that make the listener feel good. Van Rooy’s analysis, however, relies on the notion that polite language is costly (in a social way e.g., by reducing one’s social status or incurring social debt to one’s conversational partner) but it’s not clear how the polite behaviors explored in our experiments (not polite requests) would incur any cost to speaker or listener. Our model derives its predictions by construing the speaker utility as a collection of possible goals (here, epistemic and social goals). The speech-acts themselves are not costly. Will machines ever be polite? Politeness requires more than merely saying conventionalized words (please, thank you) at the right moments; it requires a balance of informativity and kindness. Politeness is not an exception to rational communication; it is one important element of rational communication, serving a key social function of maintaining relationships. We extended the Rational Speech Acts framework to include social utility as a motive for utterance production. This work takes a concrete step toward quantitative models of the nuances of polite speech. And it moves us closer to courteous computation—to computers that communicate with tact."