How did you change my view? A corpus-based study of consessions' persuasive role

Elena Musi, Columbia


In everyday communicative arenas we engage in critical discussions to persuade others to change their views about issues of personal as well as public interest. In Linguistics, Rhetorics and Discourse Studies concessions are deemed as means to increase the hearer’s positive attitude towards the speaker’s opinion. But are concessions truly perceived as persuasive by language users? If so, which semantic and pragmatic properties make them persuasive strategies?

To answer these questions, I present the results of a corpus-based study on Changemyview, a subreddit where users discuss a wide range of issues and award a so-called “Delta point” to the posts that managed to change their view. The working hypothesis is that if concessions are more frequent in the winning threads than in the others, it is reasonable to conclude that they work as persuasive strategies. The retrieval of concessions in a large quantity of texts is in principle manageable since they are the most frequently explicitly signaled coherence relations. However we show that the automatic identification of the concessive role played by connectives (e.g. (al)though, however) through state of the art parsers achieves a low accuracy, due to the polysemy of these markers. The presentation will be structured in two parts, a theoretically oriented discussion and an empirically oriented one.

In the first part, drawing from a semantic definition of concessions as propositional operators linking a nucleus and a satellite, I propose a corpus-based motivated typology of concessive relations. I single out those having an argumentative value adopting a constructional perspective: what semantic and pragmatic features are shared by and distinguish the sentences “it is December but there is no snow” / “I agree that this flat is gorgeous but it is expensive” / “I am not expert, but Mark’s theory is more reasonable”?

In the second part, using a crowdsourced annotated corpus, I present a bootstrapping method to build a lexicon of linguistic constructions attested in concessions that bear an argumentative value. This lexicon as well as the other hypothesized semantic and pragmatic properties of argumentative concessions are used as features in a classification experiment. Having obtained a sufficiently reliable accuracy, we comment on the distribution of argumentative concessions in persuasive and non-persuasive threads to verify their persuasive value. As a conclusive remark, the relevance of a qualitative fine-grained semantic and pragmatic analysis to improve quantitative based computational approaches will be brought to the fore.