Assessing lexical, syntactic, and conceptual turn-by-turn alignment in conversations involving conflict and deception

Activity: Talk or presentation typesLecture and oral contribution

See relations at Aarhus University

Riccardo Fusaroli - Lecturer

Our goal is to present a novel systematic approach for automatically assessing conversational alignment across turn-by-turn exchanges, examining linguistic behavior at increasing levels of abstractness through lexical, syntactic, and conceptual similarity. Using the latest advances in Python-based NLP tools, the procedure begins by taking conversational partners' turns and converting each into a lemmatized sequence of words, assigning part-of-speech tags and computing high-dimensional semantic vectors per each utterance. Words and part-of-speech tags are further sequenced into n-g! rams of increasing length (from uni- to quad-grams) to allow a range of linguistic structures to be examined. Lexical, syntactic, and conceptual alignment values are then calculated on a turn-by-turn basis as cosine scores. To showcase our approach, and to demonstrate its effectiveness in capturing turn-level linguistic alignment, we turn to a unique conversational context: one in which participants disagree or agree with each other about contentious sociopolitical topics, with the added element of one partner secretly taking a "devil's advocate" position. Our findings reveal that high-level intentional factors can modulate alignment processes consistently across multiple levels of linguistic abstraction. Contrary to previous findings on non-verbal coordination (Duran and Fusaroli, under review), deception disrupts verbal alignment, and alignment generally decreases over time. Moreover, for a subset of lexical and syntactic measures, this decrease is most pro! nounced in truth conversations. This suggests that a hypothesi! zed role of alignment, whereby mutual understanding is facilitated, is established early and is required less as truth conversations progress. Implications for current models of interpersonal coordination will be discussed.
19 Nov 2015

Event (Conference)

TitleScip 2015
Date19/11/2015 → …
CountryUnited States



ID: 94252791