Shared knowledge in natural conversations: can entropy metrics shed light on information transfers?

Proceedings of the 26th Conference on Computational Natural Language Learning (CoNLL), pages 213 - 227 December 7-8, 2022 ©2022 Association for Computational Linguistics Shared knowledge in natural conversations: can entropy metrics shed light on information transfers? Eliot Maës1,3 Philippe Blache2,3 1Aix Marseille Univ, Université de Toulon, CNRS, LIS, Marseille, France 2LPL, CNRS & Aix-Marseille University 3ILCB (Institute of Language, Communication and the Brain) {eliot.maes,leonor.becerra}@lis-lab.fr, blache@ilcb.fr Leonor Becerra-Bonache1,3 Abstract The mechanisms underlying human commu- nication have been under investigation for decades, but the answer to how understanding between locutors emerges remains incomplete. Interaction theories suggest the development of a structural alignment between the speak- ers, allowing for the construction of a shared knowledge base (common ground). In this pa- per, we propose to apply metrics derived from information theory to quantify the amount of information exchanged between participants, the dynamics of information exchanges, to pro- vide an objective way to measure the common ground instantiation. We focus on a corpus of free conversations augmented with prosodic segmentation and an expert annotation of the- matic episodes. We show that during free con- versations, the amount of information remains globally constant at the scale of the conver- sation, but varies depending on the thematic structuring, underlining the role of the speaker introducing the theme. We propose an original methodology applied to uncontrolled material. 1 Introduction Theories of interaction explain how participants elaborate their discourse in the perspective of ex- changing information, executing a task, establish- ing a joint action, etc. These theories stipulate in particular that such activity is correlated with the building of a shared knowledge between partici- pant, also called common ground (Pickering and Garrod, 2004, 2021). In these frameworks, the quality of an interaction depends on the capacity of building such mutual knowledge, which to its turn depends on the alignment of linguistic representa- tions between participants. These mechanisms are based on different levels of convergence between the participants, that can occur at any level: lexical, syntactic, prosodic, as well as gestures, behaviors, etc. One hypothesis is that this phenomenon is also visible at the semantic level, showing a coordina- tion between participants in terms of information exchange that can be uncovered by studying the amount of such information and its dynamics dur- ing a conversation. The goal of this work is therefore to eval- uate these questions by means of information- theoretical measures (Shannon, 1948): sharing information relies on the use of simple symbols which can be combined, concatenated to transfer increasingly complex knowledge. Moreover, it is possible to analyze the dynamics of this process, whether the amount of transfer vary during a con- versation, at what position, and whether an align- ment between participants can also be observed at this level. An estimation of the quantity of in- formation exchanged between participants and its dynamics could therefore constitute an objective way to measure the common ground instantiation. Several works have been done in this direction, based on lexical information measured by entropy, and showing a convergence between participants. …

Content truncated to 500 words