Information density converges in dialogue: Towards an information-theoretic model

Research output: Contribution to journalArticle

8 Citations (Scopus)

Abstract

The principle of entropy rate constancy (ERC) states that language users distribute information such that words tend to be equally predictable given previous contexts. We examine the applicability of this principle to spoken dialogue, as previous findings primarily rest on written text. The study takes into account the joint-activity nature of dialogue and the topic shift mechanisms that are different from monologue. It examines how the information contributions from the two dialogue partners interactively evolve as the discourse develops. The increase of local sentence-level information density (predicted by ERC) is shown to apply to dialogue overall. However, when the different roles of interlocutors in introducing new topics are identified, their contribution in information content displays a new converging pattern. We draw explanations to this pattern from multiple perspectives: Casting dialogue as an information exchange system would mean that the pattern is the result of two interlocutors maintaining their own context rather than sharing one. Second, we present some empirical evidence that a model of Interactive Alignment may include information density to explain the effect. Third, we argue that building common ground is a process analogous to information convergence. Thus, we put forward an information-theoretic view of dialogue, under which some existing theories of human dialogue may eventually be unified.

Original languageEnglish (US)
Pages (from-to)147-163
Number of pages17
JournalCognition
Volume170
DOIs
StatePublished - Jan 1 2018

Fingerprint

Entropy
Theoretical Models
dialogue
Data Display
Information Systems
Language
Joints
entropy
information content
information exchange
discourse
language
evidence

All Science Journal Classification (ASJC) codes

  • Experimental and Cognitive Psychology
  • Language and Linguistics
  • Developmental and Educational Psychology
  • Linguistics and Language
  • Cognitive Neuroscience

Cite this

@article{2f68b86613ae444c8e0fab7648cd5541,
title = "Information density converges in dialogue: Towards an information-theoretic model",
abstract = "The principle of entropy rate constancy (ERC) states that language users distribute information such that words tend to be equally predictable given previous contexts. We examine the applicability of this principle to spoken dialogue, as previous findings primarily rest on written text. The study takes into account the joint-activity nature of dialogue and the topic shift mechanisms that are different from monologue. It examines how the information contributions from the two dialogue partners interactively evolve as the discourse develops. The increase of local sentence-level information density (predicted by ERC) is shown to apply to dialogue overall. However, when the different roles of interlocutors in introducing new topics are identified, their contribution in information content displays a new converging pattern. We draw explanations to this pattern from multiple perspectives: Casting dialogue as an information exchange system would mean that the pattern is the result of two interlocutors maintaining their own context rather than sharing one. Second, we present some empirical evidence that a model of Interactive Alignment may include information density to explain the effect. Third, we argue that building common ground is a process analogous to information convergence. Thus, we put forward an information-theoretic view of dialogue, under which some existing theories of human dialogue may eventually be unified.",
author = "Yang Xu and Reitter, {David T.}",
year = "2018",
month = "1",
day = "1",
doi = "10.1016/j.cognition.2017.09.018",
language = "English (US)",
volume = "170",
pages = "147--163",
journal = "Cognition",
issn = "0010-0277",
publisher = "Elsevier",

}

Information density converges in dialogue : Towards an information-theoretic model. / Xu, Yang; Reitter, David T.

In: Cognition, Vol. 170, 01.01.2018, p. 147-163.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Information density converges in dialogue

T2 - Towards an information-theoretic model

AU - Xu, Yang

AU - Reitter, David T.

PY - 2018/1/1

Y1 - 2018/1/1

N2 - The principle of entropy rate constancy (ERC) states that language users distribute information such that words tend to be equally predictable given previous contexts. We examine the applicability of this principle to spoken dialogue, as previous findings primarily rest on written text. The study takes into account the joint-activity nature of dialogue and the topic shift mechanisms that are different from monologue. It examines how the information contributions from the two dialogue partners interactively evolve as the discourse develops. The increase of local sentence-level information density (predicted by ERC) is shown to apply to dialogue overall. However, when the different roles of interlocutors in introducing new topics are identified, their contribution in information content displays a new converging pattern. We draw explanations to this pattern from multiple perspectives: Casting dialogue as an information exchange system would mean that the pattern is the result of two interlocutors maintaining their own context rather than sharing one. Second, we present some empirical evidence that a model of Interactive Alignment may include information density to explain the effect. Third, we argue that building common ground is a process analogous to information convergence. Thus, we put forward an information-theoretic view of dialogue, under which some existing theories of human dialogue may eventually be unified.

AB - The principle of entropy rate constancy (ERC) states that language users distribute information such that words tend to be equally predictable given previous contexts. We examine the applicability of this principle to spoken dialogue, as previous findings primarily rest on written text. The study takes into account the joint-activity nature of dialogue and the topic shift mechanisms that are different from monologue. It examines how the information contributions from the two dialogue partners interactively evolve as the discourse develops. The increase of local sentence-level information density (predicted by ERC) is shown to apply to dialogue overall. However, when the different roles of interlocutors in introducing new topics are identified, their contribution in information content displays a new converging pattern. We draw explanations to this pattern from multiple perspectives: Casting dialogue as an information exchange system would mean that the pattern is the result of two interlocutors maintaining their own context rather than sharing one. Second, we present some empirical evidence that a model of Interactive Alignment may include information density to explain the effect. Third, we argue that building common ground is a process analogous to information convergence. Thus, we put forward an information-theoretic view of dialogue, under which some existing theories of human dialogue may eventually be unified.

UR - http://www.scopus.com/inward/record.url?scp=85030785452&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85030785452&partnerID=8YFLogxK

U2 - 10.1016/j.cognition.2017.09.018

DO - 10.1016/j.cognition.2017.09.018

M3 - Article

C2 - 29024915

AN - SCOPUS:85030785452

VL - 170

SP - 147

EP - 163

JO - Cognition

JF - Cognition

SN - 0010-0277

ER -