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

Area of Science:

• Linguistics
• Computational Linguistics
• Cognitive Science

Background:

• The principle of entropy rate constancy (ERC) suggests information is distributed evenly in language.
• Previous research on ERC primarily focused on written text, leaving its application in spoken dialogue under-explored.
• Spoken dialogue involves joint activity and distinct topic shift mechanisms compared to monologues.

Purpose of the Study:

• To investigate the applicability of entropy rate constancy (ERC) to spoken dialogue.
• To analyze how information contributions evolve between dialogue partners during discourse.
• To explore the interaction between information density and dialogue dynamics.

Main Methods:

• Analysis of spoken dialogue data, considering the interactive nature of conversation.
• Examination of information density at the sentence level.
• Identification of topic shifts and interlocutor roles in information contribution.

Main Results:

• The study confirms that local sentence-level information density increases in spoken dialogue, aligning with ERC predictions.
• A novel converging pattern in information contribution was observed between dialogue partners, particularly when new topics are introduced.
• Interlocutor roles and their impact on information exchange dynamics were identified.

Conclusions:

• Spoken dialogue generally follows the entropy rate constancy (ERC) principle regarding information density.
• Dialogue partners exhibit a converging information contribution pattern, suggesting mechanisms beyond shared context, potentially related to interactive alignment or common ground building.
• An information-theoretic framework can unify existing theories of human dialogue.