Cognitive and social delays in the initiation of conversational repair

2026-04-19 · Source: Paper Index on ACL Anthology · Field: Science & Research — Social Sciences & Behavioral Studies, Mathematics & Computational Sciences · Depth: Expert, medium

Summary

A study by Julia Beret Mertens and Jan Peter De Ruiter, published in "Dialogue & Discourse" in March 2021, investigates the relative contributions of cognitive and social factors to delays in conversational turn-taking. While most turns are initiated within 250ms, certain types, particularly "other-initiations of repair" (OIRs) like "What?", are often delayed due to their cognitive demands and social dispreference. The researchers analyzed the Floor Transfer Offsets of 456 OIRs. Their findings indicate that interlocutors initiated OIRs later when the source of the comprehension problem had weaker discourse context, was shorter, or when the OIR itself was more "face-threatening." This suggests that both cognitive processes (attention, prediction, planning) and social preferences significantly influence the timing of delayed utterances in human conversation.

Key takeaway

For NLP engineers designing conversational AI, understanding that both cognitive processing and social dynamics impact turn-taking delays is crucial. You should account for these factors when modeling human-like interaction, particularly for repair mechanisms. Incorporate variable response times based on the complexity of the query and the social implications of the AI's response to create more natural and contextually aware dialogue systems.

Key insights

Conversational delays in repair turns are influenced by both cognitive load and social considerations.

Principles

• Turn initiation within 250ms is typical.
• Cognitive load and social dispreference delay turns.
• Face-threatening OIRs increase delay.

Method

The study analyzed Floor Transfer Offsets of 456 other-initiations of repair (OIRs) to quantify delays, correlating them with discourse context, trouble source length, and social face-threat.

In practice

• Analyze turn-taking for cognitive load.
• Identify socially dispreferred turn types.
• Quantify "face-threat" in interactions.

Topics

• Conversational Repair
• Turn-Taking Timing
• Other-Initiations of Repair
• Cognitive Processing
• Social Preference

Best for: NLP Engineer, AI Scientist, Research Scientist

Related on AIssential

• ‘Alexa, tell me a joke’: how talking to AI impacts young children’s development — Routine AI interactions shape children's language habits and conversational expectations differently than human exchanges.
• Talking to a Know-It-All GPT or a Second-Guesser Claude? How Repair reveals unreliable Multi-Turn Behavior in LLMs — LLM multi-turn conversational repair behavior is unreliable and highly model-specific, challenging consistent user interaction.
• How to Handle Interruptions in Voice AI Calls 🤫🎙️ — Prioritize task completion over conversational perfection in AI interactions, adapting agent speech to caller behavior.
• The Dynamics of Human and AI-Generated Language: How Semantics Fluctuates across Different Timescales — The ACW-0 semantic timescale analysis effectively distinguishes temporal organization of generic versus specific content in human and AI speech.
• LVLMs and Humans Ground Differently in Referential Communication — LVLMs fail to build common ground and adapt communication in multi-turn referential tasks, unlike humans.
• MT-PingEval: Evaluating Multi-Turn Collaboration with Private Information Games — Language models struggle with multi-turn collaboration, often failing to improve over non-interactive baselines despite potential.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Paper Index on ACL Anthology.