Can AI systems detect and correct misunderstandings after responding?
How do conversational systems recognize when their previous response was based on a misunderstanding, and what mechanism allows them to correct it retroactively rather than restart?
Conversation Analysis identifies a highly systematic mechanism for handling miscommunication called Third Position Repair (TPR). The sequence: a speaker says something (T1, the trouble source), the addressee misunderstands and responds based on that misunderstanding (T2), which reveals the misunderstanding to the original speaker, who then corrects it (T3).
This is fundamentally different from other repair mechanisms. Insert-expansions (since When should AI agents ask users instead of just searching?) are pre-emptive — they detect potential misunderstanding before committing to a response. TPR is reactive — it detects misunderstanding only after the erroneous response makes it visible. Both are necessary. Pre-emptive repair catches uncertainty before it compounds. Reactive repair catches failures that pre-emptive mechanisms missed.
Current AI systems have neither. They don't proactively probe (the passivity problem) and they don't retroactively correct when their responses reveal misunderstanding. When a user says "No, that's not what I meant," the model typically starts from scratch rather than diagnosing what specific misunderstanding occurred and correcting it structurally.
The REPAIR-QA dataset is the first large dataset of TPRs in a conversational QA setting. The key challenge: to handle TPR, a system must (a) recognize that its previous response was based on a misunderstanding, (b) identify specifically what was misunderstood, and (c) generate a corrected response. This requires maintaining a model of what it assumed and being able to revise that assumption — a form of dynamic belief revision that current single-pass generation architectures don't support.
Since Do language models actually build shared understanding in conversation?, TPR is precisely the mechanism for CORRECTING false common ground after it has been acted on. And since Why do language models sound fluent without grounding?, TPR is a specific form of communicative work that fluent models skip -- the reactive repair that would make misunderstanding visible and correctable rather than silently compounding. And since Why do AI assistants get worse at longer conversations?, TPR addresses the "can't recover" half — it is the formal mechanism for recovery after a wrong turn has already been made.
Inquiring lines that use this note as a source 18
This note is a source for these synthesized inquiries. Follow a line forward into its question, or open it to trace back to all of its sources.
- How does AI lose correct information under conversational persuasive pressure?
- How do time gaps between conversations change what chatbots should remember?
- What does a receiver project onto AI that the system never performed?
- Why does static grounding prevent AI systems from supporting dialectical reconciliation?
- How do conversation repair patterns handle user corrections and interruptions?
- Can social conversation retroactively govern claims that were never addressed to anyone?
- Can conversation analysis predict when agents should ask users for clarification?
- Can AI systems recover from premature assumptions made early in multi-turn conversations?
- What structural changes enable agents to ask clarifying questions?
- Why do chatbots fail to recognize when someone is ambivalent about change?
- How do customer service chatbots get systematically misled by users?
- Can a system without an addressee ever truly tell a joke?
- Can grammar alone repair misunderstanding without ritual correction work?
- What architectural changes help AI avoid adding interpretations users didn't express?
- How should conversational AI balance world knowledge with avoiding false expertise?
- What communicative work do fluent conversations perform that AI systems skip?
- What prevents AI from recovering after conversations take a wrong turn?
- Why does AI that mirrors arguments still fail to build rapport?
Related concepts in this collection 8
This note in its neighbourhood — explore the map, then jump to a related concept in the list below.
Click a node to walk · click center to open · click Open in graph to see this note in the full knowledge graph
-
When should AI agents ask users instead of just searching?
Explores whether tool-enabled LLMs should probe users for clarification when uncertain, rather than silently chaining tool calls that drift from intent. Examines conversation analysis patterns as a formal alternative.
complementary mechanism: insert-expansions are pre-emptive, TPR is reactive; together they cover the full repair space
-
Do language models actually build shared understanding in conversation?
When LLMs respond fluently to prompts, do they perform the communicative work humans do to establish mutual understanding? Research suggests they skip the grounding acts that make dialogue reliable.
TPR is the mechanism for correcting false common ground after acting on it
-
Why do AI assistants get worse at longer conversations?
Explores why LLM performance drops 25 points when instructions span multiple turns instead of one message, and whether models can recover from early wrong assumptions.
TPR directly addresses the recovery mechanism
-
What six problems must every conversation solve?
Schegloff's Conversation Analysis identifies six universal organizational challenges that speakers navigate in all talk-in-interaction. Understanding these helps explain why current AI dialogue systems fall short of human fluency.
TPR is a specific instantiation of the trouble-handling generic order
-
Why do language models sound fluent without grounding?
Explores whether LLM fluency masks the absence of communicative work—the clarifying questions, acknowledgments, and understanding checks that humans perform. Why does skipping these acts make models sound more confident?
TPR is a specific form of communicative work that fluent models skip; its absence contributes to the illusion of fluency
-
Why do language models skip the calibration step?
Current LLMs assume shared understanding rather than building it through dialogue. This explores why that design choice persists and what breaks when it fails.
TPR is a core mechanism of dynamic grounding: it corrects false common ground after it has been acted on, the reactive complement to insert-expansions' pre-emptive clarification
-
Why don't conversational AI systems mirror their users' word choices?
Explores whether current dialogue models exhibit lexical entrainment—the human tendency to align vocabulary with conversation partners—and what's needed to bridge this gap in AI communication.
TPR and lexical entrainment are complementary grounding mechanisms: entrainment builds shared vocabulary proactively, TPR repairs shared understanding reactively; AI systems lack both
-
Does agreeable AI actually help people resolve conflicts better?
When AI affirms users' positions in interpersonal disputes, does it support better decision-making or undermine the outside perspective users most need? Two large experiments tested whether sycophancy shifts how people handle real conflicts.
contradicts: sycophancy suppresses the very repair behavior third-position repair tries to enable
Related papers in this collection 8
Papers most semantically related to this note, ranked by cosine similarity in the embedding space.
- No that's not what I meant: Handling Third Position Repair in Conversational Question Answering
- Task-Oriented Dialogue with In-Context Learning
- Conversational Alignment with Artificial Intelligence in Context
- Intent Mismatch Causes LLMs to Get Lost in Multi-Turn Conversation
- Probing the Multi-turn Planning Capabilities of LLMs via 20 Question Games
- LLMs can implicitly learn from mistakes in-context
- Beyond Hallucinations: The Illusion of Understanding in Large Language Models
- A recipe for annotating grounded clarifications
Original note title
third position repair addresses misunderstanding correction after an erroneous response reveals it — a systematic repair mechanism absent from current AI systems