Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Collective Narrative Grounding: Community-Coordinated Data Contributions to Improve Local AI Systems

Zihan Gao, Mohsin Y. K. Yousufi, Jacob Thebault-Spieker

TL;DR

The paper addresses the problem of local knowledge gaps and epistemic injustice in AI question answering by introducing Collective Narrative Grounding, a participatory protocol that converts community stories into structured narrative units governed by locals. It combines elicitation, structuring, and governance to create a provenance-visible, retrieval-augmented QA layer trained and administered under community oversight. Empirical evidence includes a county-scale LocalBench audit ($14{,}782$ QA pairs across $526$ counties) showing $76.7\%$ of errors arise from local-specific gaps, and a participatory QA baseline where an LLM achieved less than $21\%$ accuracy without local context; most missing facts resided in the collected narratives, suggesting the protocol can directly close key error modes. The work also articulates governance, representation, privacy, and consent tensions, offering concrete requirements for retrieval-first, provenance-visible, locally governed QA systems and laying a foundation for equitable, community-grounded AI.

Abstract

Large language model (LLM) question-answering systems often fail on community-specific queries, creating "knowledge blind spots" that marginalize local voices and reinforce epistemic injustice. We present Collective Narrative Grounding, a participatory protocol that transforms community stories into structured narrative units and integrates them into AI systems under community governance. Learning from three participatory mapping workshops with N=24 community members, we designed elicitation methods and a schema that retain narrative richness while enabling entity, time, and place extraction, validation, and provenance control. To scope the problem, we audit a county-level benchmark of 14,782 local information QA pairs, where factual gaps, cultural misunderstandings, geographic confusions, and temporal misalignments account for 76.7% of errors. On a participatory QA set derived from our workshops, a state-of-the-art LLM answered fewer than 21% of questions correctly without added context, underscoring the need for local grounding. The missing facts often appear in the collected narratives, suggesting a direct path to closing the dominant error modes for narrative items. Beyond the protocol and pilot, we articulate key design tensions, such as representation and power, governance and control, and privacy and consent, providing concrete requirements for retrieval-first, provenance-visible, locally governed QA systems. Together, our taxonomy, protocol, and participatory evaluation offer a rigorous foundation for building community-grounded AI that better answers local questions.

Collective Narrative Grounding: Community-Coordinated Data Contributions to Improve Local AI Systems

TL;DR

The paper addresses the problem of local knowledge gaps and epistemic injustice in AI question answering by introducing Collective Narrative Grounding, a participatory protocol that converts community stories into structured narrative units governed by locals. It combines elicitation, structuring, and governance to create a provenance-visible, retrieval-augmented QA layer trained and administered under community oversight. Empirical evidence includes a county-scale LocalBench audit ( QA pairs across counties) showing of errors arise from local-specific gaps, and a participatory QA baseline where an LLM achieved less than accuracy without local context; most missing facts resided in the collected narratives, suggesting the protocol can directly close key error modes. The work also articulates governance, representation, privacy, and consent tensions, offering concrete requirements for retrieval-first, provenance-visible, locally governed QA systems and laying a foundation for equitable, community-grounded AI.

Abstract

Large language model (LLM) question-answering systems often fail on community-specific queries, creating "knowledge blind spots" that marginalize local voices and reinforce epistemic injustice. We present Collective Narrative Grounding, a participatory protocol that transforms community stories into structured narrative units and integrates them into AI systems under community governance. Learning from three participatory mapping workshops with N=24 community members, we designed elicitation methods and a schema that retain narrative richness while enabling entity, time, and place extraction, validation, and provenance control. To scope the problem, we audit a county-level benchmark of 14,782 local information QA pairs, where factual gaps, cultural misunderstandings, geographic confusions, and temporal misalignments account for 76.7% of errors. On a participatory QA set derived from our workshops, a state-of-the-art LLM answered fewer than 21% of questions correctly without added context, underscoring the need for local grounding. The missing facts often appear in the collected narratives, suggesting a direct path to closing the dominant error modes for narrative items. Beyond the protocol and pilot, we articulate key design tensions, such as representation and power, governance and control, and privacy and consent, providing concrete requirements for retrieval-first, provenance-visible, locally governed QA systems. Together, our taxonomy, protocol, and participatory evaluation offer a rigorous foundation for building community-grounded AI that better answers local questions.
Paper Structure (13 sections, 2 figures, 2 tables)

This paper contains 13 sections, 2 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Background and Related Work
  3. Local Knowledge and Information Gaps in AI
  4. Integrating Local Knowledge into AI Systems
  5. Participatory Approaches and Narrative in HCI
  6. The Narrative Grounding Approach: Participatory grounding for narrative unit
  7. Protocol Design Principles
  8. Narrative Elicitation Methods
  9. From Stories to Structured Knowledge
  10. Local Knowledge Gaps: From Failure Taxonomy to Participatory Grounding
  11. Narrative Grounding: A system for Constructive Algorithmic Collective Action
  12. System architecture
  13. Design Tensions

Figures (2)

  • Figure 1: Collective Narrative Grounding Framework. (a) Global LLMs exhibit systematic local knowledge gaps across factual, cultural, geographic, and temporal dimensions. (b) Community members share place-based stories through participatory mapping; stories are transformed into structured, validated narrative units. (c) Narrative units form a queryable knowledge layer that powers a retrieval-augmented local QA system under community governance.
  • Figure 2: System architecture of the Collective Narrative Grounding Framework. The system combines community input and human-in-the-loop processing to build a structured narrative knowledge layer supporting local question answering and accountable governance. Input modules collect local stories; processing modules structure them into validated narrative units; output modules power local AI applications and community review.