Surfacing and Applying Meaning: Supporting Hermeneutical Autonomy for LGBTQ+ People in Taiwan

Paper Title

Surfacing and Applying Meaning: Supporting Hermeneutical Autonomy for LGBTQ+ People in Taiwan

Publication Info

• Topic area: LGBTQ+ identity exploration and support through AI in social media contexts.
• Keywords: LGBTQ+, hermeneutical autonomy, social media, epistemic infrastructure, large language models, participatory design, Taiwan, identity exploration, community governance, AI ethics.

Background and Problem

• Problem / challenge: LGBTQ+ individuals in Taiwan face hermeneutical injustice on social media, where interpretive resources necessary for understanding and articulating their identities are inaccessible or obscured. Social media platforms often fail to support nuanced identity exploration, leaving users isolated or exposed to hostility.
• Significance: Addressing hermeneutical injustice is crucial for fostering LGBTQ+ individuals' identity formation, resilience, and participation in public discourse. Effective technological solutions could mitigate emotional and cognitive burdens while amplifying marginalized voices.
• Motivation and related work: Prior research highlights the role of online communities in creating interpretive resources for LGBTQ+ people but notes limitations in their accessibility and application in hostile or mixed environments. Existing AI systems often lack cultural sensitivity and fail to address the epistemic needs of marginalized groups, particularly in non-Western contexts like Taiwan.

Solution

• Proposed approach: Queerbot, a retrieval-augmented, LLM-powered chatbot designed to scaffold hermeneutical autonomy by supporting identity exploration, reframing hostile narratives, and facilitating community-driven knowledge sharing.
• Novelty:
  1. Introduction of hermeneutical autonomy as a design goal for AI systems.
  2. Development of a chatbot with four interaction modes (reflection, validation, discussion, allyship) tailored to LGBTQ+ users' needs.
  3. Integration of community-curated hermeneutical resources to enhance cultural relevance and inclusivity.
  4. Emphasis on user control and transparency in system understanding and personalization.
• Procedure and key techniques:
  • Conducted a multi-stage study: formative interviews (10 participants), design workshops (8 participants), and evaluation workshops (11 participants).
  • Designed Queerbot with features like user-curated hermeneutical resources, transparent user knowledge, and contextualized response modes.
  • Implemented the system using retrieval-augmented generation (RAG) and natural language personalization techniques.

Results

• Concrete findings:
  • Queerbot facilitated identity exploration, validated lived experiences, and provided actionable guidance for navigating hostile environments.
  • Participants valued descriptive self-statements over rigid labels and appreciated the system's ability to synthesize and contextualize interpretive resources.
  • The chatbot helped participants engage in public discourse with confidence and knowledge, reducing hermeneutical labor.
• Advantage over baselines:
  • Unlike traditional social media platforms, Queerbot offered personalized, context-sensitive support and integrated community-driven resources, addressing gaps in existing epistemic infrastructures.
  • The system's transparency and user control over its understanding of identity were novel and empowering features.
• Experiments / evaluation:
  • Evaluation workshops involved 11 LGBTQ+ participants interacting with Queerbot across four modes. Participants reflected on its capacity to support identity expression, interpretive empowerment, and discourse participation.
  • Data analysis revealed recurring themes such as the importance of validation, actionable guidance, and community-driven governance of hermeneutical resources.
• Limitations and future work:
  • Limited participant diversity (mostly university students in northern Taiwan) and absence of a field deployment.
  • Future work should explore multimodal retrieval, broader participant demographics, and longitudinal studies to assess real-world integration.
  • Risks include potential bias in LLMs, hallucinations, and emotional dependence on AI systems.

Summary

This study addresses hermeneutical injustice faced by Taiwanese LGBTQ+ individuals on social media by designing and evaluating Queerbot, an LLM-powered chatbot that supports identity exploration, reframes hostile narratives, and amplifies marginalized voices. Through participatory design and evaluation, the system demonstrated its ability to scaffold hermeneutical autonomy by integrating community-curated resources and offering transparent, personalized support. While promising, the approach highlights challenges such as AI bias, emotional dependence, and the need for inclusive community governance. Future work should expand participant diversity, explore multimodal capabilities, and ensure ethical deployment in real-world contexts.