After Taiwan's legalization of same-sex marriage in 2019, LGBTQ+ communities continue to face hostility on social media. Using the lens of hermeneutical injustice and autonomy, we examine how technological conditions affect LGBTQ+ individuals' identity exploration, narrative seeking, and community resilience. We conducted a multi-stage study with Taiwanese LGBTQ+ individuals, including in-depth interviews, participatory design workshops, and evaluation sessions. Participants described fragile yet creative strategies such as seeking validation in online interactions, reframing hostile content through theory, and relying on allies. Building on these insights, we designed and evaluated a retrieval-augmented, LLM-powered chatbot with four modes of interaction: reflection, validation, discussion, and allyship. Findings show that the system fosters hermeneutical autonomy by helping participants reframe hostile narratives, validate lived experiences, and scaffold identity exploration, while reducing the hermeneutical labor of navigating social media hostility. We conclude by outlining design implications for AI systems that advance hermeneutical autonomy through fluid self-representation, contextualized dialogue, and inclusive community participation.
Alex: Welcome to another episode of ResearchPod.
Sam: Today we're looking at a study on how people in Taiwan's LGBTQ+ communities handle tough online spaces. The paper is called "Surfacing and Applying Meaning: Supporting Hermeneutical Autonomy for LGBTQ+ People in Taiwan." It explores how social media often fails these communities, even after same-sex marriage became legal there in 2019.
Alex: So the core problem is that online hostility persists, and platforms make it hard for supportive ideas to reach people?
Sam: Yes. People face comments that dismiss or attack their experiences, but algorithms push those down and highlight the negative stuff instead. This leaves folks without the right words or stories—called hermeneutical resources—to make sense of what they're going through or push back confidently. The study asks how AI chatbots might help by pulling in helpful community knowledge on the spot.
Alex: Right, like a portable set of tools from trusted sources. But first—those resources... you mean shared stories and ideas from the community itself?
Sam: Exactly. Imagine trying to explain a feeling to someone who doesn't get it, and you lack the everyday words everyone else has. Communities build those words through talks and posts to fight isolation. Without easy access across social media, it creates hermeneutical injustice, where marginalized voices stay stuck.
Alex: And in Taiwan, with family pressures and uneven acceptance, that hits extra hard.
Sam: It does. The researchers interviewed Taiwanese LGBTQ+ people and found they use fragile tricks, like hunting for validation in replies or reframing hate with theories. But it's exhausting. Their idea: a chatbot that grabs community-vetted stories and concepts right when needed, acting like a pocket ally.
Alex: Huh... so turning AI into something that travels with you through hostile feeds.
Alex: Those fragile tricks—like checking comments or sharing screenshots—sound unreliable. How did the researchers turn that into an actual tool?
Sam: They started with in-depth interviews of ten Taiwanese LGBTQ+ people, mostly students in their early twenties, to map out real experiences. Participants described using online communities for identity stories, scanning comment sections for backup views—which helped most of them feel less alone—and sharing screenshots in safe groups for advice. But support was spotty; algorithms often hid positive replies. This pointed to a need for steady access to community stories right in hostile spots. To fill it, they designed a chatbot using retrieval-augmented generation, or RAG—like having a librarian scan shelves of vetted books for exactly what you need before answering.
Alex: Okay, so it grabs specific community stuff on the fly, not generic advice. But how does it know what fits you personally?
Sam: They added a way to track user details in plain words, like "prefers casual discussions, identifies as pansexual with medium confidence." Users can edit these notes and rate how sure they feel, so the system updates based on chats. This makes replies tailored—for example, in reflection mode, it helps unpack feelings using pulled resources; validation mode affirms with matching stories; discussion builds arguments; allyship suggests ways to support others. The paper suggests this setup cuts reliance on buried comments by making community knowledge portable and bias-resistant.
Alex: That addresses the inconsistency head-on.
Alex: Building that chatbot sounds promising, but how did they make sure it actually fit what people wanted? Did they just guess, or involve the community?
Sam: They held three design workshops with eight LGBTQ+ participants from Taiwan, many students in their early twenties. First, they shared ten real-life scenarios from the interviews—like struggling to find supportive replies or feeling drained from explaining identities—to spark ideas. Participants wrote questions starting with "how might we" fix these, grouped them into themes, sketched solutions in rounds, and picked top design goals together. This participatory approach ensured the tool matched lived needs, not outsider assumptions.
Alex: So real people brainstorming fixes step by step. What themes came out that shaped the chatbot?
Sam: One big theme was the "self"—people wanted ways to describe their identities fluidly, like in casual chat, not rigid boxes. They pictured editing a clear summary of what the system knows about them, such as confidence in labels. Another was "people and information," stressing links to communities with similar stories and tailored resources like theories or narratives to interpret stress. A third, "environment," saw the bot as an ally filtering hostility or helping craft strong replies without exhausting effort.
Alex: Huh—that ties back to easing the labor of reaching out or fighting comments. Like a teammate who knows your style.
Sam: Precisely. These led to key design goals: first, make the system's grasp of the user transparent, showing and letting edits to natural-language notes on identity and preferences. Second, pull relevant community stories, theories, or media matched to the moment, especially for underrepresented views, to scaffold understanding without bias.
Alex: So it becomes a steady partner in messy online spaces... grounded in what users themselves built.
Alex: Right, grounded in user input. But what exactly does this partner do in practice—walk me through its main features.
Sam: They built it around four clear goals from the workshops. One focuses on helping craft smart replies to cut down on the effort of arguing back, like suggesting ways to respond while pointing others to helpful info. Another ensures community members add and check the stories and ideas it uses, keeping everything trusted and relevant. Queerbot has a section showing what it knows about you—like notes on your identity or preferences—in simple words you can edit anytime. After chats, it sums up changes so you see how your profile evolves.
Alex: Okay, editable notes for control. And it changes based on talks... that keeps it personal without assuming too much.
Sam: Yes. It also switches into different chat styles depending on what you need. One style asks questions to learn more about you, filling in gaps. Another, for tough comments, just listens, confirms your feelings, and suggests similar stories from others. A deeper one pulls in ideas from community writings or theories to help sort out big issues. The last helps build replies that fit your style, offering a few options backed by solid resources.
Alex: So modes like tools in a kit—validation when hurt, arguments when ready. How does it pick the right stories without going off track?
Sam: It stores info in organized spots: your profile with notes rated by how sure it is; quick summaries of past chats; short clips from trusted sources; and full texts from community articles. When you chat, it grabs only what's relevant to your mode and profile, mixes it with guides for tone, and builds a response. After, it reviews the talk to tweak your profile.
Alex: Huh... like a notebook that updates itself from conversations.
Sam: They tested this in five group sessions with eleven new Taiwanese LGBTQ+ folks, mostly in their twenties. People tried scenarios and discussed deeper wants—like more fluid ways to handle shifting identities amid family norms. The paper notes it resonated, surfacing resilience tricks like using online spots for self-exploration, but highlights needs for non-Western cultural tweaks in queer tech design.
Alex: That builds real staying power online.
Alex: You mentioned testing in group sessions—walk me through how they set those up and what people actually said about using Queerbot.
Sam: They ran five workshops with eleven Taiwanese LGBTQ+ participants, aged nineteen to thirty-one, including students, public servants, and others with diverse identities. Beforehand, people shared journal entries, social posts, and community materials via chat to personalize the prototype, plus curated examples of real discriminatory content mocked up like social media threads. Sessions started with icebreakers, then intros to the four modes, followed by tasks: checking and editing what the bot knew about them, chatting to build understanding, handling hurtful posts for support, unpacking tricky ones, and crafting ally replies. After each, folks noted reflections on group posters, ending with a debrief.
Alex: So hands-on with their own stuff and fake-but-real scenarios. What patterns emerged from those talks?
Sam: Researchers reviewed transcripts to spot common themes, like how the bot acted as a mirror, pulling together users' inputs into descriptions that sparked fresh thoughts on identity—better than stiff labels. People valued it pushing deeper, like suggesting "misogyny" for an encounter, giving a new angle to explore. It validated feelings by affirming "others feel this too" with matching stories, easing doubt, and pointed to hidden communities or practical tips, making paths seem doable—though some noted machine comfort felt a bit empty next to humans.
Alex: Huh... so less alone, with real-world pointers. Did it help in arguments too?
Sam: Yes, it broadened views by weaving in global or historical queer ideas, challenging local norms like family roles. Users saw it aiding rational replies backed by theories, spotting argument gaps to stay strong in debates. It helped balance discussions—not just self-defense, but boosting minority views for lurkers, while advising when to skip draining fights.
Alex: That selective push sounds practical.
Sam: Participants stressed community control over the resources too—curating experiences into a safe bridge for comparison and dialogue, trusted because queer-led. The paper suggests this scaffolds resilience amid cultural tensions like Confucian family views, but calls for more non-Western tweaks in queer tech.
Alex: A meaningful way to navigate those spaces, grounded in their voices.
Alex: Pulling it all together, this chatbot seems like a practical bridge—making community knowledge handy right where it's toughest to find.
Sam: The paper shows how tools like Queerbot can reshape online setups so they help people interpret their lives on their own terms, rather than letting algorithms bury useful stories. It highlights a two-way street: users shape the system through edits and community input, and it shapes them back by surfacing fitting ideas in real time. In Taiwan, this counters ongoing tensions from family expectations and uneven acceptance, even after legal changes.
Alex: Right, turning platforms from obstacles into allies. But what about downsides—any risks they flag?
Sam: Yes, large language models can carry built-in prejudices from their training data, slipping in unfair views despite safeguards. They might also invent facts—called hallucinations—like fake stories or theories, which could mislead someone feeling vulnerable. Long-term, chatbots risk creating emotional reliance if they always agree too easily, or attract bad actors poisoning the shared resources. The prototype sticks to text only, missing images or videos that often carry key clues. Most testers were young urban students, so it may not fit older or rural folks fully. No long-term real-world use means we lack data on daily habits.
Alex: Those are real hurdles.
Sam: Still, it points to embedding such systems in messaging apps or browsers for proactive help, always with community oversight to handle inner-group differences. The work advances designs that prioritize user control and diverse resources, fostering steadier ways to handle digital stress without overpromising.
Alex: A solid step for building resilient online spaces. That's our look at supporting LGBTQ+ voices in Taiwan's digital world. Thanks for listening to ResearchPod.