Haven't read anything recently? Come anyway: bring something you read a while ago, or join to listen and respond.
Bring a Positive Claim
What claim would you defend?
Identify a claim from the paper, or from NLP research and practice more broadly, that you find meaningful and are willing to defend, along with any known critiques you choose to engage. The group interrogates the claim, its assumptions, and the boundaries of its validity.
See examples
- Reasoning benchmarks have diagnostic value even if they fail as measures of LLMs' general reasoning ability.
- Chain-of-thought reasoning improves AI safety classification performance.
- Integrating labels or preferences from more than one annotator makes models generate more diverse responses.
- Fine-tuning with culturally diverse datasets improves models' cultural coverage.
Practices on Trial
What practice does it take for granted?
Identify a method, metric, or framing that the paper treats as standard, neutral, or simply the way things are done in NLP. The discussion moves beyond asking whether the practice works, and instead examines it as a social and epistemic phenomenon: why it emerged, what assumptions it encodes, whose interests it serves, and what it makes harder to see.
See examples
- Why do math, coding, and deductive logic tasks stand in for "reasoning"? Are they sufficient?
- How did step-by-step decomposition (chain-of-thought) become synonymous with reasoning, and what kinds of reasoning does it sideline?
- Why are annotators treated as interchangeable labor rather than socially situated knowers?
- How did minimally paid crowdwork become "cost-effective" standard infrastructure?
- Why is "English first, then multilingual" the default, and what gets baked in that can't be fixed later?
Epistemic Shift
What idea from outside NLP would reframe it?
Bring a concept, method, or framework from another discipline and make the case for its relevance. The discussion unpacks what the idea actually means in its home discipline, traces where it may already be quietly shaping NLP, and explores what it would look like to apply it more deliberately, whether to expose blind spots, reframe questions, or suggest novel methodologies.
See examples
- Informal, pragmatic, and rhetorical reasoning from informal logic
- Situated cognition from cognitive science
- Standpoint epistemology from feminist theory
- Pluriversality from decolonial theory
