qproj wikiNLP and argumentation
Quell is meant to let players practise real philosophical debate. That needs two kinds of groundwork. One is a model of argument structure, detailed enough for the game to understand what a player has built. The other is a way to tell what a player’s notes are about.
Start with Debate, the intended mechanic and what is still undecided. Argument maps describes the reviewed graphs of each text that debates would be built from, and Reading notes covers how a player’s highlights and notes would be interpreted.
The other articles explain the argumentation theory and language-processing techniques this work draws on, grouped by topic below.
Topics
- Debate
- Argument maps · Debate · Reading notes
- Argumentation theory
- Argument mining · CAPNA · Abstract argumentation · Inference Anchoring Theory and xAIF · Periodic Table of Arguments · Value-based argumentation · oAMF
- Understanding text
- Cross-encoders · Dependency parsing · Embeddings and cosine similarity · GLiNER and GLiNER2 · Natural language inference
- Generating text
- Constrained decoding · In-context learning · LoRA fine-tuning · Prompt caching
- Running models
- ONNX and ort · Quantization and llama.cpp
- Measuring results
- Top-k accuracy and calibration · Stage-wise evaluation
Records
The debate plan’s pages, the ATIP algorithm note and the reading-notes design are drafts in the NLP document library.
See also
- Research labs — note-lab and the xAIF editor.