Topic: Understanding text

Part of NLP and argumentation.

This topic contains 5 articles.

• Cross-encoders — Models that score a pair of texts by reading them together: slower than embeddings, but usually more accurate.
• Dependency parsing — Recovering a sentence’s grammatical structure with a small, fast, deterministic model.
• Embeddings and cosine similarity — Representing text as vectors so that similar meanings can be found by simple arithmetic.
• GLiNER and GLiNER2 — Small extraction models that take their labels at run time, with no retraining.
• Natural language inference — Classifying whether one text entails, contradicts or is neutral towards another, and using that for zero-shot classification.