Researchers propose using the circular structure of Schwartz values as an explicit output-space geometry to improve human value detection, moving beyond treating labels as independent. They compare training-time geometry-aware objectives with a post-hoc Schwartz-aware energy decoder that scores whole label sets jointly on a DeBERTa-v3-base classifier.
- The decoder makes label sets more coherent with the continuum on theory-aware coherence metrics without affecting Macro-F1 or Micro-F1.
- Gains are specific to the true Schwartz ordering and do not appear for random permutations or empirical co-occurrence graphs.
- A bounded Qwen2.5-72B-Instruct diagnostic shows that supplying the continuum at inference shifts behavior but does not match supervised structured prediction.
Theory-aware decoding offers a lightweight, controllable way to make value detection faithful to its label space.