Presence Requires Continuity: Cross-Session Hysteresis in Transformer Generation
Abstract
We investigate whether a witness session geometrically modifies the trajectory of a subsequent emotional session in a large language model, as measured by the SERC framework. Using three session orderings and two experimental architectures (with and without continuous linguistic context), we find that the witness→emotional effect (ΔC ≈ −0.06 to −0.08) is replicated across computational regimes when context is continuous, but disappears entirely when context is absent. These findings support the thesis that cross-session hysteresis in transformer generation is a property of continuous linguistic context, not of session type in isolation.
Fields of Study & Subjects
Topic Modeling
Natural Language Processing Techniques
Speech and dialogue systems
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 11 March 2026
- Language
- EN
- Document type
- Preprint