Temporal Anchor Injection: A Semantic Method for Simulated Causal Time in Transformer Embeddings
Fuente:
Zenodo
Salvato in:
| Autore principale: | |
|---|---|
| Natura: | Recurso digital |
| Lingua: | inglese |
| Pubblicazione: |
Zenodo
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866902218925408256 |
|---|---|
| author | Schetnikov, Aleksey Sergeevich |
| author_facet | Schetnikov, Aleksey Sergeevich |
| contents | <p>Transformer-based language models lack an intrinsic understanding of time and causality. This work introduces Temporal Anchor Injection (TAI), a novel method that simulates semantic time by computing vector deltas between cause-effect events in embedding space. These delta vectors reveal an emergent temporal structure, enabling causal classification without modifying the underlying model. TAI is lightweight, model-agnostic, and effective for detecting narrative flow and temporal inconsistencies in language.</p> <p> </p> <p>Contact:<br>Telegram: <a target="_new" rel="noopener">@Alex_larinov</a><br>Email: <a rel="noopener">alex21259alex@gmail.com</a></p> |
| format | Recurso digital |
| id | zenodo_https___doi_org_10_5281_zenodo_15576522 |
| institution | Zenodo |
| language | eng |
| publishDate | 2025 |
| publisher | Zenodo |
| record_format | zenodo |
| spellingShingle | Temporal Anchor Injection: A Semantic Method for Simulated Causal Time in Transformer Embeddings Schetnikov, Aleksey Sergeevich causality temporal reasoning embeddings transformer models interpretability NLP semantic vectors <p>Transformer-based language models lack an intrinsic understanding of time and causality. This work introduces Temporal Anchor Injection (TAI), a novel method that simulates semantic time by computing vector deltas between cause-effect events in embedding space. These delta vectors reveal an emergent temporal structure, enabling causal classification without modifying the underlying model. TAI is lightweight, model-agnostic, and effective for detecting narrative flow and temporal inconsistencies in language.</p> <p> </p> <p>Contact:<br>Telegram: <a target="_new" rel="noopener">@Alex_larinov</a><br>Email: <a rel="noopener">alex21259alex@gmail.com</a></p> |
| title | Temporal Anchor Injection: A Semantic Method for Simulated Causal Time in Transformer Embeddings |
| topic | causality temporal reasoning embeddings transformer models interpretability NLP semantic vectors |
| url | https://doi.org/10.5281/zenodo.15576522 |