TokenShapley: Token Level Context Attribution with Shapley Value
Yingtai Xiao, Yuqing Zhu, Sirat Samyoun, Wanrong Zhang, Jiachen T. Wang, Jian Du · 2025
Read TokenShapley: Token Level Context Attribution with Shapley Value by Yingtai Xiao, Yuqing Zhu, Sirat Samyoun, Wanrong Zhang, Jiachen T. Wang, Jian Du (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2507-05261.txt.
Description
Yingtai Xiao and co-authors present TokenShapley, a method that attributes a language model's answer to specific tokens in its context. Their 2025 paper deals with a gap in checking model output: sentence-level attribution cannot credit a single number, year, or name.
TokenShapley joins Shapley value data attribution with nearest-neighbor retrieval, using a precomputed datastore for context. Shapley values quantify how much each token matters for the answer. That gives fine-grained credit at the token level, not just the sentence level. Tests on four benchmarks show the method beats prior baselines at token-level attribution. Such credit can help users verify a response piece by piece.
Catalog record
| CREATOR | Yingtai Xiao, Yuqing Zhu, Sirat Samyoun, Wanrong Zhang, Jiachen T. Wang, Jian Du |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 44 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-05261