Cell-level attribution on ToTTo (gold set)
56.89PrecisionTRACEBACK
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TRACEBACKvariant=Full2026.02 | 56.89 | 48.39 | |
| TRACEBACK-Litevariant=Lite2026.02 | 53.87 | 40.2 | |
| Fewshot + CoTmode=Few-shot, reasoning=Chain-of-Thought2026.02 | 30.81 | 13.25 | |
| SBERT + GPT-4oembedding=SBERT, model=GPT-4o2026.02 | 20.51 | 16.84 | |
| INSEQ2026.02 | 16.85 | 18.95 | |
| GENERATIONPROGRAMSalias=GP2026.02 | 14.96 | 11.32 |