Narrative Similarity Judgments on SemEval Task 4 Track A 2026 (dev)
81AccuracyCascadeMind
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| CascadeMindBackbone=Gemini 2.5 Flash, Sample size=n=1002026.01 | 81 | — | |
| CascadeMind + Symbolic TiebreakerBackbone=Gemini 2.5 Flash, Component=Symbolic Tiebreaker, Sample size=n=1002026.01 | 81 | — | |
| Majorityk=3 calls, Backbone=Gemini 2.5 Flash, Sample size=n=2002026.01 | 78 | 3 | |
| Self-Consistencyk=8, Backbone=Gemini 2.5 Flash, Sample size=n=2002026.01 | 76.5 | 1 | |
| Single VoteBackbone=Gemini 2.5 Flash, Sample size=n=2002026.01 | 68 | 1 |