ELMABench
ELMA: Benchmarking Anaphoric Compositions for Long-Term Text-to-Motion Generation
Evaluates dependency resolution and state persistence in long-term text-to-motion generation using 30-second clips with anaphoric multi-segment instructions.
How to run
$ npm i -g lettokenburn-cli$ ltb bench info elmabench
External harness. No Harbor adapter has been generated for this benchmark yet. `ltb bench info` prints the known links.
Runs happen on your machine with your own subscription; only the score summary is pushed. CLI docs
Scores
5 sourced entries| Harness | Source | Confidence | Link | |||
|---|---|---|---|---|---|---|
| Gemini 2.5 Pro | 0.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Qwen2.5-VL Instruct | 0.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| Gemini 2.5 Flash | 0.9 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| PoseScript | 0.8 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
| PoseFix | 0.8 | N/A | Independent | medium | Oct 9, 2026 | arxiv.org |
Every number links to where it was published. Low-confidence rows are muted and need per-cell verification.
Reproduced by the community
All runsAbout
ELMA is a dataset of continuous 30-second 3D motions split into six 5-second segments with context-aware captions and dependency labels, extracted from in-the-wild videos. ELMABench evaluates models on their ability to resolve cross-segment dependencies and maintain physical realism and semantic alignment under anaphoric constraints.