Competitions — where to submit the model
Mova is submitted, where possible, to linguistic competitions. The candidates below have their papers in the library; folder names are given in parentheses.
| competition | what it measures | where we fit |
|---|---|---|
| CRAC — multilingual coreference on CorefUD | mention chains. 2025: "Can LLMs Dethrone Traditional Approaches?", a separate LLM track with prompts and agentic systems (novak-2025-findings-fourth-shared-task-multilingual). 2026: distant entities, 4 of 10 systems LLM-based, CorPipe won (novak-2026-findings-fifth-shared-task-multilingual) | context and pronouns, the story layer. SLM (small language model) + LLM is exactly the 2025 question |
| UniDive — morphosyntactic parsing | parsing without the notion of "word": LAS only over content nodes, node features, abstract nodes (goldman-2025-findings-unidive-2025-shared-task) | the metadialect "not even UD", converters |
| PARSEME 2.0 — multiword expressions | MWE identification based on UD (bogdanova-2026-pmi-mwe-scorer-parseme-2) | multiword expressions in seeds and domain snapshots |
| DISRPT — discourse | discourse units and relations (ju-2025-dedisco-disrpt-2025-shared-task) | the story layer, pragmatics |
| SemEval — new tasks every year | for example, intended sarcasm (farha-2022-semeval-2022-task-6-isarcasmeval), event abduction (cao-2026-semeval-2026-task-12-abductive) | pragmatics of "what was meant" |
| BabyLM — a small model on limited data | including Gricean pragmatics (askari-2025-are-babylms-deaf-gricean-maxims) | the SLM learner, the five-year-old test |
What in the rules concerns us
- Tracks. The closed track accepts only the competition data; the open or LLM track accepts any resources. SLM + LLM goes to the open or LLM track; SLM alone goes to the closed one.
- Measurement. We already measure parsing with the CoNLL 2018 script (LAS, MLAS, BLEX), so the numbers are comparable.
- Competition data is for research. It sometimes has NC parts, as GUM does. So the competition build is separate and for research, like our GUM measurement. It does not go into the production model:
commercial_gatewill stop it. - A system paper is usually mandatory. We write it from the usage log (
docs/citations.md). - Calendar and deadlines are checked when the time comes: "where possible".
This page in Markdown → https://md.mova.elamur.ai/free/en/docs/docs/competitions/