--- license: cc-by-nc-sa-4.0 language: [ca, cs, cu, de, en, es, fr, grc, hbo, hi, hu, ko, lt, no, pl, ru, tr] tags: [ CorPipe, CorPipe 25, CRAC Shared Task, CRAC 2025, CorefUD, CorefUD 1.3, Coreference Resolution, Minnt] base_model: google/mt5-large --- # The `corpipe25-corefud1.3-large-251101` Model The `corpipe25-corefud1.3-large-251101` is a `mT5-large`-based multilingual model for coreference resolution usable in CorPipe 25 . It is released on [LINDAT/CLARIAH-CZ](https://hdl.handle.net/11234/1-6078) and on [HuggingFace](https://huggingface.co/ufal/corpipe25-corefud1.3-large-251101) under the CC BY-NC-SA 4.0 license. The model is downloaded automatically from HuggingFace when running prediction with the `--load ufal/corpipe25-corefud1.3-large-251101` argument. The model is language agnostic, so it can be in theory used to predict coreference in any `mT5` language; for zero-shot cross-lingual evaluation, please refer to the CRAC 2025 paper. The model expects empty nodes to be already present on input, predicted by https://github.com/ufal/crac2025_empty_nodes_baseline. The model was trained using the following command (see the CorPipe 25 repository for more information): ```sh tbs="ca_ancora cs_pcedt cs_pdt cu_proiel de_potsdamcc en_gum en_litbank es_ancora fr_ancor fr_democrat grc_proiel hbo_ptnk hi_hdtb hu_korkor hu_szegedkoref ko_ecmt lt_lcc no_bokmaalnarc no_nynorsknarc pl_pcc ru_rucor tr_itcc" python3 corpipe25.py --train --dev --treebanks $(for c in $tbs; do echo data/$c/$c-corefud-train.conllu; done) --batch_size=8 --learning_rate=6e-4 --learning_rate_decay --adafactor --encoder=google/mt5-large --exp=corpipe25-corefud1.3-large --compile ``` ## CorefUD 1.3 Test Sets Results The model achieves the following CorefUD 1.3 test set results (as reported in the paper); segment size 2560 was used, with the exception for `cu_proiel` and `grc_proiel` where it was 512: | avg | ca | cs_pce| cs_pdt| cu | de_pot| en_gum| en_lit| es | fr_anc| fr_dem| grc | hbo_pt| hi | hu_kor| hu_sze| ko_emc| lt | no_bok| no_nyn| pl | ru | tr | |-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------|-------| | 72.84 | 80.1 | 74.6 | 78.0 | 58.5 | 67.2 | 73.3 | 77.4 | 82.0 | 72.1 | 68.5 | 71.2 | 67.9 | 76.3 | 67.3 | 68.0 | 69.8 | 74.4 | 75.2 | 74.0 | 77.5 | 81.2 | 67.7 | ## Running the Model on Plain Text To run the model on plain text, first the plain text needs to be tokenized and converted to CoNLL-U (and optionally parsed if you also want mention heads), by using for example UDPipe 2: ```sh curl -F data="Eve came home and Peter greeted her there. Then Peter and Paul set out to a trip and Eve waved them off." \ -F model=english -F tokenizer= -F tagger= -F parser= https://lindat.mff.cuni.cz/services/udpipe/api/process \ | python -X utf8 -c "import sys,json; sys.stdout.write(json.load(sys.stdin)['result'])" >input.conllu ``` Then the CoNLL-U file can be processed by CorPipe 25, by using for example ```sh python3 corpipe25.py --load ufal/corpipe25-corefud1.3-large-251101 --exp . --epoch 0 --test input.conllu ``` which would generate the following predictions in `input.00.conllu`: ``` # generator = UDPipe 2, https://lindat.mff.cuni.cz/services/udpipe # udpipe_model = english-ewt-ud-2.17-251125 # udpipe_model_licence = CC BY-NC-SA # newdoc # global.Entity = eid-etype-head-other # newpar # sent_id = 1 # text = Eve came home and Peter greeted her there. 1 Eve Eve PROPN NNP Number=Sing 2 nsubj _ Entity=(c1--1) 2 came come VERB VBD Mood=Ind|Number=Sing|Person=3|Tense=Past|VerbForm=Fin 0 root _ _ 3 home home ADV RB _ 2 advmod _ Entity=(c2--1) 4 and and CCONJ CC _ 6 cc _ _ 5 Peter Peter PROPN NNP Number=Sing 6 nsubj _ Entity=(c3--1) 6 greeted greet VERB VBD Mood=Ind|Number=Sing|Person=3|Tense=Past|VerbForm=Fin 2 conj _ _ 7 her she PRON PRP Case=Acc|Gender=Fem|Number=Sing|Person=3|PronType=Prs 6 obj _ Entity=(c1--1) 8 there there ADV RB PronType=Dem 6 advmod _ Entity=(c2--1)|SpaceAfter=No 9 . . PUNCT . _ 2 punct _ _ # sent_id = 2 # text = Then Peter and Paul set out to a trip and Eve waved them off. 1 Then then ADV RB PronType=Dem 5 advmod _ _ 2 Peter Peter PROPN NNP Number=Sing 5 nsubj _ Entity=(c4--1(c3--1) 3 and and CCONJ CC _ 4 cc _ _ 4 Paul Paul PROPN NNP Number=Sing 2 conj _ Entity=(c5--1)c4) 5 set set VERB VBD Mood=Ind|Number=Plur|Person=3|Tense=Past|VerbForm=Fin 0 root _ _ 6 out out ADP RP _ 5 compound:prt _ _ 7 to to ADP IN _ 9 case _ _ 8 a a DET DT Definite=Ind|PronType=Art 9 det _ Entity=(c6--2 9 trip trip NOUN NN Number=Sing 5 obl _ Entity=c6) 10 and and CCONJ CC _ 12 cc _ _ 11 Eve Eve PROPN NNP Number=Sing 12 nsubj _ Entity=(c1--1) 12 waved wave VERB VBD Mood=Ind|Number=Sing|Person=3|Tense=Past|VerbForm=Fin 5 conj _ _ 13 them they PRON PRP Case=Acc|Number=Plur|Person=3|PronType=Prs 12 obj _ Entity=(c4--1) 14 off off ADP RP _ 12 compound:prt _ SpaceAfter=No 15 . . PUNCT . _ 5 punct _ SpaceAfter=No ``` ## How to Cite ``` @inproceedings{straka-2025-corpipe, title = "{C}or{P}ipe at {CRAC} 2025: Evaluating Multilingual Encoders for Multilingual Coreference Resolution", author = "Straka, Milan", editor = "Ogrodniczuk, Maciej and Novak, Michal and Poesio, Massimo and Pradhan, Sameer and Ng, Vincent", booktitle = "Proceedings of the Eighth Workshop on Computational Models of Reference, Anaphora and Coreference", month = nov, year = "2025", address = "Suzhou, China", publisher = "Association for Computational Linguistics", url = "https://aclanthology.org/2025.crac-1.11/", doi = "10.18653/v1/2025.crac-1.11", pages = "130--139", } ```