UDPipe 2 Models

1. Universal Dependencies 2.12 Models
2. Universal Dependencies 2.10 Models
3. Universal Dependencies 2.6 Models
4. Czech PDT-C 1.0 Model
5. EvaLatin 2020 Models

Universal Dependencies 2.12 Models

Universal Dependencies 2.12 Models are distributed under the CC BY-NC-SA licence. The models are based solely on Universal Dependencies 2.12 treebanks, and additionally use multilingual BERT and RobeCzech.

The models require UDPipe 2.

Download

The latest version 230717 of the Universal Dependencies 2.12 models can be downloaded from LINDAT/CLARIN repository.

The models are also available in the REST service.

Acknowledgements

This work has been supported by the Ministry of Education, Youth and Sports of the Czech Republic, Project No. LM2018101 LINDAT/CLARIAH-CZ.

The models were trained on Universal Dependencies 2.12 treebanks.

For the UD treebanks which do not contain original plain text version, raw text is used to train the tokenizer instead. The plain texts were taken from the W2C -- Web to Corpus.

Finally, multilingual BERT and RobeCzech are used to provide contextualized word embeddings.

Publications

• Milan Straka, Jakub Náplava, Jana Straková, David Samuel (2020): RobeCzech: Czech RoBERTa, a monolingual contextualized language representation model. In: Text, Speech, and Dialogue. TSD 2021. Lecture Notes in Computer Science, vol 12848. Springer, Cham.
• Milan Straka, Jana Straková, Jan Hajič (2019): Evaluating Contextualized Embeddings on 54 Languages in POS Tagging, Lemmatization and Dependency Parsing. In: ArXiv.org Computing Research Repository, ISSN 2331-8422, 1904.02099
• Milan Straka (2018): UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task. In: Proceedings of CoNLL 2018: The SIGNLL Conference on Computational Natural Language Learning, pp. 197-207, Association for Computational Linguistics, Stroudsburg, PA, USA, ISBN 978-1-948087-72-8

Model Description

The Universal Dependencies 2.12 models contain 131 models of 72 languages, each consisting of a tokenizer, tagger, lemmatizer and dependency parser, all trained using the UD data. We used the original train-dev-test split, but for treebanks with only train and no dev data we used last 10% of the train data as dev data. We produce models only for treebanks with at least 1000 training words.

The tokenizer is trained using the SpaceAfter=No features. If the features are not present in the data, they can be filled in using raw text in the language in question.

The tagger, lemmatizer and parser are trained using gold UD data.

Model Performance

We present the tokenizer, tagger, lemmatizer and parser performance, measured on the testing portion of the data, evaluated both on the raw text and using the gold tokenization. The results are F1 scores measured by the conll18_ud_eval.py script.

Model	Mode	Words	Sents	UPOS	XPOS	UFeats	AllTags	Lemma	UAS	LAS	MLAS	BLEX
afrikaans-afribooms-ud-2.12-230717	Raw text	99.94	99.65	98.65	95.80	98.35	95.62	98.29	90.41	87.46	78.92	79.72
afrikaans-afribooms-ud-2.12-230717	Gold tokenization	—	—	98.71	95.85	98.41	95.67	98.33	90.58	87.63	79.08	79.85
ancient_greek-proiel-ud-2.12-230717	Raw text	99.98	49.19	97.63	97.93	92.13	90.72	94.76	81.74	77.92	61.96	66.22
ancient_greek-proiel-ud-2.12-230717	Gold tokenization	—	—	97.87	98.14	92.58	91.30	94.82	86.59	82.72	68.31	72.13
ancient_greek-perseus-ud-2.12-230717	Raw text	99.97	98.85	92.98	85.76	91.42	84.89	86.72	80.13	74.55	54.46	55.76
ancient_greek-perseus-ud-2.12-230717	Gold tokenization	—	—	93.02	85.78	91.44	84.91	86.74	80.30	74.71	54.56	55.87
ancient_hebrew-ptnk-ud-2.12-230717	Raw text	71.14	98.06	69.03	69.17	67.09	66.13	67.47	48.21	46.75	31.14	32.42
ancient_hebrew-ptnk-ud-2.12-230717	Gold tokenization	—	—	96.70	96.86	90.36	89.22	96.60	91.29	87.61	68.26	76.48
arabic-padt-ud-2.12-230717	Raw text	94.58	82.09	91.67	89.00	89.12	88.65	90.34	78.87	74.83	66.04	68.16
arabic-padt-ud-2.12-230717	Gold tokenization	—	—	96.96	94.34	94.50	94.01	95.20	88.24	83.65	74.73	76.32
armenian-armtdp-ud-2.12-230717	Raw text	99.28	95.70	96.29	—	91.22	90.21	94.93	86.55	81.73	68.65	73.67
armenian-armtdp-ud-2.12-230717	Gold tokenization	—	—	96.86	—	91.86	90.71	95.54	88.21	83.30	69.30	74.60
armenian-bsut-ud-2.12-230717	Raw text	99.79	98.73	97.38	—	91.96	91.20	96.79	90.06	85.83	71.10	79.05
armenian-bsut-ud-2.12-230717	Gold tokenization	—	—	97.60	—	92.16	91.41	96.99	90.62	86.38	71.65	79.57
basque-bdt-ud-2.12-230717	Raw text	99.94	99.83	96.33	—	93.32	91.44	96.38	87.81	84.80	74.93	79.49
basque-bdt-ud-2.12-230717	Gold tokenization	—	—	96.39	—	93.37	91.48	96.41	87.88	84.86	74.97	79.52
belarusian-hse-ud-2.12-230717	Raw text	99.37	86.58	98.21	97.59	94.52	93.57	93.25	87.32	85.41	76.98	76.66
belarusian-hse-ud-2.12-230717	Gold tokenization	—	—	98.81	98.16	95.10	94.13	93.82	89.79	87.60	78.81	78.42
bulgarian-btb-ud-2.12-230717	Raw text	99.91	94.17	99.20	97.25	97.96	96.87	98.01	94.41	91.71	85.92	86.61
bulgarian-btb-ud-2.12-230717	Gold tokenization	—	—	99.33	97.37	98.09	96.99	98.09	95.22	92.48	86.61	87.35
catalan-ancora-ud-2.12-230717	Raw text	99.95	99.08	99.11	97.25	98.72	96.98	99.40	94.92	93.43	88.06	89.51
catalan-ancora-ud-2.12-230717	Gold tokenization	—	—	99.17	97.36	98.79	97.09	99.46	95.08	93.59	88.27	89.70
chinese-gsdsimp-ud-2.12-230717	Raw text	90.29	99.10	87.23	87.15	89.70	86.47	90.23	72.74	70.28	63.38	66.93
chinese-gsdsimp-ud-2.12-230717	Gold tokenization	—	—	96.10	96.00	99.43	95.33	99.93	87.08	83.97	78.23	82.55
chinese-gsd-ud-2.12-230717	Raw text	90.27	99.10	87.18	87.09	89.68	86.42	90.20	72.57	70.14	63.02	66.73
chinese-gsd-ud-2.12-230717	Gold tokenization	—	—	96.16	96.05	99.41	95.38	99.92	87.18	84.07	78.07	82.55
classical_chinese-kyoto-ud-2.12-230717	Raw text	97.94	46.37	89.68	88.94	91.59	86.06	97.51	71.39	66.13	62.36	64.70
classical_chinese-kyoto-ud-2.12-230717	Gold tokenization	—	—	93.43	92.14	94.63	89.89	99.52	84.71	79.40	75.55	78.24
coptic-scriptorium-ud-2.12-230717	Raw text	75.26	33.63	73.02	72.92	73.15	72.08	73.90	51.82	50.05	37.34	39.84
coptic-scriptorium-ud-2.12-230717	Gold tokenization	—	—	97.05	96.91	97.81	95.90	97.41	90.46	87.70	76.41	79.79
croatian-set-ud-2.12-230717	Raw text	99.93	94.79	98.49	95.82	96.31	95.53	97.78	92.41	89.57	81.89	84.77
croatian-set-ud-2.12-230717	Gold tokenization	—	—	98.57	95.90	96.39	95.62	97.83	92.91	90.07	82.34	85.21
czech-pdt-ud-2.12-230717	Raw text	99.93	93.37	99.30	98.46	98.78	98.25	99.38	95.01	93.64	90.75	92.30
czech-pdt-ud-2.12-230717	Gold tokenization	—	—	99.38	98.55	98.87	98.34	99.46	95.81	94.43	91.41	92.97
czech-cac-ud-2.12-230717	Raw text	99.99	99.68	99.69	98.25	98.07	97.74	99.30	96.20	94.95	90.84	93.11
czech-cac-ud-2.12-230717	Gold tokenization	—	—	99.70	98.26	98.08	97.74	99.31	96.20	94.95	90.86	93.13
czech-cltt-ud-2.12-230717	Raw text	99.32	96.92	98.95	93.72	93.99	93.65	98.48	92.47	90.77	82.87	88.60
czech-cltt-ud-2.12-230717	Gold tokenization	—	—	99.40	94.18	94.49	94.07	99.03	93.17	91.20	82.92	88.86
czech-fictree-ud-2.12-230717	Raw text	99.99	98.95	99.14	97.07	97.93	96.88	99.29	96.17	94.67	89.47	92.40
czech-fictree-ud-2.12-230717	Gold tokenization	—	—	99.16	97.08	97.95	96.89	99.30	96.24	94.73	89.57	92.50
danish-ddt-ud-2.12-230717	Raw text	99.82	89.80	97.98	—	97.36	96.60	97.38	88.60	86.46	79.10	81.13
danish-ddt-ud-2.12-230717	Gold tokenization	—	—	98.17	—	97.59	96.83	97.52	89.82	87.64	80.18	82.15
dutch-alpino-ud-2.12-230717	Raw text	99.75	89.10	97.77	96.66	97.68	96.28	94.91	92.90	90.52	83.71	80.13
dutch-alpino-ud-2.12-230717	Gold tokenization	—	—	98.02	96.82	97.89	96.45	95.16	94.31	91.90	85.01	81.32
dutch-lassysmall-ud-2.12-230717	Raw text	99.77	74.71	97.33	96.15	96.86	95.50	96.18	90.99	88.49	80.77	79.51
dutch-lassysmall-ud-2.12-230717	Gold tokenization	—	—	97.67	96.84	97.49	96.29	96.51	94.63	91.91	85.25	84.07
english-ewt-ud-2.12-230717	Raw text	99.09	87.82	96.69	96.32	96.70	95.01	97.26	90.71	88.81	82.26	84.29
english-ewt-ud-2.12-230717	Gold tokenization	—	—	97.50	97.17	97.56	95.84	98.08	93.03	91.04	84.70	86.64
english-atis-ud-2.12-230717	Raw text	99.98	80.49	99.04	—	98.59	98.14	99.63	94.45	93.02	87.97	90.18
english-atis-ud-2.12-230717	Gold tokenization	—	—	99.07	—	98.65	98.21	99.64	96.17	94.59	90.22	92.42
english-eslspok-ud-2.12-230717	Raw text	100.00	92.21	98.59	98.54	—	98.10	—	95.37	93.87	91.26	93.01
english-eslspok-ud-2.12-230717	Gold tokenization	—	—	98.68	98.68	—	98.19	—	96.20	94.79	91.98	93.74
english-gum-ud-2.12-230717	Raw text	99.63	95.72	98.00	98.04	97.91	97.02	98.84	92.87	90.96	85.35	87.10
english-gum-ud-2.12-230717	Gold tokenization	—	—	98.34	98.39	98.27	97.36	99.15	93.72	91.79	86.11	87.79
english-lines-ud-2.12-230717	Raw text	99.92	87.45	97.68	96.86	97.03	94.48	98.33	91.32	88.55	80.61	83.70
english-lines-ud-2.12-230717	Gold tokenization	—	—	97.78	96.95	97.13	94.58	98.39	92.18	89.38	81.36	84.51
english-partut-ud-2.12-230717	Raw text	99.72	100.00	97.43	97.29	96.44	95.41	98.17	94.24	92.33	83.28	87.38
english-partut-ud-2.12-230717	Gold tokenization	—	—	97.68	97.54	96.68	95.69	98.44	94.42	92.52	83.71	87.64
erzya-jr-ud-2.12-230717	Raw text	99.18	94.15	87.94	87.38	78.90	73.49	84.89	72.92	63.24	41.32	48.34
erzya-jr-ud-2.12-230717	Gold tokenization	—	—	88.66	88.07	79.54	74.02	85.52	74.08	64.24	41.90	48.91
estonian-edt-ud-2.12-230717	Raw text	99.94	92.23	97.67	98.21	96.42	95.26	95.45	88.56	85.96	80.03	79.59
estonian-edt-ud-2.12-230717	Gold tokenization	—	—	97.79	98.26	96.50	95.38	95.52	89.43	86.81	80.81	80.32
estonian-ewt-ud-2.12-230717	Raw text	98.63	78.03	94.88	96.14	94.10	91.81	93.75	83.23	80.01	72.15	73.37
estonian-ewt-ud-2.12-230717	Gold tokenization	—	—	96.22	97.46	95.35	93.07	95.00	87.27	83.71	74.95	76.20
faroese-farpahc-ud-2.12-230717	Raw text	99.74	92.77	97.45	93.00	94.24	92.32	99.74	86.01	82.29	68.12	75.32
faroese-farpahc-ud-2.12-230717	Gold tokenization	—	—	97.67	93.17	94.49	92.53	100.00	86.96	83.20	69.20	76.51
finnish-tdt-ud-2.12-230717	Raw text	99.70	90.82	97.67	98.31	96.08	95.23	92.11	90.42	88.48	82.43	78.28
finnish-tdt-ud-2.12-230717	Gold tokenization	—	—	98.00	98.59	96.39	95.54	92.39	91.72	89.75	83.44	79.26
finnish-ftb-ud-2.12-230717	Raw text	99.91	86.84	96.70	95.08	96.78	94.01	95.76	90.17	87.35	80.13	80.78
finnish-ftb-ud-2.12-230717	Gold tokenization	—	—	97.08	95.29	96.87	94.35	95.90	92.40	89.53	82.71	83.28
french-gsd-ud-2.12-230717	Raw text	98.84	94.93	97.33	—	97.25	96.57	97.72	93.15	91.20	84.70	86.92
french-gsd-ud-2.12-230717	Gold tokenization	—	—	98.48	—	98.32	97.65	98.85	94.94	93.11	86.58	88.24
french-parisstories-ud-2.12-230717	Raw text	99.73	93.08	97.20	—	93.07	91.20	98.02	79.94	76.66	62.32	71.73
french-parisstories-ud-2.12-230717	Gold tokenization	—	—	97.48	—	93.30	91.45	98.26	81.17	77.85	63.20	72.63
french-partut-ud-2.12-230717	Raw text	99.42	98.64	97.43	96.97	95.28	94.51	97.89	94.44	92.83	82.58	87.48
french-partut-ud-2.12-230717	Gold tokenization	—	—	98.12	97.62	95.89	95.12	98.50	95.35	93.89	83.73	88.49
french-rhapsodie-ud-2.12-230717	Raw text	99.16	99.82	97.31	97.37	96.16	93.38	98.19	88.06	84.92	75.51	80.31
french-rhapsodie-ud-2.12-230717	Gold tokenization	—	—	98.19	98.11	97.02	94.15	99.00	89.33	86.14	76.49	80.94
french-sequoia-ud-2.12-230717	Raw text	99.15	89.53	98.40	—	97.19	96.84	98.30	94.06	92.75	86.39	89.39
french-sequoia-ud-2.12-230717	Gold tokenization	—	—	99.25	—	98.01	97.63	99.14	95.63	94.37	88.23	90.53
galician-treegal-ud-2.12-230717	Raw text	98.74	87.99	95.93	93.63	94.83	92.83	96.76	83.46	79.60	67.95	71.94
galician-treegal-ud-2.12-230717	Gold tokenization	—	—	97.19	94.72	95.89	93.88	97.90	86.99	82.78	71.43	75.90
galician-ctg-ud-2.12-230717	Raw text	99.22	97.22	97.24	97.07	99.05	96.65	98.12	85.14	82.72	71.07	75.57
galician-ctg-ud-2.12-230717	Gold tokenization	—	—	97.97	97.79	99.83	97.35	98.86	86.86	84.31	73.03	77.52
german-gsd-ud-2.12-230717	Raw text	99.76	82.68	96.16	97.53	90.78	88.15	96.91	87.04	83.20	65.39	75.45
german-gsd-ud-2.12-230717	Gold tokenization	—	—	96.47	97.80	91.21	88.63	97.18	88.79	84.99	66.97	77.23
german-hdt-ud-2.12-230717	Raw text	99.90	92.39	98.55	98.46	94.21	93.81	97.69	96.90	96.00	84.87	90.50
german-hdt-ud-2.12-230717	Gold tokenization	—	—	98.66	98.59	94.34	93.95	97.79	97.60	96.71	85.54	91.20
gothic-proiel-ud-2.12-230717	Raw text	100.00	31.12	96.17	96.70	89.88	87.80	94.71	78.72	72.74	58.81	63.43
gothic-proiel-ud-2.12-230717	Gold tokenization	—	—	96.85	97.22	91.01	89.27	94.78	86.66	80.81	68.63	72.61
greek-gdt-ud-2.12-230717	Raw text	99.87	90.19	98.19	98.21	95.72	95.10	96.09	92.99	91.18	81.67	81.75
greek-gdt-ud-2.12-230717	Gold tokenization	—	—	98.32	98.34	95.83	95.20	96.17	93.80	91.91	82.25	82.38
greek-gud-ud-2.12-230717	Raw text	99.92	94.98	97.01	96.26	94.24	90.55	95.76	92.94	90.06	75.94	80.42
greek-gud-ud-2.12-230717	Gold tokenization	—	—	97.11	96.32	94.32	90.65	95.83	93.59	90.68	76.44	80.91
hebrew-htb-ud-2.12-230717	Raw text	85.10	99.69	82.96	82.95	81.30	80.69	83.02	70.71	68.21	55.84	60.10
hebrew-htb-ud-2.12-230717	Gold tokenization	—	—	97.64	97.62	95.79	95.27	97.36	92.45	89.94	79.28	82.12
hebrew-iahltwiki-ud-2.12-230717	Raw text	88.54	97.16	85.97	85.97	81.45	80.46	87.15	76.11	74.26	57.99	67.30
hebrew-iahltwiki-ud-2.12-230717	Gold tokenization	—	—	97.09	97.09	92.18	91.10	98.29	93.66	91.26	75.14	85.53
hindi-hdtb-ud-2.12-230717	Raw text	100.00	98.72	97.74	97.35	94.23	92.39	98.93	95.31	92.43	79.64	87.78
hindi-hdtb-ud-2.12-230717	Gold tokenization	—	—	97.74	97.34	94.25	92.40	98.94	95.43	92.55	79.77	87.94
hungarian-szeged-ud-2.12-230717	Raw text	99.85	95.89	96.68	—	94.18	93.47	94.89	88.56	84.89	74.96	78.33
hungarian-szeged-ud-2.12-230717	Gold tokenization	—	—	96.77	—	94.31	93.57	95.01	88.99	85.30	75.22	78.65
icelandic-modern-ud-2.12-230717	Raw text	99.37	94.59	97.58	95.34	88.49	85.62	96.87	86.05	83.30	64.64	75.54
icelandic-modern-ud-2.12-230717	Gold tokenization	—	—	98.15	95.92	88.93	86.07	97.45	87.03	84.17	65.30	76.47
icelandic-gc-ud-2.12-230717	Raw text	99.72	94.64	94.72	82.28	85.01	79.83	91.64	83.22	78.78	58.56	68.85
icelandic-gc-ud-2.12-230717	Gold tokenization	—	—	95.06	82.71	85.52	80.34	91.81	84.14	79.66	59.21	69.49
icelandic-icepahc-ud-2.12-230717	Raw text	99.80	92.67	96.90	93.31	92.01	87.13	96.24	87.30	83.46	66.91	74.57
icelandic-icepahc-ud-2.12-230717	Gold tokenization	—	—	97.08	93.55	92.18	87.35	96.39	87.85	83.95	67.40	75.14
indonesian-gsd-ud-2.12-230717	Raw text	99.49	92.35	94.35	94.03	95.77	89.10	98.12	87.62	81.71	72.43	76.99
indonesian-gsd-ud-2.12-230717	Gold tokenization	—	—	94.79	94.41	96.17	89.42	98.52	88.41	82.43	73.18	77.77
indonesian-csui-ud-2.12-230717	Raw text	99.45	91.01	95.88	96.07	96.66	95.33	98.11	85.95	81.63	76.18	78.21
indonesian-csui-ud-2.12-230717	Gold tokenization	—	—	96.34	96.58	97.15	95.78	98.74	87.26	82.71	77.14	79.20
irish-idt-ud-2.12-230717	Raw text	99.88	97.58	96.04	94.90	90.84	87.69	95.85	86.48	80.95	64.53	71.46
irish-idt-ud-2.12-230717	Gold tokenization	—	—	96.13	95.08	90.97	87.85	95.97	86.75	81.19	64.56	71.53
irish-twittirish-ud-2.12-230717	Raw text	98.50	46.62	90.58	—	—	90.58	88.41	78.58	72.34	58.38	56.97
irish-twittirish-ud-2.12-230717	Gold tokenization	—	—	91.80	—	—	91.80	89.57	85.75	79.31	66.72	64.16
italian-isdt-ud-2.12-230717	Raw text	99.74	99.07	98.44	98.38	98.14	97.64	98.68	94.73	93.05	86.79	88.06
italian-isdt-ud-2.12-230717	Gold tokenization	—	—	98.71	98.64	98.39	97.89	98.95	95.14	93.47	87.19	88.54
italian-markit-ud-2.12-230717	Raw text	99.62	98.24	96.96	97.13	94.12	92.60	88.18	88.60	84.72	70.64	77.87
italian-markit-ud-2.12-230717	Gold tokenization	—	—	97.35	97.52	94.39	92.88	88.50	89.39	85.51	71.25	78.65
italian-parlamint-ud-2.12-230717	Raw text	99.42	94.12	98.59	97.96	97.95	97.05	98.63	91.93	89.97	84.20	86.02
italian-parlamint-ud-2.12-230717	Gold tokenization	—	—	99.17	98.52	98.50	97.58	99.16	93.41	91.44	85.80	87.60
italian-partut-ud-2.12-230717	Raw text	99.73	100.00	98.41	98.41	98.19	97.64	98.57	96.15	94.15	87.84	88.90
italian-partut-ud-2.12-230717	Gold tokenization	—	—	98.52	98.52	98.27	97.72	98.82	96.18	94.09	87.68	88.80
italian-postwita-ud-2.12-230717	Raw text	99.36	49.53	96.58	96.33	96.33	94.80	96.62	82.80	79.03	68.81	70.55
italian-postwita-ud-2.12-230717	Gold tokenization	—	—	97.17	96.95	96.90	95.44	97.25	87.96	83.79	75.19	76.95
italian-twittiro-ud-2.12-230717	Raw text	98.94	46.67	95.91	95.74	94.94	93.42	94.57	83.09	78.65	66.30	66.97
italian-twittiro-ud-2.12-230717	Gold tokenization	—	—	96.84	96.57	95.87	94.15	95.43	88.34	83.43	71.82	72.62
italian-vit-ud-2.12-230717	Raw text	99.74	94.87	98.11	97.29	97.65	96.15	98.87	92.22	89.43	81.28	84.07
italian-vit-ud-2.12-230717	Gold tokenization	—	—	98.35	97.65	97.87	96.51	99.10	92.94	90.12	81.99	84.78
japanese-gsdluw-ud-2.12-230717	Raw text	95.18	99.72	93.82	93.50	95.18	93.44	93.56	86.27	85.58	76.26	76.41
japanese-gsdluw-ud-2.12-230717	Gold tokenization	—	—	98.36	98.01	100.00	97.90	97.78	95.16	94.19	86.28	84.89
japanese-gsd-ud-2.12-230717	Raw text	96.17	100.00	94.97	94.18	96.16	93.85	95.03	87.91	87.07	80.80	80.98
japanese-gsd-ud-2.12-230717	Gold tokenization	—	—	98.59	97.52	99.99	97.20	98.47	94.93	93.94	88.80	88.47
korean-kaist-ud-2.12-230717	Raw text	100.00	100.00	96.19	87.78	—	87.58	94.18	88.85	86.92	82.77	80.35
korean-kaist-ud-2.12-230717	Gold tokenization	—	—	96.19	87.78	—	87.58	94.18	88.85	86.92	82.77	80.35
korean-gsd-ud-2.12-230717	Raw text	99.87	93.93	96.54	90.07	99.67	87.94	93.62	87.88	83.98	80.68	76.82
korean-gsd-ud-2.12-230717	Gold tokenization	—	—	96.72	90.24	99.79	88.12	93.74	88.69	84.76	81.49	77.58
latin-ittb-ud-2.12-230717	Raw text	99.99	91.21	99.01	96.65	97.07	95.62	99.16	90.25	88.31	82.53	85.95
latin-ittb-ud-2.12-230717	Gold tokenization	—	—	99.03	96.66	97.12	95.64	99.18	91.28	89.35	83.17	86.55
latin-llct-ud-2.12-230717	Raw text	99.99	99.49	99.75	97.14	97.15	96.87	97.76	95.37	94.37	89.06	90.39
latin-llct-ud-2.12-230717	Gold tokenization	—	—	99.75	97.14	97.16	96.87	97.77	95.39	94.39	89.07	90.41
latin-perseus-ud-2.12-230717	Raw text	99.95	98.99	92.88	81.11	84.60	77.45	88.86	78.92	71.78	53.50	59.27
latin-perseus-ud-2.12-230717	Gold tokenization	—	—	92.95	81.14	84.65	77.49	88.89	79.08	71.91	53.59	59.31
latin-proiel-ud-2.12-230717	Raw text	99.85	37.40	96.60	96.69	90.66	89.40	96.19	76.66	72.46	59.50	64.96
latin-proiel-ud-2.12-230717	Gold tokenization	—	—	97.02	97.12	91.43	90.31	96.42	83.88	79.55	67.88	73.35
latin-udante-ud-2.12-230717	Raw text	99.61	98.81	90.94	75.50	84.14	72.23	86.97	75.88	68.63	47.63	51.46
latin-udante-ud-2.12-230717	Gold tokenization	—	—	91.18	75.56	84.38	72.29	87.20	75.95	68.65	47.73	51.44
latvian-lvtb-ud-2.12-230717	Raw text	99.29	97.80	96.79	90.92	94.75	90.16	96.57	88.79	85.85	77.56	81.04
latvian-lvtb-ud-2.12-230717	Gold tokenization	—	—	97.44	91.56	95.44	90.77	97.21	89.95	86.94	78.74	82.16
lithuanian-alksnis-ud-2.12-230717	Raw text	99.91	87.87	95.95	90.31	91.09	89.50	93.45	82.74	78.94	68.03	71.20
lithuanian-alksnis-ud-2.12-230717	Gold tokenization	—	—	96.08	90.44	91.27	89.64	93.56	83.94	80.08	68.93	72.06
lithuanian-hse-ud-2.12-230717	Raw text	97.30	97.30	89.93	90.03	81.92	79.03	88.16	71.85	62.63	44.48	53.82
lithuanian-hse-ud-2.12-230717	Gold tokenization	—	—	91.32	91.42	83.40	80.09	90.75	75.00	65.00	45.87	55.84
maghrebi_arabic_french-arabizi-ud-2.12-230717	Raw text	91.65	7.00	78.81	71.59	82.65	69.81	50.63	57.90	49.93	36.22	24.59
maghrebi_arabic_french-arabizi-ud-2.12-230717	Gold tokenization	—	—	86.69	78.89	90.82	77.88	54.66	76.32	65.86	47.72	31.57
maltese-mudt-ud-2.12-230717	Raw text	99.84	86.29	95.73	95.79	—	95.31	—	85.08	80.25	68.88	73.05
maltese-mudt-ud-2.12-230717	Gold tokenization	—	—	95.87	95.92	—	95.46	—	85.70	80.81	69.28	73.43
manx-cadhan-ud-2.12-230717	Raw text	97.36	98.25	94.18	—	95.78	93.37	93.43	87.42	84.02	77.75	78.03
manx-cadhan-ud-2.12-230717	Gold tokenization	—	—	96.77	—	98.39	95.93	95.98	92.47	89.13	83.16	82.07
marathi-ufal-ud-2.12-230717	Raw text	94.16	92.63	82.73	—	74.21	70.80	84.91	68.13	60.83	39.75	47.62
marathi-ufal-ud-2.12-230717	Gold tokenization	—	—	87.14	—	76.94	73.06	87.86	73.79	65.29	41.82	50.10
naija-nsc-ud-2.12-230717	Raw text	99.95	100.00	98.04	—	98.92	97.53	99.33	93.02	90.46	87.48	88.98
naija-nsc-ud-2.12-230717	Gold tokenization	—	—	98.08	—	98.96	97.56	99.39	93.10	90.53	87.55	89.03
north_sami-giella-ud-2.12-230717	Raw text	99.87	98.79	91.63	93.51	89.29	85.24	86.91	75.78	70.85	60.16	58.42
north_sami-giella-ud-2.12-230717	Gold tokenization	—	—	91.78	93.65	89.42	85.38	87.01	76.01	71.08	60.38	58.60
norwegian-bokmaal-ud-2.12-230717	Raw text	99.82	97.27	98.38	98.94	97.47	96.81	98.58	93.88	92.64	86.95	88.78
norwegian-bokmaal-ud-2.12-230717	Gold tokenization	—	—	98.58	99.14	97.65	97.01	98.78	94.54	93.28	87.54	89.41
norwegian-nynorsk-ud-2.12-230717	Raw text	99.93	94.54	98.43	99.16	97.35	96.55	98.46	93.79	92.37	85.84	88.08
norwegian-nynorsk-ud-2.12-230717	Gold tokenization	—	—	98.62	99.27	97.51	96.75	98.60	94.58	93.19	86.82	89.08
old_church_slavonic-proiel-ud-2.12-230717	Raw text	100.00	40.05	96.11	96.39	89.72	88.01	90.18	78.06	73.64	60.96	61.97
old_church_slavonic-proiel-ud-2.12-230717	Gold tokenization	—	—	96.66	96.98	90.32	88.99	90.19	85.05	80.53	68.59	69.15
old_east_slavic-torot-ud-2.12-230717	Raw text	100.00	34.53	95.40	95.48	89.92	87.64	88.09	76.94	72.12	58.23	57.89
old_east_slavic-torot-ud-2.12-230717	Gold tokenization	—	—	95.89	95.94	90.67	88.72	88.14	85.27	80.38	67.99	65.94
old_east_slavic-birchbark-ud-2.12-230717	Raw text	99.99	16.66	88.50	99.37	76.09	72.03	65.58	64.39	57.66	32.84	27.49
old_east_slavic-birchbark-ud-2.12-230717	Gold tokenization	—	—	89.13	99.38	76.70	72.84	65.59	76.43	69.19	41.15	32.90
old_east_slavic-rnc-ud-2.12-230717	Raw text	97.64	60.48	92.65	86.46	78.10	68.76	77.22	64.58	60.03	36.90	37.52
old_east_slavic-rnc-ud-2.12-230717	Gold tokenization	—	—	93.99	88.74	79.24	69.78	78.28	70.79	65.17	39.63	40.04
old_french-srcmf-ud-2.12-230717	Raw text	99.70	100.00	96.74	96.58	97.76	95.78	99.66	90.92	87.17	80.65	84.10
old_french-srcmf-ud-2.12-230717	Gold tokenization	—	—	97.06	96.91	98.07	96.08	99.96	91.34	87.62	81.11	84.56
persian-perdt-ud-2.12-230717	Raw text	99.66	99.83	97.51	97.39	97.64	95.63	98.91	93.56	91.32	86.30	88.68
persian-perdt-ud-2.12-230717	Gold tokenization	—	—	97.81	97.68	97.94	95.92	99.23	94.08	91.84	86.86	89.27
persian-seraji-ud-2.12-230717	Raw text	99.65	98.75	97.89	97.91	97.90	97.44	98.26	91.78	88.92	84.47	84.56
persian-seraji-ud-2.12-230717	Gold tokenization	—	—	98.22	98.23	98.22	97.74	98.54	92.47	89.62	85.12	85.22
polish-pdb-ud-2.12-230717	Raw text	99.85	97.33	98.86	95.76	95.89	95.08	98.08	94.21	92.16	85.14	88.25
polish-pdb-ud-2.12-230717	Gold tokenization	—	—	99.02	95.92	96.04	95.22	98.21	94.72	92.67	85.55	88.68
polish-lfg-ud-2.12-230717	Raw text	99.85	99.65	98.97	96.08	96.49	95.15	98.24	96.78	95.40	89.61	92.26
polish-lfg-ud-2.12-230717	Gold tokenization	—	—	99.13	96.26	96.67	95.32	98.38	97.19	95.81	90.02	92.60
pomak-philotis-ud-2.12-230717	Raw text	99.98	94.49	98.80	—	95.54	95.26	96.71	88.17	83.04	70.60	73.82
pomak-philotis-ud-2.12-230717	Gold tokenization	—	—	98.82	—	95.54	95.26	96.73	88.63	83.54	71.09	74.23
portuguese-bosque-ud-2.12-230717	Raw text	99.68	89.73	97.88	—	96.86	95.87	98.27	92.13	89.83	80.75	84.19
portuguese-bosque-ud-2.12-230717	Gold tokenization	—	—	98.16	—	97.11	96.11	98.57	93.28	90.94	81.78	85.38
portuguese-cintil-ud-2.12-230717	Raw text	99.41	78.66	97.42	96.01	95.29	93.21	97.66	85.13	81.88	71.77	75.51
portuguese-cintil-ud-2.12-230717	Gold tokenization	—	—	98.00	96.61	95.91	93.80	98.23	87.54	84.20	74.06	77.83
portuguese-petrogold-ud-2.12-230717	Raw text	99.59	93.11	98.75	—	98.70	98.20	99.09	94.69	93.57	88.77	90.11
portuguese-petrogold-ud-2.12-230717	Gold tokenization	—	—	99.05	—	99.00	98.46	99.51	95.62	94.42	89.68	91.18
romanian-rrt-ud-2.12-230717	Raw text	99.70	95.50	97.88	97.14	97.39	96.91	98.00	91.92	88.46	81.93	83.37
romanian-rrt-ud-2.12-230717	Gold tokenization	—	—	98.15	97.40	97.65	97.16	98.25	92.77	89.25	82.56	83.97
romanian-nonstandard-ud-2.12-230717	Raw text	98.83	96.77	96.12	91.86	90.52	89.16	94.82	88.67	84.67	68.20	76.17
romanian-nonstandard-ud-2.12-230717	Gold tokenization	—	—	97.25	92.87	91.51	90.13	95.88	90.46	86.41	69.79	77.51
romanian-simonero-ud-2.12-230717	Raw text	99.84	100.00	98.41	97.97	97.51	97.20	98.89	93.95	92.03	85.30	88.19
romanian-simonero-ud-2.12-230717	Gold tokenization	—	—	98.56	98.12	97.66	97.34	99.04	94.29	92.35	85.57	88.47
russian-syntagrus-ud-2.12-230717	Raw text	99.67	98.31	98.50	—	94.02	93.76	98.19	93.80	91.66	82.80	88.87
russian-syntagrus-ud-2.12-230717	Gold tokenization	—	—	98.83	—	94.33	94.07	98.48	94.51	92.33	83.33	89.40
russian-gsd-ud-2.12-230717	Raw text	99.50	96.49	98.05	97.56	94.57	93.51	96.87	91.64	88.70	80.95	84.53
russian-gsd-ud-2.12-230717	Gold tokenization	—	—	98.50	97.97	95.01	93.90	97.25	92.83	89.81	81.85	85.46
russian-taiga-ud-2.12-230717	Raw text	98.07	86.01	95.59	—	93.02	92.05	94.62	82.89	79.47	70.47	73.61
russian-taiga-ud-2.12-230717	Gold tokenization	—	—	97.25	—	94.80	93.67	96.34	85.53	81.91	72.70	76.02
sanskrit-vedic-ud-2.12-230717	Raw text	100.00	27.18	89.20	—	81.19	76.40	87.11	61.05	50.02	41.42	44.77
sanskrit-vedic-ud-2.12-230717	Gold tokenization	—	—	90.07	—	82.70	78.17	87.40	73.69	61.55	51.56	54.91
scottish_gaelic-arcosg-ud-2.12-230717	Raw text	97.43	61.26	93.66	89.50	91.07	88.34	94.92	80.79	76.33	64.78	69.63
scottish_gaelic-arcosg-ud-2.12-230717	Gold tokenization	—	—	96.40	92.49	94.04	91.41	97.47	87.25	82.84	71.75	76.18
serbian-set-ud-2.12-230717	Raw text	99.99	93.00	99.03	95.94	96.13	95.71	97.82	93.48	90.99	83.45	86.89
serbian-set-ud-2.12-230717	Gold tokenization	—	—	99.05	95.95	96.16	95.74	97.83	94.18	91.65	84.17	87.56
slovak-snk-ud-2.12-230717	Raw text	100.00	81.69	97.68	90.26	93.35	89.42	96.51	91.48	89.73	80.19	84.62
slovak-snk-ud-2.12-230717	Gold tokenization	—	—	97.83	90.34	93.44	89.56	96.54	93.88	92.00	82.42	86.89
slovenian-ssj-ud-2.12-230717	Raw text	99.94	98.95	98.96	97.09	97.26	96.78	98.57	94.23	92.86	87.37	89.22
slovenian-ssj-ud-2.12-230717	Gold tokenization	—	—	99.02	97.15	97.33	96.84	98.61	94.40	93.02	87.52	89.34
slovenian-sst-ud-2.12-230717	Raw text	99.97	24.74	95.82	93.37	93.58	91.64	97.69	66.76	61.98	52.04	55.84
slovenian-sst-ud-2.12-230717	Gold tokenization	—	—	96.10	93.49	93.79	91.91	97.73	78.52	73.20	63.83	68.34
spanish-ancora-ud-2.12-230717	Raw text	99.95	98.78	99.06	96.12	98.76	95.71	99.39	93.68	91.92	86.79	88.29
spanish-ancora-ud-2.12-230717	Gold tokenization	—	—	99.11	96.16	98.81	95.75	99.43	93.87	92.10	86.95	88.45
spanish-gsd-ud-2.12-230717	Raw text	99.72	94.90	97.10	—	96.74	95.07	98.58	92.51	90.28	78.76	84.27
spanish-gsd-ud-2.12-230717	Gold tokenization	—	—	97.36	—	97.01	95.32	98.83	93.38	91.10	79.58	85.17
swedish-talbanken-ud-2.12-230717	Raw text	99.84	96.53	98.37	97.23	97.31	96.43	98.17	92.08	89.72	83.69	85.90
swedish-talbanken-ud-2.12-230717	Gold tokenization	—	—	98.53	97.42	97.49	96.61	98.33	92.53	90.15	84.20	86.44
swedish-lines-ud-2.12-230717	Raw text	99.96	88.00	97.61	95.49	90.93	88.18	97.82	90.48	87.14	71.66	81.94
swedish-lines-ud-2.12-230717	Gold tokenization	—	—	97.71	95.50	90.96	88.23	97.86	91.19	87.78	72.25	82.63
tamil-ttb-ud-2.12-230717	Raw text	94.26	97.52	84.29	82.92	84.09	77.76	88.79	70.63	61.98	50.53	54.69
tamil-ttb-ud-2.12-230717	Gold tokenization	—	—	89.14	87.13	89.19	81.80	93.87	78.38	68.88	56.48	60.89
telugu-mtg-ud-2.12-230717	Raw text	99.58	96.62	92.94	92.94	98.61	92.94	—	90.72	84.07	76.19	80.19
telugu-mtg-ud-2.12-230717	Gold tokenization	—	—	93.48	93.48	99.03	93.48	—	91.68	85.02	77.22	81.03
turkish-boun-ud-2.12-230717	Raw text	96.57	86.25	90.03	86.05	81.12	71.69	90.47	73.32	67.11	49.09	60.77
turkish-boun-ud-2.12-230717	Gold tokenization	—	—	92.97	88.91	83.07	73.35	93.52	80.79	73.91	52.78	66.50
turkish-atis-ud-2.12-230717	Raw text	100.00	80.20	99.02	—	98.57	98.40	99.11	89.03	87.24	84.91	85.71
turkish-atis-ud-2.12-230717	Gold tokenization	—	—	99.04	—	98.57	98.42	99.11	90.96	89.09	86.83	87.69
turkish-framenet-ud-2.12-230717	Raw text	100.00	100.00	96.52	—	94.75	93.87	96.39	93.73	84.80	74.12	77.96
turkish-framenet-ud-2.12-230717	Gold tokenization	—	—	96.52	—	94.75	93.87	96.39	93.73	84.80	74.12	77.96
turkish-imst-ud-2.12-230717	Raw text	97.94	97.70	93.70	93.46	90.63	88.13	94.46	75.22	66.28	55.69	60.80
turkish-imst-ud-2.12-230717	Gold tokenization	—	—	95.46	95.27	92.41	89.75	96.32	78.94	69.42	57.62	62.96
turkish-kenet-ud-2.12-230717	Raw text	100.00	98.12	93.80	—	92.10	90.85	93.50	83.94	71.51	62.32	65.35
turkish-kenet-ud-2.12-230717	Gold tokenization	—	—	93.83	—	92.12	90.89	93.51	84.10	71.63	62.46	65.49
turkish-penn-ud-2.12-230717	Raw text	99.34	80.59	95.50	—	94.48	93.29	94.14	84.40	71.73	62.08	64.36
turkish-penn-ud-2.12-230717	Gold tokenization	—	—	96.14	—	95.12	93.94	94.71	86.84	73.89	63.46	65.77
turkish-tourism-ud-2.12-230717	Raw text	99.96	99.86	98.92	—	94.98	94.67	98.27	97.04	91.43	81.58	87.09
turkish-tourism-ud-2.12-230717	Gold tokenization	—	—	98.96	—	95.02	94.72	98.31	97.10	91.50	81.66	87.18
turkish_german-sagt-ud-2.12-230717	Raw text	98.91	99.44	90.21	—	80.24	75.45	90.80	71.42	61.22	41.16	50.92
turkish_german-sagt-ud-2.12-230717	Gold tokenization	—	—	91.09	—	80.82	75.93	91.49	72.76	62.20	41.63	51.55
ukrainian-iu-ud-2.12-230717	Raw text	99.81	96.23	97.84	94.28	94.25	93.16	97.47	90.37	87.94	78.30	82.74
ukrainian-iu-ud-2.12-230717	Gold tokenization	—	—	98.03	94.44	94.39	93.31	97.67	90.97	88.52	78.70	83.18
urdu-udtb-ud-2.12-230717	Raw text	100.00	98.31	94.09	92.20	82.76	78.43	97.41	88.02	82.68	57.25	74.71
urdu-udtb-ud-2.12-230717	Gold tokenization	—	—	94.06	92.19	82.76	78.41	97.41	88.13	82.81	57.30	74.88
uyghur-udt-ud-2.12-230717	Raw text	99.54	81.87	89.77	91.72	88.23	80.82	94.71	75.32	64.44	50.04	57.14
uyghur-udt-ud-2.12-230717	Gold tokenization	—	—	90.23	92.21	88.65	81.27	95.22	77.05	66.02	50.95	58.33
vietnamese-vtb-ud-2.12-230717	Raw text	86.06	93.73	78.61	77.61	—	77.50	85.76	56.86	50.02	41.40	45.54
vietnamese-vtb-ud-2.12-230717	Gold tokenization	—	—	90.02	88.88	—	88.73	99.50	76.31	65.90	55.36	61.24
welsh-ccg-ud-2.12-230717	Raw text	99.46	97.68	95.27	94.28	89.57	87.25	94.43	86.73	80.81	63.09	69.84
welsh-ccg-ud-2.12-230717	Gold tokenization	—	—	95.74	94.73	90.03	87.66	94.94	87.83	81.85	64.04	70.94
western_armenian-armtdp-ud-2.12-230717	Raw text	99.89	98.68	96.67	—	92.31	91.63	97.14	89.26	84.68	69.80	76.24
western_armenian-armtdp-ud-2.12-230717	Gold tokenization	—	—	96.75	—	92.40	91.71	97.23	89.51	84.91	69.99	76.44
wolof-wtb-ud-2.12-230717	Raw text	99.23	91.95	94.16	94.17	93.56	91.48	95.18	83.75	78.61	66.55	69.91
wolof-wtb-ud-2.12-230717	Gold tokenization	—	—	95.11	95.08	94.34	92.33	95.88	85.90	80.57	68.67	71.71

Universal Dependencies 2.10 Models

Universal Dependencies 2.10 Models are distributed under the CC BY-NC-SA licence. The models are based solely on Universal Dependencies 2.10 treebanks, and additionally use multilingual BERT and RobeCzech.

The models require UDPipe 2.

Download

The latest version 220711 of the Universal Dependencies 2.10 models can be downloaded from LINDAT/CLARIN repository.

The models are also available in the REST service.

Acknowledgements

This work has been supported by the Ministry of Education, Youth and Sports of the Czech Republic, Project No. LM2018101 LINDAT/CLARIAH-CZ.

The models were trained on Universal Dependencies 2.10 treebanks.

For the UD treebanks which do not contain original plain text version, raw text is used to train the tokenizer instead. The plain texts were taken from the W2C -- Web to Corpus.

Finally, multilingual BERT and RobeCzech are used to provide contextualized word embeddings.

Publications

• Milan Straka, Jakub Náplava, Jana Straková, David Samuel (2020): RobeCzech: Czech RoBERTa, a monolingual contextualized language representation model. In: Text, Speech, and Dialogue. TSD 2021. Lecture Notes in Computer Science, vol 12848. Springer, Cham.
• Milan Straka, Jana Straková, Jan Hajič (2019): Evaluating Contextualized Embeddings on 54 Languages in POS Tagging, Lemmatization and Dependency Parsing. In: ArXiv.org Computing Research Repository, ISSN 2331-8422, 1904.02099
• Milan Straka (2018): UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task. In: Proceedings of CoNLL 2018: The SIGNLL Conference on Computational Natural Language Learning, pp. 197-207, Association for Computational Linguistics, Stroudsburg, PA, USA, ISBN 978-1-948087-72-8

Model Description

The Universal Dependencies 2.10 models contain 123 models of 69 languages, each consisting of a tokenizer, tagger, lemmatizer and dependency parser, all trained using the UD data. We used the original train-dev-test split, but for treebanks with only train and no dev data we used last 10% of the train data as dev data. We produce models only for treebanks with at least 1000 training words.

The tokenizer is trained using the SpaceAfter=No features. If the features are not present in the data, they can be filled in using raw text in the language in question.

The tagger, lemmatizer and parser are trained using gold UD data.

Model Performance

We present the tokenizer, tagger, lemmatizer and parser performance, measured on the testing portion of the data, evaluated both on the raw text and using the gold tokenization. The results are F1 scores measured by the conll18_ud_eval.py script.

Model	Mode	Words	Sents	UPOS	XPOS	UFeats	AllTags	Lemma	UAS	LAS	MLAS	BLEX
afrikaans-afribooms-ud-2.10-220711	Raw text	99.78	98.59	98.58	95.46	98.13	95.33	97.43	90.10	87.23	78.64	78.59
afrikaans-afribooms-ud-2.10-220711	Gold tokenization	—	—	98.77	95.62	98.31	95.50	97.53	90.72	87.80	79.23	78.99
ancient_greek-perseus-ud-2.10-220711	Raw text	99.97	98.85	92.83	85.55	91.45	84.87	86.68	80.13	74.36	54.62	55.72
ancient_greek-perseus-ud-2.10-220711	Gold tokenization	—	—	92.88	85.60	91.47	84.90	86.70	80.32	74.53	54.73	55.87
ancient_greek-proiel-ud-2.10-220711	Raw text	100.00	48.02	97.77	98.05	92.35	91.05	94.71	79.82	76.06	60.08	65.75
ancient_greek-proiel-ud-2.10-220711	Gold tokenization	—	—	97.87	98.14	92.49	91.26	94.73	86.05	82.14	67.03	71.90
ancient_hebrew-ptnk-ud-2.10-220711	Raw text	68.76	98.06	56.80	56.94	55.13	50.80	49.88	38.73	34.67	18.47	17.69
ancient_hebrew-ptnk-ud-2.10-220711	Gold tokenization	—	—	68.03	67.97	66.97	56.15	53.35	63.31	51.61	28.08	24.34
arabic-padt-ud-2.10-220711	Raw text	94.58	82.09	91.72	89.01	89.14	88.69	90.41	78.63	74.54	65.84	67.88
arabic-padt-ud-2.10-220711	Gold tokenization	—	—	97.02	94.38	94.53	94.08	95.31	88.11	83.49	74.57	76.13
armenian-armtdp-ud-2.10-220711	Raw text	99.28	95.70	96.07	—	91.39	90.28	95.04	86.84	82.22	69.53	74.39
armenian-armtdp-ud-2.10-220711	Gold tokenization	—	—	96.63	—	92.03	90.77	95.70	88.50	83.81	70.18	75.42
armenian-bsut-ud-2.10-220711	Raw text	99.79	98.73	97.31	—	92.01	91.24	96.62	90.02	85.75	71.20	78.86
armenian-bsut-ud-2.10-220711	Gold tokenization	—	—	97.53	—	92.24	91.48	96.82	90.56	86.29	71.73	79.32
basque-bdt-ud-2.10-220711	Raw text	99.94	99.83	96.25	—	92.69	90.69	96.36	87.40	84.28	73.94	79.81
basque-bdt-ud-2.10-220711	Gold tokenization	—	—	96.30	—	92.73	90.72	96.39	87.48	84.36	73.99	79.86
belarusian-hse-ud-2.10-220711	Raw text	99.47	83.97	98.30	96.26	94.38	92.37	93.35	86.84	84.85	76.00	76.01
belarusian-hse-ud-2.10-220711	Gold tokenization	—	—	98.81	96.74	94.94	92.87	93.86	89.55	87.38	78.23	78.11
bulgarian-btb-ud-2.10-220711	Raw text	99.91	94.17	99.19	97.20	97.97	96.85	97.99	94.41	91.67	85.89	86.31
bulgarian-btb-ud-2.10-220711	Gold tokenization	—	—	99.29	97.30	98.07	96.96	98.09	95.24	92.44	86.52	87.03
catalan-ancora-ud-2.10-220711	Raw text	99.95	99.08	99.07	97.21	98.70	96.96	99.40	94.86	93.14	87.45	88.92
catalan-ancora-ud-2.10-220711	Gold tokenization	—	—	99.14	97.32	98.78	97.07	99.46	95.02	93.30	87.69	89.13
chinese-gsdsimp-ud-2.10-220711	Raw text	90.29	99.10	87.21	87.16	89.74	86.42	90.29	73.11	70.62	63.58	67.09
chinese-gsdsimp-ud-2.10-220711	Gold tokenization	—	—	96.14	96.04	99.45	95.30	99.99	87.28	84.07	78.56	82.64
chinese-gsd-ud-2.10-220711	Raw text	90.27	99.10	87.15	87.05	89.71	86.36	90.27	72.85	70.29	63.41	66.89
chinese-gsd-ud-2.10-220711	Gold tokenization	—	—	96.21	96.08	99.40	95.34	99.99	87.15	83.96	78.41	82.59
classical_chinese-kyoto-ud-2.10-220711	Raw text	97.26	40.71	87.40	86.48	89.97	83.27	96.78	67.56	62.17	58.02	60.60
classical_chinese-kyoto-ud-2.10-220711	Gold tokenization	—	—	92.30	90.87	93.94	88.19	99.47	83.16	77.63	73.15	76.42
coptic-scriptorium-ud-2.10-220711	Raw text	74.49	33.87	72.43	72.34	72.53	71.54	72.91	51.25	49.43	36.55	39.14
coptic-scriptorium-ud-2.10-220711	Gold tokenization	—	—	96.94	96.78	97.49	95.50	97.02	90.48	87.70	76.04	79.57
croatian-set-ud-2.10-220711	Raw text	99.93	94.79	98.48	95.72	96.23	95.49	97.60	92.17	89.27	81.53	84.23
croatian-set-ud-2.10-220711	Gold tokenization	—	—	98.54	95.80	96.30	95.56	97.68	92.67	89.75	81.92	84.69
czech-pdt-ud-2.10-220711	Raw text	99.94	93.74	99.37	98.40	98.33	98.02	99.21	94.90	93.50	90.28	91.88
czech-pdt-ud-2.10-220711	Gold tokenization	—	—	99.45	98.47	98.40	98.09	99.28	95.63	94.23	90.88	92.50
czech-cac-ud-2.10-220711	Raw text	99.99	99.68	99.72	98.57	98.37	98.12	99.18	96.12	94.76	91.09	92.67
czech-cac-ud-2.10-220711	Gold tokenization	—	—	99.73	98.58	98.38	98.13	99.19	96.12	94.76	91.11	92.69
czech-cltt-ud-2.10-220711	Raw text	99.71	97.79	99.22	95.32	95.23	95.03	99.18	90.77	89.24	81.35	86.22
czech-cltt-ud-2.10-220711	Gold tokenization	—	—	99.47	95.47	95.40	95.18	99.47	91.20	89.68	81.67	86.73
czech-fictree-ud-2.10-220711	Raw text	99.99	98.95	99.17	97.06	97.83	96.86	99.35	96.38	94.91	89.61	92.81
czech-fictree-ud-2.10-220711	Gold tokenization	—	—	99.18	97.08	97.84	96.88	99.36	96.46	94.97	89.71	92.91
danish-ddt-ud-2.10-220711	Raw text	99.81	89.78	97.95	—	97.29	96.54	97.26	88.27	86.25	79.22	80.96
danish-ddt-ud-2.10-220711	Gold tokenization	—	—	98.16	—	97.53	96.79	97.45	89.46	87.42	80.42	82.17
dutch-alpino-ud-2.10-220711	Raw text	99.83	88.98	97.86	96.79	97.80	96.29	95.11	92.95	90.58	83.15	79.88
dutch-alpino-ud-2.10-220711	Gold tokenization	—	—	97.97	96.87	97.91	96.41	95.26	94.00	91.63	84.17	80.83
dutch-lassysmall-ud-2.10-220711	Raw text	99.80	74.93	96.98	95.62	96.63	94.88	95.70	90.61	87.94	79.67	78.25
dutch-lassysmall-ud-2.10-220711	Gold tokenization	—	—	97.25	96.43	97.36	95.83	95.97	94.51	91.66	84.48	83.08
english-ewt-ud-2.10-220711	Raw text	98.95	87.02	96.39	96.13	96.53	94.80	97.13	90.07	88.10	81.47	83.42
english-ewt-ud-2.10-220711	Gold tokenization	—	—	97.35	97.06	97.52	95.71	98.07	92.62	90.56	84.02	85.98
english-atis-ud-2.10-220711	Raw text	100.00	81.96	98.97	—	98.54	98.13	99.94	94.39	92.92	87.85	90.39
english-atis-ud-2.10-220711	Gold tokenization	—	—	98.97	—	98.56	98.15	99.94	95.88	94.26	89.80	92.40
english-gum-ud-2.10-220711	Raw text	99.64	95.36	97.95	97.91	97.88	96.91	98.77	92.30	90.35	84.63	86.31
english-gum-ud-2.10-220711	Gold tokenization	—	—	98.27	98.26	98.22	97.24	99.09	93.17	91.19	85.42	87.04
english-lines-ud-2.10-220711	Raw text	99.92	87.45	97.71	96.77	97.02	94.41	98.40	91.17	88.22	80.27	83.45
english-lines-ud-2.10-220711	Gold tokenization	—	—	97.79	96.84	97.07	94.48	98.47	92.10	89.17	81.05	84.36
english-partut-ud-2.10-220711	Raw text	99.72	100.00	97.23	97.11	96.35	95.26	98.14	94.24	92.21	83.35	87.34
english-partut-ud-2.10-220711	Gold tokenization	—	—	97.48	97.36	96.60	95.51	98.42	94.48	92.46	83.74	87.62
estonian-edt-ud-2.10-220711	Raw text	99.95	92.03	97.68	98.31	96.28	95.07	95.36	88.81	86.16	79.92	79.56
estonian-edt-ud-2.10-220711	Gold tokenization	—	—	97.81	98.36	96.36	95.19	95.43	89.71	87.03	80.77	80.37
estonian-ewt-ud-2.10-220711	Raw text	98.82	75.26	95.41	96.29	94.06	91.92	93.86	82.62	79.30	71.40	72.35
estonian-ewt-ud-2.10-220711	Gold tokenization	—	—	96.65	97.43	95.15	93.10	94.97	86.76	83.25	74.79	75.57
faroese-farpahc-ud-2.10-220711	Raw text	99.74	92.77	97.44	93.04	94.43	92.50	99.74	85.76	82.13	68.07	75.34
faroese-farpahc-ud-2.10-220711	Gold tokenization	—	—	97.64	93.28	94.68	92.72	100.00	86.82	83.10	69.17	76.51
finnish-tdt-ud-2.10-220711	Raw text	99.70	90.82	97.58	98.18	95.99	95.10	92.14	90.20	88.18	82.19	78.16
finnish-tdt-ud-2.10-220711	Gold tokenization	—	—	97.92	98.49	96.29	95.43	92.46	91.51	89.46	83.20	79.17
finnish-ftb-ud-2.10-220711	Raw text	99.91	86.84	96.69	95.14	96.83	94.02	95.57	89.80	87.18	80.04	80.49
finnish-ftb-ud-2.10-220711	Gold tokenization	—	—	97.00	95.36	96.92	94.32	95.67	91.91	89.23	82.55	82.84
french-gsd-ud-2.10-220711	Raw text	98.78	94.69	97.26	—	97.35	96.63	97.55	92.76	90.82	84.55	86.32
french-gsd-ud-2.10-220711	Gold tokenization	—	—	98.44	—	98.47	97.71	98.75	94.55	92.71	86.34	87.59
french-parisstories-ud-2.10-220711	Raw text	99.49	87.87	96.24	—	94.41	92.17	97.55	79.95	74.84	61.23	68.35
french-parisstories-ud-2.10-220711	Gold tokenization	—	—	96.81	—	94.90	92.68	97.98	81.67	76.50	62.46	69.32
french-partut-ud-2.10-220711	Raw text	99.48	100.00	97.26	96.76	94.72	93.96	97.33	94.72	92.81	81.09	86.22
french-partut-ud-2.10-220711	Gold tokenization	—	—	97.89	97.35	95.27	94.51	97.89	95.62	93.85	82.18	87.24
french-rhapsodie-ud-2.10-220711	Raw text	99.22	99.47	97.20	97.45	96.12	93.30	98.26	88.71	84.99	75.15	79.88
french-rhapsodie-ud-2.10-220711	Gold tokenization	—	—	98.00	98.13	96.89	93.97	98.99	89.89	86.08	75.91	80.35
french-sequoia-ud-2.10-220711	Raw text	99.15	84.02	98.32	—	97.15	96.68	98.33	93.60	92.22	86.08	89.00
french-sequoia-ud-2.10-220711	Gold tokenization	—	—	99.24	—	97.95	97.54	99.13	95.43	94.11	88.00	90.34
galician-ctg-ud-2.10-220711	Raw text	99.22	97.22	97.28	97.05	99.06	96.70	98.04	85.59	83.20	72.11	76.94
galician-ctg-ud-2.10-220711	Gold tokenization	—	—	98.01	97.78	99.84	97.41	98.79	87.31	84.80	74.04	78.88
galician-treegal-ud-2.10-220711	Raw text	98.74	87.99	96.00	93.69	94.85	92.82	96.67	83.44	79.36	67.82	71.68
galician-treegal-ud-2.10-220711	Gold tokenization	—	—	97.19	94.83	95.94	93.91	97.86	86.75	82.40	71.30	75.54
german-hdt-ud-2.10-220711	Raw text	99.90	92.34	98.51	98.45	94.08	93.70	97.16	96.94	96.04	84.79	90.40
german-hdt-ud-2.10-220711	Gold tokenization	—	—	98.62	98.58	94.22	93.83	97.26	97.63	96.75	85.46	91.08
german-gsd-ud-2.10-220711	Raw text	99.81	81.12	95.78	97.68	90.23	87.27	96.75	87.32	83.12	63.79	75.00
german-gsd-ud-2.10-220711	Gold tokenization	—	—	95.94	97.87	90.60	87.60	96.96	89.28	85.04	65.33	76.75
gothic-proiel-ud-2.10-220711	Raw text	100.00	31.12	96.48	96.98	90.08	88.19	94.62	74.17	68.40	55.39	62.02
gothic-proiel-ud-2.10-220711	Gold tokenization	—	—	96.97	97.42	90.90	89.33	94.71	84.11	78.29	65.73	71.10
greek-gdt-ud-2.10-220711	Raw text	99.87	90.19	98.09	98.10	95.60	95.01	95.61	93.05	91.24	81.58	81.04
greek-gdt-ud-2.10-220711	Gold tokenization	—	—	98.23	98.24	95.79	95.20	95.70	93.85	92.04	82.28	81.75
hebrew-htb-ud-2.10-220711	Raw text	85.05	99.39	82.78	82.80	81.23	80.53	82.91	70.63	68.13	55.31	59.44
hebrew-htb-ud-2.10-220711	Gold tokenization	—	—	97.44	97.42	95.73	95.05	97.34	92.71	90.24	78.77	81.80
hebrew-iahltwiki-ud-2.10-220711	Raw text	88.54	97.16	85.97	86.00	80.55	79.47	87.15	76.16	74.19	56.91	66.92
hebrew-iahltwiki-ud-2.10-220711	Gold tokenization	—	—	97.09	97.10	91.59	90.41	98.24	93.88	91.45	74.27	85.44
hindi-hdtb-ud-2.10-220711	Raw text	100.00	98.90	97.57	97.12	94.16	92.23	98.92	95.30	92.32	79.20	87.66
hindi-hdtb-ud-2.10-220711	Gold tokenization	—	—	97.58	97.14	94.18	92.26	98.92	95.42	92.44	79.35	87.81
hungarian-szeged-ud-2.10-220711	Raw text	99.85	95.89	96.68	—	94.22	93.53	94.92	88.81	85.09	75.22	78.19
hungarian-szeged-ud-2.10-220711	Gold tokenization	—	—	96.79	—	94.36	93.64	95.04	89.31	85.54	75.51	78.47
icelandic-icepahc-ud-2.10-220711	Raw text	99.82	92.15	96.90	93.24	91.32	86.42	95.99	87.21	83.36	65.98	74.25
icelandic-icepahc-ud-2.10-220711	Gold tokenization	—	—	97.08	93.45	91.47	86.61	96.15	87.78	83.87	66.49	74.84
icelandic-modern-ud-2.10-220711	Raw text	99.92	99.22	99.07	98.14	98.38	97.88	98.91	94.41	93.17	89.31	90.07
icelandic-modern-ud-2.10-220711	Gold tokenization	—	—	99.14	98.21	98.45	97.95	98.98	94.50	93.26	89.41	90.16
indonesian-gsd-ud-2.10-220711	Raw text	99.48	92.90	94.23	93.81	95.53	88.78	98.13	87.65	81.59	72.35	77.02
indonesian-gsd-ud-2.10-220711	Gold tokenization	—	—	94.66	94.26	95.99	89.17	98.53	88.57	82.42	73.21	77.88
indonesian-csui-ud-2.10-220711	Raw text	99.45	91.01	96.05	96.14	96.85	95.43	98.23	86.38	82.10	76.54	78.80
indonesian-csui-ud-2.10-220711	Gold tokenization	—	—	96.56	96.72	97.37	95.99	98.87	87.77	83.28	77.62	79.92
irish-idt-ud-2.10-220711	Raw text	99.72	97.25	95.63	94.76	90.33	87.14	95.30	86.74	81.10	64.20	71.52
irish-idt-ud-2.10-220711	Gold tokenization	—	—	95.89	95.07	90.60	87.46	95.54	87.28	81.64	64.56	71.85
italian-isdt-ud-2.10-220711	Raw text	99.84	98.76	98.57	98.50	98.25	97.67	98.79	94.66	93.01	86.61	88.00
italian-isdt-ud-2.10-220711	Gold tokenization	—	—	98.72	98.65	98.41	97.83	98.95	94.96	93.34	86.97	88.40
italian-markit-ud-2.10-220711	Raw text	99.59	98.24	96.76	97.00	93.80	92.08	88.18	88.36	84.51	69.95	77.77
italian-markit-ud-2.10-220711	Gold tokenization	—	—	97.15	97.40	94.10	92.35	88.54	89.13	85.26	70.51	78.46
italian-partut-ud-2.10-220711	Raw text	99.73	100.00	98.43	98.43	98.35	97.61	98.68	96.21	94.18	87.87	89.09
italian-partut-ud-2.10-220711	Gold tokenization	—	—	98.54	98.57	98.49	97.69	98.93	96.26	94.15	87.68	89.07
italian-postwita-ud-2.10-220711	Raw text	99.40	28.11	96.43	96.18	96.30	94.79	96.72	80.61	76.89	65.29	66.90
italian-postwita-ud-2.10-220711	Gold tokenization	—	—	97.04	96.82	96.80	95.29	97.31	88.34	84.19	75.32	77.32
italian-twittiro-ud-2.10-220711	Raw text	99.14	39.36	95.92	95.92	95.07	93.46	94.50	82.23	77.79	64.50	65.42
italian-twittiro-ud-2.10-220711	Gold tokenization	—	—	96.91	96.61	96.00	94.15	95.16	88.07	83.53	71.89	72.69
italian-vit-ud-2.10-220711	Raw text	99.76	96.73	98.14	97.39	97.64	96.21	98.89	92.08	89.16	80.93	83.70
italian-vit-ud-2.10-220711	Gold tokenization	—	—	98.36	97.71	97.85	96.53	99.10	92.88	89.97	81.91	84.63
japanese-gsd-ud-2.10-220711	Raw text	96.17	100.00	94.93	94.18	96.16	93.81	95.05	87.68	86.85	80.43	80.78
japanese-gsd-ud-2.10-220711	Gold tokenization	—	—	98.55	97.50	99.99	97.13	98.47	94.73	93.75	88.50	88.34
japanese-gsdluw-ud-2.10-220711	Raw text	95.18	99.72	93.81	93.54	95.18	93.46	93.66	86.22	85.54	76.27	76.58
japanese-gsdluw-ud-2.10-220711	Gold tokenization	—	—	98.36	98.05	100.00	97.93	97.89	95.23	94.18	86.38	85.19
korean-kaist-ud-2.10-220711	Raw text	100.00	100.00	95.88	87.74	—	87.56	94.17	89.33	87.47	82.15	80.14
korean-kaist-ud-2.10-220711	Gold tokenization	—	—	95.88	87.74	—	87.56	94.17	89.33	87.47	82.15	80.14
korean-gsd-ud-2.10-220711	Raw text	99.87	93.93	96.57	90.27	99.67	88.02	93.57	88.54	84.91	80.73	77.23
korean-gsd-ud-2.10-220711	Gold tokenization	—	—	96.73	90.43	99.80	88.20	93.69	89.27	85.61	81.45	77.93
latin-ittb-ud-2.10-220711	Raw text	99.99	91.21	98.91	96.58	96.75	95.19	99.18	90.53	88.53	82.07	86.07
latin-ittb-ud-2.10-220711	Gold tokenization	—	—	98.92	96.57	96.78	95.20	99.18	91.50	89.51	82.63	86.59
latin-llct-ud-2.10-220711	Raw text	100.00	99.49	99.68	97.14	97.26	96.89	97.78	95.55	94.56	89.80	90.95
latin-llct-ud-2.10-220711	Gold tokenization	—	—	99.68	97.15	97.27	96.90	97.78	95.55	94.57	89.81	90.97
latin-perseus-ud-2.10-220711	Raw text	100.00	98.46	91.83	80.66	86.12	78.56	88.13	77.98	68.59	52.30	55.51
latin-perseus-ud-2.10-220711	Gold tokenization	—	—	91.85	80.66	86.12	78.55	88.16	78.14	68.71	52.39	55.58
latin-proiel-ud-2.10-220711	Raw text	99.87	36.81	96.69	96.87	90.56	89.54	96.21	74.07	69.56	56.74	63.93
latin-proiel-ud-2.10-220711	Gold tokenization	—	—	97.12	97.32	91.19	90.27	96.44	83.20	78.50	66.34	73.00
latin-udante-ud-2.10-220711	Raw text	99.61	98.81	90.58	75.59	81.31	71.62	87.25	75.26	67.81	43.95	50.36
latin-udante-ud-2.10-220711	Gold tokenization	—	—	90.82	75.70	81.53	71.70	87.44	75.50	67.97	44.08	50.51
latvian-lvtb-ud-2.10-220711	Raw text	99.31	97.83	96.51	89.83	93.86	89.08	95.92	88.75	85.79	76.04	80.25
latvian-lvtb-ud-2.10-220711	Gold tokenization	—	—	97.14	90.43	94.50	89.67	96.55	89.84	86.82	77.09	81.31
lithuanian-alksnis-ud-2.10-220711	Raw text	99.91	87.87	95.94	90.44	91.03	89.52	93.60	82.45	78.64	67.97	71.37
lithuanian-alksnis-ud-2.10-220711	Gold tokenization	—	—	96.04	90.52	91.16	89.63	93.69	83.70	79.88	68.98	72.36
lithuanian-hse-ud-2.10-220711	Raw text	97.30	97.30	89.28	90.21	83.13	78.38	88.16	70.27	61.79	45.67	54.04
lithuanian-hse-ud-2.10-220711	Gold tokenization	—	—	91.23	92.36	85.19	80.09	90.57	73.96	64.53	47.54	56.10
maltese-mudt-ud-2.10-220711	Raw text	99.84	86.29	95.80	95.79	—	95.35	—	84.96	80.07	68.98	72.86
maltese-mudt-ud-2.10-220711	Gold tokenization	—	—	95.95	95.92	—	95.48	—	85.65	80.70	69.40	73.33
marathi-ufal-ud-2.10-220711	Raw text	90.25	92.63	76.50	—	65.25	60.75	80.75	60.75	50.75	28.39	38.00
marathi-ufal-ud-2.10-220711	Gold tokenization	—	—	82.52	—	67.96	62.86	80.83	68.93	58.50	29.46	38.17
naija-nsc-ud-2.10-220711	Raw text	99.94	100.00	98.03	—	98.94	97.53	99.32	93.65	90.99	88.13	89.60
naija-nsc-ud-2.10-220711	Gold tokenization	—	—	98.08	—	99.00	97.58	99.38	93.75	91.08	88.21	89.68
north_sami-giella-ud-2.10-220711	Raw text	99.87	98.79	91.77	93.54	89.30	85.36	87.01	75.16	70.43	59.76	58.27
north_sami-giella-ud-2.10-220711	Gold tokenization	—	—	91.91	93.67	89.45	85.52	87.13	75.47	70.76	60.05	58.56
norwegian-bokmaal-ud-2.10-220711	Raw text	99.77	96.05	98.35	—	97.43	96.82	98.57	93.62	92.16	86.91	88.74
norwegian-bokmaal-ud-2.10-220711	Gold tokenization	—	—	98.61	—	97.68	97.07	98.82	94.40	92.91	87.59	89.43
norwegian-nynorsk-ud-2.10-220711	Raw text	99.93	94.17	98.24	—	97.34	96.55	98.40	93.89	92.18	86.03	88.36
norwegian-nynorsk-ud-2.10-220711	Gold tokenization	—	—	98.41	—	97.50	96.73	98.53	94.63	92.93	86.93	89.20
norwegian-nynorsklia-ud-2.10-220711	Raw text	99.91	99.53	96.61	—	95.71	93.75	98.05	81.18	76.61	66.01	69.68
norwegian-nynorsklia-ud-2.10-220711	Gold tokenization	—	—	96.72	—	95.80	93.85	98.14	81.42	76.84	66.23	69.90
old_church_slavonic-proiel-ud-2.10-220711	Raw text	100.00	41.43	96.72	96.90	90.37	89.19	93.13	77.71	73.92	63.82	68.87
old_church_slavonic-proiel-ud-2.10-220711	Gold tokenization	—	—	97.08	97.28	91.06	89.93	93.14	88.30	84.18	74.01	77.39
old_french-srcmf-ud-2.10-220711	Raw text	99.70	100.00	96.68	96.50	97.70	95.72	99.65	91.17	87.38	80.76	84.40
old_french-srcmf-ud-2.10-220711	Gold tokenization	—	—	96.99	96.82	98.01	96.03	99.95	91.58	87.82	81.20	84.85
old_russian-torot-ud-2.10-220711	Raw text	100.00	29.60	94.39	94.70	87.56	85.23	85.92	71.00	65.32	51.64	53.64
old_russian-torot-ud-2.10-220711	Gold tokenization	—	—	95.06	95.29	88.50	86.60	85.96	83.30	77.24	64.09	62.94
old_russian-rnc-ud-2.10-220711	Raw text	97.48	84.03	90.94	86.55	76.51	67.15	75.31	61.28	55.93	33.24	34.04
old_russian-rnc-ud-2.10-220711	Gold tokenization	—	—	93.29	88.93	78.48	68.86	76.77	67.13	61.08	37.15	37.24
old_east_slavic-birchbark-ud-2.10-220711	Raw text	99.98	16.73	89.24	99.35	76.11	72.43	65.88	63.41	56.50	32.53	27.14
old_east_slavic-birchbark-ud-2.10-220711	Gold tokenization	—	—	89.37	99.37	76.54	72.82	66.05	76.31	69.00	41.63	33.60
persian-perdt-ud-2.10-220711	Raw text	99.66	99.83	97.48	97.36	97.61	95.60	98.88	93.63	91.42	86.18	88.66
persian-perdt-ud-2.10-220711	Gold tokenization	—	—	97.78	97.65	97.90	95.89	99.19	94.18	91.95	86.72	89.23
persian-seraji-ud-2.10-220711	Raw text	99.65	98.75	97.91	97.94	97.95	97.48	96.52	91.68	88.84	84.21	82.83
persian-seraji-ud-2.10-220711	Gold tokenization	—	—	98.24	98.28	98.28	97.78	96.80	92.36	89.48	84.82	83.40
polish-pdb-ud-2.10-220711	Raw text	99.85	97.33	98.89	95.89	96.11	95.26	98.10	94.22	92.19	85.44	88.36
polish-pdb-ud-2.10-220711	Gold tokenization	—	—	99.05	96.03	96.24	95.40	98.24	94.72	92.69	85.83	88.78
polish-lfg-ud-2.10-220711	Raw text	99.85	99.65	99.00	96.08	96.57	95.16	98.24	96.86	95.51	89.80	92.34
polish-lfg-ud-2.10-220711	Gold tokenization	—	—	99.17	96.25	96.74	95.33	98.38	97.25	95.89	90.19	92.66
pomak-philotis-ud-2.10-220711	Raw text	99.98	94.49	98.86	—	95.62	95.30	96.67	88.24	83.26	71.19	74.14
pomak-philotis-ud-2.10-220711	Gold tokenization	—	—	98.90	—	95.65	95.33	96.69	88.68	83.75	71.48	74.42
portuguese-gsd-ud-2.10-220711	Raw text	99.87	97.28	98.51	98.51	99.74	98.41	99.27	94.50	93.41	88.76	89.96
portuguese-gsd-ud-2.10-220711	Gold tokenization	—	—	98.65	98.64	99.89	98.55	99.40	94.90	93.81	89.23	90.36
portuguese-bosque-ud-2.10-220711	Raw text	99.68	89.89	97.87	—	96.95	96.00	98.35	92.35	90.07	81.38	84.69
portuguese-bosque-ud-2.10-220711	Gold tokenization	—	—	98.22	—	97.23	96.28	98.66	93.50	91.16	82.47	85.87
romanian-nonstandard-ud-2.10-220711	Raw text	98.83	96.77	96.18	91.87	90.53	89.18	94.90	88.85	84.82	68.21	76.36
romanian-nonstandard-ud-2.10-220711	Gold tokenization	—	—	97.30	92.86	91.49	90.10	95.99	90.57	86.50	69.69	77.68
romanian-rrt-ud-2.10-220711	Raw text	99.71	95.16	97.90	97.21	97.40	96.98	97.96	91.97	88.44	81.66	83.13
romanian-rrt-ud-2.10-220711	Gold tokenization	—	—	98.19	97.45	97.65	97.22	98.22	92.72	89.13	82.15	83.70
romanian-simonero-ud-2.10-220711	Raw text	99.84	100.00	98.45	97.97	97.56	97.25	98.91	94.08	92.13	85.52	88.32
romanian-simonero-ud-2.10-220711	Gold tokenization	—	—	98.61	98.12	97.70	97.40	99.07	94.42	92.45	85.81	88.62
russian-syntagrus-ud-2.10-220711	Raw text	99.67	98.31	98.46	—	93.96	93.71	98.18	93.84	91.70	82.72	88.90
russian-syntagrus-ud-2.10-220711	Gold tokenization	—	—	98.79	—	94.28	94.03	98.46	94.56	92.39	83.28	89.44
russian-gsd-ud-2.10-220711	Raw text	99.50	96.49	98.11	97.55	94.71	93.61	97.01	91.44	88.55	81.04	84.62
russian-gsd-ud-2.10-220711	Gold tokenization	—	—	98.58	97.98	95.17	94.01	97.43	92.67	89.69	82.00	85.65
russian-taiga-ud-2.10-220711	Raw text	98.12	86.33	95.65	—	93.13	92.06	94.73	83.08	79.57	70.60	73.88
russian-taiga-ud-2.10-220711	Gold tokenization	—	—	97.34	—	94.90	93.72	96.37	85.64	81.92	72.82	76.10
sanskrit-vedic-ud-2.10-220711	Raw text	100.00	27.18	89.16	—	81.61	76.76	87.05	60.92	50.04	41.66	44.99
sanskrit-vedic-ud-2.10-220711	Gold tokenization	—	—	89.97	—	83.02	78.34	87.34	73.74	62.01	52.00	55.41
scottish_gaelic-arcosg-ud-2.10-220711	Raw text	97.47	60.89	93.78	89.29	90.91	88.21	95.08	81.24	75.60	62.73	69.22
scottish_gaelic-arcosg-ud-2.10-220711	Gold tokenization	—	—	96.62	92.24	94.02	91.39	97.59	87.33	81.65	69.25	75.23
serbian-set-ud-2.10-220711	Raw text	99.99	93.00	99.09	96.00	96.21	95.75	97.76	93.63	91.20	83.76	87.00
serbian-set-ud-2.10-220711	Gold tokenization	—	—	99.13	96.01	96.20	95.75	97.78	94.26	91.80	84.32	87.60
slovak-snk-ud-2.10-220711	Raw text	100.00	81.69	97.65	90.35	93.50	89.56	96.46	91.39	89.65	80.43	84.44
slovak-snk-ud-2.10-220711	Gold tokenization	—	—	97.88	90.55	93.69	89.80	96.50	93.91	92.08	82.89	86.95
slovenian-ssj-ud-2.10-220711	Raw text	99.94	98.95	98.97	96.97	97.15	96.63	98.58	93.99	92.60	86.83	88.91
slovenian-ssj-ud-2.10-220711	Gold tokenization	—	—	99.03	97.02	97.23	96.69	98.63	94.15	92.76	86.99	89.02
slovenian-sst-ud-2.10-220711	Raw text	99.85	23.14	94.82	92.71	92.43	89.84	97.38	65.69	60.84	50.88	54.78
slovenian-sst-ud-2.10-220711	Gold tokenization	—	—	95.62	93.09	92.84	90.89	97.56	78.39	73.07	63.39	68.33
spanish-ancora-ud-2.10-220711	Raw text	99.95	98.78	99.06	96.02	98.74	95.59	99.37	93.70	91.79	86.41	87.88
spanish-ancora-ud-2.10-220711	Gold tokenization	—	—	99.11	96.07	98.79	95.63	99.42	93.88	91.97	86.59	88.04
spanish-gsd-ud-2.10-220711	Raw text	99.75	95.62	97.15	—	96.94	95.27	98.72	91.87	89.57	78.63	84.25
spanish-gsd-ud-2.10-220711	Gold tokenization	—	—	97.39	—	97.19	95.53	98.97	92.66	90.32	79.43	85.04
swedish-talbanken-ud-2.10-220711	Raw text	99.84	96.53	98.44	97.33	97.32	96.51	98.15	92.23	89.85	83.92	85.97
swedish-talbanken-ud-2.10-220711	Gold tokenization	—	—	98.61	97.52	97.51	96.72	98.32	92.68	90.30	84.48	86.54
swedish-lines-ud-2.10-220711	Raw text	99.96	88.00	97.66	95.51	90.84	88.14	97.72	90.60	87.38	71.82	82.17
swedish-lines-ud-2.10-220711	Gold tokenization	—	—	97.73	95.52	90.87	88.15	97.76	91.44	88.19	72.50	82.95
tamil-ttb-ud-2.10-220711	Raw text	94.26	97.52	84.29	83.18	84.64	78.22	89.45	70.43	61.88	50.61	55.39
tamil-ttb-ud-2.10-220711	Gold tokenization	—	—	89.29	87.78	89.99	82.70	94.42	78.13	68.78	56.87	61.48
telugu-mtg-ud-2.10-220711	Raw text	99.58	96.62	93.63	93.63	98.61	93.49	—	90.72	84.63	77.14	81.14
telugu-mtg-ud-2.10-220711	Gold tokenization	—	—	94.04	94.04	99.03	93.90	—	91.68	85.58	77.98	81.98
turkish-boun-ud-2.10-220711	Raw text	98.83	86.93	91.56	92.51	91.72	86.56	93.23	78.48	72.40	59.77	65.11
turkish-boun-ud-2.10-220711	Gold tokenization	—	—	92.53	93.47	92.67	87.31	94.26	81.07	74.73	61.33	66.92
turkish-atis-ud-2.10-220711	Raw text	100.00	80.20	98.96	—	98.46	98.25	99.15	89.22	87.49	85.12	86.08
turkish-atis-ud-2.10-220711	Gold tokenization	—	—	99.02	—	98.52	98.32	99.13	91.11	89.30	86.98	87.93
turkish-framenet-ud-2.10-220711	Raw text	100.00	100.00	96.86	—	94.89	94.21	96.66	93.39	84.25	73.98	77.64
turkish-framenet-ud-2.10-220711	Gold tokenization	—	—	96.86	—	94.89	94.21	96.66	93.39	84.25	73.98	77.64
turkish-imst-ud-2.10-220711	Raw text	98.30	96.97	94.38	93.98	90.92	88.60	94.54	74.73	69.04	58.25	63.10
turkish-imst-ud-2.10-220711	Gold tokenization	—	—	95.94	95.49	92.40	89.97	96.13	78.07	72.09	60.26	65.33
turkish-kenet-ud-2.10-220711	Raw text	100.00	98.12	93.71	—	92.05	90.86	93.33	83.91	71.18	61.81	64.77
turkish-kenet-ud-2.10-220711	Gold tokenization	—	—	93.72	—	92.06	90.87	93.33	84.07	71.29	61.92	64.89
turkish-penn-ud-2.10-220711	Raw text	99.34	80.59	95.60	—	94.41	93.33	94.36	84.22	71.67	62.21	64.53
turkish-penn-ud-2.10-220711	Gold tokenization	—	—	96.30	—	95.11	94.02	95.01	86.76	73.91	63.63	66.02
turkish-tourism-ud-2.10-220711	Raw text	99.96	99.86	98.80	—	95.08	94.67	98.36	97.20	91.52	81.98	87.38
turkish-tourism-ud-2.10-220711	Gold tokenization	—	—	98.85	—	95.12	94.73	98.40	97.25	91.58	82.04	87.45
turkish_german-sagt-ud-2.10-220711	Raw text	98.91	99.44	90.21	—	80.32	75.60	90.82	71.14	60.98	41.12	51.00
turkish_german-sagt-ud-2.10-220711	Gold tokenization	—	—	91.09	—	80.89	76.08	91.52	72.69	62.06	41.64	51.71
ukrainian-iu-ud-2.10-220711	Raw text	99.81	96.61	97.90	94.35	94.18	93.12	97.34	90.61	88.27	78.92	83.01
ukrainian-iu-ud-2.10-220711	Gold tokenization	—	—	98.08	94.54	94.34	93.29	97.53	91.12	88.72	79.21	83.36
urdu-udtb-ud-2.10-220711	Raw text	100.00	98.31	93.91	92.15	82.83	78.40	97.41	88.15	82.49	56.62	74.68
urdu-udtb-ud-2.10-220711	Gold tokenization	—	—	93.93	92.17	82.86	78.43	97.41	88.23	82.58	56.67	74.77
uyghur-udt-ud-2.10-220711	Raw text	99.54	81.81	89.33	91.75	88.12	79.98	94.67	76.66	64.87	46.84	55.29
uyghur-udt-ud-2.10-220711	Gold tokenization	—	—	89.71	92.30	88.59	80.50	95.14	78.38	66.49	47.83	56.56
vietnamese-vtb-ud-2.10-220711	Raw text	85.37	93.46	78.21	76.76	85.12	76.57	85.16	52.68	47.84	41.55	44.29
vietnamese-vtb-ud-2.10-220711	Gold tokenization	—	—	90.36	88.55	99.72	88.32	99.59	72.88	65.41	58.76	62.51
welsh-ccg-ud-2.10-220711	Raw text	99.42	97.37	95.33	94.40	89.82	87.61	93.93	86.61	80.67	63.31	69.02
welsh-ccg-ud-2.10-220711	Gold tokenization	—	—	95.84	94.87	90.31	88.07	94.44	87.85	81.83	64.36	70.21
western_armenian-armtdp-ud-2.10-220711	Raw text	99.89	98.68	96.82	—	92.51	91.83	97.14	89.39	84.66	69.84	76.01
western_armenian-armtdp-ud-2.10-220711	Gold tokenization	—	—	96.90	—	92.60	91.93	97.22	89.64	84.89	70.07	76.23
wolof-wtb-ud-2.10-220711	Raw text	99.23	91.95	94.20	94.15	93.50	91.41	95.20	84.15	78.69	66.75	70.23
wolof-wtb-ud-2.10-220711	Gold tokenization	—	—	95.17	95.07	94.32	92.31	95.96	86.27	80.75	68.70	72.06

Universal Dependencies 2.6 Models

Universal Dependencies 2.6 Models are distributed under the CC BY-NC-SA licence. The models are based solely on Universal Dependencies 2.6 treebanks, and additionally use multilingual BERT.

The models require UDPipe 2.

Download

The latest version 200831 of the Universal Dependencies 2.6 models can be downloaded from LINDAT/CLARIN repository.

The models are also available in the REST service.

Acknowledgements

This work has been supported by the Ministry of Education, Youth and Sports of the Czech Republic, Project No. LM2018101 LINDAT/CLARIAH-CZ.

The models were trained on Universal Dependencies 2.6 treebanks.

For the UD treebanks which do not contain original plain text version, raw text is used to train the tokenizer instead. The plain texts were taken from the W2C -- Web to Corpus.

Finally, multilingual BERT is used to provide contextualized word embeddings.

Publications

• Milan Straka, Jana Straková, Jan Hajič (2019): Evaluating Contextualized Embeddings on 54 Languages in POS Tagging, Lemmatization and Dependency Parsing. In: ArXiv.org Computing Research Repository, ISSN 2331-8422, 1904.02099
• Milan Straka (2018): UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task. In: Proceedings of CoNLL 2018: The SIGNLL Conference on Computational Natural Language Learning, pp. 197-207, Association for Computational Linguistics, Stroudsburg, PA, USA, ISBN 978-1-948087-72-8

Model Description

The Universal Dependencies 2.6 models contain 99 models of 63 languages, each consisting of a tokenizer, tagger, lemmatizer and dependency parser, all trained using the UD data. We used the original train-dev-test split, but for treebanks with only train and no dev data we used last 10% of the train data as dev data. We produce models only for treebanks with at least 1000 training words.

The tokenizer is trained using the SpaceAfter=No features. If the features are not present in the data, they can be filled in using raw text in the language in question.

The tagger, lemmatizer and parser are trained using gold UD data.

Model Performance

We present the tokenizer, tagger, lemmatizer and parser performance, measured on the testing portion of the data, evaluated both on the raw text and using the gold tokenization. The results are F1 scores measured by the conll18_ud_eval.py script.

Model	Mode	Words	Sents	UPOS	XPOS	UFeats	AllTags	Lemma	UAS	LAS	MLAS	BLEX
afrikaans-afribooms-ud-2.6-200830	Raw text	99.82	98.25	98.55	95.42	98.27	95.33	97.52	90.34	87.93	80.33	79.91
afrikaans-afribooms-ud-2.6-200830	Gold tokenization	—	—	98.70	95.56	98.41	95.48	97.61	90.80	88.40	80.78	80.31
ancient_greek-perseus-ud-2.6-200830	Raw text	99.97	98.85	93.20	86.01	91.59	85.27	86.81	79.57	73.90	54.80	55.63
ancient_greek-perseus-ud-2.6-200830	Gold tokenization	—	—	93.24	86.03	91.62	85.30	86.84	79.74	74.06	54.92	55.72
ancient_greek-proiel-ud-2.6-200830	Raw text	100.00	48.02	97.73	98.04	92.36	91.01	94.71	79.98	75.99	60.15	65.88
ancient_greek-proiel-ud-2.6-200830	Gold tokenization	—	—	97.91	98.18	92.59	91.30	94.76	85.95	81.90	66.72	71.84
arabic-padt-ud-2.6-200830	Raw text	94.58	82.09	91.68	88.96	89.14	88.65	90.36	78.86	74.85	66.06	68.12
arabic-padt-ud-2.6-200830	Gold tokenization	—	—	96.87	94.20	94.36	93.82	95.23	87.60	83.14	74.51	76.09
armenian-armtdp-ud-2.6-200830	Raw text	99.34	97.85	95.64	—	90.30	88.94	94.45	85.07	79.97	66.54	71.61
armenian-armtdp-ud-2.6-200830	Gold tokenization	—	—	96.11	—	90.90	89.37	95.04	86.31	81.18	67.00	72.22
basque-bdt-ud-2.6-200830	Raw text	99.94	99.83	96.44	—	93.60	91.69	96.40	87.24	84.15	74.97	79.94
basque-bdt-ud-2.6-200830	Gold tokenization	—	—	96.48	—	93.64	91.72	96.43	87.30	84.21	75.00	79.96
belarusian-hse-ud-2.6-200830	Raw text	99.84	78.70	96.14	31.78	82.07	26.98	81.48	75.81	71.05	49.78	50.93
belarusian-hse-ud-2.6-200830	Gold tokenization	—	—	96.39	31.85	82.19	27.08	81.53	80.18	75.18	52.92	53.60
bulgarian-btb-ud-2.6-200830	Raw text	99.91	94.17	99.15	97.19	97.95	96.84	97.97	94.35	91.61	85.92	86.43
bulgarian-btb-ud-2.6-200830	Gold tokenization	—	—	99.27	97.30	98.05	96.95	98.07	95.17	92.41	86.62	87.17
catalan-ancora-ud-2.6-200830	Raw text	99.98	99.43	99.05	98.99	98.63	98.14	99.31	94.53	92.86	87.63	89.24
catalan-ancora-ud-2.6-200830	Gold tokenization	—	—	99.09	99.03	98.67	98.18	99.34	94.60	92.95	87.73	89.35
chinese-gsdsimp-ud-2.6-200830	Raw text	90.29	99.10	87.32	87.20	89.73	86.54	90.29	72.68	70.32	63.38	66.94
chinese-gsdsimp-ud-2.6-200830	Gold tokenization	—	—	96.32	96.15	99.43	95.50	99.99	86.89	83.93	78.52	82.60
chinese-gsd-ud-2.6-200830	Raw text	90.27	99.10	87.27	87.18	89.74	86.50	90.27	72.99	70.50	63.83	67.21
chinese-gsd-ud-2.6-200830	Gold tokenization	—	—	96.30	96.16	99.42	95.45	99.99	87.30	84.22	78.63	82.84
classical_chinese-kyoto-ud-2.6-200830	Raw text	99.46	46.22	90.91	90.91	93.43	88.00	99.42	72.75	67.18	63.67	66.02
classical_chinese-kyoto-ud-2.6-200830	Gold tokenization	—	—	93.55	93.24	95.01	90.86	99.96	85.49	80.20	76.42	79.25
coptic-scriptorium-ud-2.6-200830	Raw text	71.91	35.97	69.61	68.00	63.06	60.16	70.51	47.75	45.89	25.42	35.81
coptic-scriptorium-ud-2.6-200830	Gold tokenization	—	—	96.15	92.53	87.75	81.98	96.70	89.14	85.79	57.57	76.42
croatian-set-ud-2.6-200830	Raw text	99.95	94.41	98.18	95.91	96.40	95.27	97.58	92.20	88.40	80.16	83.07
croatian-set-ud-2.6-200830	Gold tokenization	—	—	98.23	96.00	96.52	95.38	97.64	92.72	88.89	80.66	83.53
czech-pdt-ud-2.6-200830	Raw text	99.93	93.35	99.23	97.61	97.59	97.13	99.09	93.81	92.03	87.79	89.88
czech-pdt-ud-2.6-200830	Gold tokenization	—	—	99.30	97.71	97.70	97.24	99.17	94.60	92.81	88.45	90.57
czech-cac-ud-2.6-200830	Raw text	99.98	99.68	99.52	97.33	97.05	96.64	98.93	94.31	92.48	87.56	89.76
czech-cac-ud-2.6-200830	Gold tokenization	—	—	99.54	97.36	97.07	96.67	98.95	94.37	92.54	87.63	89.83
czech-fictree-ud-2.6-200830	Raw text	99.99	98.95	98.68	95.80	96.79	95.38	99.20	94.83	92.66	85.35	89.58
czech-fictree-ud-2.6-200830	Gold tokenization	—	—	98.69	95.82	96.80	95.40	99.21	94.92	92.74	85.47	89.69
czech-cltt-ud-2.6-200830	Raw text	99.65	97.40	99.21	95.00	94.98	94.76	99.06	91.37	89.67	82.08	86.96
czech-cltt-ud-2.6-200830	Gold tokenization	—	—	99.49	95.19	95.16	94.95	99.30	91.91	90.21	82.25	87.31
danish-ddt-ud-2.6-200830	Raw text	99.81	89.78	98.01	—	97.52	96.72	97.31	88.56	86.46	79.62	81.12
danish-ddt-ud-2.6-200830	Gold tokenization	—	—	98.26	—	97.73	96.99	97.53	89.82	87.67	80.73	82.27
dutch-alpino-ud-2.6-200830	Raw text	99.83	88.59	97.41	95.98	97.02	95.36	97.32	92.79	90.38	81.53	83.18
dutch-alpino-ud-2.6-200830	Gold tokenization	—	—	97.57	96.13	97.18	95.53	97.46	93.93	91.53	82.72	84.42
dutch-lassysmall-ud-2.6-200830	Raw text	99.83	75.40	96.58	95.42	96.41	94.73	97.21	90.36	87.66	78.84	80.17
dutch-lassysmall-ud-2.6-200830	Gold tokenization	—	—	96.79	96.05	96.97	95.40	97.33	94.26	91.24	83.56	84.84
english-ewt-ud-2.6-200830	Raw text	98.95	86.60	96.36	96.06	96.56	94.88	97.64	89.55	87.43	80.50	83.29
english-ewt-ud-2.6-200830	Gold tokenization	—	—	97.29	97.03	97.57	95.84	98.57	92.24	90.05	83.33	86.07
english-gum-ud-2.6-200830	Raw text	99.81	83.66	96.79	96.76	97.55	95.88	97.35	90.02	87.52	79.41	80.43
english-gum-ud-2.6-200830	Gold tokenization	—	—	96.99	96.93	97.75	96.09	97.56	91.93	89.36	81.20	82.25
english-lines-ud-2.6-200830	Raw text	99.92	87.45	97.60	95.86	96.88	93.39	98.34	89.36	86.45	79.35	82.87
english-lines-ud-2.6-200830	Gold tokenization	—	—	97.67	95.90	96.92	93.41	98.41	90.26	87.36	80.24	83.79
english-partut-ud-2.6-200830	Raw text	99.72	100.00	97.37	97.08	96.29	95.38	98.23	94.12	92.09	83.04	87.20
english-partut-ud-2.6-200830	Gold tokenization	—	—	97.62	97.33	96.54	95.63	98.50	94.40	92.37	83.44	87.48
estonian-edt-ud-2.6-200830	Raw text	99.96	91.56	97.65	98.25	96.44	95.19	95.34	88.75	86.18	80.12	79.65
estonian-edt-ud-2.6-200830	Gold tokenization	—	—	97.75	98.29	96.48	95.28	95.40	89.66	87.06	80.93	80.44
estonian-ewt-ud-2.6-200830	Raw text	98.96	70.09	95.00	96.30	93.74	91.31	93.81	81.07	77.55	69.14	70.69
estonian-ewt-ud-2.6-200830	Gold tokenization	—	—	96.22	97.37	94.65	92.37	94.83	86.37	82.55	73.03	74.39
finnish-tdt-ud-2.6-200830	Raw text	99.70	88.64	97.63	98.25	96.05	95.11	92.06	90.11	88.10	82.04	77.91
finnish-tdt-ud-2.6-200830	Gold tokenization	—	—	97.97	98.56	96.37	95.48	92.38	91.69	89.63	83.30	79.18
finnish-ftb-ud-2.6-200830	Raw text	99.91	86.84	96.52	95.08	96.72	93.82	95.73	89.93	87.32	80.13	80.74
finnish-ftb-ud-2.6-200830	Gold tokenization	—	—	96.85	95.31	96.87	94.16	95.83	91.99	89.34	82.64	83.05
french-gsd-ud-2.6-200830	Raw text	98.87	94.67	97.23	98.86	96.65	96.00	97.69	92.77	90.82	83.14	86.08
french-gsd-ud-2.6-200830	Gold tokenization	—	—	98.29	99.99	97.63	96.94	98.80	94.46	92.63	84.72	87.21
french-sequoia-ud-2.6-200830	Raw text	99.09	87.50	98.33	—	97.25	96.79	98.16	93.90	92.45	86.54	89.25
french-sequoia-ud-2.6-200830	Gold tokenization	—	—	99.32	—	98.19	97.78	99.09	95.80	94.43	88.78	90.78
french-partut-ud-2.6-200830	Raw text	99.42	100.00	97.28	96.93	94.17	93.63	95.59	94.71	92.71	80.18	83.34
french-partut-ud-2.6-200830	Gold tokenization	—	—	97.89	97.54	94.74	94.20	96.20	95.47	93.62	81.20	84.28
french-spoken-ud-2.6-200830	Raw text	99.06	21.15	96.49	96.44	—	93.98	97.48	79.23	74.91	64.48	66.67
french-spoken-ud-2.6-200830	Gold tokenization	—	—	97.63	97.31	—	95.00	98.28	87.27	82.51	74.23	75.56
galician-ctg-ud-2.6-200830	Raw text	99.22	97.22	97.30	97.07	99.05	96.71	98.07	85.45	83.07	72.03	76.75
galician-ctg-ud-2.6-200830	Gold tokenization	—	—	98.04	97.79	99.83	97.43	98.82	87.22	84.73	74.05	78.78
galician-treegal-ud-2.6-200830	Raw text	98.74	87.99	95.99	93.58	94.72	92.63	96.71	83.26	79.23	67.54	71.73
galician-treegal-ud-2.6-200830	Gold tokenization	—	—	97.23	94.65	95.76	93.73	97.89	86.57	82.30	71.04	75.71
german-hdt-ud-2.6-200830	Raw text	99.91	92.34	98.51	98.45	94.09	93.69	97.23	96.88	95.96	84.87	90.41
german-hdt-ud-2.6-200830	Gold tokenization	—	—	98.62	98.57	94.21	93.81	97.32	97.57	96.67	85.53	91.10
german-gsd-ud-2.6-200830	Raw text	99.58	80.90	94.39	97.51	91.14	85.97	96.58	87.06	82.93	62.33	74.97
german-gsd-ud-2.6-200830	Gold tokenization	—	—	94.73	97.96	91.65	86.51	96.95	89.36	85.31	64.33	77.26
gothic-proiel-ud-2.6-200830	Raw text	100.00	31.12	96.39	96.90	90.18	88.05	94.70	74.10	68.48	55.16	62.26
gothic-proiel-ud-2.6-200830	Gold tokenization	—	—	96.81	97.26	91.12	89.28	94.77	83.73	77.93	65.37	70.85
greek-gdt-ud-2.6-200830	Raw text	99.87	90.19	97.99	98.00	95.57	94.91	95.53	93.00	91.16	81.28	80.73
greek-gdt-ud-2.6-200830	Gold tokenization	—	—	98.14	98.14	95.69	95.02	95.61	93.82	91.95	82.03	81.53
hebrew-htb-ud-2.6-200830	Raw text	85.04	99.39	82.79	82.76	81.31	80.57	82.97	69.85	67.39	54.79	59.16
hebrew-htb-ud-2.6-200830	Gold tokenization	—	—	97.48	97.48	96.03	95.36	97.23	91.83	89.25	78.52	81.02
hindi-hdtb-ud-2.6-200830	Raw text	100.00	98.90	97.64	97.29	94.18	92.32	98.78	95.32	92.37	79.24	87.69
hindi-hdtb-ud-2.6-200830	Gold tokenization	—	—	97.65	97.29	94.21	92.35	98.78	95.44	92.49	79.41	87.84
hungarian-szeged-ud-2.6-200830	Raw text	99.85	95.89	96.77	—	94.32	93.51	94.97	87.78	84.24	74.80	77.84
hungarian-szeged-ud-2.6-200830	Gold tokenization	—	—	96.87	—	94.45	93.61	95.09	88.28	84.73	75.27	78.26
indonesian-gsd-ud-2.6-200830	Raw text	100.00	94.13	93.89	94.28	95.55	89.00	99.61	86.07	79.97	69.25	77.74
indonesian-gsd-ud-2.6-200830	Gold tokenization	—	—	93.90	94.26	95.52	88.98	99.61	86.32	80.18	69.51	78.00
irish-idt-ud-2.6-200830	Raw text	99.71	97.36	94.35	94.30	73.43	70.38	93.18	84.47	77.88	40.78	65.74
irish-idt-ud-2.6-200830	Gold tokenization	—	—	94.59	94.60	73.65	70.63	93.41	84.98	78.30	40.94	65.87
italian-isdt-ud-2.6-200830	Raw text	99.84	98.76	98.52	98.44	98.23	97.66	98.65	94.77	93.12	86.91	87.85
italian-isdt-ud-2.6-200830	Gold tokenization	—	—	98.68	98.60	98.38	97.81	98.81	95.07	93.44	87.20	88.19
italian-partut-ud-2.6-200830	Raw text	99.73	100.00	98.41	98.52	98.27	97.77	98.74	96.07	93.90	87.45	88.95
italian-partut-ud-2.6-200830	Gold tokenization	—	—	98.54	98.65	98.38	97.88	98.93	96.18	93.98	87.48	89.15
italian-postwita-ud-2.6-200830	Raw text	99.47	30.49	96.53	96.28	96.43	94.89	96.76	80.97	76.94	65.79	67.44
italian-postwita-ud-2.6-200830	Gold tokenization	—	—	97.06	96.79	96.89	95.41	97.18	88.04	83.76	75.23	76.98
italian-twittiro-ud-2.6-200830	Raw text	99.06	36.80	95.99	95.86	95.22	93.37	94.68	81.69	77.38	64.34	65.32
italian-twittiro-ud-2.6-200830	Gold tokenization	—	—	97.01	96.77	96.14	94.42	95.50	87.84	83.43	71.64	72.68
italian-vit-ud-2.6-200830	Raw text	99.69	94.69	97.86	97.07	97.38	95.76	98.64	92.03	89.20	80.39	83.83
italian-vit-ud-2.6-200830	Gold tokenization	—	—	98.16	97.49	97.66	96.16	98.92	92.77	89.91	81.15	84.53
japanese-gsd-ud-2.6-200830	Raw text	95.34	94.61	93.67	93.56	95.32	92.74	95.02	85.11	84.01	76.23	77.83
japanese-gsd-ud-2.6-200830	Gold tokenization	—	—	98.03	97.71	99.99	96.83	99.61	94.73	93.41	87.64	89.28
korean-kaist-ud-2.6-200830	Raw text	99.95	100.00	95.89	87.82	—	87.62	94.23	89.41	87.58	82.32	80.34
korean-kaist-ud-2.6-200830	Gold tokenization	—	—	95.94	87.85	—	87.66	94.27	89.51	87.67	82.42	80.42
korean-gsd-ud-2.6-200830	Raw text	99.87	93.93	96.61	90.19	99.69	88.03	93.51	88.68	85.04	80.93	77.36
korean-gsd-ud-2.6-200830	Gold tokenization	—	—	96.74	90.32	99.82	88.16	93.64	89.50	85.84	81.76	78.14
latin-ittb-ud-2.6-200830	Raw text	99.99	92.44	98.54	96.35	96.92	95.12	98.94	90.31	88.16	82.19	85.37
latin-ittb-ud-2.6-200830	Gold tokenization	—	—	98.52	96.37	96.92	95.11	98.93	91.24	89.07	82.62	85.88
latin-llct-ud-2.6-200830	Raw text	100.00	99.49	99.60	97.13	97.11	96.63	97.68	95.48	94.35	89.31	90.44
latin-llct-ud-2.6-200830	Gold tokenization	—	—	99.60	97.14	97.11	96.63	97.68	95.54	94.40	89.40	90.53
latin-proiel-ud-2.6-200830	Raw text	99.87	36.81	96.67	96.81	90.71	89.59	96.16	74.44	69.97	57.51	64.96
latin-proiel-ud-2.6-200830	Gold tokenization	—	—	97.07	97.16	91.53	90.52	96.42	83.78	79.04	67.58	73.88
latin-perseus-ud-2.6-200830	Raw text	100.00	98.46	91.65	81.18	86.33	78.75	88.05	78.09	68.97	52.82	56.03
latin-perseus-ud-2.6-200830	Gold tokenization	—	—	91.64	81.17	86.33	78.74	88.04	78.21	69.07	52.84	55.99
latvian-lvtb-ud-2.6-200830	Raw text	99.32	98.74	96.28	89.64	93.79	88.84	95.81	88.31	85.26	75.23	79.56
latvian-lvtb-ud-2.6-200830	Gold tokenization	—	—	96.92	90.24	94.40	89.43	96.45	89.33	86.23	76.29	80.60
lithuanian-alksnis-ud-2.6-200830	Raw text	99.91	87.87	95.97	90.37	91.07	89.41	93.61	82.54	78.70	67.95	71.30
lithuanian-alksnis-ud-2.6-200830	Gold tokenization	—	—	96.04	90.40	91.18	89.49	93.70	83.93	80.08	69.02	72.43
lithuanian-hse-ud-2.6-200830	Raw text	97.30	97.30	89.66	89.28	81.45	77.07	87.98	70.92	62.53	44.26	53.76
lithuanian-hse-ud-2.6-200830	Gold tokenization	—	—	91.23	91.32	83.21	78.40	90.28	73.77	64.53	45.25	54.68
maltese-mudt-ud-2.6-200830	Raw text	99.84	86.29	95.77	95.66	—	95.30	—	84.76	79.76	68.39	72.24
maltese-mudt-ud-2.6-200830	Gold tokenization	—	—	95.88	95.77	—	95.40	—	85.46	80.38	68.69	72.66
marathi-ufal-ud-2.6-200830	Raw text	90.25	92.63	78.50	—	65.25	61.50	80.00	61.25	53.50	31.73	40.92
marathi-ufal-ud-2.6-200830	Gold tokenization	—	—	84.22	—	68.69	63.83	80.10	70.39	60.92	31.95	42.32
naija-nsc-ud-2.6-200830	Raw text	100.00	99.56	98.14	—	99.16	97.77	99.27	92.46	89.81	84.18	86.20
naija-nsc-ud-2.6-200830	Gold tokenization	—	—	98.14	—	99.16	97.77	99.27	92.50	89.84	84.25	86.26
north_sami-giella-ud-2.6-200830	Raw text	99.87	98.79	92.35	93.57	89.40	85.61	86.85	76.66	71.84	60.71	58.95
north_sami-giella-ud-2.6-200830	Gold tokenization	—	—	92.47	93.70	89.56	85.75	86.96	76.97	72.16	60.95	59.19
norwegian-bokmaal-ud-2.6-200830	Raw text	99.83	95.63	98.37	—	97.52	96.86	98.55	93.74	92.26	87.03	88.76
norwegian-bokmaal-ud-2.6-200830	Gold tokenization	—	—	98.57	—	97.71	97.05	98.75	94.48	93.00	87.67	89.43
norwegian-nynorsk-ud-2.6-200830	Raw text	99.91	94.11	98.36	—	97.38	96.67	98.37	93.86	92.11	86.07	88.11
norwegian-nynorsk-ud-2.6-200830	Gold tokenization	—	—	98.50	—	97.51	96.80	98.50	94.66	92.93	86.95	89.01
norwegian-nynorsklia-ud-2.6-200830	Raw text	99.91	99.53	96.45	—	95.71	93.62	98.05	80.90	76.53	65.74	69.55
norwegian-nynorsklia-ud-2.6-200830	Gold tokenization	—	—	96.55	—	95.79	93.72	98.14	81.15	76.76	65.94	69.80
old_church_slavonic-proiel-ud-2.6-200830	Raw text	100.00	41.43	96.58	96.83	90.44	89.17	93.19	77.42	73.57	63.51	68.53
old_church_slavonic-proiel-ud-2.6-200830	Gold tokenization	—	—	96.89	97.09	91.22	89.97	93.20	87.95	83.81	73.92	77.26
old_french-srcmf-ud-2.6-200830	Raw text	99.93	100.00	96.40	96.27	97.80	95.58	—	92.28	87.74	81.08	84.17
old_french-srcmf-ud-2.6-200830	Gold tokenization	—	—	96.47	96.33	97.86	95.64	—	92.36	87.81	81.17	84.26
old_russian-torot-ud-2.6-200830	Raw text	100.00	29.60	94.33	94.39	87.51	85.16	85.82	70.66	65.18	51.26	53.18
old_russian-torot-ud-2.6-200830	Gold tokenization	—	—	94.93	94.99	88.44	86.35	85.77	83.15	77.17	63.78	62.66
old_russian-rnc-ud-2.6-200830	Raw text	98.15	85.46	91.80	87.74	75.83	66.63	74.94	63.08	57.53	33.85	35.04
old_russian-rnc-ud-2.6-200830	Gold tokenization	—	—	93.34	89.43	77.09	67.76	76.13	66.86	60.73	36.05	37.07
persian-seraji-ud-2.6-200830	Raw text	99.65	98.75	97.69	97.66	97.75	97.29	96.67	91.09	88.15	83.43	82.26
persian-seraji-ud-2.6-200830	Gold tokenization	—	—	97.98	97.97	98.07	97.60	96.94	91.74	88.76	84.00	82.82
polish-pdb-ud-2.6-200830	Raw text	99.85	97.33	98.88	95.73	95.84	95.03	98.05	94.02	92.01	84.93	88.08
polish-pdb-ud-2.6-200830	Gold tokenization	—	—	99.04	95.88	95.99	95.18	98.19	94.51	92.50	85.36	88.52
polish-lfg-ud-2.6-200830	Raw text	99.85	99.65	98.92	95.99	96.51	95.06	98.27	96.89	95.52	89.73	92.45
polish-lfg-ud-2.6-200830	Gold tokenization	—	—	99.09	96.18	96.70	95.25	98.41	97.29	95.91	90.12	92.77
portuguese-gsd-ud-2.6-200830	Raw text	99.84	97.50	98.53	98.52	99.71	98.43	99.33	94.57	93.47	88.69	90.02
portuguese-gsd-ud-2.6-200830	Gold tokenization	—	—	98.69	98.69	99.87	98.59	99.49	94.94	93.82	89.11	90.36
portuguese-bosque-ud-2.6-200830	Raw text	99.55	90.64	97.19	—	96.17	94.79	97.98	92.32	89.72	79.29	84.22
portuguese-bosque-ud-2.6-200830	Gold tokenization	—	—	97.60	—	96.49	95.11	98.42	93.53	90.80	80.42	85.51
romanian-rrt-ud-2.6-200830	Raw text	99.69	95.28	97.79	97.18	97.32	96.81	98.20	91.83	87.56	80.00	82.17
romanian-rrt-ud-2.6-200830	Gold tokenization	—	—	98.08	97.44	97.60	97.08	98.49	92.74	88.38	80.82	82.88
romanian-nonstandard-ud-2.6-200830	Raw text	98.35	96.73	95.61	91.38	90.03	88.67	94.23	88.89	84.47	67.59	75.81
romanian-nonstandard-ud-2.6-200830	Gold tokenization	—	—	97.21	92.90	91.53	90.13	95.74	91.00	86.49	69.53	77.28
russian-syntagrus-ud-2.6-200830	Raw text	99.60	98.80	98.86	—	97.60	97.38	98.33	94.22	92.97	89.27	90.35
russian-syntagrus-ud-2.6-200830	Gold tokenization	—	—	99.27	—	97.98	97.76	98.68	94.99	93.72	89.90	90.95
russian-gsd-ud-2.6-200830	Raw text	99.50	96.22	98.03	97.51	94.76	93.60	96.89	91.66	88.38	80.67	84.18
russian-gsd-ud-2.6-200830	Gold tokenization	—	—	98.49	97.98	95.17	93.97	97.27	92.77	89.43	81.44	85.05
russian-taiga-ud-2.6-200830	Raw text	97.16	82.69	94.13	95.72	90.01	87.50	93.05	81.17	76.99	65.28	69.94
russian-taiga-ud-2.6-200830	Gold tokenization	—	—	96.47	98.56	92.72	89.87	95.68	85.57	80.81	68.93	73.90
sanskrit-vedic-ud-2.6-200830	Raw text	100.00	27.18	89.50	—	81.72	77.12	87.11	60.79	49.75	41.65	44.67
sanskrit-vedic-ud-2.6-200830	Gold tokenization	—	—	90.01	—	83.11	78.58	87.24	73.34	61.55	51.87	54.91
scottish_gaelic-arcosg-ud-2.6-200830	Raw text	99.58	55.57	93.63	87.07	89.78	85.43	95.41	77.66	71.86	55.15	60.51
scottish_gaelic-arcosg-ud-2.6-200830	Gold tokenization	—	—	94.26	87.84	90.23	86.30	95.85	83.77	77.61	62.05	68.26
serbian-set-ud-2.6-200830	Raw text	99.99	93.00	98.98	95.75	95.92	95.35	97.82	93.66	91.18	83.18	86.80
serbian-set-ud-2.6-200830	Gold tokenization	—	—	99.01	95.78	95.94	95.39	97.83	94.33	91.82	83.84	87.45
slovak-snk-ud-2.6-200830	Raw text	100.00	85.28	97.19	87.79	92.66	86.71	96.52	91.71	89.60	78.75	84.54
slovak-snk-ud-2.6-200830	Gold tokenization	—	—	97.30	88.06	92.84	86.98	96.60	93.68	91.57	80.55	86.59
slovenian-ssj-ud-2.6-200830	Raw text	97.99	67.98	96.93	94.35	94.56	93.95	96.59	88.09	86.65	80.90	83.58
slovenian-ssj-ud-2.6-200830	Gold tokenization	—	—	98.86	96.44	96.69	96.01	98.54	94.41	92.96	86.68	89.30
slovenian-sst-ud-2.6-200830	Raw text	99.85	23.14	94.70	92.70	92.52	89.74	97.14	64.23	59.57	49.25	52.95
slovenian-sst-ud-2.6-200830	Gold tokenization	—	—	95.71	93.11	92.94	90.90	97.46	77.81	72.24	62.71	67.18
spanish-ancora-ud-2.6-200830	Raw text	99.95	98.32	99.09	99.02	98.87	98.33	99.36	93.62	91.78	86.82	88.06
spanish-ancora-ud-2.6-200830	Gold tokenization	—	—	99.14	99.06	98.91	98.37	99.40	93.83	91.97	87.01	88.24
spanish-gsd-ud-2.6-200830	Raw text	99.76	94.54	97.17	—	97.05	95.32	98.80	92.00	89.70	79.23	84.49
spanish-gsd-ud-2.6-200830	Gold tokenization	—	—	97.40	—	97.27	95.54	99.02	92.73	90.38	79.93	85.13
swedish-talbanken-ud-2.6-200830	Raw text	99.89	96.13	98.41	97.26	97.33	96.46	98.19	91.99	89.68	83.63	85.82
swedish-talbanken-ud-2.6-200830	Gold tokenization	—	—	98.51	97.38	97.44	96.57	98.30	92.46	90.14	84.12	86.33
swedish-lines-ud-2.6-200830	Raw text	99.96	87.20	97.71	95.47	90.89	88.10	97.76	89.14	85.80	71.44	81.67
swedish-lines-ud-2.6-200830	Gold tokenization	—	—	97.75	95.48	90.91	88.09	97.79	89.91	86.52	72.13	82.48
tamil-ttb-ud-2.6-200830	Raw text	94.51	97.52	88.39	82.92	85.30	82.11	89.15	70.28	64.91	54.93	58.46
tamil-ttb-ud-2.6-200830	Gold tokenization	—	—	93.36	87.28	90.10	86.22	93.97	78.03	71.79	61.09	64.80
telugu-mtg-ud-2.6-200830	Raw text	99.58	96.62	93.63	93.63	98.48	93.63	—	90.17	83.52	76.00	79.62
telugu-mtg-ud-2.6-200830	Gold tokenization	—	—	94.04	94.04	98.89	94.04	—	91.12	84.47	76.84	80.46
turkish-imst-ud-2.6-200830	Raw text	98.30	96.97	94.48	93.69	92.06	89.95	94.41	72.63	66.80	58.31	61.57
turkish-imst-ud-2.6-200830	Gold tokenization	—	—	96.10	95.32	93.66	91.50	96.00	76.10	69.93	60.33	63.83
ukrainian-iu-ud-2.6-200830	Raw text	99.81	96.61	97.89	94.22	94.18	93.13	97.39	90.59	88.24	78.76	83.19
ukrainian-iu-ud-2.6-200830	Gold tokenization	—	—	98.10	94.42	94.34	93.30	97.56	91.11	88.75	79.11	83.59
urdu-udtb-ud-2.6-200830	Raw text	100.00	98.31	94.10	92.27	82.89	78.41	97.38	88.27	82.63	56.79	74.77
urdu-udtb-ud-2.6-200830	Gold tokenization	—	—	94.08	92.26	82.92	78.43	97.39	88.37	82.74	56.90	74.92
uyghur-udt-ud-2.6-200830	Raw text	99.54	81.81	89.24	91.70	88.47	80.04	94.76	76.58	64.72	46.67	55.08
uyghur-udt-ud-2.6-200830	Gold tokenization	—	—	89.67	92.21	88.92	80.47	95.27	78.39	66.27	47.53	56.23
vietnamese-vtb-ud-2.6-200830	Raw text	85.37	93.46	78.19	76.69	85.11	76.53	85.15	52.80	47.90	41.56	44.31
vietnamese-vtb-ud-2.6-200830	Gold tokenization	—	—	90.56	88.69	99.72	88.47	99.58	72.63	65.26	58.85	62.42
welsh-ccg-ud-2.6-200830	Raw text	99.42	96.28	94.02	92.96	89.04	86.39	92.88	85.79	79.16	60.98	66.81
welsh-ccg-ud-2.6-200830	Gold tokenization	—	—	94.54	93.51	89.52	86.83	93.46	87.04	80.35	62.14	68.05
wolof-wtb-ud-2.6-200830	Raw text	99.23	91.95	94.25	94.12	93.37	91.19	95.22	83.79	78.59	66.50	70.09
wolof-wtb-ud-2.6-200830	Gold tokenization	—	—	95.19	95.03	94.22	92.10	95.97	85.98	80.75	68.61	72.08

Czech PDT-C 1.0 Model

PDT-C 1.0 Model is distributed under the CC BY-NC-SA licence. The model is trained on PDT-C 1.0 treebank using RobeCzech model, and performs morphological analysis using the MorfFlex CZ 2.0 morphological dictionary via MorphoDiTa.

The model requires UDPipe 2.1, together with Python packages ufal.udpipe version at least 1.3.1.1 and ufal.morphodita version at least 1.11.2.1.

Download

The latest version 231116 of the Czech PDT-C 1.0 model can be downloaded from the LINDAT/CLARIN repository.

The model is also available in the REST service.

PDT-C 1.0 Morphological System

PDT-C 1.0 uses the PDT-C tag set from MorfFlex CZ 2.0, which is an evolution of the original PDT tag set devised by Jan Hajič (Hajič, 2004). The tags are positional with 15 positions corresponding to part of speech, detailed part of speech, gender, number, case, etc. (e.g. NNFS1-----A----). Different meanings of same lemmas are distinguished and additional comments can be provided for every lemma meaning. The complete reference can be found in the Manual for Morphological Annotation, Revision for the Prague Dependency Treebank - Consolidated 2020 release and quick reference is available in the PDT-C positional morphological tags overview.

The PDT-C 1.0 emply dependency relations from the PDT analytical level, with a quick reference available in the PDT-C analytical functions and clause segmentation overview.

In the CoNLL-U format, the

• tags are filled in the XPOS column, and
• the dependency relations are filled in the DEPREL, even if they are different from the universal dependency relations.

PDT-C 1.0 Train/Dev/Test Splits

The PDT-C corpus consists of four datasets, but some of them do not have an official train/dev/test split. We therefore used the following split:

• PDT dataset is already split into train, dev (dtest), and test (etest).
• PCEDT dataset is a translated version of the Wall Street Journal, so we used the usual split into train (sections 0-18), dev (sections 19-21), and test (sections 22-24).
• PDTSC and FAUST datasets have no split, so we split it into dev (documents with identifiers ending with 6), test (documents with identifiers ending with 7), and train (all the remaining documents).

Acknowledgements

This work has been supported by the LINDAT/CLARIAH-CZ project funded by Ministry of Education, Youth and Sports of the Czech Republic (project LM2023062).

Publications

• Milan Straka, Jakub Náplava, Jana Straková, David Samuel (2020): RobeCzech: Czech RoBERTa, a monolingual contextualized language representation model. In: Text, Speech, and Dialogue. TSD 2021. Lecture Notes in Computer Science, vol 12848. Springer, Cham.
• Jan Hajič, Eduard Bejček, Jaroslava Hlavacova, Marie Mikulová, Milan Straka, Jan Štěpánek, and Barbora Štěpánková (2020). Prague Dependency Treebank - Consolidated 1.0. In: Proceedings of the 12th Language Resources and Evaluation Conference, pages 5208–5218, Marseille, France. European Language Resources Association.
• Milan Straka (2018): UDPipe 2.0 Prototype at CoNLL 2018 UD Shared Task. In: Proceedings of CoNLL 2018: The SIGNLL Conference on Computational Natural Language Learning, pp. 197-207, Association for Computational Linguistics, Stroudsburg, PA, USA, ISBN 978-1-948087-72-8
• Straková Jana, Straka Milan and Hajič Jan (2014): Open-Source Tools for Morphology, Lemmatization, POS Tagging and Named Entity Recognition. In: Proceedings of 52nd Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 13-18, Baltimore, Maryland, June 2014. Association for Computational Linguistics.

Model Performance

Tagging and Lemmatization

We evaluate tagging and lemmatization on the four datasets of PDT-C 1.0, and we also compute a macro-average. For lemmatization, we use the following metrics:

• Lemmas: a primary metric comparing the lemma proper, which is the lemma with an optional lemma number (but we ignore the additional lemma comments like “this is a given name”);
• LemmasEM: an exact match comparing also the lemma comments. This metric is less or equal to Lemmas. Our model directly predicts only lemma proper (no additional comments), and relies on the morphological dictionary to supply the comments, so it fails to generate comments for unknown words (like an unknown given name).

We perform the evaluation using the udpipe2_eval.py, which is a minor extension of the CoNLL 2018 Shared Task evaluation script.

Because the model also include a rule-based tokenizer and sentence splitter, we evaluate both:

• using raw input text, which must first be tokenized and split into sentences. The resulting scores are in fact F1-scores. Note that the FAUST dataset does not contain any discernible sentence boundaries.
• using gold tokenization.

Treebank	Mode	Tokens	Sents	XPOS	Lemma	LemmaEM
PDT	Raw text	99.91	88.00	98.69	99.10	98.86
PDT	Gold tokenization	—	—	98.78	99.19	98.96
PCEDT	Raw text	99.97	94.06	98.77	99.36	98.75
PCEDT	Gold tokenization	—	—	98.80	99.40	98.78
PDTSC	Raw text	100.0	98.31	98.77	99.23	99.16
PDTSC	Gold tokenization	—	—	98.77	99.23	99.16
FAUST	Raw text	100.0	10.98	97.05	98.88	98.43
FAUST	Gold tokenization	—	—	97.42	98.78	98.30
MacroAvg	Gold tokenization	—	—	98.44	99.15	98.80

Dependency Parsing

In PDT-C 1.0, the only manually annotated dependency parsing dataset is a subset of the PDT dataset. We perform the evaluation as in the previous section.

Treebank	Mode	Tokens	Sents	XPOS	Lemma	LemmaEM	UAS	LAS
PDT subset	Raw text	99.94	88.49	98.74	99.16	98.97	93.45	90.32
PDT subset	Gold tokenization	—	—	98.81	99.23	99.03	94.41	91.48

EvaLatin 2020 Models

EvaLatin 2020 Models are distributed under the CC BY-NC-SA licence. The models are based solely on EvaLatin 2020 treebanks, and additionally use multilingual BERT.

The models require UDPipe 2.

Download

The latest version 200831 of the EvaLatin 2020 models can be downloaded from LINDAT/CLARIN repository.

The models are also available in the REST service.

Acknowledgements

This work was supported by the grant no. GX20-16819X of the Grant Agency of the Czech Republic, and has been using language resources stored and distributed by the LINDAT/CLARIAH-CZ project of the Ministry of Education, Youth and Sports of the Czech Republic (project LM2018101).

The models were trained on EvaLatin 2020 treebanks.

Finally, multilingual BERT is used to provide contextualized word embeddings.

Publications

• Milan Straka, Jana Straková (2020): UDPipe at EvaLatin 2020: Contextualized Embeddings and Treebank Embeddings. In: ArXiv.org Computing Research Repository, ISSN 2331-8422, 2006.03687

Model Performance

Model	Dataset	UPOS	Lemma
latin-evalatin20-200830	test classical	96.73	96.39
latin-evalatin20-200830	test cross-genre	90.47	86.89
latin-evalatin20-200830	test cross-time	87.58	90.59