🪐 Weasel Project: la_core_web_xx

Code required to train spaCy-compatible sm, md, lg, and trf core models for Latin, i.e pipeline with POS tagger, morphologizer, lemmatizer, dependency parser, and NER trained on all available Latin UD treebanks, i.e. Perseus, PROIEL, ITTB, UDante, and LLCT (as currently normalized by Gamba/Zeman 2003; see assets below); the lemmatizer and tagger components are also trained on LASLA Corpus connlup files (see assets below). The md and lg models contains floret vectors trained on Wikipedia, Oscar, UD and—for lg—CC100-Latin data. The trf model uses bert-base-multilingual-uncased. NER is based on custom tagged data based on tagger output and manual annotation, supplemented by data from the Herodotos Project; this data is included in assets/ner/.

📋 project.yml

The project.yml defines the data assets required by the project, as well as the available commands and workflows. For details, see the Weasel documentation.

⏯ Commands

The following commands are defined by the project. They can be executed using weasel run [name]. Commands are only re-run if their inputs have changed.

Command	Description
assets	Download assets
preprocess	Convert different UD treebanks so that they use shared formatting, feature defs, etc.
convert	Convert the data to spaCy's format
init-labels	Initialize labels for components
train_sm	Train tagger/parser/etc. on Latin UD treebanks
train-lemma_sm	Retrain tagger/parser/etc. on Latin UD treebanks and LASLA
train_md	Train tagger/parser/etc. on Latin UD treebanks
train-lemma_md	Retrain tagger/parser/etc. on Latin UD treebanks and LASLA
train_lg	Train tagger/parser/etc. on Latin UD treebanks
train-lemma_lg	Retrain tagger/parser/etc. on Latin UD treebanks and LASLA
train_trf	Train tagger/parser/etc. on Latin UD treebanks
train-lemma_trf	Retrain tagger/parser/etc. on Latin UD treebanks and LASLA
evaluate_sm	Evaluate model on the test data and save the metrics
evaluate_md	Evaluate model on the test data and save the metrics
evaluate_lg	Evaluate model on the test data and save the metrics
evaluate_trf	Evaluate model on the test data and save the metrics
convert-ner	Convert the NER data to spaCy's binary format
train-ner_sm	Train the NER model for the model
train-ner_md	Train the NER model for the model
train-ner_lg	Train the NER model for the model
train-ner_trf	Train the NER model for the model
assemble_sm	Assemble core model, i.e. add NER component to dep model
assemble_md	Assemble core model, i.e. add NER component to dep model
assemble_lg	Assemble core model, i.e. add NER component to dep model
assemble_trf	Assemble core model, i.e. add NER component to dep model
assemble-meta_sm	Assemble meta.json files so that all metrics are included
assemble-meta_md	Assemble meta.json files so that all metrics are included
assemble-meta_lg	Assemble meta.json files so that all metrics are included
assemble-meta_trf	Assemble meta.json files so that all metrics are included
package_sm	Package the trained core model
package_md	Package the trained core model
package_lg	Package the trained core model
package_trf	Package the trained core model
document	Document core_web_xx
clean	Remove intermediate files

⏭ Workflows

The following workflows are defined by the project. They can be executed using weasel run [name] and will run the specified commands in order. Commands are only re-run if their inputs have changed.

Workflow	Steps
all	train-ner_trf → assemble_trf → assemble-meta_trf → package_sm → package_md → package_lg → package_trf → document

🗂 Assets

The following assets are defined by the project. They can be fetched by running weasel assets in the project directory.

File	Source	Description
assets/original/ud	Git
assets/original/lasla	Git

Install

• To install the current version of the lg model...
  • pip install "la-core-web-lg @ https://huggingface.co/latincy/la_core_web_lg/resolve/main/la_core_web_lg-any-py3-none-any.whl"

Use in spaCy

• To load the lg model in spaCy...

import spacy
nlp = spacy.load("la_core_web_lg")

Model repository

• The models can be found here:
  • https://huggingface.co/latincy

Changelog

• v3.7.7: Adds CIRCSE UD treebank to training data; fixes issue with sentence segmentation [7.31.2024]
• v3.7.6: Adds lookups as requirement; updates metadata, etc. [7.14.2024]
• v3.7.5: Replaces UD training data with "harmonized" treebanks (cf. Gamba and Zeman 2023); introduces “backoff”-style Lookup Lemmatizer to catch large numbers of unambiguous forms (+850K); NB: uses a custom fork of spaCy’s space-lookup-data, cf. https://github.com/diyclassics/spacy-lookups-data/tree/master; support for Latin-specific verb tenses, i.e. perf (perfect) or imp (imperfect) as opposed to past; simplification and speed-up of LatinCy-specific UD preprocessing workflow. [07.03.2024]
• v3.7.4: Retrain with release of spaCy v3.7.4 [04.12.2023]
• v3.6.0: Retrain with release of spaCy v3.6.0 [07.08.2023]
• v3.5.3: Add Verbform to morph labels; allows better handling of infinitives, gerunds, and gerundives [6.22.2023]

Training details

• The v.3.7.7 models are trained using a single RTX8000 GPU with 32GB of memory.

Acknowledgements

This work was supported in part through the NYU IT High Performance Computing resources, services, and staff expertise; for more information, see https://sites.google.com/nyu.edu/nyu-hpc/about/acknowledgement-statement.

Bibliography

• Cecchini, F.M., Passarotti, M., Marongiu, P., and Zeman, D. 2018. “Challenges in Converting the Index Thomisticus Treebank into Universal Dependencies.” In Proceedings of the Second Workshop on Universal Dependencies (UDW 2018). 27–36.
• Cecchini, F.M., Sprugnoli, R., Moretti, G., and Passarotti, M. 2020. “UDante: First Steps Towards the Universal Dependencies Treebank of Dante’s Latin Works.” In Dell’Orletta, F., Monti, J., and Tamburini, F. eds. Proceedings of the Seventh Italian Conference on Computational Linguistics CLiC-It 2020. Accademia University Press. 99–105. doi:10.4000/books.aaccademia.8653. http://books.openedition.org/aaccademia/8653.
• Celano, G.G.A., Crane, G., Almas, B., and et al. 2014. “The Ancient Greek and Latin Dependency Treebank v.2.1.” https://perseusdl.github.io/treebank_data/.
• Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. 2019. “BERT: Pre-Training of Deep Bidirectional Transformers for Language Understanding.” In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). Minneapolis, Minnesota: Association for Computational Linguistics. 4171–86.
• Erdmann, A., Wrisley, D.J., Allen, B., Brown, C., Bodénès, S.C., Elsner, M., Feng, Y., Joseph, B., Joyeaux-Prunel, B., and Marneffe, M.-C. 2019. “Practical, Efficient, and Customizable Active Learning for Named Entity Recognition in the Digital Humanities.” In Proceedings of North American Association of Computational Linguistics (NAACL 2019). Minneapolis, Minnesota.
• Fantoli, M., Passarotti, M.C., Litta Modignani Picozzi, E., Ruffolo, P., and Moretti, G. 2023. CIRCSE/LASLA: LASLA Corpus (version v1.0.1). Zenodo. doi:10.5281/zenodo.8370759. https://zenodo.org/records/8370759.
• Gamba, F., and Zeman, D. 2023. “Universalising Latin Universal Dependencies: A Harmonisation of Latin Treebanks in UD.” In Grobol, L. and Tyers, F. eds. Proceedings of the Sixth Workshop on Universal Dependencies (UDW, GURT/SyntaxFest 2023). Washington, D.C.: Association for Computational Linguistics. 7–16. https://aclanthology.org/2023.udw-1.2.
• Haug, D.T., and Jøhndal, M. 2008. “Creating a Parallel Treebank of the Old Indo-European Bible Translations.” In Proceedings of the Second Workshop on Language Technology for Cultural Heritage Data (LaTeCH 2008). 27–34.
• Honnibal, M., and Montani, I. 2023. “SpaCy: Industrial-Strength Natural Language Processing in Python” (version v. 3.5.0). https://spacy.io/.
• Korkiakangas, T. 2021. “Late Latin Charter Treebank: Contents and Annotation.” Corpora 16 (2) (August 1): 191–203. doi:10.3366/cor.2021.0217.
• Passarotti, M., and Dell’Orletta, F. 2010. “Improvements in Parsing the Index Thomisticus Treebank. Revision, Combination and a Feature Model for Medieval Latin.” In Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC’10). Valletta, Malta: European Language Resources Association (ELRA). http://www.lrec-conf.org/proceedings/lrec2010/pdf/178_Paper.pdf
• Ströbel, P. 2022. “CC100-Latin.” https://huggingface.co/datasets/pstroe/cc100-latin.