Pie Model for Lemmatization, POS Tagging, and Morphological Analysis of Classical Armenian

The models were trained for lemmatization, POS-tagging, and morphological analysis of Classical Armenian. The training dataset used was the Classical Armenian corpus from Universal Dependencies (09/2024 release), comprising 63,271 wordforms for training and 8,686 for evaluation. Although the training data primarily consists of the Classical Armenian Gospels, the model exhibits strong performance across both in-domain and out-of-domain tests (refer to the linked publication). Note that the input data should be pre-tokenized.

The model development was part of the ANR project ANR-21-CE38-0006 "DALiH - Digitizing Armenian Linguistic Heritage", led by Victoria Khurshudyan (Inalco, SeDyL, CNRS, IRD), with initial contributions from Calfa and GREgORI. Models have been developed for the EMNLP 2024 conference (NLP4DH workshop), and rely on the PIE framework.

Data :

For the training dataset, see:

• Kocharov Petr, Kharatyan Lilit: GitHub Repository

Results :

For detailed experiments and results, please refer to the linked publication. The following table displays accuracy (f1-score).

Models can be used on Deucalion, the lemmatization service from École nationale des chartes-PSL.

Selected Bibliography:

Vidal-Gorène, C., & Kindt, B. (2020, May). Lemmatization and POS-tagging process by using joint learning approach. Experimental results on Classical Armenian, Old Georgian, and Syriac. In Proceedings of LT4HALA 2020-1st Workshop on Language Technologies for Historical and Ancient Languages(pp. 22-27).

Vidal-Gorène, C., Khurshudyan, V., & Donabédian-Demopoulos, A. (2020, December). Recycling and comparing morphological annotation models for Armenian diachronic-variational corpus processing. In Proceedings of the 7th Workshop on NLP for Similar Languages, Varieties and Dialects (pp. 90-101).

Vidal-Gorène C., Tomeh N., and Khurshudyan V. (2024, November). Cross-Dialectal Transfer and Zero-Shot Learning for Armenian Varieties: A Comparative Analysis of RNNs, Transformers and LLMs. In Proceedings of the 4th International Conference on Natural Language Processing for Digital Humanities, pages 438–449, Miami, USA. Association for Computational Linguistics.