Exploring the Potential of Machine Translation for Generating Named Entity Datasets: A Case Study between Persian and English
Amir Sartipi, Afsaneh Fatemi
TL;DR
The paper tackles the scarcity of annotated Persian NER data by generating Persian datasets via machine translation of English benchmarks. To preserve entity alignment during translation, it masks named-entity spans with a special format and aligns by index. Evaluations with a monolingual Persian model (Pars-Bert) and a multilingual model (xlm-roberta-base) show that CoNLL 2003 translated data achieves an F1 of 85.11%, while WNUT 2017 drops to 40.02%. The results indicate MT-based data generation is a promising strategy for bootstrapping NER resources in low-resource languages and can be used to augment and diversify training data, with caveats on complex entity types. Overall, the paper provides a practical pipeline and empirical insights into the impact of translation quality on cross-lingual NER performance.
Abstract
This study focuses on the generation of Persian named entity datasets through the application of machine translation on English datasets. The generated datasets were evaluated by experimenting with one monolingual and one multilingual transformer model. Notably, the CoNLL 2003 dataset has achieved the highest F1 score of 85.11%. In contrast, the WNUT 2017 dataset yielded the lowest F1 score of 40.02%. The results of this study highlight the potential of machine translation in creating high-quality named entity recognition datasets for low-resource languages like Persian. The study compares the performance of these generated datasets with English named entity recognition systems and provides insights into the effectiveness of machine translation for this task. Additionally, this approach could be used to augment data in low-resource language or create noisy data to make named entity systems more robust and improve them.