The EXLMR model is a multilingual transformer that expands the XLM-RoBERTa tokenizer by adding vocabulary for low-resource languages such as Tigrinya and Amharic. It solves issues like out-of-vocabulary words and over-tokenization, enhancing the model's ability to represent languages written in the Ge'ez script. The model can be fine-tuned for various multilingual tasks, including sentiment analysis, question answering, named entity recognition, and paraphrase detection. These improvements make EXLMR highly effective for low-resource languages, while still supporting a broad range of languages with strong overall performance.
Model Description
This model is an extended version of XLM-R (XLM-RoBERTa), specifically adapted for Geez script languages. Leveraging the robust architecture of XLM-RoBERTa, it incorporates additional tokens and vocabulary to handle the unique linguistic features and characters of Geez script languages. This extension enhances the model’s ability to perform a variety of NLP tasks, including classification, sentiment analysis, and more, within the Geez script context.
Tokenizer Description
The tokenizer for this extended XLM-R model has been customized to include new tokens that represent the characters and symbols specific to Geez script languages. It is designed to preprocess text data by converting Geez script characters into tokens compatible with the extended model. This ensures that text in the Geez script is accurately and efficiently processed for downstream tasks.
