Enemray improves Hassaniya language understanding and translation accuracy
Enemray: Toward Capable Language Models for Hassaniya
Computation and LanguageArtificial Intelligence
Summary
Hassaniya is a language spoken in Mauritania with limited digital language tools available. The authors created Enemray, a language model specially trained to understand and use Hassaniya well, while keeping the skills of a general advanced AI model. They carefully collected and assembled various types of Hassaniya text data and used a training process that balances learning new language skills with keeping general AI abilities. Enemray performs better than other models in translating English to Hassaniya and detecting translation errors involving Mauritanian language content, while still handling tasks like math and code well.
What this means in practice
- •For machine translation teams: Provide more accurate English to Hassaniya translation capabilities for translation tools focused on Mauritania.
- •For conversational ai developers: Build Hassaniya-capable chatbots and virtual assistants that understand cultural and task-specific contexts for users.
Authors
Cheikh Ahmed
Abstract
We introduce Enemray, a Hassaniya-centric language model that enables general-purpose interaction in Hassaniya. Enemray is trained around a stability--plasticity objective: acquire strong Hassaniya linguistic and cultural competence while preserving the general reasoning, multilingual, instruction-following, and safety behaviors of a capable instruction-tuned model. The development pipeline separates language acquisition from behavioral specialization. A separately assembled continual-pretraining corpus provides broad exposure to natural Hassaniya and Mauritanian text; layer-selective continual pretraining learns a compact language-specific parameter update; that update is transferred into the instruction-tuned parameter space; and supervised post-training develops conversational, cultural, literary, task-oriented, and cross-lingual behavior. The supervised corpus integrates selected public Hassaniya and Mauritanian resources with a substantially larger body of newly collected, reconstructed, curated, and constructed instruction data, while policy-generated replay provides a retention signal from the reference model's own behavior distribution. The resulting collection is substantially larger and broader in purpose than existing Hassaniya text resources. In evaluation, Enemray achieves the strongest English to Hassaniya translation among the compared open and proprietary models and the highest overall score on Mauritanian translation error detection, while retaining most of the general capabilities of its instruction-tuned base model on mathematical reasoning, knowledge, code generation, and function calling. This report describes the motivation, data construction, model design, training methodology, and evaluation of Enemray.