Specialized machine translation improves voice for a single nahuatl variant
Unfragmenting Nawatl NMT: Single-Variant Specialization for Masewalkopa
Computation and Language
Summary
Translating closely related language variants together can create confusing mixes that don't sound natural to any community. The authors focus on a specific variety of the Nawatl language spoken in Mexico to build a translation model that feels authentic for that speech community. They improved the model by using special grammar information and expert transcriptions, making translations clearer and faster to run on regular computers. This work helps preserve and use a rare indigenous language more accurately.
What this means in practice
- •For language technology developers: Create accurate translation models specializing in a single indigenous language variant instead of mixing multiple variants.
- •For cloud service operators: Deploy optimized low-latency translation models for rare languages on cloud platforms with reduced computing costs.
Authors
Luis A. Mercado-Campos, Alexis Palmer
Abstract
In minoritized (low-resource) language technologies, a common response to data sparsity is to aggregate text across multiple varieties and treat the combined data as a single corpus. However, mixed corpora of closely related indigenous variants can produce an artificial register that is accountable to no single speaker community, a structural failure we call variant fragmentation. This issue is acute for minoritized languages native to Mexico, such as Nawatl. Building on a 70,000-pair multi-variant foundation, we specialize an mT5-large checkpoint to Masewalkopa (nhn), the Nawatl variety spoken in Santa Ana Tlacotenco. We curate 2,665 parallel pairs, reserve all sentence-level data for evaluation, and incorporate 192 elder-transcribed target sentences via asymmetric backtranslation. A custom finite-state analyzer reduces unanalyzed surface forms from 82.0% to 19.9% and supplies functional syntax tags as plain-text prefixes, allowing the model to reuse existing subword embeddings rather than relying on frozen special tokens. The resulting LoRA-v9c-b model achieves 58.46 dev chrF++ and 58.20/57.90 directional test chrF++. Finally, CTranslate2 int8 quantization reduces CPU latency from 137.5 to 1.8-2.6 seconds per request, enabling cost-bounded Cloud Run deployment.