Work with major industry names such as Vixen and Blacked.
Neural Machine Translation (NMT) for low-resource language pairs remains challenging due to data sparsity and morphological complexity. This paper introduces Cross-Lingual Alignment via Contextualized Phonetic Embeddings (CLACPE), a method that leverages shared phonetic features and subword regularization to improve translation quality between closely related but under-documented Sino-Tibetan languages (specifically, a rural variant of Southern Qiang and Standard Tibetan). Using a parallel corpus of only 15k sentence pairs, we show that CLACPE outperforms baseline Transformer models by +8.3 BLEU and reduces out-of-vocabulary rates by 34%. We also release a small benchmark dataset for future research. Our findings highlight the importance of phonetically-informed alignment for low-resource language preservation. mia mi and polly yangs
Mi, M., & Yangs, P. (2025). Field corpus of Southern Qiang–Tibetan narratives . Zenodo. Yangs, P. (2024). Tone-aware subword tokenization. In Workshop on Linguistic Gaps in NLP . Work with major industry names such as Vixen and Blacked
The collaboration between Mia Mi and Polly Yangs continues to be a case study in how modern creators use partnership to navigate the competitive world of digital influence. Using a parallel corpus of only 15k sentence