Statistical Machine Translation Systems of English-Pnar Language Pair : Some Insights of the Emperical Study
Researchers have conducted the first study on translating English to Pnar, a language spoken by approximately 0.4 million people in India's Jaintia Hills. They built a parallel corpus of 10,234 sentences from newspaper articles and trained machine translation systems using Moses and other tools. The best-performing system achieved a BLEU score of 14.97 for translating Pnar to English and 11.16 for translating English to Pnar. The study also found that lexicalized reordering i
Researchers have conducted the first study on translating English to Pnar, a language spoken by approximately 0.4 million people in India's Jaintia Hills. They built a parallel corpus of 10,234 sentences from newspaper articles and trained machine translation systems using Moses and other tools. The best-performing system achieved a BLEU score of 14.97 for translating Pnar to English and 11.16 for translating English to Pnar. The study also found that lexicalized reordering improved translation quality by 3.73 BLEU points, but MERT tuning degraded performance under low-resource conditions.
---
Why it matters: This research matters because it establishes a quantitative benchmark for the English-Pnar language pair, which is essential for developing NLP resources and digital corpora for the Pnar language. It also highlights the challenges of machine translation in low-resource languages and provides insights into improving translation quality.
Source: https://arxiv.org/abs/2608.23120
This article was originally published at: https://arxiv.org/abs/2608.23120