Mizo translation improves access to Indian courts
Summary and headline written by AI from the source article. How we work
Researchers created the first English-Mizo parallel dataset of Indian court judgments, drawing on 13,168 sentence pairs from the Gauhati High Court and the Supreme Court of India.
Professional translators rendered complete documents into Mizo, and reviewers verified the alignment of individual sentences. The team observed that legal and official terminology generally remained in English during translation, while common words were typically expressed in Mizo.
The new corpus revealed inconsistencies in the use of Mizo diacritics, marks added to letters to indicate pronunciation, across the translated texts. When used to refine the NLLB 200 1.3B model, the dataset improved BLEU scores, a measure of translation quality, by 7.16 points on a legal test set. However, the model still incorporated English words at a higher rate than human translators.
This work could broaden access to India’s legal system for speakers of Mizo, a language spoken by approximately 800,000 people in northeastern India and Myanmar.
