Tamil NLP
Annotation guidelines, evaluation metrics, and a management portal for Tamil translation and transliteration.
.png)
Low-resource language work lives or dies on annotation quality. Before you can evaluate a Tamil translation system, you need agreement on what a correct output even looks like — and that agreement has to be written down precisely enough that different annotators reach the same verdict.
This project covered that full loop: drafting the guidelines, wiring up the evaluation, and building the interface that made the results legible to the people running the work.
-
Annotation guidelines
Created annotation guidelines for Tamil translation and transliteration, drawing on a range of existing sources to define consistent judgments.
-
Evaluation metrics
Through API integration and data analysis, computed metrics including BLEU score and exact match to quantify system output against the annotated references.
-
Portal interface
Developed a user-friendly interface for the Tamil NLP portal, enabling effective management of runs and viewing of results.