The question behind the work
A translation can look fluent while changing the meaning. These experiments explore translation generation and the behavior of reference-based evaluation across languages.
LANGUAGE / TRANSLATION
Translate. Then question the score.
Translation and evaluation experiments across English and Indian languages using FLORES-200, IndicTRANS2 and LLMs.
THE PROCESSINPUT → UNDERSTANDING
01Sample
Select parallel examples from FLORES-200.
02Translate
Generate translations with a selected model.
03Score
Compare output with reference translations.
04Analyze
Export results and inspect language directions.
THE PROJECT
Machine Translation & Evaluation
Collection of related implementations
How it comes together
The collection uses IndicTRANS2 and LLM variants with FLORES-200. BLEU with smoothing provides a reference-based score; it is distinct from an LLM judgment. Pandas and exported results support further analysis.
What to keep in mind
Model variants cover different language directions. FLORES-200 is an existing dataset, credited to its original creators; these projects use it for experiments.
Language changes.
The question remains.
A small, prewritten language illustration. It does not run the project’s models or report experimental results.
ENGLISH SOURCE
TRANSLATION
Illustrative examples · inspect the notebooks for generated outputs and evaluation.