MNLP-HW2

Translation from Non-Modern Italian to Modern Italian

The Task

Develop a translation model from ancient to modern italian, evaluate it through a qualitative analysis, and use a LLM as a judge following this evaluation grid:

1. Completely unacceptable translation: the translation has no pertinence with the original meaning, the generated sentence is either gibberish or something that makes no sense
2. Severe semantic errors, omissions or substantial add ons on the original sentence. The errors are of semantic and syntactic nature. It’s still something no human would ever write
3. Partially wrong translation, the translation is lackluster, it contains errors, but are mostly minor errors, like typos, or small semantic errors.
4. Good translation. The translation is mostly right, substantially faithful to the original text, but the style does not perfectly match the original sentence, still fluent and comprehensible, and could semantically acceptable
5. Perfect translation. The translation is accurate, fluent, complete and coherent. It retained the original meaning as much as it could.

Project Links

• Google Drive
• Notion Page
• Report

Useful Links

Datasets

• Login Hugging-Face

Hugging-Face

• Transformers
• PyTorch Based

Wandb

• Wandb Project