CodeMix Generation – NLP Project

English ↔ CodeMix (English-Hindi) translation using Seq2Seq neural models with attention. Explored the impact of Language Identification (LID) tags on translation performance.

---

Key Features

• Seq2Seq model with attention mechanism.

• Translation directions:

  • English → CodeMix
  • CodeMix → English

• Experimented with and without LID tags.

• Evaluated using perplexity and BLEU scores.

---

Results (Highlights)

• Without LID tags: BLEU ≈ 0.03 (English→CodeMix), 0.02 (CodeMix→English)
• With LID tags: BLEU ≈ 0.06 (English→CodeMix), 0.04 (CodeMix→English)

---

Files

• Notebooks for all translation directions (with/without LID tags)
• Final Report

---

Technologies

Python • PyTorch • Spacy • TorchText

---

Authors

• Vishal Patel
• Pawan Patidar