Urdu poems to Shakespearean English: Machine Translation
summary
Claims an automated pipeline that converts Urdu poetry into Shakespearean English by chaining models: Urdu→Modern English via Helsinki trained on ~14,371 Bible sentence pairs, then Modern English→Shakespeare via fine-tuned GPT-2 trained on ~16,000 lines from 17 plays. Uses seq2seq Transformers and tokenizers, yields a demo and generated lines that read stylistically Shakespearean but score modestly (BLEU ≈ 13 versus 23.2 baseline), and hits limits from sparse poetry parallel data, transliteration issues, and compute/memory constraints. Models and code are available on GitHub/Google Drive.
AI-written summary · about
chapters (9)
- 0:00Machine translation intro
- 0:51Motivation for translation
- 2:38Challenges in translation
- 3:40Data sources and training
- 5:07Fine-tuning translation models
- 8:26Live translation demo
- 11:54Model limitations and evaluation
- 13:00Future work opportunities
- 15:47Q&A
related talks
24mPyambic Pentameter: generating rhyming and metered poems
33mTranslation Augmented Generation - Breaking Language Barriers in RAG...
38mMachine Learning for Poets: The eMiLy App
43mPoetry (almost) From Scratch - Teaching versification to a neural network
56mValidation by Listening
45mTranscend Language Barriers with Custom Machine Learning APIs: um...
29mData driven literary analysis
20mMeta's AI Translations Inference