Neural_Machine_Translation

Best model metrics

Metric	Score
Corpus BLEU	37.0347
Dev ppl	61.4084

• You can find the model weights here

---

Try Translating yourself!

• The translation demo is available here on Streamlit Sharing.
• Even is you don't know Spanish you can use the demo as there is Google Translate which will help you to convert your English sentences to Spanish.

---

About NMT model

Hybrid Word-Character Seq2Seq Machine Translation

• It is a Seq2Seq Model that translates Spanish sentences into English based on Luong et al. 2015.
• It consists of a bidirectional LSTM encoder and unidirectional LSTM decoder.
• It also uses attention mechanism to boost its performance on the translation task.
• The pipeline and the implementations is inspired by the Open-NMT package.

• The model becomes more powerful as we combine character-level with word-level language modelling.
• The idea is that whenever the NMT model generates a <unk> token we run a character-level language model and generate a word in the output character by character.
• This hybrid word-character approach was proposed by Luong and Manning 2016 and turned out to be effective in increasing the performance of the NMT model (+1.2 BLEU).

---

Installation

Install from source:

git clone https://github.com/sahilkhose/Neural_Machine_Translation
cd Neural_Machine_Translation
pip3 install -r requirements.txt

To run the translation demo:

streamlit run stream_translate.py

Or just go here on Streamlit Sharing.

---

Contributing

If you find a bug, create a GitHub issue, or even better, submit a pull request. Similarly, if you have questions, simply post them as GitHub issues.