Neural Machine Translation (NMT) Portal

A Simple English-Chinese NMT System

I created a simple NMT system based on 150,000 English/Chinese pairs. It took 4 hours to complete the training (13 epochs). It's true that many translated sentences still required extensive post-editing, but certain sentences were well translated (see examples on the right) despite our short training time and small number of examples. The system "learnt" how to change the word order. I'll enhance the NMT system and share with you the latest findings.

NMT Samples

Useful Tools for NMT Pre-processing

Tokenisation and sentence alignments are crucial to Chinese/English NMT. We may consider well-known pre-processing tools such as Stanford Tokeniser. Meanwhile, I've developed a few pre-processing tools for NMT. Stay tuned for details.

Tool Wall

Introduction to NMT

Here is a very brief presentation on neural machine translation for my TRA3105 students.

nmt_steps.png