A Pytorch-based Seq2Seq model with Attention translating English to Bahasa Indonesian
The intention of this project is academic, for us to learn about Seq2Seq as well as how attention works by deep-diving to the PyTorch encoder/decoder and attention implementations.
Bahasa Indonesia, or Bahasa is the main language of Indonesia which is the largest economy in South East Asia. Hence, there is a demand for translation from English to Indonesian for various business needs. In this work we evaluate two different neural machine translation models on a simple English to Indonesian text corpus. We compare our approaches using both the Bilingual Evaluation Understudy Score (BLEU score) and a Turing test. We see that the attention-based sequence to sequence (Seq2Seq) translation model performs better than the vanilla Seq2Seq model on both the metrics.
Data source - Manythings_anki (Tatoeba) https://www.manythings.org/anki/ind-eng.zip.
Full report here : https://github.com/lppier/Seq2Seq_Eng2Indo-Translation/blob/master/Seq2Seq_Eng2Indo_Report.pdf