Convolutional Sequence to Sequence Learning for English-Khasi Neural Machine Translation
Aiusha Vellintihun Hujon, Khwairakpam Amitab, Thoudam Doren Singh · 2023
The convolutional neural network is relatively prevalent for visual recognition tasks. Implementing CNN for neural machine translation is still challenging compared to the predominant approach such as recurrent neural networks and transformers. This paper presents an empirical study of the convolutional sequence to sequence learning model to translate English to Khasi. The approach uses convolution for both encoders and decoders with multi-step attention. The study also includes a model using a transfer learning approach on which the parent model is trained using the English-French dataset. The quantitative and qualitative results are significant thus suggesting that the convolutional approach for neural machine translation tasks can perform remarkably well as other tasks using convolution neural networks in the case of English-Khasi language pair.