Word-Character Hybrid Machine Translation Model
Mani Bansal, Daya Krishan Lobiyal · 2020
In this paper, we propose a method which combines word and character attention information to improve neural machine translation. For this, two attention mechanisms are used. Firstly, with the help of Gated Recurrent Unit character-level attention has taken input as sequence of characters. After that, word-level pays attention to the words composed from character-level attention. With the fusion of two attentions, encoder encodes information simultaneously. On the other side, decoder decodes the information based on word-level only. The experimental results proved the proposed method outperformed the baseline system for English-Hindi language pair.