Word-based Domain Adaptation for Neural Machine Translation
This addresses the problem of limited in-domain data for e-commerce translation, offering incremental gains in translation quality.
The paper tackles domain adaptation for neural machine translation in e-commerce by applying word-level weights to leverage out-of-domain data, resulting in improvements of up to 2.11% BLEU and 1.59% TER compared to baselines.
In this paper, we empirically investigate applying word-level weights to adapt neural machine translation to e-commerce domains, where small e-commerce datasets and large out-of-domain datasets are available. In order to mine in-domain like words in the out-of-domain datasets, we compute word weights by using a domain-specific and a non-domain-specific language model followed by smoothing and binary quantization. The baseline model is trained on mixed in-domain and out-of-domain datasets. Experimental results on English to Chinese e-commerce domain translation show that compared to continuing training without word weights, it improves MT quality by up to 2.11% BLEU absolute and 1.59% TER. We have also trained models using fine-tuning on the in-domain data. Pre-training a model with word weights improves fine-tuning up to 1.24% BLEU absolute and 1.64% TER, respectively.