Neural Machine Translation NMT is a machine translation framework that uses pure neural networks to translate from source language to target language in an end-to-end manner.

NMT is consistent with the idea of statistical machine translation. Its essence is to solve probability problems, but the implementation method is different. The latter relies on a large number of data models, while NMT does not.

Problems facing NMT development

• Limited vocabulary
• Source language translation coverage
• Inaccurate translation

NMT development trend

• Rare word problem
• Single language data usage
• Multilingual Translation/Multilingual NMT
• Memory mechanism
• Language Fusion
• Coverage Issues
• Training process
• Prior knowledge integration
• Multimodal Translation