Neural machine translation: Challenges, progress and future
- PDF / 1,241,829 Bytes
- 23 Pages / 612 x 792 pts (letter) Page_size
- 38 Downloads / 268 Views
Print-CrossMark
October 2020 Vol. 63 No. 10: 2028–2050 https://doi.org/10.1007/s11431-020-1632-x
Special Topic: Natural Language Processing Technology
. Review .
Neural machine translation: Challenges, progress and future ZHANG JiaJun1,2* & ZONG ChengQing1,2,3* 1 National
Laboratory of Pattern Recognition, CASIA, Beijing 100190, China; of Chinese Academy of Sciences, Beijing 100190, China; 3 CAS Center for Excellence in Brain Science and Intelligence Technology, Shanghai 200031, China 2 University
Received March 10, 2020; accepted May 9, 2020; published online September 15, 2020
Machine translation (MT) is a technique that leverages computers to translate human languages automatically. Nowadays, neural machine translation (NMT) which models direct mapping between source and target languages with deep neural networks has achieved a big breakthrough in translation performance and become the de facto paradigm of MT. This article makes a review of NMT framework, discusses the challenges in NMT, introduces some exciting recent progresses and finally looks forward to some potential future research trends. neural machine translation, Transformer, multimodal translation, low-resource translation, document translation Citation:
Zhang J J, Zong C Q. Neural machine translation: https://doi.org/10.1007/s11431-020-1632-x
Challenges, progress and future.
1 Introduction The concept of machine translation (MT) was formally proposed in 1949 by Weaver [1] who believed it is possible to use modern computers to automatically translate human languages. From then on, machine translation has become one of the most challenging task in the area of natural language processing and artificial intelligence. Many researchers of several generations dedicated themselves to realize the dream of machine translation. From the viewpoint of methodology, approaches to MT mainly fall into two categories: rule-based method and datadriven approach. Rule-based methods were dominant and preferable before 2000s. In this kind of methods, bilingual linguistic experts are responsible to design specific rules for source language analysis, source-to-target language transformation and target language generation. Since it is very subjective and labor intensive, rule-based systems are difficult to be scalable and they are fragile when rules cannot cover the *Corresponding authors (email: [email protected]; [email protected])
Sci China Tech Sci, 2020, 63:
2028–2050,
unseen language phenomena. In contrast, the data-driven approach aims at teaching computers to learn how to translate from lots of human-translated parallel sentence pairs (parallel corpus). The study of datadriven approach has experienced three periods. In the middle of 1980s, ref. [2] proposed example-based MT which translates a sentence by retrieving the similar examples in the human-translated sentence pairs. From early 1990s, statistical machine translation (SMT) has been proposed and word or phrase level translation rules can be automatically learned from parallel corpora using pro
Data Loading...
