Books & Rights Marketplace 
日本語  |  ENGLISH 
Welcome Question?
Privacy Policy | Terms & Conditions  
First time user?
Please register online today !  
Register Now! 
MemberLog-in: Password: Login
Affiliated Partners
Author(Surname by initial)
This week's Top 10 Ranking 
1.  LOOPS ― The Seven ...
2.  The spiritual way ...

Become a fan of BRM

Promote Your Page Too
News & Events
2018.11.02 Researchers at the University of Zurich: Machine translation has yet to match humans
This spring, Microsoft announced that AI translation had managed to match the same level of accuracy as human translation. However as the results of this research only reflect results of translations of news articles, there are still many challenges that remain unaddressed with regard to its full-scale application. One of the challenges as pointed out by Samuel Laubli of Zurich University was that the quality of machine translation remains inferior to translations performed by a human.

In the research and development of AI translation recently, neural network learning has been adopted, and the accuracy of machine translation has dramatically improved. The current machine translation evaluation method is a method in which a professional translator evaluates the appropriateness of a translated sentence and an evaluator who can only understand English evaluates fluency. Just looking at the result it would seem that the accuracy of machine translation has reached an area comparable to human translation. However, this method merely compares at the sentence level, and lacks evaluation on the entire document.

Mr. Laubli asked professional translators to evaluate the appropriateness and fluency of the entire document in addition to comparisons made sentence by sentence. The result showed that machine translation and translation by humans shared a similar level of grammatical accuracy, but that in regards to the overall quality of the document, translations performed by humans were still overwhelmingly better. Mr. Laubli assumes that translation errors that appear in a sentence are more easily exposed when looking at the document as a whole. This is because the translation of more ambiguous terms or proper nouns may not be standardized throughout the document.

When a person reads a sentence, it understands the contents according to the context before and after. Therefore, in some situations merely replacing word by word may not convey the actual message to the reader. Laubli points out that in order to improve the accuracy of machine translation in the future, not only sentence comparison, but also evaluation at the document level is necessary.

For details, please visit the following website.