On March 19, 2020, the automatic translation service provided by German company DeepL became available in Japanese and has been gaining attention for its high accuracy. On June 16, 2020, the service was officially launched in Japan, further increasing its popularity.
At Human Science, we held a web seminar on June 9, 2020 with the theme "How is the quality of DeepL automatic translation service? - Comparison with competing services and introduction methods -" and had many participants.
At the seminar, we presented the accuracy verification results of DeepL and Google for each field such as medical and machinery.
Here, we will introduce the questions and answers received at the seminar.
We are constantly conducting seminars on machine translation.
Please check here for future event schedules.
Table of Contents
3. About MTrans Team, MTrans for Trados/Memsource
1. About DeepL
Q At DeepL, is it often the case that appropriate words are supplemented according to the context compared to other company's products, but are there also many cases where incorrect or incorrect words are supplemented?
A Not many, but there may be supplementary notes that could result in mistranslation.
Q What are some common errors seen in machine translation using neural networks other than DeepL, such as translation omissions or terminology errors, and what are some frequent errors specific to DeepL?
A When translating a paragraph containing multiple sentences, there may be cases where one sentence is completely missing in the translation.
Q Is there a difference in performance (translation speed, etc.)?
A DeepL takes longer than other translation engines. For example, if it takes 0.3 seconds to translate a sentence on Google, it may take 2 seconds on DeepL.
Q There was a press release stating that it is now possible to teach proprietary terminology for proper nouns in European and American languages on DeepL. Do you have any information on when it will be possible to learn proprietary terminology for Japanese ⇔ English?
A I don't understand. With MTrans for Trados/Memsource, you can use our unique terminology feature with DeepL.
Q What is the secret to high quality? Please tell me something other than paragraph-by-paragraph translation.
A Details are unknown, but DeepL provides the bilingual web search engine Linguee. I think it was trained based on the huge amount of translation data collected there.
Q Is it possible to contract DeepL individually from Japan?
A Service has started from June 16, 2020. Please visit the official website for more information.
Q Will the number of Vietnamese and other translation languages increase?
A When looking at the target languages for DeepL's bilingual web search engine Linguee, only Chinese and Japanese are available for Asian languages. Since DeepL seems to be using data from Linguee, it is unlikely that Vietnamese or other Asian languages will be added to DeepL.
Q Will the engine be trained even in the case of DeepL Pro?
A From the perspective of confidentiality, the data will not be reused as stated in the terms and conditions, so it will not be learned. From the perspective of whether the user can create their own engine, the learning function of the engine is not currently provided.
Q Is it generally true that Google excels in Bleu scores, but DeepL has an advantage in human evaluations?
A DeepL tends to translate in a more natural way, and although this may not be reflected in the Bleu score, it is often preferred by humans in certain industries and types of documents.
Q What are the future plans for language support?
A DeepL is believed to be using data from Linguee, so it is expected that Linguee's target languages (mainly European) will be added in the future.
Q Have you considered comparing Japanese and Chinese at your company? Do you have any plans to do so?
A We will consider a comparison if you have any requests. Alternatively, you can try it out for yourself with a trial.
Q Is it safe to assume that you are using machine learning for paragraph-level translation?
A The details of DeepL's algorithm are unknown.
Q Is there a difference in translation process for file formats containing tags such as HTML and XML compared to DeepL and other machine translation engines?
A Although it supports tags, there is a slight impression that the accuracy is inferior compared to other engines. It is recommended to translate without tags as there is a possibility that the translation quality may decrease if the text contains tags.
Q Have you compared the translation accuracy of IT and electronic engineering using four engines (DeepL, Google, Microsoft, Amazon)?
A In this seminar, we compare business emails, IT, medical, machinery, contracts, and patents.
Q Other than the quality score and mechanism differences compared to other Machine Learning-based machine translation systems (such as T-4OO), please let us know if you are aware of any.
A I don't understand.
Q In the verification results, there is a significant difference in the Bleu Score between machine Japanese-English and English-Japanese. Why is this?
A There may be a lack of machine-related documents in Japanese, and there may not be enough training.
Q GoogleNMT and other leading NMT providers have reported a decrease in accuracy compared to when our service was launched, as heard at last year's translation festival. We suspect that the expansion of users may be a contributing factor to this decline. How is DeepL attempting to address this issue?
A Some sentences may have decreased accuracy, but overall it seems that accuracy is improving year by year. The method for avoiding this issue is unknown, but it is believed that usage from Japan has significantly increased since DeepL added support for Japanese in March. However, the accuracy has not decreased at this time.
2. About MTrans for Office
Q Is there a limitation on the version of Office?
A Windows desktop version of Office 2013/2016/2019, Microsoft 365 is supported.
Q Is there a limit on the number of documents (file translations) and character data that can be translated per month?
A Not available.
Q Is there a limit to the amount of text that can be translated at once? Is there a limit to the amount of text that can be translated?
A Not available.
Q Is it possible to use all engines, including DeepL, Google, Microsoft, and Amazon, with MTrans for Office?
A Yes.
Q Can I register a glossary and example sentence collection? Is it possible to create a TM? What kind of management can be done for them? Is there a limit on the number of registrations? (Can categories be created?)
A MTrans for Office does not support this. By using the automatic string replacement function on the translation side, you can achieve a similar process to the terminology function. For MTrans for Memsource/Trados, the terminology function is available. Please use the function of the CAT tool for example sentences and TMs.
Q If you have already introduced MTrans Team, is there a discount?
A We do not offer discounts, but please feel free to contact us for consultation.
3. MTrans Team, MTrans for Trados/ Memsource
Q Please remind me of the price when using DeepL with SDL Trados Studio.
A Initial cost is 100,000 yen + monthly fee of 30,000 yen.
Q MTrans for Memsource, Trados - Do I need a separate contract for Memsource or Trados?
A Yes. It is also possible to contract with our company regarding Memsource.
Q Is DeepL supported on the browser version of MTrans Team?
A Not available. Currently under consideration for support.
Q Is security guaranteed with MTrans for Office? Will the text entered be saved even when using Google?
A MTrans and DeepL combined provide secure usage. Text is not saved or reused by Google or Microsoft. Data may be reused for service improvement by Amazon. It is possible to remove Amazon from the engine selection list for MTrans for Office.
Next Seminar
We are constantly conducting seminars on machine translation.
Please check here for future event schedules.
Human Science's DeepL Solution
MTrans for Office
https://www.science.co.jp/nmt/service/mtrans_office.html
If you are already using translation support (CAT) tools such as Memsource or Trados, you can immediately use "DeepL Pro" by contracting our MTrans for Memsource or MTrans for Trados.
MTrans for Memsource
https://www.science.co.jp/nmt/service/memsource.html
MTrans for Trados
https://www.science.co.jp/nmt/service/nmt.html
DeepL Related Blog
How to introduce the DeepL automatic translation service into your business?
Is confidentiality maintained with DeepL translation? What about security?
Download here
Return to the top of the page
+81-3-5321-3111[Tokyo]
Contact Us