Best machine translation company for Japanese Thread poster: ichibanmike
|
|---|
Hi,
We are trying to determine which machine translation (MT) provider (a company that actually develops their own MT engine as opposed to repackaging and selling some other company's MT engine) provides the best results for Japanese to English translation.
Does anyone have any experience in this?
Feel free to email me privately if you'd prefer that instead of replying directly to this topic.
Regards,
Mike Holland
| | | | German Legal
Canada
Local time: 23:12
English to German
+ ...
| Best machine translation system for German-English | Aug 4, 2012 |
The same question as the previous poster (about Japanese): is there a system that handles the German language particularly well? The standard and well known systems out there don't, they are much better with Roman languages, but German still seems to be a challenge.
Thank you!
Inge
German Legal Translations in Canada
[Edited at 2012-08-04 13:09 GMT]
| | | | Nicole Schnell 
United States
Local time: 19:12
English to German
+ ...
In memoriam | Which is the best thing ever. :-) | Aug 5, 2012 |
German Legal wrote:
but German still seems to be a challenge.
Because of the reversed grammar, because German adds a gender to every noun, because of the highly elegant extended attribute phrases in educated German and so much more, you will get nothing but gibberish from any MT. Never. Ever.
Nice! 
| | | | | There's no such thing | Aug 5, 2012 |
ichibanmike wrote:
Hi,
We are trying to determine which machine translation (MT) provider (a company that actually develops their own MT engine as opposed to repackaging and selling some other company's MT engine) provides the best results for Japanese to English translation.
Does anyone have any experience in this?
Feel free to email me privately if you'd prefer that instead of replying directly to this topic.
Regards,
Mike Holland
There's no such thing as best results for Japanese-English, since the word order is totally different. Why don't you try placing a few sentences in different machine translation tools and compare the results? I bet they all come out the same way. This is same for English-Japanese due to the reason above.
P.S.
@Nicole
And the separable verbs are also a challenge to machine translation too, aren't they?
[Edited at 2012-08-05 06:20 GMT]
| | |
|
|
|
Nicole Schnell
United States
Local time: 19:12
English to German
+ ...
In memoriam
Yasutomo Kanazawa wrote: @Nicole And the separable verbs are also a challenge to machine translation too, aren't they?
| | | | Heinrich Pesch
Finland
Local time: 05:12
Member (2003)
Finnish to German
+ ...
Nicole Schnell wrote: German Legal wrote:
but German still seems to be a challenge.
Because of the reversed grammar, because German adds a gender to every noun, because of the highly elegant extended attribute phrases in educated German and so much more, you will get nothing but gibberish from any MT. Never. Ever. Nice!
I currently translate for my own purpose an English book (about history) from 1917 into German, using Google translate. It is true that German grammar represents a challenge for MT, but quite often the output is astonishingly good. I had even sentences of more than 10 words which I was unable to improve. And the source text is rather non-standard and individual.
And I notice the system is learning on the fly. When it first thought barracks were Baracke, soon it learned that it should be Kaserne. But up to now I haven't make it understand that Mr. Yarrow is not Herr Schafgarbe
According to my experience some simple rules like "a title followed by an upper case word should not be translated" would improve MT immensely. Don't know why nobody seems to implement them.
Quite often though you get an output with most verbs at the wrong place, but I noticed with research papers that MT really is a valuable tool. It knows much more than any dictionary.
| | | | | German-English machine translation | Aug 5, 2012 |
German Legal wrote:
The same question as the previous poster (about Japanese): is there a system that handles the German language particularly well? The standard and well known systems out there don't, they are much better with Roman languages, but German still seems to be a challenge.
It is very difficult to compare MT system. And the results my vary a lot depending of the domain of texts. And also complexness of the texts. Rule-based MTs will usually be better with simpler sentences, and statistical MTs will do better with more complex sentences. But on the other hand statistical MTs may alter tne meaning in some strange way. One characteristic of statistical MTs is that they quite frequently lose negation in translation and so reverse the meaning. "no" is just a word like all others for statistics.
So the best way to find the best MT is to test them on real texts you want to translate. You will able to quickly eliminate the worst ones and conduct some more test on the best and so you will know what you can expect from MT.
One short test:
Original text:
Der Ausbau der Energieerzeugung durch Photovoltaikanlagen wird Deutschlands Stromkunden mindestens 111 Milliarden Euro kosten. Das ergeben neue Berechnungen des Rheinisch-Westfälischen Instituts für Wirtschaftsforschung (RWI), die der F.A.Z. vorliegen. Allein der Ausbau bis zu dem neuen Förderdeckel von 52 Gigawatt (GW) Leistung wird trotz stark sinkender Einspeisevergütung noch Zahlungen von 11,8 Milliarden Euro auslösen.
Google
The expansion of energy production by photovoltaic systems will cost German electricity consumers at least 111 billion euros. The resulting new calculations of the Rhenish-Westphalian Institute for Economic Research (RWI), the FAZ . available But the expansion up to the new funding cap of 52 gigawatts (GW) of power will cause strong despite declining feed-in tariff payments of 11.8 billion euros yet.
Bing
The expansion of the energy by photovoltaic systems will cost at least EUR 111 billion Germany of customers. New calculations of Rhine-Westphalian Institute for economic research (RWI), be provided in the F.A.Z. shown that. Alone the expansion until performance is on the cover of the new funding of 52 gigawatts (GW) despite sharply declining feed-in remuneration still trigger payments of €11.8 billion.
Trident
Development of power production through Photovoltaikanlagen will on a less measure cost Germany of stream client a 111 milliard of euro. The New calculations of Rheinisch-Westfälischen of institute make for research(RWI) of economy, F.A.Z. exist. Development to the new haulage deckle of 52 Gigawatt(GW) of work will only yet cause in spite of very pulldown Einspeisevergütung payments of 11,8 milliards of euro.
Prompt
The removal of the energy production by Photovoltaikanlagen will cost stream customers of Germany to at least 111 milliard euros. New calculations of the Rhenish-Westphalian institute of economic research (RWI) , him prove this FOLLOWING A. Z. be given. Only the removal up to the new conveyor lid of 52 gigawatt (GW) achievement will release in spite of very sinking Einspeisevergütung still payments of 11.8 milliard euros.
Systran
The development of the energy production by photovoltaic systems will cost of Germany current customers at least 111 billion euros. New calculations of the Rhenish-Westphalian institute for economic research (RWI) show, which the F.A.Z is present. However the development up to the new promotion cover of 52 Gigawatt achievement will release despite strongly sinking compensation for electricity fed into the grid still payments of 11.8 billion euros.
Skycode
The extension of Energieerzeugung across voltaic systems of picture will announce Federal Republic of Germany current at least to 111 costses of billions euros. This give newcomer accounts of Rheinisch-Westfälisch in institutes for economy research (RWI), these of f. a. s. lies vor. Of the extension until to new Förderdeckel Von to 52 Gigawatt (GW) accomplishment the contempt becomes strong a payments yet of speisevergütung won falls 11, 8 billions euros release alone.
Lingenio
The development of the energy generation by photovoltaic plants will cost Germany's current customers at least 111 billion euros. New calculations of the Rhenish-Westphalian institute for economy research (RWI) which are with the F.A.Z. yield this. Alone the development up to the new promoting lid of 52 gigawatts of (GW) performance still will trigger payments of 11.8 billion euros despite one dish compensation dropping strongly.
Linguatec
The development of the energy generation by Photovoltaikanlagen will cost Germany's current customers at least 111 billion euros. New calculations of the Rhenish Westphalian institute for commercial research (RWI) which are with the F. A. Z. yield this. Alone the development up to the new promoting lid of 52 gigawatts of (GW) performance still will trigger payments of 11.8 billion euros despite one dish compensation dropping strongly.
| | | | Jeff Allen
France
Local time: 04:12
Multiplelanguages
+ ...
| contact Jeff about MT | Aug 5, 2012 |
Mike,
You can contact me by message about MT for all languages. Just a glance at my profile here on ProZ will make sense why I know about many MT systems for a wide variety of languages.
Jeff
| | |
|
|
|
German Legal
Canada
Local time: 23:12
English to German
+ ...
| German-English MT - reply to Peter Holozan's post - legal translations with MT? | Aug 5, 2012 |
Thank you very much for running the test with the various systems, Peter! Quite telling!
I am in the legal translation business and bought Systran Premium Translator two years ago. I can't use any cloud system for confidentiality reasons, and need to buy a license for my own system, which is fine with me. Systran, however, is really not usable for any of my work. We tried it again for large briefs (written pleadings) from German into English, but the sentences in German are too conv... See more Thank you very much for running the test with the various systems, Peter! Quite telling!
I am in the legal translation business and bought Systran Premium Translator two years ago. I can't use any cloud system for confidentiality reasons, and need to buy a license for my own system, which is fine with me. Systran, however, is really not usable for any of my work. We tried it again for large briefs (written pleadings) from German into English, but the sentences in German are too convoluted and even if I sat down for several months and fed 'the beast' with all kinds of terminology, I am quite convinced that the sentences would still be so off that post-editing would take longer than translating - at least this type of text - from scratch.
I heard from another colleague that Romance languages have been Systran's strongpoint since it evolved from a purely rule-based system into (what the developers claim) is a hybrid system. He also said that the German-English language pair is just so difficult for developers that very few systems have even started approaching it anywhere near effectively.
And there is the need to train the system on high-quality bilingual corpora. That, I'm ready to do if the outcome is usable with only a few changes, but not with what I get right now.
I am asking the specialists: Do you think that legal texts, such as contracts (moderately long sentence, complex subjects), legal opinions and briefs with run-on sentences and very complex topics are *at all* good candidates for MT?
What we did with the pleadings I mentioned above was to summarize them, so that was quick and less expensive, exactly what the client needed. But they were drafted by a 'human brain', were idiomatic and read well, which cannot be said of the samples I had run with this material on Systran, or some of Peter's samples.
Again, I would be happy to hear more about specifically this language pair (DE into EN) from the specialists. Thank you so much!
Inge Noeninger, certified translator, Canada ▲ Collapse
| | | | | Customization | Aug 6, 2012 |
The best results are obtained by customization and you can get a sense of the options in the market based on a detailed independent Enterprise evaluation of the issues/difficulties of doing this here:
Implementing large scale Machine Translation in Patent Information by Andrew Ruffner CTO Lexis Nexis on comparison of Japanese Patent M... See more The best results are obtained by customization and you can get a sense of the options in the market based on a detailed independent Enterprise evaluation of the issues/difficulties of doing this here:
Implementing large scale Machine Translation in Patent Information by Andrew Ruffner CTO Lexis Nexis on comparison of Japanese Patent MT systems http://dotsub.com/view/159ce97c-dbd4-4d6a-90c2-427a3a3e755f ▲ Collapse
| | | | | Why not responding to the forum? | Aug 7, 2012 |
Jeff Allen wrote:
You can contact me by message about MT for all languages. Just a glance at my profile here on ProZ will make sense why I know about many MT systems for a wide variety of languages.
Why don't you share your wisdom with the rest of us? I thought this was the aim of using such forums.
| | | | | Advertisement rule, perhaps? | Aug 7, 2012 |
Dominique Pivard wrote: Jeff Allen wrote:
You can contact me by message about MT for all languages. Just a glance at my profile here on ProZ will make sense why I know about many MT systems for a wide variety of languages.
Why don't you share your wisdom with the rest of us? I thought this was the aim of using such forums.
Perhaps because he is affiliated with some of the MT companies, and it may be seen as advertisement, which is prohibited by the forum rules. I think he is just being considerate, that's all.
| | |
|
|
|
Jeff Allen
France
Local time: 04:12
Multiplelanguages
+ ...
| not advertising related, but just sparing irrevelant info for freelancer readers | Aug 7, 2012 |
Jeff Allen wrote:
You can contact me by message about MT for all languages. Just a glance at my profile here on ProZ will make sense why I know about many MT systems for a wide variety of languages.
Dominique Pivard wrote:
Why don't you share your wisdom with the rest of us? I thought this was the aim of using such forums.
Dominique, I regularly share the wisdom on MT here in the forums. You can use the advanced search function on ProZ and look up the word "japanese" or the word "asian" for the user "mtpostediting" and see all that I've already written on the topic in the various ProZ forums.
However, Mike's initial post which starts with "we are...." and then follows with "machine translation (MT) provider (a company that actually develops their own MT engine as opposed to repackaging and selling some other company's MT engine)" is clearly the case not of an individual single freelancer trying to decide which best software package to invest in. When people know the difference between system source providers and repackagers, then this is almost always someone looking to cross-evaluate enterprise level installations.
And when you begin looking at that level, there are many factors to consider, which would bore the majority of freelance readers in this forum. This is why during the annual Virtual Conferences there have been different sessions about MT for freelancers, for translation agencies, and for enterprise level system. It is much more than just comparing raw output of online non-configured, non-trained systems.
The needs to consider are quite different, and many times when I have responded in the forums to points about enterprise level systems, quite a few freelancers overtly say that they are not interested in that.
As Mike said in his post that he could continue the conversation outside the forum, and I know that the answer can be very dependent on his specific context which also might include info that he likely wouldn't want to share publicly either, it seemed appropriate to spare the irrelevant post bandwidth for most people reading this forum.
Katalin Horváth McClure wrote:
Perhaps because he is affiliated with some of the MT companies, and it may be seen as advertisement, which is prohibited by the forum rules. I think he is just being considerate, that's all.
Rather "used to be affiliated with". The last one I worked for was back in 2007. But yes, over the past 15+ years I have worked for several companies and institutes that develop a range of MT systems types for many different languages, and also previously worked as a market research analyst on natural language processing systems and language datasets to help the European Commission plan for future co-funding initiatives in the field.
I do not directly work in the field any longer but rather for a company that produces other types of software.
Jeff
[Edited at 2012-08-07 20:17 GMT]
| | | | Jeff Allen
France
Local time: 04:12
Multiplelanguages
+ ...
| questions on DE/EN MT by Inge | Aug 7, 2012 |
German Legal wrote:
Thank you very much for running the test with the various systems, Peter! Quite telling!
I am in the legal translation business and bought Systran Premium Translator two years ago. I can't use any cloud system for confidentiality reasons, and need to buy a license for my own system, which is fine with me. Systran, however, is really not usable for any of my work. We tried it again for large briefs (written pleadings) from German into English, but the sentences in German are too convoluted and even if I sat down for several months and fed 'the beast' with all kinds of terminology, I am quite convinced that the sentences would still be so off that post-editing would take longer than translating - at least this type of text - from scratch.
.....
Again, I would be happy to hear more about specifically this language pair (DE into EN) from the specialists. Thank you so much!
Inge Noeninger, certified translator, Canada
Hi Inge,
As your enquiry is more focused on a freelancer perpective (which many others will identify with) and since your questions are those that I have already written about in the forums, it is fine to answer your questions here.
I'm currently on vacation with sporadic and limited internet access each day and quite a few other activities going on, so I'll take some time to answer your specific questions. I would like to look up a few previous posts on ProZ where I've already written about this, but need to log off now to put the kids to sleep.
Looking up previous thread comments simply optimizes my time in answering such questions.
Kirti's statement about "customization" is right on, and it applies at all levels of use (from freelancer to enterprise systems) and also for different MT system types.
Jeff
| | | | ichibanmike
Local time: 19:12
Japanese to English
TOPIC STARTER | Best machine translation | Aug 7, 2012 |
Hi,
Thanks, everyone, for your posts. It seems that some are interested in integrating MT into their processes (free-lance processes included).
I don't know about German, but off-the-shelf (or untrained) MT systems are pretty much worthless for Japanese-to-English translation, unless it is for the simplest and shortest of sentences.
I think it is worthwhile for freelance translators to continually sample different products out there, because the technology... See more Hi,
Thanks, everyone, for your posts. It seems that some are interested in integrating MT into their processes (free-lance processes included).
I don't know about German, but off-the-shelf (or untrained) MT systems are pretty much worthless for Japanese-to-English translation, unless it is for the simplest and shortest of sentences.
I think it is worthwhile for freelance translators to continually sample different products out there, because the technology is constantly evolving, and see if MT can be integrated into a part of the workflow, perhaps for simpler stuff or translation of short phrases. If nothing else, it might be fun.
Thanks,
Mike ▲ Collapse
| | | | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » Best machine translation company for Japanese | Trados Studio 2022 Freelance | The leading translation software used by over 270,000 translators.
Designed with your feedback in mind, Trados Studio 2022 delivers an unrivalled, powerful desktop
and cloud solution, empowering you to work in the most efficient and cost-effective way.
More info » |
| | Anycount & Translation Office 3000 | Translation Office 3000
Translation Office 3000 is an advanced accounting tool for freelance translators and small agencies. TO3000 easily and seamlessly integrates with the business life of professional freelance translators.
More info » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
