| Pages in topic: < [1 2 3 4 5 6 7] > | The "Crime" of Using MT Thread poster: Michelangela
|
|---|
| Linguee.de: Is the problem really MT (auto)cannibalism? | Jun 18, 2014 |
I don't know how many people here are familiar with linguee.de, and I don't know how many languages it covers, but it searches the Internet for previously translated texts containing words or combinations of words entered by the user. I have heard before that the results for translations out of English are often better, but I know that for German>English the results are generally absurdly bad.
I mention this, because the problem of Google search engines being fed by Google Translate... See more I don't know how many people here are familiar with linguee.de, and I don't know how many languages it covers, but it searches the Internet for previously translated texts containing words or combinations of words entered by the user. I have heard before that the results for translations out of English are often better, but I know that for German>English the results are generally absurdly bad.
I mention this, because the problem of Google search engines being fed by Google Translate results is a soluble problem (and I think also a problem that has already been solved by inserting non-displayed identifiers into the translated text): The problem that the vast majority of human translations in the Internet are garbage is not so easily solved. ▲ Collapse
| | | | Phil Hand 
China
Local time: 14:14
Chinese to English
| Thank you, Samuel | Jun 18, 2014 |
First: Orrin
I think that we should all take a minute to remember that comparing MT for different language pairs, and to a somewhat lesser extent to different fields, doesn't really work.
Absolutely. My thing is that I've literally never seen MT that I would want to work from, and I'm curious to know why that is, when other people do seem to benefit from it in some fields. So if I can see where and how others are benefiting, I might learn something.
Michael
I think that Microsoft makes very extensive use of raw MT (maybe including minimal user/crowd post-editing) in its online help. I have a German version of Word, so I think that a lot of the results I get are the results of MT and these are are generally helpful. Anyone interested can take a look at Microsoft help in their non-English language or languages and form an opinion for themselves.
The Chinese version drives me up the bleeding wall. Deeply unintuitive writing, which really doesn't help a flustered IT user. Sometimes I can get the information I want out of it, though, so that's better than nothing. A good example.
Samuel:
Thanks, that's a great example. Personally, I still wouldn't want to edit that, but I have realised that part of the reason is my own reading limitations. I am still not as fluent a reader of Chinese as I am of English (and probably never will be), so if I am presented with a Chinese segment and an *English segment, my eyes automatically privilege the *English. It takes extra effort for me to go back and read the source text, then compare the meaning.
I still think that just typing the sentences would be simpler than making that many edits - it's an average of five per sentence or something. But this exercise has helped me to see part of the issue. Very instructive, I appreciate you do running that all through for us.
| | | | Samuel Murray
Netherlands
Local time: 08:14
Member (2006)
English to Afrikaans
+ ...
Phil Hand wrote:
Personally, I still wouldn't want to edit that ...
If a client sent me that quality of work and told me that it was a human who did it, I would write back and tell him that the translation is a disaster, is unsalvageable, and must be done from scratch. But if I know that it is a machine translation, then I'm willing to edit it, because the machine makes more predictable mistakes and does not interpret the text (potentially incorrectly) as a human would.
| | | | Tim Drayton
Cyprus
Local time: 09:14
Turkish to English
+ ...
| So much depends on the language pair | Jun 18, 2014 |
I can see from the examples quoted by Samuel how this could work in Afrikaans>English.
However, as Orrin rightly says, so much depends on the language pair and the extent to which the languages are related and have cognate syntactic structures. The results simply are not so good with totally unrelated languages as, for example, Turkish and English. If I enter a portion of a newspaper article, hardly the most demanding kind of text, in Turkish that I myself recently translated as:... See more I can see from the examples quoted by Samuel how this could work in Afrikaans>English.
However, as Orrin rightly says, so much depends on the language pair and the extent to which the languages are related and have cognate syntactic structures. The results simply are not so good with totally unrelated languages as, for example, Turkish and English. If I enter a portion of a newspaper article, hardly the most demanding kind of text, in Turkish that I myself recently translated as:
"Ankara University Law Faculty member Prof. Dr. Hasan İşgüzar has said that, with regard to penal, civil and administrative law, the call for the dean’s resignation was totally without foundation in statute or law. İşgüzar, pointing out that according to both the Universal Declaration of Human Rights and the Constitution, crime and punishment is personal, said, “There is no legality to calling on the dean to resign. According to both the Universal Declaration of Human Rights and the Constitution, crime and punishment is ‘personal’. Nobody can be punished for a crime alleged to have been committed by a relative. It is inconceivable for a spouse, child, mother or father to be held responsible. There is no such arrangement or provision in either Council of Higher Education (YÖK) legislation or penal law.”
into Google translate, it comes up with:
"Ankara University Faculty of Law, Prof. Hasan meddlesome, criminal law and administrative law in terms of the legality of the dean asked him to resign and said that there was no basis in law. Officious, and the Universal Human Rights Declaration and the Constitution based on crime and punishment be personal, said: "Legally the dean's resignation will not be required. Both the Universal Human Rights Declaration as well as according to the Constitution crime and jazz 'are private.' No one loved allegedly committed the shall be punished for the crime. spouse, child, mother or father any closer as he is responsible for the crime is unthinkable. YOK legislation and such a practice in criminal law, there are no regulations, "he said."
which is so far removed from what the text says as to render it useless as a basis for editing.
I will leave it to your imagination as to what Google translate makes of a demanding text like a court judgment in Turkish!
For the record, the source text was:
"Ankara Üniversitesi Hukuk Fakültesi Öğretim Üyesi Prof.Dr. Hasan İşgüzar, ceza hukuku ve idare hukuku açısından dekanın istifasının istenmesinin yasal ve hukuki hiçbir dayanağı olmadığını söyledi. İşgüzar, hem İnsan Hakları Evrensel Beyannamesi hem de Anayasa'ya göre suç ve cezanın şahsi olduğunu belirterek, "Yasal olarak dekanın istifası istenemez. Hem İnsan Hakları Evrensel Beyannamesi hem de Anayasamıza göre suç ve caza 'şahsidir'. Hiç kimse bir yakının işlediği iddia edilen suçtan dolayı cezalandırılamaz. Eş, çocuk, anne veya baba gibi herhangi bir yakının o suçtan sorumlu olması düşünülemez. YÖK mevzuatında ve ceza hukukunda da böyle bir uygulama, düzenleme yoktur" dedi."
(The above example was taken at random and was not in way cherry-picked.)
[Edited at 2014-06-18 09:51 GMT] ▲ Collapse
| | |
|
|
|
Giles Watson
Italy
Local time: 08:14
Italian to English
In memoriam
Samuel Murray wrote:
But if I know that it is a machine translation, then I'm willing to edit it, because the machine makes more predictable mistakes and does not interpret the text (potentially incorrectly) as a human would.
This is not my experience.
After you have been editing a particular human translator's work for a while, you begin to know what to look out for. We human translators have our own little quirks and the inaccuracies of MT versions look extremely random in comparison.
But I can see the potential advantage of using an MT plug-in with a CAT if the corpus on which the MT is based is germane to your translation.
| | | | | What MT is and what it isn't, for purposes of this discussion | Jun 18, 2014 |
Vladimir Pochinov wrote:
I would never doubt your professional skills but you might be wrong here on both counts.
1. You can add one of more MT tools to your project to be translated using SDL Trados Studio, whereby this MT would serve as a provider of translation candidates, just like a regular TM does. The quality of MT-supplied translation candidates depends on the quality of the main dictionary; add-on industry-, project-, or client-specific dictionaries; and, primarily, on the underlying algorithms and rules.
2. If a specific MT tool has been fed with the approved terminology as used by various UN system entities, the MT-supplied translation candidates may be of surprisingly high-quality, provided that the algorithms are up to the mark as well.
Hi Vladimir,
Likewise, you have my full respect as a translator of the highest caliber. Moreover, you are one of the couple of people I mentioned at the beginning of this thread who, as I know, have been using MT to good results.
I see your point about merging CATs and MT. However, I feel we must strictly define what MT is and what it isn't before going any further and trying to prove or disprove any claims.
The MT-based tools and practices that you have mentioned in your last few posts, in my opinion, go far beyond what MT is for most people, and most translators. Kudoz to you for developing and effectively applying such advanced working methods. However, let's please call them what they are: smart MT-based solutions drawing on various corpora of human-produced translations. Not simple machine translation per se, which is what I was referring to and arguing against.
I see you are taking the arguments presented here very close to heart. But I don't see how they apply to you. Your case is entirely different.
| | | | | Interpreting text | Jun 18, 2014 |
Samuel Murray wrote:
...because the machine... does not interpret the text (potentially incorrectly) as a human would.
Maybe you misspoke, but this doesn't stand up to scrutiny. MT does NOT interpret any text as there's no true AI in it, i.e. it does not understand the *meaning* of text. Therefore, it can and often will pick wildly wrong target-language terms for any source-language term.
Yes, human translations can also be extremely inaccurate, but that is irrelevant.
[Edited at 2014-06-18 13:03 GMT]
| | | | | MT can indeed increase productivity (but not necessarily quality) | Jun 18, 2014 |
Mikhail Kropotov wrote:
Hi Vladimir,
Likewise, you have my full respect as a translator of the highest caliber. Moreover, you are one of the couple of people I mentioned at the beginning of this thread who, as I know, have been using MT to good results.
I see your point about merging CATs and MT. However, I feel we must strictly define what MT is and what it isn't before going any further and trying to prove or disprove any claims.
The MT-based tools and practices that you have mentioned in your last few posts, in my opinion, go far beyond what MT is for most people, and most translators. Kudoz to you for developing and effectively applying such advanced working methods. However, let's please call them what they are: smart MT-based solutions drawing on various corpora of human-produced translations. Not simple machine translation per se, which is what I was referring to and arguing against.
I agree with Vladimir and Mikhail. Smart MT-based solutions based on various corpora can help translators translate more and faster. As someone said, MT can be used as a sophisticated dictionary that replaces most of the source words with their target equivalents (and proper names/numbers in the process). Next, it's a question of replacing the terminology used by the MT tool with the terminology appropriate for the source context (and the same source term could be translated differently in the same document), and re-arranging the target text to flow smoothly. The translator can focus on researching the subject of the text and its terminology, ensuring the text reads well, and thoroughly proof-reading the text at a conceptual level.
The entire typing/dictation process is skipped. This is the phase I have always found the least appealing as a translator.
In the end, what matters is a high-quality translation in less time. MT helps me achieve this in less time than a traditional approach. No less than a TM helps in producing a translation in less time. I do NOT rely on MT to produce a quality translation.
Of course, due to issues of confidentiality, MT tools such as Google Translate are a strict no-no for me. I use a desktop based tool exclusively.
| | |
|
|
|
Samuel Murray
Netherlands
Local time: 08:14
Member (2006)
English to Afrikaans
+ ...
Mikhail Kropotov wrote: Samuel Murray wrote:
...because the machine... does not interpret the text (potentially incorrectly) as a human would. Maybe you misspoke ... MT does NOT interpret any text as there's no true AI in it, i.e. it does not understand the *meaning* of text.
I think that that is precisely my point:
A human interprets the text and understands it. A human who is a very bad translator will misinterpret the text and misunderstand it and still try to translate it. Therefore if I have to choose between editing a lousy machine translation and editing a lousy human translation, I'd rather edit the machine translation because there is less of a risk that the text will look beautiful, lulling me into a false sense of confidence. Either way, I'd rather edit a good translation than a lousy one.
[Edited at 2014-06-18 13:29 GMT]
| | | | Samuel Murray
Netherlands
Local time: 08:14
Member (2006)
English to Afrikaans
+ ...
Giles Watson wrote: Samuel Murray wrote:
But if I know that it is a machine translation, then I'm willing to edit it, because the machine makes more predictable mistakes and does not interpret the text (potentially incorrectly) as a human would. After you have been editing a particular human translator's work for a while, you begin to know what to look out for.
I agree, if the translator is known to me. But if a translator is known to me to be a lousy translator, then I will not continue to accept editing jobs where he is known to have been the translator.
| | | |
Samuel Murray wrote:
Either way, I'd rather edit a good translation than a lousy one.
Finally something we do agree on!
And sorry if I took your words out of context. I did not mean to.
If only more translators (alert: not talking about Vladimir and select others like him) refused to continuously fall into the trap of accepting poor translations to salvage, and all other kinds of unfavorable working conditions in general, the industry would have been in a much better shape today.
| | | | | my 2 bani/pennies/cents/euro-cents | Jun 19, 2014 |
I have just noticed that the original poster is a fellow Romanian translator, living abroad, and I honestly hope she is not using MT (G or B or whatever else there is) in this language pair.
My worst experience with MT was the one time an end-client's employee decided to copy the (already quite well translated in English) Romanian form, add his/her own input (do not get me started on his/her knowledge of Romanian), and then use MT to translate it to English. And then they send it ov... See more I have just noticed that the original poster is a fellow Romanian translator, living abroad, and I honestly hope she is not using MT (G or B or whatever else there is) in this language pair.
My worst experience with MT was the one time an end-client's employee decided to copy the (already quite well translated in English) Romanian form, add his/her own input (do not get me started on his/her knowledge of Romanian), and then use MT to translate it to English. And then they send it over to my client for translation.
The end result? The MT English version was so terrible, I had to refuse doing detective work to see where do various parts of the form fit into the existing Romanian documentation and fill in the missing parts.
So, no MT in this language pairs for me. I'd rather start converting my hobbies into an income source, than post-edit MT. ▲ Collapse
| | |
|
|
|
| Not universally applicable but ... | Jun 19, 2014 |
Michelangela wrote:
What's wrong with that?
....you could well be in breach of any confidentiality agreement/NDA you've signed.
Depends on the system you're using, natch. Some bespoke, in-house beast is probably fine, but then those typically deliver better output, so I'm led to believe.
I think most people on this thread are mostly talking about Google Translate and Bing and whatnot, where your text is sent over the internet to their infrastructure. Regardless of the T&C of service, we all know Google for one has been found collecting data it said it wouldn't, so frankly, I don't believe them, whatever they say. If I gave you a confidential document and you used GT and I found out, I would be quite annoyed (!), as I would consider that a breach.
In contrast, I find it astonishing how dogmatic people can be in rejecting outright others' observations of time saved, when the conditions are right. I once reported elsewhere that I reckoned it saved me maybe 30% of first draft time, when I did a couple of experiments a few years back. Several people simply refused to believe it. Point blank. And naturally, as has happened here, basically said if I'd saved 30% of the time, the end result must have been crap. Ho hum.
I agree with other comments on the first page - I really don't give a flying one at a rolling doughnut *how* output is produced, what matters is whether it's fit for use (whatever that use is). With the proviso that confidentiality is respected - and I'm not at all convinced that online MT applications meet that proviso.
[Edited at 2014-06-19 09:44 GMT]
| | | | Michelangela
United States
Local time: 23:14
German to English
+ ...
TOPIC STARTER
Charlie Bavington wrote: Michelangela wrote:
What's wrong with that?
....you could well be in breach of any confidentiality agreement/NDA you've signed I think most people on this thread are mostly talking about Google Translate and Bing and whatnot ... In contrast, I find it astonishing how dogmatic people can be in rejecting outright others' observations of time saved, when the conditions are right. I agree with other comments on the first page - I really don't give a flying one at a rolling doughnut *how* output is produced, what matters is whether it's fit for use (whatever that use is)
Exactly, breaching confidentiality is a very good answer to the original question.
I found it amusing how "dogmatic people", the best of the best translators, lacking technological sophistication, were the first to judge others as inexperienced, incompetent, etc.
I would like to thank the true professionals who contributed with meaningful and insighful comments.
| | | | | Not using MT indicates lack of technological sophistication? | Jun 20, 2014 |
Michelangela wrote:
I found it amusing how "dogmatic people", the best of the best translators, lacking technological sophistication, were the first to judge others as inexperienced, incompetent, etc.
So, to you, not using MT indicates lack of technological sophistication?
Could you please explain how so?
In my more than 21 years working as a translator (20 of them as an ATA Certified Translator), I have worked with many excellent technical translators and editors, both men and women, who do not use MT.
The last thing anybody could say about those brilliant minds is that they lack technological sophistication, because to me, what characterizes them is precisely their technological sophistication and knowledge, given their academic backgrounds and experience, as well as the technical fields we have worked in together.
[Edited at 2014-06-20 17:32 GMT]
| | | | | Pages in topic: < [1 2 3 4 5 6 7] > | To report site rules violations or get help, contact a site moderator: You can also contact site staff by submitting a support request » The "Crime" of Using MT | Trados Business Manager Lite | Create customer quotes and invoices from within Trados Studio
Trados Business Manager Lite helps to simplify and speed up some of the daily tasks, such as invoicing and reporting, associated with running your freelance translation business.
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 
Translation news
