Pages in topic:   < [1 2 3] >
Are my assumptions about MT wrong? And an invitation to share a guess...
Thread poster: Henry Dotterer
Bernhard Sulzer
Bernhard Sulzer  Identity Verified
United States
Local time: 11:02
English to German
+ ...
MT can help but doesn't replace the human translator Jun 27, 2020

Answers:

- MT is helpful to translators in some cases (I'm talking about using MT within your CAT tool, this post is not about PEMT)
yes: get relatively good MT first, then edit it by comparing it to OT.


- Most translators are aware of this
most likely yes


- Many translators are successfully using MT, within their CAT tools, to boost their productivity productivity as in more product in less time - <
... See more
Answers:

- MT is helpful to translators in some cases (I'm talking about using MT within your CAT tool, this post is not about PEMT)
yes: get relatively good MT first, then edit it by comparing it to OT.


- Most translators are aware of this
most likely yes


- Many translators are successfully using MT, within their CAT tools, to boost their productivity productivity as in more product in less time - yes and no

yes: cutting down on original translation time and possibly on finishing as long as the text is not too complicated or repetitive or when you have a good TM and/or MT engine or output;

no: it's an additional step, esp. if used without a CAT tool. It adds time and so you might not be faster, but the additional step can improve your focus on accuracy and style. You know the machine can't get it perfectly right.


- Some translators have some sorts of work where MT is either not useful or should not be used. So there is still some work that does not benefit from MT.

yes, but: literature; but it depends on how you use it.


- Some people just choose not to use MT, even when it might be useful. Call it a preference. There is no problem with this.
yes: You can work without it. But many choose to use it for various reasons.

Am I wrong on any of these points?

What percentage of translators use MT, ever? 60-70%

What percentage of translators refuse to use MT, even when it could be used to boost productivity? 30-40%



[Edited at 2020-06-27 03:06 GMT]

[Edited at 2020-06-27 03:18 GMT]
Collapse


 
Mario Cerutti
Mario Cerutti  Identity Verified
Japan
Local time: 00:02
Italian to Japanese
+ ...
With certain languages it's much worse Jun 27, 2020

Lingua 5B wrote:
The more the syntax expands, the more the MT gets confused. If it manages to produce accurate words, it will mess up phrases, sentences, or ultimately the overall flow. It may produce a good sentence here and there but that’s not enough.

With Japanese, depending on the subject multiply this by two or three in terms of time needed to correct the MT output.


Kuochoe Nikoi-Kotei
 
Lingua 5B
Lingua 5B  Identity Verified
Bosnia and Herzegovina
Local time: 16:02
Member (2009)
English to Croatian
+ ...
Also very deceptive, makes people mentally lazy. Jun 27, 2020

Aliseo Japan wrote:

Lingua 5B wrote:
The more the syntax expands, the more the MT gets confused. If it manages to produce accurate words, it will mess up phrases, sentences, or ultimately the overall flow. It may produce a good sentence here and there but that’s not enough.

With Japanese, depending on the subject multiply this by two or three in terms of time needed to correct the MT output.


But I find this scenario even worse: When it produces a relatively good and usable output, but... For instance, two whole paragraphs running completely fine, then pop out a totally illogical word, then resume running OK. This is even more dangerous. It’s like driving in the countryside with no traffic, then a cow appears in the middle of the road, but you are already speeding too much...


Mario Cerutti
Thomas T. Frost
Maria Teresa Borges de Almeida
Daryo
Ying-Ju Fang
Edwin den Boer
Anton Konashenok
 
Milan Condak
Milan Condak  Identity Verified
Local time: 16:02
English to Czech
Translator or supply chain Jun 27, 2020

I am Slate Desktop user. A quality of Slate Connect output is depending of my choice of "engine" and covering a new topic by my previous TMX which I can feed Slate Desktop ("a factory for producting engine"). The output contains only a terminology which is in TMX used as the input.

The
... See more
I am Slate Desktop user. A quality of Slate Connect output is depending of my choice of "engine" and covering a new topic by my previous TMX which I can feed Slate Desktop ("a factory for producting engine"). The output contains only a terminology which is in TMX used as the input.

The link to an article by Tom Hoar:

https://www.slatedesktop.com/csa-linguist-supply-chain/#more-1619

CSA Linguist Supply Chain

"Linguists agree! Cloud-based MT is the lowest quality and least effective productivity tool. This report shares demograph nesouhlasíics, behaviors, attitudes, and challenges that translators and interpreters face at the turn of the 2020s.

CSA Research in cooperation with ProZ.com, Translators without Borders, and other industry associations surveyed translators and interpreters around the world."
---
Milan
Collapse


 
Milan Condak
Milan Condak  Identity Verified
Local time: 16:02
English to Czech
Local MT vs. web MT Jun 27, 2020

I am using MT for everything. Nobody is perfect. Real life sometimes brings little reason to laugh.

Manufacturing engines for SMT or NMT takes a lot of time. You need to have them prepared in advance and know what's in them.

I am a micro-euro-entrepreneur and I have been using for two month service eTranslation (of EC EU) and its output in TMX format. When used in OmegaT for translation of formatted documents (DOCX, HTML) there do not match segmentation of document and
... See more
I am using MT for everything. Nobody is perfect. Real life sometimes brings little reason to laugh.

Manufacturing engines for SMT or NMT takes a lot of time. You need to have them prepared in advance and know what's in them.

I am a micro-euro-entrepreneur and I have been using for two month service eTranslation (of EC EU) and its output in TMX format. When used in OmegaT for translation of formatted documents (DOCX, HTML) there do not match segmentation of document and TMX. Therefore, it is advisable to use the DGT-OmegaT or Wordfast Classic instead of OmegaT.

I can quickly affect /influence/induce/ MT output only in the PC Translator (RBMT) where I can add a new terminology on-the-fly singly or in batch and add the field in which it is used. PC Translator works in two modes; in other application or direct in own SW. In MS application: Outlook, Internet Explorer, Word. In Word serves as RBMT for Wordfast Classic. Direct in PC Translator is not only Directory but TM database for reading and saving translated text. Exported TMX can be used (after some editing) in CAT tools. For opened segment can get suggestion from Google Translate a Microsoft Translator.

I search a word Systran on ProZ.com: last was written 5 years ago. Searching with Google took me to their web

https://www.systransoft.com/translation-products/

and their YouTube video channel

https://www.youtube.com/channel/UCalOMpnXUPGWTNLWHV5U5hg/videos

Milan,
MT evangelist, SR evangelist, CAT evangelist, open-data evangelist (I do not sell Systran)


[Edited at 2020-06-27 06:51 GMT]
Collapse


 
Hans Lenting
Hans Lenting
Netherlands
Member (2006)
German to Dutch
Slate vs public NMT Jun 27, 2020

Milan Condak wrote:

I am Slate Desktop user.


Hello Slate!

Slate is nice for companies and institutions that translate in the same subject field over and over.

Pro NMT:
If you have to translate different subject fields every 2 or 3 days, you can benefit more from public NMT engines, since they can suggest new terminology that you haven't used yet and thus aren't present in the translation memories that you have to feed to a closed system like Slate.

Con NMT:
However, a big danger of public NMT is that it glues words together, making up new, nonsensical words. So you'll have to verify everything carefully.


[Edited at 2020-06-27 08:04 GMT]


 
Heinrich Pesch
Heinrich Pesch  Identity Verified
Finland
Local time: 17:02
Member (2003)
Finnish to German
+ ...
Your points are to the point Jun 27, 2020

I use MT quite regularly, but one has to watch the terminology, because MT translates only one phrase at a time and does not remember, what term was used in the previous phrase or unit. Even grammar has improved immensely the last few years. And another obstacle is style, which should be the same throughout the whole text.
Many customers though still think a translation with the aid of MT must be inferior to one entirely done by a human. Which is nonsense of course, do you translate better
... See more
I use MT quite regularly, but one has to watch the terminology, because MT translates only one phrase at a time and does not remember, what term was used in the previous phrase or unit. Even grammar has improved immensely the last few years. And another obstacle is style, which should be the same throughout the whole text.
Many customers though still think a translation with the aid of MT must be inferior to one entirely done by a human. Which is nonsense of course, do you translate better without a dictionary and without spellchecker?
Collapse


Sheila Wilson
Marina Steinbach
Elena Feriani
Philip Lees
neilmac
Dr. Stephan Pietzko
 
RobinB
RobinB  Identity Verified
United States
Local time: 10:02
German to English
Largely right Jun 27, 2020

Hi Henry,

I think you're largely right on all points.

Based on my conversations with professional translators both here in the U.S. and in Europe, I would guess that at least 75% of them use MT occasionally (like myself) or regularly. Ultimately, it's just another way to recycle other people's translations, and most translators do that more often than many of them are willing to admit. I freely acknowledge that I'm always happy to use somebody else's translation if I th
... See more
Hi Henry,

I think you're largely right on all points.

Based on my conversations with professional translators both here in the U.S. and in Europe, I would guess that at least 75% of them use MT occasionally (like myself) or regularly. Ultimately, it's just another way to recycle other people's translations, and most translators do that more often than many of them are willing to admit. I freely acknowledge that I'm always happy to use somebody else's translation if I think it's better than what I would have used myself.

Bet you didn't expect me to say that

You might also like to read my statement on LinkedIn:

https://www.linkedin.com/posts/robin-bonthrone-pflt_translators-translation-machinetranslation-activity-6631609951397699584-faRU

Best,
Robin
Collapse


Edwin den Boer
 
Henry Dotterer
Henry Dotterer
Local time: 11:02
SITE FOUNDER
TOPIC STARTER
Thanks, RobinB! Jun 27, 2020

RobinB wrote:
Ultimately, it's just another way to recycle other people's translations, and most translators do that more often than many of them are willing to admit. I freely acknowledge that I'm always happy to use somebody else's translation if I think it's better than what I would have used myself.

Bet you didn't expect me to say that

You might also like to read my statement on LinkedIn:

https://www.linkedin.com/posts/robin-bonthrone-pflt_translators-translation-machinetranslation-activity-6631609951397699584-faRU

I like how you are thinking about this. Multiple inputs... MT is another way to benefit from someone else's prior translation... the translator needs to "own" the final translation.

RobinB wrote:
Bet you didn't expect me to say that

Ha ha, yeah.


 
Katalin Horváth McClure
Katalin Horváth McClure  Identity Verified
United States
Local time: 11:02
Member (2002)
English to Hungarian
+ ...
@Henry, clarify, please Jun 28, 2020

"What percentage of translators use MT, ever?"

Could you clarify whether you meant "use" as in "use for translation work"?
Or did you mean "ever" as for any purpose, including work but also outside of work: daily life, travel, crossword puzzles, etc.?

Because I bet the answers would be different.


Grace Anderson
 
Edwin den Boer
Edwin den Boer  Identity Verified
Netherlands
Local time: 16:02
Member (2009)
English to Dutch
I don't use MT unless a client supplies it Jun 28, 2020

There's a problem with judging the effectiveness of machine translation: companies that rely on MT will probably target metrics that can be quantified easily, such as price and speed, while underestimating the effect of an unnatural style on sales figures etc. Until a product manager suddenly notices that MTPE doesn't produce marketing copy which appeals to the audience.

As a reviewer, I can tell when a translator has used MT. The text will read like a translation, with phrases that
... See more
There's a problem with judging the effectiveness of machine translation: companies that rely on MT will probably target metrics that can be quantified easily, such as price and speed, while underestimating the effect of an unnatural style on sales figures etc. Until a product manager suddenly notices that MTPE doesn't produce marketing copy which appeals to the audience.

As a reviewer, I can tell when a translator has used MT. The text will read like a translation, with phrases that a native speaker wouldn't use. And I don't blame those translators. MT is just one aspect of an economic system that prioritizes low price and fast turnaround at the expense of quality and job satisfaction.

For English to Dutch, correcting the word order is a common task that takes more effort than translating from scratch.

As Kay-Viktor Stegemann said, tags are surprisingly difficult for MT engines. Tags will not only be around the wrong words, but even invalid due to the number or order. And Trados isn't flexible about editing tags, so in that case, need to copy the source text and re-translate the segment, just to get the tags right.

One MT-promoting client started to neglect their TM, which was more useful. Apparently, in their setup, it was easier to pull MT from the cloud than to send huge TM files with contributions from multiple agencies. I had to align content from their website in order to create a better TM.

The unpredictability of neural nets remains a hard problem. I've rated MT results to compare engines, and the one that produced more idiomatic translations was also more likely to go off the rails occasionally.

[Edited at 2020-06-28 12:35 GMT]
Collapse


Maria Teresa Borges de Almeida
Stephanie Busch
 
neilmac
neilmac
Spain
Local time: 16:02
Spanish to English
+ ...
MT = My Take Jun 30, 2020

I've been using MT - judiciously - since a colleague gave me a copy of a Systran CD way back in the day. That was quite helpful for certain types of bureaucratic/regional and local government type texts. However, around the turn-of-the-century day it got "upgraded" and I found it more unwieldy and began to use it less and less.

A few years ago I started using online MT - again, judiciously - and I find it quite helpful. However, I don't rely on it to translate large chunks of text o
... See more
I've been using MT - judiciously - since a colleague gave me a copy of a Systran CD way back in the day. That was quite helpful for certain types of bureaucratic/regional and local government type texts. However, around the turn-of-the-century day it got "upgraded" and I found it more unwieldy and began to use it less and less.

A few years ago I started using online MT - again, judiciously - and I find it quite helpful. However, I don't rely on it to translate large chunks of text or whole documents, which is apparently possible nowadays. Instead, I tend to use it for comparisons, and to see what kind of howlers it occasionally spits out. I do a lot of post-editing for academic writers whose first language is not English, and nowadays they are obviously using MT to help them, occasionally leading to errors such as "genre" instead of "gender", which cropped up yesterday, and is the type of mistake not easily detected by a standard spellcheck.

Like most tools, it is probably only as good as the person using it. It's not like giving a monkey a typewriter - or even a word processor - and expecting it to write Shakespeare.

Collapse


 
Oliver Pekelharing
Oliver Pekelharing  Identity Verified
Netherlands
Local time: 16:02
Dutch to English
MT only works with a good source text Jun 30, 2020

One important thing about MT that I don't think gets enough attention is that it can only ever produce anything halfway useful if the source text is reasonably well written and follows accepted writing conventions. I rarely get to translate anything written lucidly and so for me the benefits of MT are generally restricted to terminology suggestions and time saved looking up and typing geographical names and the like.

Sheila Wilson
Kay-Viktor Stegemann
Elena Feriani
Anton Konashenok
 
Mario Cerutti
Mario Cerutti  Identity Verified
Japan
Local time: 00:02
Italian to Japanese
+ ...
Not only that Jun 30, 2020

Oliver Pekelharing wrote:
...it can only ever produce anything halfway useful if the source text is reasonably well written...

And possibly written by a native-level English speaker.


 
Michael Wetzel
Michael Wetzel  Identity Verified
Germany
Local time: 16:02
German to English
MT means thinking through the text one less time. Jun 30, 2020

I think DeepL grew out of Linguee, which was a search engine for finding previously translated texts. The German>English results were mostly garbage, although their stench varied and you could find the occasional helpful hit by scrolling some. I always saw this as a result of the fact that (1) most German-English bilingual sites on the Internet are translated German sites and (2) most websites are not translated well. It might be very different working English>other language because, although yo... See more
I think DeepL grew out of Linguee, which was a search engine for finding previously translated texts. The German>English results were mostly garbage, although their stench varied and you could find the occasional helpful hit by scrolling some. I always saw this as a result of the fact that (1) most German-English bilingual sites on the Internet are translated German sites and (2) most websites are not translated well. It might be very different working English>other language because, although you would still be getting translated texts instead of genuine "parallel texts" (comparable texts written independently in the target language), you would at least be starting with a genuine English term or phrase and would thus be much more likely to hit well-translated sites. That is one fundamental reason why I feel comfortable assuming that Google and DeepL are not going to be of any real help in my work.

Like others have already mentioned, the other problem with using GoopL brand-name generic MT is it would mean developing a completely different workflow: It completely removes the initial phase of interacting with the source text (raw translation), it completely reframes the editing phase comparing source and target texts, and it provides a very different starting point for the process of converting the source text into a linguistically and stylistically convincing target text. That is why I agree that using MT to buy more words per hour would certainly cost me a great deal of quality.

Slate and other Desktop-based personalized MT systems have always interested me, because I have been working more or less entirely within a single field for ten years, but it's never interested me enough to actually get around to investing the time in exploring what is out there.
Collapse


 
Pages in topic:   < [1 2 3] >


To report site rules violations or get help, contact a site moderator:


You can also contact site staff by submitting a support request »

Are my assumptions about MT wrong? And an invitation to share a guess...







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 »
TM-Town
Manage your TMs and Terms ... and boost your translation business

Are you ready for something fresh in the industry? TM-Town is a unique new site for you -- the freelance translator -- to store, manage and share translation memories (TMs) and glossaries...and potentially meet new clients on the basis of your prior work.

More info »