| Gaiari buruzko orrialdeak: < [1 2 3] > | Comparing MT engines, which MT should I choose? Haria hasi duena: Philippe Locquet
|
|---|
| customized MT | May 22, 2021 |
Hello, I'm an experienced translator/reviser (English and Spanish to French) and about to leave my job in a large institution to work on my own. Beyond the technicals I'm interested in MT from a pragmatic perspective, namely, the amount of time it can save me. I've looked into courses and they tend to be either too academic, or not focused as nearly enough on the requirements of professional translation. Still, I consider recent advancements in MT, from a pragmatic viewpoint, to have been rather... See more Hello, I'm an experienced translator/reviser (English and Spanish to French) and about to leave my job in a large institution to work on my own. Beyond the technicals I'm interested in MT from a pragmatic perspective, namely, the amount of time it can save me. I've looked into courses and they tend to be either too academic, or not focused as nearly enough on the requirements of professional translation. Still, I consider recent advancements in MT, from a pragmatic viewpoint, to have been rather remarkable.
So far I've used mostly DeepL which is valuable provided the user knows what they're doing; time gains vary widely, from marginal to substantial, 5 to 20+ % depending on documents. In addition to that less typing is a valuable gain. As good as DeepL may be it isn't evolving - not that I expect a free tool to do so.
With the prospect of working on my own, I would appreciate a more tailored tool with, mainly, the ability to use my own originals and production as source text, select preferred terminology and filter out bad translations from the public sources that would also be part of my corpus. Ideally - hypothetically - that would allow improved results at the outset and over time.
The tool that seems to most closely match those requirements seems to be ModernMT. I haven't given the free trial a shot yet. In the ad copy, they don't mention a subsequent (exorbitant) fee of £4/1000 words once subscribed, but I have seen that fee mentioned here. Or is the fee from an older version ?
A more desirable option would perhaps be to design my own system, based on code readily available on sites such as github for instance. I simply have no programming experience, and no idea how hard it could be, or how much time it could take. I wonder if any of the members here have attempted that. Thanks in advance for your feedback! ▲ Collapse
| | | | | My option, ModernMT | May 23, 2021 |
beef01 wrote:
The tool that seems to most closely match those requirements seems to be ModernMT. I haven't given the free trial a shot yet. In the ad copy, they don't mention a subsequent (exorbitant) fee of £4/1000 words once subscribed, but I have seen that fee mentioned here. Or is the fee from an older version?
No, that is clearly NOT the price for ModernMT. The subscription is far cheaper and has a very high wordcount limit within the montly rate. I would encourage you to examine the rates again.
Now, I have tried building my own in-house engines with rather large corpora I had in the form of TMs after years of working for specific account, and the results were not that good: it lacked a lot of linguistic variety/richness to tackle the idiomatic side of every translation. Not an option.
I have tried ModernMT extensively, in particular coupled with memoQ, and it yields very nice results of you train it a bit with a good-quality, reasonably extense TM, and if you enable the self-learning option so that it learns from your corrections on the fly. The TM does not need to be terribly big, but it must reflect the work in the way you want it, i.e. it needs to have a low variability and it must be consistent. With these premises, ModernMT should yield far better results than Deepl, or at least that is my experience in my tests with English and German into Spanish.
As for your initial question of what would be a reasonable expectation of saved time with the help of ModernMT, it would greatly depend on the text. The most favourable situation with ModenMT is perhaps a gain of 25-30% of the time in simpler, procedural, descriptive source texts with a lower level of idiomaticity. In my case, when I have tried it, it served me mostly as a way to avoid typing long sentences.
Good luck!
| | | | | One important caveat | May 23, 2021 |
beef01 wrote:
Beyond the technicals I'm interested in MT from a pragmatic perspective, namely, the amount of time it can save me.
I would like to mention that obviously we should respect our clients' wishes in terms of using MT tools or not, especially when it comes to options that do not guarantee privacy or store anything on the MT provider's syste,. It is of a maximum important that you examine the privacy practices and claims of the different options carefully and ensure that you are not jumping into the volcano by using something that puts your commitments to privacy in jeopardy.
Another important thing to remember is that, if you process XLIFF files, there are formats that will record the fact that you are using MT in your CAT tools and make the XLIFF files bigger when you did not count on it. Freel free to email me for an example if you plan to translate XLIFF files with MT within a CAT tool.
| | | | Philippe Locquet 
Portugal
Local time: 02:32
Bazkidea (2013)
Ingelesatik Frantsesara
+ ...
TOPIC STARTER
beef01
Hi,
Training custom MTs with your own corpus requires technical knowledge for cleaning etc. even if you start with a shell MT from a third party.
If you mean to code your own, don't. That's too big a task and RBMT models available for free do not compete with modern options.
DeepL VS Modern MT: in my experience, If german is involved as a language pair, DeepL is better. For nearly anything else, at the moment ModernMT is better.
I agree with Tomás on MMT pricing. I made a video explaining step by step what ModernMT adapts. MemoQ works with it but with some limitations on context Adapt. https://youtu.be/p1RIVC3II3w
If you are looking at documents with a big wordcount and wish to pretranslate with MT, I suggest you use Intento. They have a tier where you can use your own glossaries to adapt MT output and they have blazingly fast speeds. I have a video explaining it here: https://youtu.be/5K6raLjZZ3A
Hope you'll find what you're looking for.
| | |
|
|
|
Philippe Locquet
Portugal
Local time: 02:32
Bazkidea (2013)
Ingelesatik Frantsesara
+ ...
TOPIC STARTER | Can be removed | May 23, 2021 |
[quote]Tomás Cano Binder, BA, CT wrote:
beef01 wrote:
Another important thing to remember is that, if you process XLIFF files, there are formats that will record the fact that you are using MT in your CAT tools and make the XLIFF files bigger when you did not count on it. Freel free to email me for an example if you plan to translate XLIFF files with MT within a CAT tool.
GT4T has a "remove MT tag" feature to fix this issue. It works well with sdlxliff and txlf.
| | | |
Dear Philippe and Tómas,
First of all thank you for some very useful replies.
So glad that ModernMT is priced reasonably. I had read, perhaps skimming too much, information that seemed to be conflicting in that respect, and began suspecting a hidden fee structure post subscription, but I'm glad it isn't the case. My bad. Your feedback on building one's own engines is invaluable - no use in vain effort. I was briefly in contact with an engineer who told me it was doabl... See more Dear Philippe and Tómas,
First of all thank you for some very useful replies.
So glad that ModernMT is priced reasonably. I had read, perhaps skimming too much, information that seemed to be conflicting in that respect, and began suspecting a hidden fee structure post subscription, but I'm glad it isn't the case. My bad. Your feedback on building one's own engines is invaluable - no use in vain effort. I was briefly in contact with an engineer who told me it was doable, and it may be the case in theory but in the real world, probably not. On the other hand, I have easy enough access to about a decade of personal work which I hope to be able to put to good use. Maybe that task is a worthwhile and more accessible one to learn on its own. Tómas, you also made some excellent points about privacy. As to XLIFF files, I have not been exposed to them yet, but that might be a possibility in the future, so I will keep that in mind. Thanks for the video links, Philippe, I will check them out. ▲ Collapse
| | | | RoyMarie
Estatu Batuak
Local time: 18:32
| | Thanks for testing | Jun 21, 2021 |
Philippe Locquet wrote:
Here goes the snapshot.
Difficult to have 10 columns in the same picture... I had to remove manually some returns so it could fit
9 engines tested here.
My overall winner IMO is "ModernMT"
Engines give different results some are better than others in different aspects.
Hi Philippe, thanks for your test, that really helps me out, I never tried ModernMT before, I always use google or baidu. I'll try to use ModernMT next time, see if it suitable for me.
| | |
|
|
|
Philippe Locquet
Portugal
Local time: 02:32
Bazkidea (2013)
Ingelesatik Frantsesara
+ ...
TOPIC STARTER
CHENGRUI ZHANG wrote:
Hi Philippe, thanks for your test, that really helps me out, I never tried ModernMT before, I always use google or baidu. I'll try to use ModernMT next time, see if it suitable for me.
I think I've seen great results in chinese to english with DeepL. If that's one of your language pairs, that's worth comparing too.
| | | | | | Michael Beijer
Erresuma Batua
Local time: 02:32
Bazkidea (2009)
Nederlanderatik Ingelesara
+ ...
| communication with DeepL is a joke | Feb 17, 2023 |
Tomás Cano Binder, BA, CT wrote:
[…]
In terms of interaction with Deepl the company, despite my repeated suggestions and requests to fix endemic issues in spelling in Spanish, I have never seen any result or even had a single response back from Deepl. Despite being a paying client, I do not count in their customer relations efforts at all. Sending me their monthly invoice is all my communications with them.
[…]
Totally agree about communicating with DeepL, or in my case trying to. I have been sending them support requests for around 5 days now and have gotten zero response, other than their automated 'We have received your ticket' message. My subscription ran out and was trying to resubscribe but the payment keeps failing. Pretty pathetic for a company that is doing so well.
Am currently playing around with the Intento 2-week trial.
| | | | Michael Beijer
Erresuma Batua
Local time: 02:32
Bazkidea (2009)
Nederlanderatik Ingelesara
+ ...
| DeepL vs Intento subscription (for roughly same price) | Feb 17, 2023 |
Interestingly, you can use any of around 10 or more MT engines with Intento in memoQ, or set it so the system decides itself.
DeepL Pro Advanced:
€299.88/year
Localization Starter plan
$300.00/year (= approx. €280)
In this sense, it seems smarter to get an Intentoo subscription, which will then allow me pick and choose from many different engines, depending on my particular job. Also, since I can't actually pay for my DeepL sub (because thei... See more Interestingly, you can use any of around 10 or more MT engines with Intento in memoQ, or set it so the system decides itself.
DeepL Pro Advanced:
€299.88/year
Localization Starter plan
$300.00/year (= approx. €280)
In this sense, it seems smarter to get an Intentoo subscription, which will then allow me pick and choose from many different engines, depending on my particular job. Also, since I can't actually pay for my DeepL sub (because their checkout keeps failing with all my bank cards), and they won't get back to me, I'd say the decision is even more easily made ▲ Collapse
| | |
|
|
|
| Useful... to a certain extent? | Feb 18, 2023 |
Michael Beijer wrote:
Interestingly, you can use any of around 10 or more MT engines with Intento in memoQ, or set it so the system decides itself.
Indeed. However, what works for my office (mostly technical and scientific translation) is to be able to closely control terminology in all MT projects. Generic MT output without controlling term translations is not that useful in my case. When we are authorised by our clients to use MT, we mostly use a Deepl Pro account with specific glossaries.
Do you mean that Intento would allow us to feed our materials into "sub-plugins" like Deepl or ModernMT? (I have not tried Intento thus far.)
| | | | Philippe Locquet
Portugal
Local time: 02:32
Bazkidea (2013)
Ingelesatik Frantsesara
+ ...
TOPIC STARTER | Intento for specific workflows | Feb 18, 2023 |
Tomás Cano Binder, BA, CT wrote:
Do you mean that Intento would allow us to feed our materials into "sub-plugins" like Deepl or ModernMT? (I have not tried Intento thus far.)
The short answer is "yes and more".
Intento is primarily set up for bigger translation customers, although freelancer can benefit from the system too.
Since I made the video the offers have slightly changed (although the information is still relevant and accurate to describe the service). https://youtu.be/5K6raLjZZ3A
Now you have 3 tiers.
Localization starter = no customization
Localization Expert = can apply glossaries to MT output
Enterprise Plans = you can use massive MT not available to freelancers and use your own data to train these engines (i.e. you can train Google MT with your own data), they provide assistance and expertise to do the engine tuning and the data preparation for you. On top of this, your glossaries can be added. There are great speed gains too for that kind of tiers, as they can open several accounts at once and split you file to hit the MT engines on several "lanes" at the same time.
If you decide to approach them, please use this link: https://hubs.la/H0BTct00 so they know I sent you
For big jobs, I suggest you request a demo. They're nice knowledgeable people and since they're not tied to a specific MT provider they won't speak nonsense like telling you your MT will do everything for you and won't require human translators. That's my experience with them anyway.
Hope this helps.
Philippe
| | | |
Philippe Locquet wrote:
Tomás Cano Binder, BA, CT wrote:
Do you mean that Intento would allow us to feed our materials into "sub-plugins" like Deepl or ModernMT? (I have not tried Intento thus far.)
The short answer is "yes and more".
Thanks a lot for all the details. It might be worth looking into!
| | | | | Gaiari buruzko orrialdeak: < [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 » Comparing MT engines, which MT should I choose? | Protemos translation business management system | Create your account in minutes, and start working! 3-month trial for agencies, and free for freelancers!
The system lets you keep client/vendor database, with contacts and rates, manage projects and assign jobs to vendors, issue invoices, track payments, store and manage project files, generate business reports on turnover profit per client/manager etc.
More info » |
| | Pastey | Your smart companion app
Pastey is an innovative desktop application that bridges the gap between human expertise and artificial intelligence. With intuitive keyboard shortcuts, Pastey transforms your source text into AI-powered draft translations.
Find out more » |
|
| | | | X Sign in to your ProZ.com account... | | | | | |
Login to reply/comment 
