Call for volunteers rewarded by free MT and visibility!
Thread poster: Philippe Locquet
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 09:43
English to French
+ ...
Aug 20, 2020

I was looking at a comparative study made by Intento, very useful when you want to choose an MT (check this thread: https://www.proz.com/forum/machine_translation_mt/345577-comparing_mt_engines_which_mt_should_i_choose.html)

Upon reading the report, I saw on page 28 that they are calling for volunteers to do PEMT and
... See more
I was looking at a comparative study made by Intento, very useful when you want to choose an MT (check this thread: https://www.proz.com/forum/machine_translation_mt/345577-comparing_mt_engines_which_mt_should_i_choose.html)

Upon reading the report, I saw on page 28 that they are calling for volunteers to do PEMT and LQA. They need two linguists per language pair.
They will reward the selected translators with a limited free access to the winning MT engines through plugins.
They will also promote you as an Intento MT Evaluation Partner.

It’s been a long time since I didn’t see opportunities like this. The localization program at ProZ.com has been good for me too, and since then, it’s the best volunteer offer I’ve seen so far.

Hope this helps someone.

My bests
Collapse


 
DZiW (X)
DZiW (X)
Ukraine
English to Russian
+ ...
https://inten.to Aug 20, 2020

First, the very term Machine Translation (MT) assumes pseudo-collocation rule-based “best matching”—it’s not a real language with some context, but a generalized meta-language generating human-like phrases.

Therefore, all those PEMT-volunteers and LQAs are PEMT-guessors, no translators.

Third, the fancy x20-x70-x100+ less payment implies 'at the cost of translators'.

Who, why, and for what?—Dumping discounted cheap?... See more
First, the very term Machine Translation (MT) assumes pseudo-collocation rule-based “best matching”—it’s not a real language with some context, but a generalized meta-language generating human-like phrases.

Therefore, all those PEMT-volunteers and LQAs are PEMT-guessors, no translators.

Third, the fancy x20-x70-x100+ less payment implies 'at the cost of translators'.

Who, why, and for what?—Dumping discounted cheap?
Way to go!
Collapse


 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 09:43
English to French
+ ...
TOPIC STARTER
“the significance of adequate knowledge in adoption decisions” Aug 21, 2020

DZiW wrote:

First, the very term Machine Translation (MT) assumes pseudo-collocation rule-based “best matching”—it’s not a real language with some context, but a generalized meta-language generating human-like phrases.

Therefore, all those PEMT-volunteers and LQAs are PEMT-guessors, no translators.

Third, the fancy x20-x70-x100+ less payment implies 'at the cost of translators'.

Who, why, and for what?—Dumping discounted cheap?
Way to go!


When new technology disrupts an industry, you can’t avoid worker’s reluctance to change.
The reaction to MT is no different to the reaction translators had to TM based CATs back then. Check this 2006 article:
https://www.researchgate.net/publication/220419144_Translators_and_TM_An_investigation_of_translators'_perceptions_of_translation_memory_adoption

A very valid point is made there “the significance of adequate knowledge in adoption decisions”. Hence the vast number of trainings available around the use of CATs even if translating without CATs comes still regularly regularly in the forums. Knowledge is everything.

As LSPs compete to get jobs, the speed that adequately managed MT (right language pair, right domain, right engine, right process) affords is a very clear choice for LSPs and customers. Ultimately, they are looking for savings. MT is here, we just need to accept it.

Would we imagine the automobile industry today without automated manufacturing? Nonetheless automation do not necessarily mean robot superiority. Show cars are still painted by humans (experts of the trade) some red cars with a horse on it are handmade. True, robots do not provide high-end craftsmanship. But is it their goal?

Knowledge is everything. The goal of MT as it is used at the moment is higher speed and savings. Translators that use it for their own business certainly use it with that goal in mind, the same goes for LSPs.

Hence, I wouldn’t be surprised if many translators take the same approach as Charlie’s dad in the chocolate movie taking on a job at fixing the robots that made him loose his job in the first place.
Knowledge is everything.

Be safe


 
Merab Dekano
Merab Dekano  Identity Verified
Spain
Member (2014)
English to Spanish
+ ...
MT Aug 21, 2020

I do some MT jobs here and there and I do accept that the technology is here to stay.

Now, where the today’s technology stands, you will ALWAYS have a poorer text with an edited MT output if you compare it to a translation and editing done by professional human translators and reviewers.

When reviewing an MT output, you are limited to a style and syntax-related choices “made” by the machine. You can still correct the text, but you cannot completely delete the segm
... See more
I do some MT jobs here and there and I do accept that the technology is here to stay.

Now, where the today’s technology stands, you will ALWAYS have a poorer text with an edited MT output if you compare it to a translation and editing done by professional human translators and reviewers.

When reviewing an MT output, you are limited to a style and syntax-related choices “made” by the machine. You can still correct the text, but you cannot completely delete the segment and write the translation as you would say it (unless you decide to retranslate the text, which you will not be paid for and most likely will not have enough time to deliver the project if you retranslate every single segment).

For example, if you have a phrase like this “All third parties are bound by this rider”, a typical MT output will throw at you something like this: “Todas las terceras fiestas están atadas al conductor”. If you speak some Spanish, you will realize that this is quite easy segment to correct (Todo tercero queda vinculado por esta cláusula), but when you have a large segment with several subordinate clauses, you will start having problems “correcting” the segment, until you realise that you would be better off if you did it from scratch, and you don’t have time to do many segments from scratch...

Today, MT saves money and takes out quite some quality. Some clients accept it. Some don't. An important contract done by a machine and edited by a human in rush is a ticking bomb. A product description done by a machine and edited by a human in rush is simply unappealing and may put off some potential clients. A weather forecast done by a machine and edited by a human in rush is just fine.

[Edited at 2020-08-21 09:06 GMT]
Collapse


Philippe Locquet
Beatriz Ramírez de Haro
 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 09:43
English to French
+ ...
TOPIC STARTER
Bias Aug 21, 2020

Merab Dekano wrote:

I do some MT jobs here and there and I do accept that the technology is here to stay.

Now, where the today’s technology stands, you will ALWAYS have a poorer text with an edited MT output if you compare it to a translation and editing done by professional human translators and reviewers.

When reviewing an MT output, you are limited to a style and syntax-related choices “made” by the machine. You can still correct the text, but you cannot completely delete the segment and write the translation as you would say it (unless you decide to retranslate the text, which you will not be paid for and most likely will not have enough time to deliver the project if you retranslate every single segment).

For example, if you have a phrase like this “All third parties are bound by this rider”, a typical MT output will throw at you something like this: “Todas las terceras fiestas están atadas al conductor”. If you speak some Spanish, you will realize that this is quite easy segment to correct (Todo tercero queda vinculado por esta cláusula), but when you have a large segment with several subordinate clauses, you will start having problems “correcting” the segment, until you realise that you would be better off if you did it from scratch, and you don’t have time to do many segments from scratch...

Today, MT saves money and takes out quite some quality. Some clients accept it. Some don't. An important contract done by a machine and edited by a human in rush is a ticking bomb. A product description done by a machine and edited by a human in rush is simply unappealing and may put off some potential clients. A weather forecast done by a machine and edited by a human in rush is just fine.

[Edited at 2020-08-21 09:06 GMT]


That’s a very good observation you make there. As I mentioned in the other thread, output depends a lot on language pair, field and machine programming. So the quality the translator experiences while translating depends a lot on these factors.
What you described here is bias, many MTs tend to want to output the same thing all the time. Although this has been fixed in major engines for some language pairs/domains, it still happens. This is mainly due to the training data given to the machine. It will output according to its training. A bit like if you use an automotive TM to translate a medical document and run concordance, it’s unlikely your hits will match your needs.
Hence the process I described in the other thread to make a decision on which MT and to make sure you get quality for your investment after human editing. I’ve seen some very convincing outputs in En2Fr in the news domain. Some segments did not require editing at all. So that goes to show you can get all the spectrum in outputs ranging from absolute rubbish to no need for editing in some cases.

Thanks for the insights!

My bests


 
RWS Community
RWS Community
United Kingdom
Local time: 10:43
English
What do you base this on? Aug 21, 2020

Merab Dekano wrote:

Now, where the today’s technology stands, you will ALWAYS have a poorer text with an edited MT output if you compare it to a translation and editing done by professional human translators and reviewers.



There are studies that have been carried out that completely dispute this and find that post-edited MT is more often than not a better quality than human translated material alone. So I'm wondering, out of genuine interest, where the evidence for your statement comes from?

Regards

Paul.


Neco3104
 
DZiW (X)
DZiW (X)
Ukraine
English to Russian
+ ...
The higher technology, the lower life Aug 21, 2020

Philippe, the human communication is about mutual understanding with a certain language, register, audience, purpose, euphony, relations, biases-associations, and other stylistic means and devices. A machine only has simplified algorithms with a database, producing a surrogate GMO? translation. Why, I know several mid-big companies, which have been successfully using customized rule-based MT (Promt and some others) for internal papers since 2006. However, all important papers are o... See more
Philippe, the human communication is about mutual understanding with a certain language, register, audience, purpose, euphony, relations, biases-associations, and other stylistic means and devices. A machine only has simplified algorithms with a database, producing a surrogate GMO? translation. Why, I know several mid-big companies, which have been successfully using customized rule-based MT (Promt and some others) for internal papers since 2006. However, all important papers are only for competent engineers, lawyers, and bookkeepers respectively.

For ages, numerous perturbations caused the lack of real specialists [engineers, programmers, lawyers, doctors...] with decent foreign language skills, alas. Besides mass appearance of ‘pure’ translators, it also provoked flourishing of spongers, predators, profiteers, and others. That’s also ok.

Indeed, everything depends on who is evaluating the risks-benefits ratio with what priorities in mind. It is very similar for end clients (real businesses), middlemen, and translators—yet with opposite accents. The beauty of worth and value is in the eyes of the beholder... in perspective: Better for one is far not always better for another.

I believe even very lazy (narrow/one-specialty) people must keep learning, adapting to the new rules and demands. For what it’s worth, It’s fine that such trends make ‘pure’ translators reconsider the diversifications and prospects. Count me in!)


Paul, unlike machines, people have the needs and purposes to communicate. So. by the very design a machine is always but a lame attempt to imitate the depersonalized processes. Furthermore, most PEMTors just try to cheaply beautify algorithm-generated gibberish—even without seeing the original. The average result is obvious, yet dung beetles and bottom-feeders aka hobbyists (under-centers/ earning less than McDonald's guy) also need something to eat.
Collapse


Philippe Locquet
 
Merab Dekano
Merab Dekano  Identity Verified
Spain
Member (2014)
English to Spanish
+ ...
Evidence Aug 21, 2020

SDL Community wrote:

Merab Dekano wrote:

Now, where the today’s technology stands, you will ALWAYS have a poorer text with an edited MT output if you compare it to a translation and editing done by professional human translators and reviewers.



There are studies that have been carried out that completely dispute this and find that post-edited MT is more often than not a better quality than human translated material alone. So I'm wondering, out of genuine interest, where the evidence for your statement comes from?

Regards

Paul.


The "evidence" is coming from my own experience. Unfortunately, I cannot quantify it or carry out any study for that matter.


 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 09:43
English to French
+ ...
TOPIC STARTER
Real word and studies Aug 21, 2020

DZiW wrote:

Philippe, the human communication is about mutual understanding with a certain language, register, audience, purpose, euphony, relations, biases-associations, and other stylistic means and devices. A machine only has simplified algorithms with a database, producing a surrogate GMO? translation. Why, I know several mid-big companies, which have been successfully using customized rule-based MT (Promt and some others) for internal papers since 2006. However, all important papers are only for competent engineers, lawyers, and bookkeepers respectively.

For ages, numerous perturbations caused the lack of real specialists [engineers, programmers, lawyers, doctors...] with decent foreign language skills, alas. Besides mass appearance of ‘pure’ translators, it also provoked flourishing of spongers, predators, profiteers, and others. That’s also ok.

Indeed, everything depends on who is evaluating the risks-benefits ratio with what priorities in mind. It is very similar for end clients (real businesses), middlemen, and translators—yet with opposite accents. The beauty of worth and value is in the eyes of the beholder... in perspective: Better for one is far not always better for another.

I believe even very lazy (narrow/one-specialty) people must keep learning, adapting to the new rules and demands. For what it’s worth, It’s fine that such trends make ‘pure’ translators reconsider the diversifications and prospects. Count me in!)


Paul, unlike machines, people have the needs and purposes to communicate. So. by the very design a machine is always but a lame attempt to imitate the depersonalized processes. Furthermore, most PEMTors just try to cheaply beautify algorithm-generated gibberish—even without seeing the original. The average result is obvious, yet dung beetles and bottom-feeders aka hobbyists (under-centers/ earning less than McDonald's guy) also need something to eat.


At this point I’m not sure if we should keep this conversation here or move it to this thread https://www.proz.com/forum/machine_translation_mt/345577-comparing_mt_engines_which_mt_should_i_choose.html)

I find that discussing MT ends up very often in the same topics and misunderstandings.
Often, MT programmers only see the mechanics of what they made, and translators have a hard time seeing what are the adequate use cases. Frustration is emphasized when LSPs use it in a way that doesn’t make sense.
These are some of the reasons that pushed me to spread some facts on threads and in my YT videos.

MT is a tool and with any trade, the results depend on the tool quality but even more on the craftsmanship of those using such tools.

MT studies:
On that topic, things depend a lot on the methods and metrics. Before pointing to a study, I think looking at the aspects of methods and metrics is very important when evaluating MTs. Some automated metrics are very often used but when compared with real-world workflows, the distance between test results and actual efficiency can be widened.
I don’t consider myself an MT expert but I certainly try to learn as much as possible and test as much as I can so I can master these tools where they should fit for different scenarios.

DZiW,
You mentioned PBMT, this is a kind of model that is being used less and less. RBMT is still used for Low-Res languages, and then now you have many types of models (NMT, Transformers etc.). Translators do not necessarily need to know these details, but some models can be a tell on how modern the architecture is and on how much data it has to be based upon. Basically if you combine a modern architecture with abundant high quality training data, your results should be good.

And research is still going forward. As Paul knows, a new type of MT already exists and further developments are made as we speak for what is called “Fuzzy Match Repair”. In simple terms, it is supposed to improve TM output. So that kind of research shows you where things are leading, towards reducing the effort the human translator has to produce thus increasing productivity while having a quality output.

If there’s interest, maybe one day I’ll do a training on ProZ about MT and how and when should translators and LSPs use it.

I think MT is an interesting topic to look at anyway.

Be well


 
DZiW (X)
DZiW (X)
Ukraine
English to Russian
+ ...
Wolves eat sheep Aug 21, 2020

Philippe, thanks for your reply.

For what it's worth, as far as scientists still cannot explain even how their brains work specifically, they hardly could teach a machine imitate it alright using any tricks and gimmicks.

The more [business] educated freelancers, the more independent and prospering the market.


If you took part in the experiment, let us hear about your findings and ideas.

Cheers

[Edited at 2020-08-21 18:56
... See more
Philippe, thanks for your reply.

For what it's worth, as far as scientists still cannot explain even how their brains work specifically, they hardly could teach a machine imitate it alright using any tricks and gimmicks.

The more [business] educated freelancers, the more independent and prospering the market.


If you took part in the experiment, let us hear about your findings and ideas.

Cheers

[Edited at 2020-08-21 18:56 GMT]
Collapse


 


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


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

Call for volunteers rewarded by free MT and visibility!






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 »
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 »