Workflow science
Thread poster: Philippe Locquet
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 17:11
English to French
+ ...
Dec 17, 2020

Hi all, I’ve noticed that there is not a dedicated thread to discuss workflows that are effective or those that are a pain. It is an important topic because the way workflows are designed impacts greatly the time and efforts freelancers and other colleagues will have to invest in each project.
By workflow I mean all the processes and actors involved in the translation process from the end client’s request to the delivery of the translated work.

There will be differences be
... See more
Hi all, I’ve noticed that there is not a dedicated thread to discuss workflows that are effective or those that are a pain. It is an important topic because the way workflows are designed impacts greatly the time and efforts freelancers and other colleagues will have to invest in each project.
By workflow I mean all the processes and actors involved in the translation process from the end client’s request to the delivery of the translated work.

There will be differences between scenarios like:

_Lone freelancer dealing directly with end client
_Small LSP using only CAT software
_LSP will fully integrated system (TMS, DTP etc.)
_LSP with MT-based workflows

I don’t know if in time these will need to be separated in different threads, we’ll see how this goes I suppose 😊.
Collapse


 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 17:11
English to French
+ ...
TOPIC STARTER
State of the matter: Dec 17, 2020

I have noticed than in each of the above-mentioned situations there is rarely any science to the way workflows are built. Actors are often influenced in their decisions by believing that a specific tool with somehow magically make everything better, or they rely on personal experiences/views.
Since the number of options made available to actors is ever increasing and the technology is ever evolving, not all are well understood. The result is that decisions made and personnel training seem
... See more
I have noticed than in each of the above-mentioned situations there is rarely any science to the way workflows are built. Actors are often influenced in their decisions by believing that a specific tool with somehow magically make everything better, or they rely on personal experiences/views.
Since the number of options made available to actors is ever increasing and the technology is ever evolving, not all are well understood. The result is that decisions made and personnel training seem to be inadequate.

So, translators end-up having to do funky exotic things to adapt and meet each LSPs criteria.
Have you felt it?

Which workflow do you think is best in a given situation?
Collapse


 
Christopher Schröder
Christopher Schröder
United Kingdom
Member (2011)
Swedish to English
+ ...
Clarification please Dec 17, 2020

I’m not entirely sure what you are getting at here...

My preferred work flow is:

1. Can you do this
2. Yes
3. Here it is
4. £££


Maria Teresa Borges de Almeida
Philip Lees
Michele Fauble
matt robinson
Kay Denney
Vesa Korhonen
P.L.F. Persio
 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 17:11
English to French
+ ...
TOPIC STARTER
Breakdown Dec 17, 2020

Chris S wrote:

I’m not entirely sure what you are getting at here...

My preferred work flow is:

1. Can you do this
2. Yes
3. Here it is
4. £££





JJJJJ Love this!

Ok, I'll try to clarify

Here's an overextensive workflow example with decisions at each step:

Files ingestion, platform? Cloud? - Preprocessing of files y/n? - MT pretranslation y/n - Use TMS y/n - how many translation steps (i.e. three: 1/translation, 2/editing, 3/QA)? -use bilingual review y/n - send work to DTP team y/n? - delivery to customer, how?

_Internal review and evaluation of translator's work?
_System tied to one CAT tool or interoperability to a degree?
_Preparation of TM y/n? Align past work y/n? (same for Glossaries)
_Centralised TM system with sharing?
etc.

There are pain points at each of these steps and choices have to be made. Sometimes, the use of a specific tool with narrow a lot the options for translators to adapt, jail-breaking for which will be difficult.

Translators don't have the means to by all CATs, not all LSPs can provide temporary licenses to their linguists. Then there is the issue of the learning curve and training.
Another issue is some new platforms that LSP use where ergonomics are horrendous for translators.

All these choices have a tremendous impact, so I think it's interesting to explore our experiences and feedback with a given workflow. No need to name outsourcers, but I think we mentioning the tools used would be important.

Hope this clarifies


Christopher Schröder
 
Anton Konashenok
Anton Konashenok  Identity Verified
Czech Republic
Local time: 18:11
French to English
+ ...
No need to complicate things Dec 17, 2020

My workflow:
Sorry, no online tools and no MT. For everything else, you state your requirements and I quote my price.

Regarding the learning curve and training, an adequate professional tool in the hands of an adequate professional user should not take more than 15 minutes to start basic production work. Everything else can be looked up in the manual as you go. If it takes longer or becomes a constant source of irritation, then either the tool or the user isn't up to spec.


Christopher Schröder
Maria Teresa Borges de Almeida
Philip Lees
Vesa Korhonen
P.L.F. Persio
Peter Shortall
 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 17:11
English to French
+ ...
TOPIC STARTER
Real life examples Dec 18, 2020

I’ll give a personal example of what I’m getting at:

I’ve worked a lot with a company that holds high ISO certifications for translation quality etc.
Their workflow involves in-house PMs and in-house DTP.

Human steps:
1/Translation (reviews doc and applies changes from step 3)
2/Proofreading (corrects and give quality ratings for 1)
3/QA (evaluates quality from 2 by giving sample examples of error types)
4/DTP (fixes doc look (langua
... See more
I’ll give a personal example of what I’m getting at:

I’ve worked a lot with a company that holds high ISO certifications for translation quality etc.
Their workflow involves in-house PMs and in-house DTP.

Human steps:
1/Translation (reviews doc and applies changes from step 3)
2/Proofreading (corrects and give quality ratings for 1)
3/QA (evaluates quality from 2 by giving sample examples of error types)
4/DTP (fixes doc look (language expansion etc.)
The PMs would very often swap translators so all have a go at each step and get a review to keep everyone on their toes.

Technical steps:
Source files + examples + xliff (SDL) + TM (sometimes) + instructions, all made available on TMS Plunet for steps 1 to 3.

The strength of this workflow is the high quality it produces and how well it brings to light issues. Besides, the translator can engage in this process with the tool of his/her choice as TMs and xliffs are made available separately. TMs can be converted and sdlxliif is accepted in many CATs.
The problem with this workflow is the efforts required by the translator on step 1 that will have to review and fix his/her work based on step 3 review. Being paid by wordcount, you need to factor in the extra time.

Later, they decided to alter this process by making it tool-dependant (memoQ online). So all steps are now the same with Plunet and memo-Q online. Thus, they solve terminology and TM by centralising it.
New strengths of this process: CAT is provided to you at no cost, memoQ is quite easy to learn, centralised solution for TM and glossaries.
Weaknesses of this process: Learning curve (small I agree), extremely difficult to have you perform at your best (let me explain). Indeed, if you translate highly specialized material, besides relying on the client terminology, you will have your glossaries that you built and double or triple-checked over the years, same for your TMs and concordance checks you can perform. Beside relying on solely your brain memory, to access these precious files you built, you need to go out of the LSP provided system open other things up and go fishing. This can be done. However, I would argue that this is highly time consuming and not ergonomic at all. If rates are tight, I would not be surprised if many translators would decide to accept other jobs rather than this so they can excel in a different workflow.

So, for me this brings the point across, what the LSP did to fix some issues in their workflow created new ones. As a translator I’m impacted and have to make to choices. So, for an LSP, such a change can also trigger a change in the pool of translators they rely on.
That’s only one example, I’m sure there are many more out there.
Collapse


 
Christopher Schröder
Christopher Schröder
United Kingdom
Member (2011)
Swedish to English
+ ...
Varies hugely Dec 18, 2020

This is going to vary according to your languages, your capabilities, your customer and the text type.

My preferred workflow (receive Word file, send Word file back, get paid) works well for my market because (1) these jobs only go into English and (2) I am good at what I do. You could also add a checking step in there if it's something critical.

However, I also participate from time to time in a hugely overcomplicated workflow similar to the one you describe:
... See more
This is going to vary according to your languages, your capabilities, your customer and the text type.

My preferred workflow (receive Word file, send Word file back, get paid) works well for my market because (1) these jobs only go into English and (2) I am good at what I do. You could also add a checking step in there if it's something critical.

However, I also participate from time to time in a hugely overcomplicated workflow similar to the one you describe:

1. English text is translated into 20+ languages
2. Translation is back-translated into English
3. Third person compares back-translation with original and lists discrepancies
4. Forward translator reviews list and blames back translator
5. Back translator reviews list and overrides forward translator
6. Forward translator inputs corrections in list AND in text

All this is co-ordinated from Europe and project-managed from India, of course, for a US agency that pays via an Irish subsidiary.

Both translations are done in the agency's own software which is riddled with annoying tags. Their quality assurance software then picks up ten times more false negatives than positives (e.g. cannot cope with addresses staying the same). Steps 3-6 are added manually to an ever-expanding Excel file which can have 20 columns of text. Every e-mail is copied to four different people at the agency.

While hideously complex, this is theoretically foolproof and doubtless ticks all the boxes of the drug companies who don't want to pay out £££ in compensation if someone mistranslates what to do with their drugs.

Only it doesn't work.

True story:
English original mentions "application", as in request. Forward translator translates it as "program", as in software. Back translator (who doesn't get to see the original English) translates it back as "application". Hideous error gets through and patients die.

OK, the very last bit wasn't true. Back translator guessed the error from the context and reported it (but had to justify it on three separate occasions before anyone listened).

Wouldn't it be easier, quicker, cheaper and all-round better to pay for a decent translator to take their time and do it properly in the first place, in line with my preferred workflow?
Collapse


Philippe Locquet
Vi Pukite
Beatriz Ramírez de Haro
Kay Denney
Stephanie Busch
P.L.F. Persio
Agneta Pallinder
 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 18:11
Member (2006)
English to Afrikaans
+ ...
Another example of how it breaks Dec 18, 2020

Chris S wrote:
While hideously complex, this is theoretically foolproof and doubtless ticks all the boxes of the drug companies who don't want to pay out £££ in compensation if someone mistranslates what to do with their drugs.


My reply here may be off-topic, I don't know.

I often work for a client whose workflow is supposed to work like this:
- The agency pre-translates (leverages) file against their TM. The file has empty segments, fuzzy matches, 100% matches and ICE/101/100+ matches.
- The file goes to the translator (T) who translates it (and has the option to connect to the client's TM online).
- The file then goes to the first proofreader (R1) who checks the translation for meaning related errors.
- The file then goes to the second proofreader (R2) who checks the translation for formatting related errors.

The problem is that the R1 proofreader is an industry expert whose language skills aren't great, so he tends to make minimal edits. And the R2, who is supposed to be a secondary check, ends up doing the bulk of the editing. In addition, in some jobs, the translator is not responsible for checking 100% matches or even fuzzy matches. The R1 proofreader must check the fuzzy matches and tags, but often doesn't. So the R2 proofreader (you can guess... it's often me) gets to fix the unedited fuzzy matches and misplaced tags, lots of tags.

The upshot for me as R2 proofreader is/was that my preferences end up in the client's TM. Until two translators discovered how to store their own TMs offline and connect it to the client's files. So now, it often works like this:
- The agency pre-translates (leverages) file against their TM, yada-yada.
- The file goes to the translator (T) who connects to his own TM, overwriting most of the 100% and fuzzy matches with his own translations.
- The file then goes to the first proofreader (R1), yada-yada.
- The file then goes to the second proofreader (R2) who discovers a mishmash of translation styles consisting of the translator's own TM and whatever remained from the client's TM.

It used to be that I could jump right in an check the accuracy of the translation, but now the first task of the R2 is to manually revert dozens of terms to match the client's glossary (since they are not the terms used in the translators' own TMs). It's fantastic if the R2 translations make it to the client's TM, but only if the translators actually use the client's TM.

The translators are happy because their own TMs are much more comprehensive than the client's TM, so they earn more money for less work. But these projects come in batches with lots of files that need to be consistent, and unless they all have the same R2 working on the project, the final files are going to be nothing to write home to about. In fact, with COVID, there is often a dozen or more different Ts, R1s and R2s working on what are often very similar versions of files.

In a variant to the workflow above, the client might stipulate that the translator should not edit pre-translated 100% matches unless there is a good reason and a comment is left. But our dear translators with their own TMs merrily overwrite the client's 100% segments with their own, and this means that R1 and R2 are left with the task of examining every edited 100% match to determine if it really needed to be changed, and then manually reverting each segment and writing a comment about why.

You mention drug companies, Chris... in the projects that I work in, there are dozens of "clients" (independent committees who must approve both source texts and translations) who essentially have identical texts to translate. The original source text is written by some super regulatory body, and is sent to each independent committee who makes changes to it based on their own, duplicated expertise, and then the agency gets to translate these texts that all look remarkably similar, but because of the rush, they use multiple translator teams who (surprise!) translate the files in very different ways. Sure, the agency has a TM, but in a very short time the TM is filled with trusted junk. And because of the rush, committees' files #2 and file #3 aren't sent to the same translators who did file #1 (this isn't necessary, no, because the way the system was designed, is flawless and ISO compliant etc.), so in the end the poor users of a region get files that were translated by multiple teams who each have their own bands of purists in them. Imagine a set of three software user manuals (one for users, one for trainers, one for for administrators), and the software itself, and promotional materials, all translated by different people... and the workflow is ISO compliant and thus perfectly safe.

And just when you thought it couldn't get worse, there are small updates that are done in tracked changes, by translators who weren't involved in the original files, so now even within the file you get multiple term translations and preferences, sometimes with heated debates going on in the comment sections by translators who believe that they are right and you are wrong. And you know it's going to be one of those days when the source file has 10 edits in scope but the target file comes back with 200, with comments. Long live MS Word macros that go next edit, reject edit, next edit, reject edit, etc. I sleep like a baby.

[Edited at 2020-12-18 17:14 GMT]


Christopher Schröder
Philippe Locquet
Vi Pukite
Beatriz Ramírez de Haro
Natalia Novichenko
 
Christopher Schröder
Christopher Schröder
United Kingdom
Member (2011)
Swedish to English
+ ...
Lol Dec 18, 2020

Samuel Murray wrote:
Big Momma TM

Love it!


 
Samuel Murray
Samuel Murray  Identity Verified
Netherlands
Local time: 18:11
Member (2006)
English to Afrikaans
+ ...
@Chris Dec 18, 2020

Chris S wrote:
Samuel Murray wrote:
Big Momma TM

Love it!

AFAIK, "big momma TM" is an established industry term. It refers to a single TM that contains translations from multiple jobs and multiple clients. Such TMs are popular with translators who don't know how to manage TMs or who use CAT tools that don't allow more than one TM simultaneously. The term is not always meant pejoratively. But you're right: it's a nice term.


Christopher Schröder
 
Philippe Locquet
Philippe Locquet  Identity Verified
Portugal
Local time: 17:11
English to French
+ ...
TOPIC STARTER
Rough on the translator Dec 18, 2020

Samuel Murray wrote:


- The file then goes to the second proofreader (R2) who discovers a mishmash of translation styles consisting of the translator's own TM and whatever remained from the client's TM....

But these projects come in batches with lots of files that need to be consistent, and unless they all have the same R2 working on the project, ....

In a variant to the workflow above, the client might stipulate that the translator should not edit pre-translated 100% matches unless there is a good reason and a comment is left. But our dear translators with their own TMs merrily overwrite the client's 100% segments with their own, and this means that R1 and R2 are left with the task of examining every edited 100% match to determine if it really needed to be changed, and then manually reverting each segment and writing a comment about why.



@Samuel and @Chris
Thanks for sharing these experiences. Both show exactly what I'm getting at, if the workflow is not well-thought, well put in place and well understood you get problems (which in nearly 100% of the time when you're not running a one man show). Problems where it makes work heavy, long and frustrating for the translator. Problems that don't get the LSP the style and terminology consistency they would like.

Samuel, what is the online solution the company you mentioned uses (if I may ask)?

I don't know all systems, but in some centralized systems, you can handle that kind of confusion.
I would go about it with injecting attributes to TUs. If TUs where all produced in the same tool, it should be possible to find the author. Then someone on the LSP side should make sure that their trusted TM and Glossary should have 1 central reference author by overwriting attributes.

Then translator's machines/online accounts could be setup with a penalty. So they never get 100%. They're supposed to check anyway, so adding a split second by segment on a whole job shouldn't be too bad, especially compared having to change entries or justify etc.

Somehow, I get the felling that LSP's workflow often are an orchestra with instruments but no conductor...


Christopher Schröder
 
Vi Pukite
Vi Pukite  Identity Verified
United States
Local time: 10:11
Latvian to English
+ ...
@Chris S - "true story" Dec 23, 2020

You've hit the nail on the head as to why back translation does not work. I'm still mystified as to why it is still used so much.

Christopher Schröder
Beatriz Ramírez de Haro
Kay Denney
Philippe Locquet
 
Kay Denney
Kay Denney  Identity Verified
France
Local time: 18:11
French to English
. Dec 24, 2020

I'm exhausted just reading this!
IME any attempt to harness technology in order to "make things easier on the translator" (aka pay the translator less) generates other problems further down.

At one point I was managing a project using software called Frame Maker I think? did that ever exist?
The guy writing the manuals in the source language was a total nerd who wanted to exploit the software to the full. He devised a nifty system whereby his jargon would be automatica
... See more
I'm exhausted just reading this!
IME any attempt to harness technology in order to "make things easier on the translator" (aka pay the translator less) generates other problems further down.

At one point I was managing a project using software called Frame Maker I think? did that ever exist?
The guy writing the manuals in the source language was a total nerd who wanted to exploit the software to the full. He devised a nifty system whereby his jargon would be automatically translated, so that it was always the same term being used, for Consistency's sake. Then the translator merely had to translate all the "fluff" (aka grammar) around said jargon.

Did I mention that the client insisted on us using Trados?
Unfortunately, Trados didn't work with FM, so I had to convert the files into mifs and then pifs and then finally was able to get into the files using Trados. And hey presto all the nifty jargon thingies turned into clunky illegible tags, so a typical sentence would read something like

In order to {tag}, you'll need to first open the {tag} then {tag} the file.
Collapse


Christopher Schröder
Philippe Locquet
 
Christopher Schröder
Christopher Schröder
United Kingdom
Member (2011)
Swedish to English
+ ...
Lol again Dec 24, 2020

Love the capital C in “for Consistency’s sake”!

It’s amazing how much can be said with so little.

Presumably, though, while the new technology and systems tend to be over-complex and ineffective, the average translation must have been even worse without them...

While CAT tools seem to perpetuate bad writing, correct terminology at least is more likely to be used.

[Edited at 2020-12-24 09:43 GMT]


 


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


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

Workflow science







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 »
CafeTran Espresso
You've never met a CAT tool this clever!

Translate faster & easier, using a sophisticated CAT tool built by a translator / developer. Accept jobs from clients who use Trados, MemoQ, Wordfast & major CAT tools. Download and start using CafeTran Espresso -- for free

Buy now! »