Trados v Wordfast analysis - large discrepancy
Thread poster: Clive Phillips

Clive Phillips  Identity Verified
United Kingdom
Local time: 16:52
Member (2009)
German to English
+ ...
Feb 27, 2014

The TRADOS analysis quoted by a translation agency offering a job:
Repetitions 610 words; 100% 8038; 95-99% 919; 85-94% 643; 0-84% 2207.

My Wordfast analysis:
Repetitions 1764 words; 100% 3375; 95-99% 4337; 85-94% 716; 0-84% 1897.

The agency claims that "In contrast to our Trados analysis, the Wordfast analysis counts a lot of 95-99% matches. Now, this high number will mostly contain matches where a tag is wrong but it mostly won’t require any new translation. Hence, could we agree on a price of €x for those?"

What concerns me is the discrepancy in the fuzzy matches.

Do you agree with the agency's assertion? Any comments or suggested negotiation stance please.


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 17:52
Member (2006)
English to Afrikaans
+ ...
One thought, and another Feb 27, 2014

Clive Phillips wrote:
The TRADOS analysis quoted by a translation agency offering a job:
Repetitions 610 words; 100% 8038; 95-99% 919; 85-94% 643; 0-84% 2207.

My Wordfast analysis:
Repetitions 1764 words; 100% 3375; 95-99% 4337; 85-94% 716; 0-84% 1897.


It could be that your WFP analysis applies more heavy penalties than their Trados analysis.

What file type is the source text in?
Which program created the TM against which the files are analysed?


Direct link Reply with quote
 

Clive Phillips  Identity Verified
United Kingdom
Local time: 16:52
Member (2009)
German to English
+ ...
TOPIC STARTER
Trados/Wordfast analysis Feb 27, 2014

Samuel Murray wrote:

It could be that your WFP analysis applies more heavy penalties than their Trados analysis.

What file type is the source text in?
Which program created the TM against which the files are analysed?


I have the default TM penalties in WFP.
The source text file type is prep.tag.ttx for conversion by WFP and the agency has also provided the source text as a .pdf file for reference purposes.
I assume that the agency used TRADOS to create the TM against which the files are analysed.


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 17:52
Member (2006)
English to Afrikaans
+ ...
Well, there's the answer... Feb 28, 2014

Clive Phillips wrote:
1. The source text file type is prep.tag.ttx...
2. I assume that the agency used TRADOS to create the TM...


So, the file is not a WFP native file but it is a Trados native file, and it is likely that the TM (and accompanying tags in the segments) match the tagging that is used in the source text file. WFP, on the other hand, has to convert the file (and its tags) to its own format, and it re-tags the text differently. This means that a segment in WFP might be tagged in a way that is different from the way that very segment is tagged in the TM.

Here is a quick test to illustrate it

I downloaded ProZ.com's main page, converted it to TTX using Trados 2007, and pre-translated it (source=target) and then cleaned up the file so that all segments were added to the TM. In other words, the TM is a 100% match for the document. The clean-up analysis shows exactly what we would expect:

Total number of segments: 666
Total word count: 2962
Number of non-repeating segments: 477
Word count of non-repeating segments: 2157


Now, if WFP tagged exactly the same way as Trados, then it shold be possible to load both the TTX file and the TM into WFP, do an analysis, and get all 100% matches, right?

Alas, WFP's analysis shows this (without even calculating internal matches):

Golden segments 426
Golden words 1340

Repetitions segments 36
Repetitions words 62

100% (words) 1340
95%-99% (words) 1168
85%-94% (words) 418
75%-84% (words) 5
50%-74% (words) 0
No Match (words) 36

Total segments 677
Total words 3029


Even if I deselect all penalties in WFP before running the analysis, I still get 443 words of high fuzzy matches, 55 words of medium high fuzzy matches, and 36 words of no matches. This means that the penalties don't account for all the differences.

And the penalties are there to protect you. Without the penalty, a segment like [Add 2 eggs and 300 grams of sugar] will be considered a 100% match with a segment like [Add 100 eggs and 2 grams of sugar], and a segment like [He heard a loud bang] would be considered a 100% match with a segment like [He heard a loud BANG].


[Edited at 2014-02-28 09:02 GMT]


Direct link Reply with quote
 

Clive Phillips  Identity Verified
United Kingdom
Local time: 16:52
Member (2009)
German to English
+ ...
TOPIC STARTER
Differing analyses affect price Feb 28, 2014

[quote]Samuel Murray wrote:
...a segment in WFP might be tagged in a way that is different from the way that very segment is tagged in the TM.
...the penalties don't account for all the differences.

Thank you, Samuel, for the revealing comparison exercise.

Let us assume (the assumption would require more extensive testing) that differences in tagging methodology will always lead to wide disparity between a Trados pre-translation analysis and a WFP pre-translation analysis.

I have used Wordfast satisfactorily for several years and don't wish to learn/switch to Trados, despite the agency's urgings. I don't wish to drop this valued customer.

Since, with many agencies like this regular customer, analysis forms the basis for calculation of the translation price, my main concern and suspicion are that if I accept the agency's Trados analysis I may not be paid for non-tag-related fuzzy matches which are counted in WFP but not in Trados. In short, I would be doing unpaid work. The calculated price based on WFP analysis in this case is over 40% higher than the agency's offer.

Any suggestions on how I might seek to negotiate a compromise between the agency's Trados analysis and my WFP analysis?


Direct link Reply with quote
 

neilmac  Identity Verified
Spain
Local time: 17:52
Spanish to English
+ ...
Borrow my bargepole Mar 1, 2014

I'm afraid I wouldn't even dream of trying to "negotiate a compromise" - I'd simply tell them to take their custom elsewhere. I see agencies that operate this way as chisellers and avoid them like the plague.

Direct link Reply with quote
 
xxxnrichy
France
Local time: 17:52
French to Dutch
+ ...
Thanks Samuel for the explanation Mar 1, 2014

Samuel Murray wrote:
...a segment in WFP might be tagged in a way that is different from the way that very segment is tagged in the TM.
...the penalties don't account for all the differences.


Clive Phillips wrote:

I have used Wordfast satisfactorily for several years and don't wish to learn/switch to Trados, despite the agency's urgings. I don't wish to drop this valued customer.

Since, with many agencies like this regular customer, analysis forms the basis for calculation of the translation price, my main concern and suspicion are that if I accept the agency's Trados analysis I may not be paid for non-tag-related fuzzy matches which are counted in WFP but not in Trados. In short, I would be doing unpaid work. The calculated price based on WFP analysis in this case is over 40% higher than the agency's offer.

Any suggestions on how I might seek to negotiate a compromise between the agency's Trados analysis and my WFP analysis?


Despite the fact that I like very much Wordfast, I drew the conclusion that WF files have to be treated with WF, and Trados files with Trados. Conversion tasks always mean supplementary difficulties and repercussions on the time I spend on the file and therefore also on the price. So I swallowed my pride and bought Studio. Since then, I am directly compatible with two of my main clients and can say that, after a three months adaptation to new and heavy procedures and conversion of my TMs, it is now quite comfortable: file comes in, is translated and verified, and goes out.


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 17:52
Member (2006)
English to Afrikaans
+ ...
@Clive Mar 2, 2014

Clive Phillips wrote:
Let us assume (the assumption would require more extensive testing) that differences in tagging methodology will always lead to wide disparity between a Trados pre-translation analysis and a WFP pre-translation analysis.


Well, it's more than just that. As far as I can tell, Wordfast has been optimised for speed by reducing the number of characters to check during matching, whereas Trados is not (or less so). If you have this text:

One two three 456
One TWO three 567
one Two Three 890

...then Wordfast will add these segments to the TM only once, namely as "One two three 456". This is why the penalties are so important in Wordfast. Trados will add all three those segments to the TM, as three separate segments.

If you then take the Trados TM with the three segments and convert them to a Wordfast TM, Wordfast will ignore two of those segments during analysis (and during translation as well).

So, if a Trados TM contains all three those segments, then the analysis will show 3 segments of 100% match, but Wordfast will penalise two of those segments (by e.g. 1%), and so Wordfast will show 1 segment of 100% match and 2 segments of 99% match.

I'm writing from memory here... if things have changed, someone please correct me.

If I accept the agency's Trados analysis I may not be paid for non-tag-related fuzzy matches which are counted in WFP but not in Trados. In short, I would be doing unpaid work. The calculated price based on WFP analysis in this case is over 40% higher than the agency's offer.


I have learnt to grin and bear it. I would be so much slower using Trados, that I simply decide to work for less for such clients. You can't really expect the client to pay a lot more for the translation, because the reason why your word count is so much higher is because your tool is incapable of making full use of the TM provided by the client.


Direct link Reply with quote
 

Samuel Murray  Identity Verified
Netherlands
Local time: 17:52
Member (2006)
English to Afrikaans
+ ...
@Neilmac Mar 2, 2014

neilmac wrote:
I'd simply tell them to take their custom elsewhere. I see agencies that operate this way as chisellers and avoid them like the plague.


Isn't it more the translators who are the chisellers? Convert this case from a per-word scenario to a per-hour scenario. The client hires a typist and tells him that he calculates that it would take 10 hours do to the typing job, but the typist says "I prefer to do these jobs while sitting on my one hand, so for me it will take 20 hours to do the typing... and I expect you to pay for 20 hours."


Direct link Reply with quote
 


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


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

Trados v Wordfast analysis - large discrepancy

Advanced search


Translation news related to Wordfast





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 »
BaccS – Business Accounting Software
Modern desktop project management for freelance translators

BaccS makes it easy for translators to manage their projects, schedule tasks, create invoices, and view highly customizable reports. User-friendly, ProZ.com integration, community-driven development – a few reasons BaccS is trusted by translators!

More info »



Forums
  • All of ProZ.com
  • Term search
  • Jobs
  • Forums
  • Multiple search