Trados Studio inserts translations below the desired threshold level (%)
Thread poster: Nick Quaintmere

Nick Quaintmere  Identity Verified
Germany
Local time: 22:02
German to English
Jul 21, 2010

Did that make sense? To clarify: when I'm working through my translation Trados often inserts plausible translations where a word or two is different in the new source text and what is already in the TM. That's great.

The trouble is that I have set my minimum match value to 70% (in "All language pairs) and it is inserting translations that are under this level. These are usually sentences that have no similarity whatsoever with the desired translation.

Here's quite an extreme example:

The source text is: Wandel ist allgegenwärtig. (A simple three-word sentence)

Trados checks through my TM and finds: Alles ist möglich! and its translation Anything is possible! and inserts this automatically.

Firstly, how on earth is it possible that this is given as a 61% match!!! Only 3 characters even match!

Secondly, why is it being inserted even though I set the minimum for 70%? Is there something else I need to set?

These auto-insertions can be very useful, but the match levels are obviously a little arbitrary and need to be set somewhat higher.

Thanks for your help.


Direct link Reply with quote
 

Natalie  Identity Verified
Poland
Local time: 22:02
Member (2002)
English to Russian
+ ...

Moderator of this forum
Same here Jul 21, 2010

Hi Nick,

I noticed exactly the same, though I have even set the minimum match value to 80%.

Moreover, yesterday I had another very unpleasant situation, when Studio added the same number to all cells in the table. To clarify: my text contained a table populated with some numbers. I had to convert one of them into another system of measurements and add this value in parentheses. As soon as I did this, Studio added the same value in parentheses to all other cells of the Table.

Quite unpleasant, I need to say.

Natalia


Direct link Reply with quote
 

Jerzy Czopik  Identity Verified
Germany
Local time: 22:02
Member (2003)
Polish to German
+ ...
Did you change the value in global settings or in project settings? Jul 21, 2010

When you create a project, the settings from the general set (Tools - Options) are implemented in the project. Whe the project is already there, changes via Tools - Options have no impact on this project (not all, but most of them), so if you need to change the min. match value you need to do that in project settings rather than in general settings.

As for the question how the match is calculated: I obviously do not know, but in general Studio is not so good in matching very short sentences.


Direct link Reply with quote
 

Grzegorz Gryc  Identity Verified
Local time: 22:02
French to Polish
+ ...
Flaws in the matching algorithm... Jul 22, 2010

Nick Quaintmere wrote:

Did that make sense? To clarify: when I'm working through my translation Trados often inserts plausible translations where a word or two is different in the new source text and what is already in the TM. That's great.

The trouble is that I have set my minimum match value to 70% (in "All language pairs) and it is inserting translations that are under this level. These are usually sentences that have no similarity whatsoever with the desired translation.

Here's quite an extreme example:

The source text is: Wandel ist allgegenwärtig. (A simple three-word sentence)

Trados checks through my TM and finds: Alles ist möglich! and its translation Anything is possible! and inserts this automatically.

Firstly, how on earth is it possible that this is given as a 61% match!!!


If you test another senences like Christ ist erstanden!" (63% match against Alles ist möglich!, you'll constate the error is systematic.

Only 3 characters even match!


The most absurd suggestion I received for 3 word sentence was a 71% match for one common one letter word.
This result was probably somehow biased by the absurd weight of tags applied in Studio.
The same occurs with the numbers.
I.e. for a moment, it seems the numbers are three times more important than words in Studio while they were only two times more important in the old Trados.
E.g. see:
http://pol.proz.com/forum/sdl_trados_support/174805-trados_2007_studio_2009_prof_analysis_discrepancy.html
As you see, in some cases Studio may easily suggest matches with no common words at all.

BTW.
For a 4 word sentence like A cat of mine., I have a 61% hit for Cat.
For a 5 word sentence like A funny cat of mine., I have a 50% hit for Cat.
It seems the algotithm starts to behave in a human way above 6 (no matches for a 35% match level defined).

I.e. all matches below 5 words seem suspect, especially when tags are present.
E.g. for A cat of mine. the match level for Cat. is 71% which is completely abberrant (I expect rather a lower match leved because of a different formatting).

Cheers
GG

[Edited at 2010-07-22 07:38 GMT]


Direct link Reply with quote
 

Nick Quaintmere  Identity Verified
Germany
Local time: 22:02
German to English
TOPIC STARTER
False matches in Trados Studio 2009 Jul 22, 2010

Unfortunately Grzegorz, the problem isn't even limited to short sentences:

According to Studio, the following two sentences:

"Während und auch nach dem Verkauf unterstützen wir Sie mit erstklassigen
 Service leistungen, um Ihre Kunden nachhaltig an Sie zu
 binden."

and

"Wir wissen, dass wir mit unseren Mitarbeitern sprechen und ihnen auch zuhören müssen, um sie zu motivieren und an das Unternehmen zu binden."

also exhibit a 60% correlation although only the bold words are registered as exact matches (theoretically Trados could also be counting one of the "wir"s I suppose). There is no other correlation in vocabulary or meaning or even syntax. I have not tested this in Trados 2007, but I am almost certain that this would not count as a match at all.

@Jerzy. Thanks for the tip. It seems that in the project settings, the level for German-English was indeed set at 60% although the global settings were at 70% ("All language pairs" and the specific language pairs German-English GB and German-English US). Why this should be, I don't know. I would have thought that new projects would have automatically taken over the global settings, but maybe one wasn't as it should have been. I will check the next time I set up a new project. Maybe I overlooked something or accidentally used a project profile with different settings. Anyway, the problem is solved for the current project


Direct link Reply with quote
 

Jerzy Czopik  Identity Verified
Germany
Local time: 22:02
Member (2003)
Polish to German
+ ...
Global settings vs project settings Jul 22, 2010

Well, global settings are implementen into project settings, which then "live their own life".
And you can have a global setting for "All language pairs" of 50%, but for example a 70%-setting for just a single language pairs among them.
This is a hierarchical structure, set up as you go:
First you start with global settings.
Then you have All language pairs.
After those settings are done all language pairs get them. However, now you can go to each language pair and individualise there...
And this is then inherited by the project, where you again can start individualising.
Just perfect for someone as me, but I admit it might be a source for trouble if you have some less structured and logical way of thinking


Direct link Reply with quote
 

Grzegorz Gryc  Identity Verified
Local time: 22:02
French to Polish
+ ...
Trados matching swindle... Jul 22, 2010

Nick Quaintmere wrote:

Unfortunately Grzegorz, the problem isn't even limited to short sentences:

If I point out the short sentences it's mainly because it's easier to track, understand and reproduce the issues.

According to Studio, the following two sentences:

"Während und auch nach dem Verkauf unterstützen wir Sie mit erstklassigen
 Service leistungen, um Ihre Kunden nachhaltig an Sie zu
 binden."

and

"Wir wissen, dass wir mit unseren Mitarbeitern sprechen und ihnen auch zuhören müssen, um sie zu motivieren und an das Unternehmen zu binden."

also exhibit a 60% correlation although only the bold words are registered as exact matches (theoretically Trados could also be counting one of the "wir"s I suppose).

It does.
The legacy Trados matching algorithm was always based on the 100%/x parameter where x is the number of words in the sentence (several other principles apply).

BTW, a nice example, thanks.

There is no other correlation in vocabulary or meaning or even syntax. I have not tested this in Trados 2007, but I am almost certain that this would not count as a match at all.

I tested it for you
I receive a 33% match.
I.e., 7 x Delta for the words (or puntuation marks) in the right order + 2 x Delta/2 fo misplaced words.
As Delta is approx 4,15% (100% divided by 24 words/ponctuation marks in the target sentence), it gives approximatively 33%.
I may be somhow wrong in my calculations, I usually have problems when counting above 3, unlike the guys which create SDL Synergy 2007 setup packages who were puzzled between 2 and 3

As we see, the matching in Studio is really worse than in the old Trados.

By the other side, a lot of translation offices use to apply reductions for fuzzy matches.
If the Studio fuzzy matches are like our examples, these reductions start to be a pure extortion with the 2009 edition.

Cheers
GG


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 Studio inserts translations below the desired threshold level (%)

Advanced search







SDL Trados Studio 2017 only €415 / $495
Get the cheapest prices for SDL Trados Studio 2017 on ProZ.com

Join this translator’s group buy brought to you by ProZ.com and buy SDL Trados Studio 2017 Freelance for only €415 / $495 / £325 / ¥60,000 You will also receive FREE access to our getting started eLearning program!

More info »
Wordfast Pro
Translation Memory Software for Any Platform

Exclusive discount for ProZ.com users! Save over 13% when purchasing Wordfast Pro through ProZ.com. Wordfast is the world's #1 provider of platform-independent Translation Memory software. Consistently ranked the most user-friendly and highest value

More info »



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