SDL Analyse - troubleshooting

Hello all, 

Thank you for yor help so far.... we're getting closer!  We created this new thread as we had three separate threads for this and it's going to be easier to manage in one place.  We have created a new build which you can find here: 

BUILD REMOVED - CHECK THE THREAD FOR THE LATEST

This build has the following fixes which were applied after getting feedback from the last round of testing you carried out yesterday: 

  • Persisted project template is verified to check it actually exists. If not the default template is proposed
  • The desktop client sends the roaming folder to the analysis service instead of the local user.
  • Proper character encoding for data sent to the analysis service (failed for some “exotic” usernames)

If you receive a “project templates not defined” error on start up after reinstalling just click “try again” which you’ll see in the same message.  This may occur every time the service is restarted.  We are looking at this.

Thank you for your help again... much appreciate your time in helping us to resolve this.

Regards

Paul

Parents Reply Children
  • Hi,

    It works perfectly well now. No need to click on "Try again".
    Thank you!
  • Hi Paul,

    Thanks very much for your help! I've tried the new version and it looks better in terms of compared wordcount between the stand alone App and the Studio project.

    Regarding the exact matches and if they come from figures, as I originally asked when I started the thread, they seem to do so. I have locked the figures in the sdlxliffs and run an analysis and then compared that to the App log for non-locked figures, and I got these results:

    Locked figures_generatedReport.en-GB.es-ES.xlsx

    Figures_generatedReport.en-GB.es-ES.xlsx

    On the other hand, I have spotted also some differences with the analysis in Studio 2014. Is that normal?

    All files_generatedReport.en-GB.es-ES.xlsx

    Analysis_All files_2014.xlsx

    Analysis_All files_2017.xlsx

    And last but not least, I found another issue that I also saw in the previous version: a message when selecting a folder to save the log other than the suggested one:

    When you click Yes, it takes really long to save the log and sometimes it doesn't even work and I have to restart the analysis.

    Thanks very much for your help and please let me know in case you need further information!

    Kind regards,

  • Many questions!!

    Elisa Wenceslá said:
    I have locked the figures in the sdlxliffs and run an analysis and then compared that to the App log for non-locked figures, and I got these results:

    Are you comparing the simple view without a TM or using a Project Template?  This is important because if you start to use specific settings then you should only use the advanced view and make sure you have the project template you need.

    Elisa Wenceslá said:
    On the other hand, I have spotted also some differences with the analysis in Studio 2014. Is that normal?

    Of course... the analysis between Studio 2014/Studio 2015/Studio 2017 will all be different to some extent.  You can tweak the settings a little to get close, but you may not be able to match them at all.  But this is a completely different topic to this app!

    Elisa Wenceslá said:
    And last but not least, I found another issue that I also saw in the previous version: a message when selecting a folder to save the log other than the suggested one

    Hopefully  will be aware of this one and can try to reproduce.  Although I tested this just now and can't reproduce the problem at all.

  • Hi Paul,

    When I do the analysis/tests I always do them in the same way, same settings, no TM, just default template both in the App and Studio 2017.

    What I did to compare these two analysis was analysing the Word files with the App, where I get the exact matches (default template, no TM - log "Figures_generatedReport.en-GB.es-ES.xlsx") and then I opened the Words in the Editor View in Studio 2017 to create sdlxliff files, where I filtered by number only and locked those, to have the locked segments in a different category. I analysed those sdlxliff files with the App too (again default template, no TM - log "Locked figures_generatedReport.en-GB.es-ES.xlsx") to check if these locked segments matched the exact matches from the previous analysis, and so it seems according to the logs I provided in my previous post.

    I would like you to kindly confirm if that's the expected behavior with the stand alone App: to count numbers and any other automatically translatable segments directly as exact matches. This would be very useful but would create a discrepancy with the analysis of a Studio project, where that's not taken into account...

    Thanks for confirming the rest of queries, and please let me know in case you need further info!

    Kind regards,

  • Elisa Wenceslá said:
    Thanks for confirming the rest of queries, and please let me know in case you need further info!

    Hi Elisa,

    Can I have one of your test files please so I can get the same kind of results you are?

    Thank you

    Paul

  • Sure!

    Please find them attached. Please note that these are pseudotranslated files (with fake xxxxx content...) as I cannot show the real content.

    You will find a zip containing the Word files used to create the log "Words_generatedReport.en-GB.es-ES.xlsx" (where you can see 1275 words/segments/recognized tokens as 100% matches) and also the sdlxliff files with locked figures only used to create the log "Locked sdlxliffs_generatedReport.en-GB.es-ES.xlsx" (where the amount of locked plus exact matches equals the exact matches in the other log).

    When using the files with real content all of those are shown as locked segments, as you can see in my previous post where I included the corresponding logs. Obviously, as these files contain faked repeated content, I understand the result is not as exact as with the files with real content (you will see some differences in the logs) but it should help you have an idea of what I mean ;-)

    Test files_20170323.zip

    Please have a try yourself and let me know in case of further queries.

    Thanks,

  • Hi Elisa,

    In the meantime this is what I did.  I create a test file from en(GB) containing the following:

    Segments 2-8 are all recognised tokens in Studio for this language (I placed them with QuickPlace), but only two will be auto-translated for the purpose of pre-translation and analysis in an empty TM.  The Studio analysis on the source DOCX with an empty TM gives me this:

    SDL Analyse gives me the following if I use no TM at all and exactly the same result if I use the TM I used for Studio:

    This is exactly the same so no problem here, on the basis of my simple test.  The 100% is the first AT number, and the fuzzy seems to be a penalty on the number in segment #7 having the currency sign... treating it as a measurement (probably a bug we need to look at but irrelevant to this discussion).  The repetition is segment #3.

    If I confirm these into the TM and repeat the test with new tokens like this:

    I expected these to all be 100%/CM matches but I think there is another problem here with Studio.  So not a problem with SDL Analyse as this behaves in exactly the same way if I use the same TM.

    Finally, if I now lock some of the numbers, and check the box to exclude locked segments then I am going away from the default settings.  So if I want to compare the use of locked segments then I need to make sure I use a project template in SDL Analyse to achieve this.  If I do that then the results are still exactly the same, irrespective of whether I use Studio or SDL Analyse.

    If you like we can take a call on this and look at it together, but I think we are pretty much there now.

    Regards

    Paul

  • Thanks for looking into this, Paul!

    However, as far as I understand, if I don't get the recognised tokens as exact matches/fuzzies when doing an analysis without a TM in Studio 2017, why do I get them as exact matches when doing an analysis with the App if I'm not using a TM either? I guess the App uses a "background" TM even if we don't provide it and even if we use the Default template not including TMs (and that's why it is shown in the log as "gen-de-de.sdltm", for example).

    In our case we sometimes prepare quotes without any TMs (not even dummy/empty TMs) and the fact that we have this new App is great because we don't need to create a project just for a quote. However, it seems like the results of an analysis without TM that we get when using the App (that will use a background TM anyway) and the results we get with Studio 2017 are not exactly the same.

    My final question would be: can we trust that, when we use the App to analyse without a TM, the resulting exact matches are segments that will be automatically translated when working on a Studio project?

    Thanks and regards,

  • Hi Elisa,

    Elisa Wenceslá said:
    I guess the App uses a "background" TM even if we don't provide it and even if we use the Default template not including TMs (and that's why it is shown in the log as "gen-de-de.sdltm", for example).

    Correct.  This is needed because Studio cannot recognise any tokens at all unless the TM is present, so everything would be shown as "New".  It is debatable I suppose whether this is important or not for the analysis, but as this is the first time since the release of Studio 2009 that I have ever come across anyone who does this I think we'll stick to the way we do it now :-)

    Elisa Wenceslá said:
    However, it seems like the results of an analysis without TM that we get when using the App (that will use a background TM anyway) and the results we get with Studio 2017 are not exactly the same.

    Correct, for the reasons mentioned.  So in my examples above the file analysis without a TM would look like this:

    In many ways cleaner considering it's an empty TM but as you are probably mainly concerned with the total wordcount (or I think you would be using a TM) then as long as you understand this it should be fine.

    Certainly an interesting exercise and I have already escalated the problems encountered in the earlier analysis so that we can discuss ensuring we have some consistency in the analysis.  The problem with making changes in the analysis is the impact it always has for customers.  A small change can have a very large impact... sometimes good, sometimes not so good.

    Elisa Wenceslá said:
    My final question would be: can we trust that, when we use the App to analyse without a TM, the resulting exact matches are segments that will be automatically translated when working on a Studio project?

    Yes.  This is exactly the same as using Studio to do the analysis with an empty TM attached.

    Regards

    Paul

  • Hi Paul,

    Thanks so much for confirming and for all your help here!! I think it's all clear now ;-)

    Kind regards,

  • Still works beautifully for me, thanks! Looking forward to a signed copy being uploaded to the app store, so the annoying message about the unsigned plug-in will disappear. Kindly let us know here as soon as this happens, Paul!