Under Community Review

Add "PE" for "post-edited" next to 100% pre-translated segments that were originally machine translation

Wouldn't it be practical for translator and project manager alike to be able to distinguish between those 100%-matches that were made from scratch and those that were post-edited from NMT?

It would help to recognize the potentially more direct translations quicker and allow for revision if necessary. Adding a check box in the Analysis batch task to distinguish between these might be interesting as well.

Alternatively, one could also just add the option of penalizing the TM-match if it was generated by NMT, just like alignment matches can be.

Parents
  • Can't you do this already?

    Screenshot showing two segments in Trados Studio Ideas. The top segment displays a 100% TM match with a green checkmark, and the bottom segment shows a post-edited Machine Translation with an NMT tag.

    The top one is a post-edited TM match, and the bottom one is a post-edited Machine Translation.

    What seems to be missing I think is the ability to filter effectively on these (the AND/OR feature in the advanced display filter seems to be failing to recognise the filtered options) and the WIP report isn't really customisable in a way you can report on these.  I don't think an analysis report is much use anyway as I think it's based on source compared to TM and not edited status compared to what was in your bilingual file.

    Using the Reports Viewer Plus (from the appstore) could be a possibility if you are comfortable with stylesheets.

  • Hi Paul,

    Thanks for your reply! Hmm, maybe I wasn't specific enough after all, sorry about that!

    I'm aware I can distinguish between unedited TM-matches/machine translations and their edited ones based on the colour.

    What I was talking about was post-edited machine translations that were saved in the TM and are being pre-translated again in a future project or a different file in the same project.

    The way I see it, you'd like to stay consistent and use everything you have in your TM first and not use NMT to retranslate sgements you already postedited. I thought that being able to distinguish translations stored in the TM might be helpful.

    As for the Reports Viewer Plus, I've installed it not too long ago but did not get to testing it yet. I'm not yet familiar with stylesheets but would find it very interesing to be able to create one tailored to our agency's needs.

  • What I was talking about was post-edited machine translations that were saved in the TM and are being pre-translated again in a future project or a different file in the same project.

    This is always an interesting one for me.  If you post-edit the machine translation doesn't that infer that it is now an acceptable translation?  It may be no different to you translating the segment from scratch.  So why would you want to penalise it?  Unless you are working on the basis that a post-edited MT match is inferior to a post-edited TM match, or even a new translation from scratch?  How far do you take this?  At what point does a post-edited, post-edited, post-edited MT match become good enough?

    A bit academic I think and I wonder how valuable this is in practice when you're looking for ROI from your translations.

Comment
  • What I was talking about was post-edited machine translations that were saved in the TM and are being pre-translated again in a future project or a different file in the same project.

    This is always an interesting one for me.  If you post-edit the machine translation doesn't that infer that it is now an acceptable translation?  It may be no different to you translating the segment from scratch.  So why would you want to penalise it?  Unless you are working on the basis that a post-edited MT match is inferior to a post-edited TM match, or even a new translation from scratch?  How far do you take this?  At what point does a post-edited, post-edited, post-edited MT match become good enough?

    A bit academic I think and I wonder how valuable this is in practice when you're looking for ROI from your translations.

Children
  • Hi Paul, in my opinion a translation made from scratch by a competent translator is still better than an automatic one that has been postedited, but maybe that's just me being brainwashed by my translation studies and the fact that I work in an agency that specializes in marketing and human translation. That was up till now, as we're currently looking into offering MT as a separate service and defining the workflow for it, and that's where I got this idea from.

    From that perspective, I thought that having this "PE" or "PEMT" indication, at least for the first time a postedited NMT is pre-translated by the TM, might be a nice safety measure to have one more look at it before it just becomes indistinguishable from all other TUs. I know it doesn't make sense and that there's no time for our translators to go over the same segment multiple times, but a one-time check couldn't hurt either, so I wanted to put the idea out there and let the community decide if they see any use in it. Slight smile

  • I totally agree. My translations made from automatic pretranslation are never as natural, rich or creative as the ones I do from scratch. In suggesting solutions, pretranslation is like that tree that hides the forest of all possible options. I seldom accept such projects, and even there, I often reorganize sentences, which voids the interest of automatic pretranslation. That one just gives "acceptable" translations. I recently had to update a large document for a client. I am pretty sure it was automatically translated then post-edited. I even found errors of meaning, since the post-editor seemingly focused more on form than on meaning, which is a trap of AT. Also, neuronal automatic translation learns from a corpus of previous translations, good or bad, then tends to standardize the translations and validate poor phrases...

    So, distinguishing post-edited segments from genuinely translated documents would be useful and would call subsequent translators to pay more attention to those results.

  • Hi Paul,

    I completely agree with Guillaume and Philippe. We struggle everyday with the issue of Post-edited vs. translated and duly revised content, so much so that we finally had to save our post-edited documents into a separate TM.

    It is absolutely true that post-edited translations are NOT of the same quality as a proper translation done by a seasoned translator. We have a team of extremely artful translators internally. Post-editing does not even start to compare to what they can produce. 

    It should be remembered that in some cases, post-editing is requested in order to shorten turnaround time. That's true for us. Cut down the human processing time in half, you'll get a lower tier output. It will be "usable", but it can't serve as a reference, as a TM match would normally do. It might even still contain errors, as NMT can produce very misleading results that look good and seem right at first glance, but are in fact wrong.

    This is especially true with complex, sensitive and sub-text rich documents, which make up about 98% of what we deal with in our organisation.

    Being able to "tag" PE segments in the Editor and in TMs would be a BIG PLUS for us!