Improving SDL Edge Translation Accuracy

I am interested to hear from any SDL client users about your experience with improving accuracy using the dictionary and adaptive learning features. 

We are reviewing options to improve accuracy in a specific language pair, short of the effort and expense to build new adaptive learning models.

Happy to have direct 121 conversations on this subject. 

Best regards

Rich

  • Hi Richard, 

    Which version of Edge are you using? Which LP are you having quality issues with? 

    Thanks,

    Mark

  • We are running 8.5 with all current LP's. I have had feedback on German, Turkish, Arabic ,and Japanese. We are aware of our options with regard to making community contributions to improve the annual LP updates ( no charge) and adaptive learning models where we provide TMs and Termbases and we deploy adaptive LP ( for fee). Trying to appreciate the dictionary and new interactive features. Seems like there are a lot of moving pieces to provide direct answer. document vs browser text, specific LP, Dictionary vs interactive feature, etc ..

    .

  • Hi Richard,

    my client is currently setting up an evaluation project related to quality improvement. We will have definitive results by May and will share it with you.

    One thing, however, is obvious already now: even using adapted engines trained with several mio TUs, terminology is the biggest issue. The usage of the dictionary is not even considered due to the lack of a  morphological component and the lack of direct integration into MultiTerm. 

    What they would be happy to use is the following:

    1. define an export filter in MultiTerm according to the metadata relevant to the text

    2. export 1:1 wordlist directly into MTE dictionary format 

    3. automatically import dictionary to MTE

    4. apply dictionary in MTE, making use of a morphological component. Our partner Kaleidoscope has morphological components for a number of languages that might be used...

    I have already spoken to George Bara and other people in the MTE team about it. I am sure that my clients (the German Government) would welcome this very much. Combined with training, this feature would be a killer for all of our competitors...

    Moni

  • Moni, thank you for responding and the information included.

    We are new to trying to organize our translation tools and services. We like many, use a panel of loosely managed external providers, and previously unmanaged public machine translation activities.  Our initial goal in deploying SDL MT Edge on premise was to address client confidentiality requirements, and that lack thereof in the use of public MT. The near term goal is to identify what low cost methods we can undertake to improve the value of the SDL MT Edge beyond sufficient comprehension ( informal) use cases. With the global nature of the firm and the various maturity levels of SDL LPs we may have to further restrict this short term objective to be LP and practice specific.  

     Your response was very helpful ( and a bit over my head :) ). We have a significant presence in Germany, perhaps we should include that LP in our pilot accuracy improvement project.

     Looking forward to understanding more about the approach and outcomes.

  • Hi Richard,

    just heard from SDL that positions 1-3 in my proposal will happen in Q2 this year, so this will already help. As to the rest, we will see. 

    As to the rest, we will see. I'll keep you posted.

    Moni