MT Connector is no longer working with the new neural decoder of MMT

Issue #655 resolved
Former user created an issue

Hello support team,

I just would like to inform that the current MT connector for ModernMT is no longer working with the new API for the neural decoder. I believe it's a very easy fix though.

The new API URL now takes 2 more mandatory arguments/end points. "Source and Target".

So, the new URL would look like this:

http://localhost:8045/translate?q=world&%22%20&source=en&target=ar

Is there a quick workaround for this? How can we append these 2 parameters? I tried adding them to the context field, but apparently it didn't work. Any idea?

Thanks in advance for your support!

Mohamed Zeid

Comments (9)

  1. mzeid

    I am not sure if this is something introduced by the new neural decoder or it's my mistake. What I am sure of is that for the SMT version of MMT, I just entered the URL of the server and it worked just fine. Any idea?

    So, now if you don't add a source and target in the URL, it throws this error:

    #!
    
    {
    status: 400,
    error: {
    type: "ParameterParsingException",
    message: "Missing parameter source"
    }
    }
    

    If you add the source parameter 'source=en", you get the same error but for target this time.

    #!
    
    {
    status: 400,
    error: {
    type: "ParameterParsingException",
    message: "Missing parameter target"
    }
    }
    

    The error that shows up in Rainbow is:

    #
    
    ModernMT URL: http://192.168.1.245:8045
    ERROR: Error querying the server.
    Error: response code 400
    Bad Request
    
    Error count: 1, Warning count: 0
    Process duration: 0h 0m 0s 158ms
    === End process
    

    By the way, I reported this to the MMT team

    Thanks!

  2. mzeid

    Thank you so much for the quick fix. This is amazing. How can I get the latest version of Rainbow?

    Thanks again for your great support! I appreciate it.

  3. mzeid

    @letconex Yes, it was me who posted that issue on MMT GitHub account there and Davide did the fix, but this applies only to engines that has one language combination. If you have an engine with multiple languages, Okapi Connector won't work. So, the fix is still much needed.

    I found out that people who are behind MMT and Okapi Framework are amazing and very supportive. Thanks again and wish you all a wonderful weekend!

  4. Log in to comment