Under Community Review

Reporting - add fuzzy match as an initial translation origin

 We are trying to see how much fuzzy matches are changed by linguists. It would be beneficial to call out match types in the filtering for initial translation origin.

Screenshot of a data filtering interface showing the dataset 'Translation Productivity' with fields like 'Number of Segments' and a filter set to include 'Neural machine translation'.

Russell Banton
  • in reply to Ian Parnell

    Yes, the more granularity the better, but at least separating by 100%/CM, fuzzy, no match, MT would be great. We can already call at MT, the other pieces are missing. It would also be nice if the lock status was something to filter by as well.

  • Hi Russell, 

    I'm checking with the team on the level of detail we currently receive for segments where the initial translation origin is "translation memory". I.e. Is the data already pre-aggregated by the time we receive the information in the data warehouse or is the aggregation performed after we receive the data but before storing this in DataHub.

    If this is something we are able to enhance, what level of granularity would you expect to be available for TM matches? Would it already be useful to report matches based on 100% match, Fuzzy match, etc. or would you require an additional level of detail for the categorization of fuzzy matches?

    Cheers,

    Ian

  • Really good idea. Machine translation can often be nearer the mark than a fuzzy match, so it would be useful to analyse this in more detail.