Under Community Review

Translation quality estimation function

Hi, there.


It would be useful to have a function that calculates the similarity between the source text and the target text using BERT-type natural language processing technology, and warns the reviewer if the similarity exceeds a certain threshold as a segment for which semantic identity cannot be confirmed, so that the reviewer can focus on that segment when reviewing.

hiroshi kameya
Parents

  • Hi Hiroshi!

    Full-disclosure: I'm the CEO and co-founder of ModelFront, the leading provider of machine translation quality prediction, also know as "quality estimation" in the research world.

    ModelFront quality prediction is independent from the TMS or CAT, machine translation and of course the LSP, and provides an API for your custom model.

    So your ModelFront model can be integrated with any TMS or CAT, including RWS systems, or even in-house TMSes or other applications.

    As you suggest, under the hood, is uses multilingual Transformer-based LLMs, specifically modified for the task of translation quality prediction.

    However, in our experience, customization is absolutely required for the use case you're interested in, which is the highest-value use case.  So you probably don't want some generic model bundled with the CAT tool, but rather a model customized for your content or your customer's content.

    Because as you know, good/bad translation in an enterprise human translation workflow is extremely specific.  To achieve human quality, your quality prediction needs to be aware of the terminology, style and quality bar, according to the edits by the professional human translators in the workflow.

    Adam

Comment

  • Hi Hiroshi!

    Full-disclosure: I'm the CEO and co-founder of ModelFront, the leading provider of machine translation quality prediction, also know as "quality estimation" in the research world.

    ModelFront quality prediction is independent from the TMS or CAT, machine translation and of course the LSP, and provides an API for your custom model.

    So your ModelFront model can be integrated with any TMS or CAT, including RWS systems, or even in-house TMSes or other applications.

    As you suggest, under the hood, is uses multilingual Transformer-based LLMs, specifically modified for the task of translation quality prediction.

    However, in our experience, customization is absolutely required for the use case you're interested in, which is the highest-value use case.  So you probably don't want some generic model bundled with the CAT tool, but rather a model customized for your content or your customer's content.

    Because as you know, good/bad translation in an enterprise human translation workflow is extremely specific.  To achieve human quality, your quality prediction needs to be aware of the terminology, style and quality bar, according to the edits by the professional human translators in the workflow.

    Adam

Children
No Data