Auto Select: Identifying the Best MT Engine Match

During their webinar in September 2019, Smartling announced a new feature to their Translation Management System (TMS): Machine Translation (MT) Auto Select. This means that whenever you decide to utilize a specific MT engine in a TMS, the tool will automatically pick the best engine for your particular case.

 

Auto Select Identifying the Best Engine Match graph - Smartling - Nimdzi.com

Source: Smartling webinar

 

 

The idea to spare the linguists the effort on evaluation of the MT output directly in the TMS is not new.

For instance, in Q1 2019, Memsource released Memsource Translate which does a similar trick. Since then, they have been working on the next version that is scheduled for Q2 2020. It will feature more than 30 MT engines that are supported in Memsource, including custom ones (as opposed to only two engines being supported as of now: Google Translate and Microsoft Translator).

Here is an interesting fact shared with us by David Canek, Memsource’s CEO: “When we first launched Memsource Translate, we saw that for 70 percent of projects a sub-optimal engine was selected by the linguist. Another engine would have performed better in terms of translation quality.”

Wave a magic wand, et voilà

Memsource Translate is powered by a reinforcement learning algorithm that helps recommend the optimal MT engine based on the source and target language. Starting with the new version, the plan is to also add the domain as the third parameter.

The algorithm is constantly evaluating the performance of various MT engines for a given combination of a source language, target language and a domain that is automatically identified from the corpus of the document. It learns continuously based on this evaluation what the best MT engine is.

So, when you translate, say, technical documentation from English to Japanese in Memsource, it will recommend an engine that provides the highest quality for that specific language pair and domain.

Memsource Translate also includes an MTQE feature, which helps evaluate the quality of the MT output: scores are automatically calculated before any post-editing is done. For this one, deep neural network is used.

There are dozens of viable MT engines available and selecting the right one for your language pair and domain is not a straightforward choice. MT Auto Select features offered as part of your TMS can help you make the most of it.

Nimdzi Finger Food is the bite-sized insight you need to fuel your decision-making today.

Tags: , , , , ,

Stay up to date as Nimdzi publishes new insights.
We will keep you posted as each new report is published so that you are sure not to miss anything.

  • Sign Up
Lost your password? Please enter your username or email address. You will receive a link to create a new password via email.