When it comes to choosing an MT engine, there is no perfect answer. The final choice can come down to several possible factors. For example, some MT engines produce better results for a specific language pair or if equally capable, results might be affected by the topic.
Some companies publish detailed analysis of MT output and rank engines for some language pairs. One such company is inten.to and they post some results on a regular basis. Their latest study is available here.
Studies like this one might give some insights into what engine is better for a specific language pair. But when dealing with a large project, it might make sense to perform testing using different MT engines and collect feedback from the translation teams. The perception from translators might vary from the automated results generated for the studies. Testing a couple of thousand words with selected engines should be sufficient in most cases.
Fortunately, Smartcat makes it easy to perform such a test. In projects with popular language pairs, it is possible to assign multiple MT engines. The results will be displayed in the Editor and the translators will be able to indicate which MT engine they perceived to produce the most usable output for that project.
How it works
When setting up a project, click on the gear icon under Use machine translation at the Settings step:
As shown above, there are two options for machine translation at that step: a free MT option and paid services. The free MT version is provided by Yandex but in exchange for free MT, they collect post-editing data used to improve the engine. This could be a consideration when working with sensitive documents. The paid services are provided by different MT developers under commercial licenses with varying terms of services.
In the screenshot above, you can see that it is possible to select a different engine for each language. Simply drag and drop the language label to the MT engine you want to use for that language. By clicking on the Add option, you can add languages to a specific engine. For example, you can see that French will be processed by two different engines above.
Each engine supports a specific number of languages. When you drag a language label, some MT engines may become unavailable if there is no coverage from these engines.
Note: Using multiple engines can be useful for testing but is not practical for production. First, it adds to the cost of production as most engines are paid services; and second, it might negatively affect the linguist’s productivity as they might spend more time evaluating the MT output than editing it.
The MT output can be used to pre-populate the target segments in the Editor using the pre-translation option. This might be useful in a post-editing project or simply as a suggestion to the linguist along with matches from the translation memory and glossary terms in more traditional projects.
You can also use machine translation to pre-translate a new or existing project. To do this, set up a rule during the project creation process or in the project’s settings.
Once pre-translation is completed, target segments will be pre-populated with machine translation output as shown below.
You can buy MT pages on the Productivity services page.