Transit NXT meets machine translation - nº 1
Transit NXT meets machine translation - nº 1
"CAT meets MT" and the ingenious ways of Transit NXT to make that encounter as beneficial as possible for all users and players is too wide a field than to be dealt with in one single article. We therefore decided to split the topic into various posts. Here comes the first one.
Why machine translation?
Web 2.0 technologies, a rapidly increasing Internet community, its linguistic diversity, explosion of content and continuing globalization are the main driving forces held culprit for the mounting demand for readily available multilingual information. With the advent of statistical machine translation (SMT), refinements in rule-based machine translation (RBMT) and the combination of both technologies, the quality of MT has undeniably improved, albeit only for a limited number of communicative purposes, language combinations and text types. Be that as it may, MT with human post-editing (MTPE) is increasingly considered as a solution to help overcome language barriers.
Transit NXT for MTPE
Consequently, with Service Pack 7, Transit NXT starts to embrace MTPE. It offers a variety of secure and money saving options for translators, post-editors, reviewers, project managers and organizations to integrate different machine translation solutions within their computer-assisted translation workflows and simultaneously, into their usual multilingual document supply chain. For that purpose, Transit distinguishes between the following two MT usage scenarios, depending on the kind of MT engine that it exchanges data with:
- Customer-specific MT systems: these are used exclusively during the import of project files and are therefore referred to as Import MT.
- Unspecific online translation services: these are supported exclusively via the Transit editor and are therefore referred to as Editor MT.
In this tooltip, we will have a look at Editor MT and show how Transit uses some basic options to control the exchange of data with unspecific on-line translation services. You will learn how you as a translator and post-editor can control what you send to the machine translation engines.
Before starting to use MT, you need to make the appropriate settings in the User preferences. You can do so by clicking on User preferences Editor MT from the resource bar as seen in the above figure or by selecting the Machine translation tab in the User preferences window:
At the top of the dialogue, you have to mark the option that you want to use machine translation in Transit. This brings up a message that informs about the implications of using machine translation and prompts you to confirm explicitly that you want to send data to an MT system. You can then decide if and when you want this message to be displayed again.
Best of both worlds
With Editor MT in Transit NXT, you can use machine translation output from the chosen MT engine and your own translation memory at the same time. You as a translator and post-editor are in full control of when and how you want to use MT. You can decide, for example, to only consult the MT engine automatically in those cases where there are no fuzzy matches above a certain fuzzy match threshold. In other words, when Transit offers you high fuzzy matches from your translation memory, it does not consult the MT engines, thus saving you time and money.
The feature is also a quality assurer. It assures that you appropriately leverage client-specific legacy content, instead of alternative MT suggestions, thus contributing to terminological and phraseological consistency, important quality criteria for technical and other translations. Depending on TM availability and the quality of the MT output for your given text, you can lower or raise this threshold.
No junk please
There is another feature that allows you to control when to consult the MT engine. It is a known fact that the quality of MT output depends on the length of the translated segment. One word sentences are often ambiguous, so their meaning can only be derived from the context, a tricky task for any MT system. Equally challenging for machines are long, convoluted sentences. Transit enables you to disregard MT junk right away by consulting only segments that are within a specified length range. You can do so by determining the minimum and the maximum length in words that a segment must have to be processed by MT, as shown in the screenshot above.
GoogleTranslate and iTranslate4eu
Transit currently integrates two on-line translation services: GoogleTranslate and iTranslate4eu; you can use only one or both at the same time. There will be more services available in the future. To be able to use them, Transit needs to know their API keys. Selecting the desired service from the Machine translation menu in the resource bar brings up the corresponding dialogue that lets you introduce the API key.
Another way of introducing the API key for a given MT system is in the User preferences. Select the desired option in the Use column and configure it as seen in the following figure:
Enter your API-key for the MT system. The selected MT system is now ready to be used. How suggestions from the machine translation engine are automatically retrieved, how translators can edit them and how reviewers and project managers can track the use of MT will be explained in our next post, so stay tuned. We are looking forward to your visit and your comments!