Technology Showcase

Transit NXT and STAR MT

STAR Group is a premier developer of innovative software technologies that support the en-tire information life cycle process. This allows the seamless integration of MT technology in established and proven processes and tools for any stage of the information life cycle: MT-enhanced authoring with STAR GRIPS, MT-enhanced workflows with STAR CLM, MT-enhanced language services, and MT-enhanced human translation with STAR Transit.

Our many years of experience in data and language processing and intelligent format handling enable STAR to provide project managers, translators and post-editors with the best possible assistance from both worlds:

  • Text corpora: In Transit, formatting information is separated from the text such that – after translation – they can be merged automatically. For us it is a tried-and-tested routine to generate text corpora from any file format with which to train engines.
  • Markup handling: Transit’s unique markup handling allows markups to be positioned in the MT result automatically and linguistically correctly.
  • Terminology updates: Via Transit’s integrated terminology management system, TermStar, new terms can be added at any time to the MT result, without having to continually re-train the engine.
  • TM-validated MT: STAR’s solution uses TM matches to validate MT results, rather than repairing missing sections of fuzzy matches using MT. Translators and post-editors can see at a glance what needs to be done.
  • Quality assurance: Transit’s proven formal and linguistic QA functions can also be used for post-editing projects and “mixed projects” produced by MT and human translation.

New at AMTA Showcase 2016:

  • Morphologically generated extra information: MT quality improves noticeably by using additional training input that is generated by using Transit’s morphological technology. The morphological support for 80+ languages enables Transit to retrieve inflected forms of dictionary terms from the reference material without any additional effort. This is especially useful for morphologically rich languages.
  • Security keys for controlled access to the STAR MT server: Engines can be protected against unwanted access by generating security keys during engine training. The project manager needs to have the security key in order to see and select an MT engine in Transit.
  • Optimized project exchange: MT suggestions are automatically packed into the project package. Translators do not require access to the MT system as they receive the MT suggestions with the project.
  • Optimized display of MT suggestions in the Fuzzy window: The translator can individually adjust the position in which MT suggestions will be displayed in the Fuzzy window. Untranslated words (out-of-vocabulary words) are clearly marked in the MT suggestion.

Speaker: Nadira Hofmann, STAR Group

www.amtaweb.org

Commercial Users Track

Seamlessly integrating machine translation into existing translation processes (STAR MT and Transit NXT)

Having developed both STAR Transit (translation memory system) and STAR MT (machine translation platform), STAR has combined these into a single integrated solution.

“Typical” customers have established processes for language management, and use a TM system when translating. They are interested in MT as current solutions are unable to cope with ever-increasing translation requirements on a tight budget.

Firstly, MT should be integrated – manually or automated – into established processes in order to handle higher volumes, shorter deadlines and smaller budgets. Secondly, MT should unlock the benefits of other available resources.

Successful integration relies on being able to use the existing benefits as before (pretranslation and “classic fuzzy matches” from validated TM) and minimising change management for the staff involved (no additional tools or steps, no new workflows). This allows projects managers and translators to use their prior experience and see how MT can assist them in their work.

But how?

Transit creates training packages for the MT using existing TM and terminology. STAR routinely extracts text from any file format and leverages terminology to ensure that MT engines are optimally trained for the customer’s translations.

In the translation process, files are imported into Transit, pretranslated with TM and any untranslated segments transferred to the MT engine. MT suggestions are then sent back.

MT integration has been made as simple as possible for translators:

  • During project exchange, MT suggestions are automatically packed into the project package. Translators therefore do not require access to the MT system as they receive the MT suggestions with the project.
  • Translators benefit from two types of suggestions: Classic fuzzy matches and MT suggestions.
  • STAR's experience shows that MT quality is particularly high for segments with very good fuzzy matches, which is why we validate MT results using TM material rather than fill missing sections of fuzzy matches using MT. This creates an entirely new type of match: In the “TM-validated MT suggestion”, translators can see which portion of the suggestion comes from the validated TM and which has been added by STAR MT.
  • Transit’s markup handling feature positions markups in the MT result automatically and linguistically correctly.
  • Transit’s integrated terminology management system, TermStar, enables new terms to be added to the MT result at any time, without having to continually re-train the engine.
  • Transit’s QA functions can also be used for post-editing projects and “mixed projects” produced by MT and human translation.

These points show how STAR MT can simplify, speed up and optimise existing translation processes. Language departments can also benefit from additional applications that go beyond pure translation management.

STAR MT Translate is a browser solution that gives staff access to the trained STAR MT engines. Individual sentences or whole paragraphs are transferred to STAR MT, which then returns machine translations. The company-specific texts contain translation suggestions that use their corporate terminology and style, allowing customers to create an in-house, real-time translation solution that will be used solely by them and cannot be mined for data. Language-management departments can therefore tap new target groups within the company and increase the visibility of their language skills.

Nadira Hofmann, STAR Group

www.amtaweb.org

« Back