For the last several years, there is an MT industry trend of combining two approaches. Statistical MT aspires to use the linguistic data for translation quality improvement, while rule-based engines look for the application of statistical techniques in their technology.
In 2010, PROMT introduced the Hybrid version of its engine — PROMT DeepHybrid, which combines the flexibility and predictability of rule-based MT with the fluency and lexical acumen of statistical MT.
The PROMT DeepHybrid Engine already has the baseline rule-based quality. Further, it harvests from the client's TMs only those items that truly improve MT quality.
When a PROMT DeepHybrid profile is trained, the following components are automatically created:
- A Language Model (LM) of the target language for evaluating translation candidates
- A Statistical Post-editing (SPE) Table for replacing machine-translated n-grams with human translated n-grams that were harvested from TMs
- Additional Dictionaries with terminology harvested from TMs for more precise lexical choices
When translating, the PROMT DeepHybrid engine generates several translation candidates on all levels of translation process (this is why it is called Deep Hybrid), evaluates every candidate against the LM, and finally selects the best (most probable) output.
- Maintaining the rule-based technology benefits (creation of syntactically connected and grammatically correct text, terminological uniformity, etc.)
- More customizable and predictable than pure SMT
- Profiling (multiple profiles can be easily created in one engine)
- Getting the benefit of statistical MT (fast learning ability, data acquisition from parallel corpora in automatic mode, text smoothness, etc.)
- Engine training is faster than pure RBMT
- MT output is more fluent and “human-like” than pure RBMT
PROMT DeepHybrid was initially designed for the Localization Industry and Language Service Providers (LSPs) but also works for any client wishing to translate large text volumes on a regular basis and that has amassed enough content for engine training.
PROMT is a world leading provider of automated translation software. The company develops PROMT translation software and dictionaries that have received numerous awards from leading computer magazines. PROMT offers translation solutions for the home and business use, as well as for Internet and corporate intranets.
PROMT provides machine translation for 16 languages: English, Russian, German, French, Spanish, Italian, Portuguese, Chinese in two variants, Ukrainian, Kazakh, Turkish, Bulgarian, Latvian, Japanese and Polish, with a total of 64 translation directions
The Ukrainian, Latvian and Polish language pairs included in the PROMT solutions are developed by Trident Software.