SYSTRAN Hybrid Technology
SYSTRAN's Innovative Hybrid Engine
SYSTRAN's hybrid machine translation engine combines the strengths of rule-based and statistical machine translation. It delivers high translation quality for any domain.
- Rule-based components guarantee predictable and consistent translations, compliance with corporate terminology, out-of-domain usability, and high performance.
- Statistical components learn from existing monolingual and multilingual corpora which drastically reduce customization costs and further improve translation quality within specified domains.
SYSTRAN's renowned rule-based translation software is the backbone of its new hybrid MT and provides a solid translation framework. The new statistical techniques learn from existing monolingual and bilingual data to improve every phase of the translation process and enhance the customization process.
The combination of rule-based and statistical technologies sharply reduces the amount of data required to train the software. It also reduces the size of the statistical models while maintaining high performance.
Hybrid MT Meets Customer Expectations
SYSTRAN's hybrid engine meets customer expectations for translation quality, cost-effectiveness and productivity.
SYSTRAN's hybrid engine combines the predictability and language consistency of rule-based MT with the fluency and flexibility of statistical MT to reach customer quality requirements.
Customizing the software to a particular domain involves a variety of resources - dictionaries, glossaries, translation memories, and existing monolingual and bilingual data – to improve overall quality. SYSTRAN's hybrid engine leverages all available language assets to reach the quality threshold for each target domain, reducing customization costs. It can be trained on existing corpora and integrated glossaries. It also leverages corpora to generate translation models and reuses them to build custom dictionaries.
Translation predictability is enhanced and quality is improved with a limited initial investment. Ongoing maintenance of resources ensures incremental quality while customization costs are contained.
SYSTRAN's hybrid engine delivers high performance with standard hardware per the recommended system requirements. Human translators save significant time revising proposed translations instead of translating from scratch, which guarantees highly consistent and reusable translations. Post-edition results are easily re-integrated into the software through the dictionary or by additional training so the system never makes the same mistake twice.
Out-of-the-Box Quality, Compliance with Corporate Terminology and High Performance
Each language pair is built on the same translation engine and integrates the latest natural language processing technologies. SYSTRAN makes use of a broad range of linguistic resources and is therefore able to guarantee precise translations.
Built-in linguistic knowledge guarantees out-of-the-box translation quality for 52 language pairs. Users can add their own linguistic resources such as dictionaries to improve the quality of specific domains or business objectives. This ensures that translations are compliant with corporate terminology. Hybrid translation engines still provide the same high level of performance that is valued by SYSTRAN customers for many years.
Improved Quality, Productivity and Cost-Effective Customization
Statistical techniques are embedded at every stage of the translation process. Built-in or customer specific language models increase disambiguation of the source text. Translation models guarantee that translation is compliant with existing translations learned from human translated texts.
SYSTRAN's hybrid engine can be trained to specific domains or business objectives with available – even limited - monolingual or bilingual corpora based on previously translated material to improve translation quality.
Statistical techniques are also used to build linguistic resources such as dictionaries which are used by the translation engine. This is a very cost-effective way to increase linguistic assets and control corporate terminology.