Adaptable Language Pairs
An Adaptable Language Pair is a language pair that can be customized to the customer's domain and content using SDL Machine Translation Edge Trainer.
Benefits
- Data Privacy
- Data never leaves customers' secured perimeter, so data privacy is guaranteed. As a result, enterprises can minimize their digital footprint, and protect their company's IP, financial information, employee and customer data. This level of control is a great advantage and most of the time, a must, making SDL Machine Translation Edge particularly appealing in today's digitized world for sectors like Finance, Legal, Government, Communications.
- Continuous Adaptation
- Customers can train models as much as needed, according to their data growth. Their models constantly improve as they get trained with new data, without the need to contact SDL.
- Deployment Flexibility
- Customized models can be deployed on premises in SDL Machine Translation Edge. The models are encrypted and cannot be reverse-engineered to extract data. Therefore, the models can be safely deployed to SDL BeGlobal through our Edge-Cloud hybrid architecture. In a nutshell, customers can train on premises and use the data in the cloud.
- Data Reusability
- Customers can leverage old, domain-specific translated data to deliver new high quality translations that are in line with their business's area of expertise and variety of services offered.
- Higher Translation Accuracy
- An Adaptable Language Pair built on a sufficient amount of valid data normally results in higher translation accuracy for domain-specific terminology.
- Unique Translation Assets
- SDL Machine Translation Edge Trainer offers customers the ability to create a unique set of customized Language Pairs with translation capabilities that cannot be duplicated by anybody else without access to the same data resources used to build them.
- Quality Improvement Cycle
- As soon as sufficient new parallel data is available for a certain domain, customers can perform new customizations by adding it as customization input together with the older domain data previously used to improve the quality of output translations.
- Fast Customization Process
- Customers can create customized Language Pairs whenever needed, on their own schedule, without investing too much time in the process (performing a customization takes about 2 to 4 hours, depending on the hardware used).
- Domain and Language Pair Coverage
- If customers have valid parallel data from different domains and for different Language Pairs, they can get access to create as many customized Language Pairs as required, to cover a multitude of use cases, expand their outreach and facilitate their business success as part of the global market.
Differences between Generic and Adaptable Language Pairs
Generic NMT Language Pairs are built with generic data and are meant for translating different types of information on a wide variety of topics, without necessarily being able to always accurately convey highly specialized terminology pertaining to very specific domains or unique to a restricted number of businesses. For customers who require a quality boost, Adaptable Language Pairs are a highly recommended solution.
NMT Adaptable Language Pairs are built using additional, domain-specific parallel data that customers translated previously as learning materials. As a result, Adaptable Language Pairs can translate not only generic information, but also highly specialized vocabulary, unique to an industry or a customer.
Can we create Adaptable Language Pairs from scratch?
SDL Machine Translation Edge Trainer can create custom models by training an existing Language Pair using additional data available. It is not meant to create a totally new language pair from scratch.
SDL Machine Translation Edge Trainer supports the customization of NMT (Neural Machine Translation) Adaptable Language Pairs only.