Language pair adaptation
An adaptable language pair is a language pair that can be customized to the customer's domain and content type using Language Weaver Edge Adaptation. The result is either an adapted language pair or an auto adaptive language pair.
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 intellectual property, as well as their financial information and employee and customer data. This level of control is a great advantage and most of the time, a must, making Language Weaver Edge particularly appealing in today's digitized world for sectors like Finance, Legal, Government, or Communications.
- Flexible adaptation
- Customers can train models as much as needed, according to their data growth. Adaptation can be done manually or automatically. The models constantly improve as they get trained with new data, without the need to contact RWS.
- Deployment flexibility
- Adapted models can be deployed on premises in Language Weaver Edge. The models are encrypted and cannot be reverse-engineered to extract data. Therefore, the models can be safely deployed to Language Weaver 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 existing, domain-specific translated data to deliver new high quality translations that are in line with their area of expertise and variety of offered services.
- Higher translation accuracy
- An adapted language pair built on a sufficient amount of valid data normally results in higher translation accuracy for domain-specific terminology.
- Unique translation assets
- Language Weaver Edge Adaptation 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 a new adaptation by adding the new data to the existing one, thus improving the quality of the adapted language pair.
- Fast customization process
- Customers can create new adapted language pairs whenever needed, on their own schedule, without investing too much time in the process: performing an adaptation 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 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 adapted language pairs
Generic 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 from very specific domains or unique to a restricted number of businesses. For customers who require a quality boost, using adapted language pairs is a highly recommended solution.
Adapted language pairs are built using additional, domain-specific parallel data that customers have translated previously. As a result, adapted language pairs can translate not only generic information, but also highly specialized vocabulary, unique to an industry or a customer.
Using generic models as a baseline to create new adapted language pairs
Language Weaver Edge Adaptation can create adapted models by training an existing language pair using a bilingual corpus. It is not meant to create a totally new language pair from scratch.
Only those language pairs that support Language Weaver Edge Adaptation can be used as a baseline to create new adapted language pairs.