STAR was tasked with finding potential solutions for Endress+Hauser’s ultimate objective. After several selection rounds and a proof of concept, STAR was able to start implementation. Following numerous customer-specific adaptations, the system went live. A few components had to be developed, and processes defined and tested before the fully automated process was ready.
The Salesforce connection to STAR CLM is the result of close collaboration with the Salesforce team and is established via the SF2CLM connector, which packages up the information to be translated in Salesforce and transfers this to downstream systems. A middleware, in the form of a bridge API developed by STAR, then prepares the incoming packages in a COTI-compatible format and passes this on to STAR CLM, which is used to control the MT translation process. Project setup, receipt and delivery of the translation packages are fully automated by means of predefined workflows.
KBA texts are machine-translated using neural MT engines that are trained by STAR specifically for Endress+Hauser. Unlike generic systems that only take their training material from publicly available data sources, Endress+Hauser engines are primarily based on customer-specific materials, resulting in a tone and terminology application that is better adapted for this use case.
Through rounds of feedback with the Endress+Hauser contact partners, the initial quality of the MT engines was established and continually improved through a series of re-training measures and acceptance processes. There are now nine languages available in total, which can be combined with each other however needed. Depending on the language combination, English may be used as what is known as a pivot language in order to increase the translation quality in languages that have fewer resources available.