A translation agency requested development of a multi-tenant translation quality assessment solution dynamically adjusting consumed cloud resources to the amount of files processed at any point in time.
CAT stands for Computer-Added Translation
Our Customer is a global provider of translation and localization services with an impressive client roster of about 800 brands. Over 5,000 of the company’s employees operate cloud-based CAT tools to help businesses ensure consistency of communications worldwide while preserving local relevancy and technical usability.
To further polish the existing localization workflows the Customer needed to integrate a cloud-based fully automated translation quality analysis tool into the day-to-day operations of their localization managers. Performing quality assessment in the background mode would free the company’s employees from manual operations, thus, streamlining time-to-consumer while improving the overall translation quality.
Itransition’s dedicated team delivered a flexible, highly elastic solution with no single point of failure. The system’s ability to seamlessly adapt to the varying workload in real time ensures the Customer only pays for the resources currently consumed, with rising workload demands proactively met.
Having set clearly articulated requirements to the system at the outset of the cooperation, our Customer was fully satisfied with the achieved results. Currently, further extension for real-time translation quality analysis in the Customer’s CAD tools is considered.
The solution is used to perform the overall quality analysis of machine or human translation, from spelling and vocabulary analysis to punctuation analysis, segments consistency checks, etc.
The software package comprises:
From a technology perspective, the solution is represented by a distributed asynchronous multi-agent computing platform operating translation quality algorithms, with Rabbit MQ as a service bus. The system provides RESTful API, which is available to external systems.
Tough requirements were set to system reliability, availability, performance and elasticity.
A critical requirement to the system was the ability to dynamically adjust provisioned cloud servers to the amount of files processed at each moment of operation. The delivered solution scales up to 100 distributed nodes, processing 30-100 execution orders per minute (i.e. 15 Mb/s).
Running the algorithms takes from 2-3 sec to 3-4 hours per task, depending on configuration and input files varying from 100 kb to 10 Mb each.
The system operates translation quality algorithms originally developed by the Customer’s in-house team. Itransition’s specialists performed source code refactoring in core algorithms to enable multithreading analysis while preserving backward compatibility with existing applications.
Besides, our team enabled to update and extend algorithm functionality independently in another branch, with the ability to easily merge changes to multithreading version.