Removing Language Barriers with AI Machine Translation

Increasing confidence in deciding to publish or edit machine translation content

Blindly trusting a Machine Translation (MT) system to publish content to international markets is not an option. A company needs to have quick oversight of the quality and accuracy of the output of MT. Multiple types of MT systems are available, offering varied degrees of speed, customisation and reliability. The performance of a particular MT approach is dependent on the data context and goals for product positioning and reach.

Industry Challenge

Challenge: Confidently evaluate a range of different MT technologies comparing translation accuracy and quality estimation to the computational complexity of the approach.

ADAPT Solution

ADAPT’s solutions in Neural Quality Estimation (NQE) and Neural Automatic Post-editing (NPE) take the best technologies available to put content owners at the helm of MT content publication and editing, these solutions are compatible with business demands in terms of cost and resource efficiency.

– MT post-edited data for training (different languages, different documents, content, etc.)
– New MT content for testing

Technologies: Open-source Neural Quality Estimation (NQE) and Neural Automatic Post-editing (NPE) systems

– MT outputs with reduced editing
– Visibility over content with high publishability and content with high editing requirements

Diagram showing how AI can be used to remove language barriers

Results & Benefits

– Neural methods can be used not just to translate but also to control the quality of translation.
– This project established that neural estimation of MT quality is ready to be deployed in corporate environments, it achieves high improvements over current methods, and there are cost-efficient solutions.

Industry Benefits

– Control the risk of publishing MT content
– Control the cost of editing MT content
– Identify and select higher quality MT content
– Increased knowledge over types of content and their adaptability to MT
– Develop cost-effective processes to deploy NMT
– Enhance use, reuse and creation of MT content

Use Cases

– Internationalisation and localisation of physical and digital products and services
– Real-time translation
– International customer service

Download PDF