This article was originally published in October 2020 and has been updated.
Terminology management is a critical and key component to ensuring optimal translation quality for all localized content. It is of notable importance to our Manufacturing and Automotive clients who require an exceptional level of accuracy from their technical translation services provider. Key terms and client specific lexicon should be fully integrated in a technical translation workflow.
Effective terminology management also supports our clients’ style, brand voice and tone, and maintains these points of differentiation even in technical manuals and support content. Any trademarks, slogans or brand names should be incorporated consistently across all published content and not siloed as “marketing” language.
Not Performing Terminology Management Risks the Quality of Your Machine Translation
Without robust terminology management processes, a Manufacturing client’s documentation and/or brand is put at unnecessary risk. In a space where Machine Translation has become a cornerstone of technical content localization, special attention should be paid to terminology and ensuring generic engines are not used in isolation or without customization. Neural Machine Translation (NMT) systems are typically trained with large sets of generic bilingual data. While this process yields good quality for generic texts, it does not provide the necessary accuracy expected for customized domains or client-preferred vocabulary. Using only generic engines also runs the risk of diluting well-earned brand recognition.
Terminology Management Leads to More Targeted Machine Translation
To address and improve machine translation customization and targeted results, neural machine translation models are trained with business or client-specific data using Translation Memories. Nevertheless, over-training of generic NMT models with specific data does not guarantee that the correct terminology is applied in all cases. For this, linguistic corpora needs to be analyzed to identify specific categories of content within the data and ensure that each category has the necessary set of terms to support it.
For one of our clients in the agricultural manufacturing space, examples of common categories for term lists would be agricultural software and UI, crop science, service and diagnostics manuals, user instruction guides, and legal and warranty categories. ULG’s Language Asset Management (LAM) team has the skills to analyze and create the most effective category groupings for our clients.
ULG's Approach to Machine Translation
ULG’s LAM team consists of data scientists and computational linguists who are fully dedicated to training our neural machine translation solutions to meet target language expectations. Using our customized technology, the team manages data selection, terminology extraction and analysis, data categorization, TM analysis and cleaning, retraining of existing NMT baseline models, and testing the quality of the resulting translations. Ongoing testing is key to ensuring that our neural machine translation training is effective and producing quality, customized content for our clients. We use KPI’s to report regularly on performance and invest in the assessment of linguistic assets to improve our inputs and outputs during the process.
Successful terminology incorporation is only one of the KPI’s measured when rolling-out a successful neural machine translation workflow, but it is a critical one. It should be monitored and reviewed regularly by both language service providers and clients. For more information on how to effectively measure or implement a terminology management program, please contact your ULG representative or submit an inquiry through our website.