Two distinct methodologies for Machine Translation (MT) stand out for enhancing translation accuracy and relevance: MT Customisation and MT Training.

As organisations increase their adoption of Machine Translation (MT) and drive programmes to integrate multilingual content, there is a great need to make the output of those MT engines more relevant to the brand, product, or service for which the content will be used. There is also a need to ensure translation accuracy. These are achieved by implementing a complete programme involving multiple stages.

We regularly refer to two of these stages, which are often confused: MT Customisation and MT Training. While both approaches aim to improve machine-generated translations, they employ different strategies and serve unique needs.

MT Customisation and MT Training are powerful tools for improving the accuracy and relevance of machine-generated translations. While MT Customisation offers a quicker, more accessible way to refine translations through specific linguistic instructions, MT Training provides a deeper, data-driven method for tailoring the MT engine's output to a company's unique requirements.

Let’s delve into the nuances of MT Customisation and MT Training.

MT Customisation - Tweaking Existing Translations

MT Customisation involves modifying an existing Machine Translation engine to better align with specific linguistic preferences and requirements. This is achieved through the integration of a translation glossary and a Do Not Translate (DNT) list. A translation glossary comprises critical terms specific to a company and their corresponding translations, while a DNT list includes terms that the company wishes to exclude from translations, such as product names, branding, logos, etc.

The customisation process enhances the MT engine's accuracy by instructing it on how to translate (or not translate) specific terms. This targeted intervention ensures that translations reflect the company's brand identity, adhere to preferred terminologies, and accommodate regional linguistic variations. As a result, the quality of translations improves, reducing the necessity for extensive post-editing.

Implementing MT Customisation is generally more straightforward than undertaking MT Training. However, the effectiveness of customisation significantly depends on the expertise of the MT specialist managing the process. Skills in selecting the right terms, as well as handling input and output normalisation rules, DNT lists, and glossaries, are crucial. A lack of experience can lead to low-quality suggestions from the MT engine, negatively impacting the overall quality of translations. It's also important to note that not all MT platforms support customisation, which can influence the choice of MT engine.

MT Training - Teaching the Computer from Scratch

MT Training takes a more foundational approach by creating and refining an MT engine using extensive bilingual data. This data is sourced from corpora and Translation Memories (TMs), which are archives of previously translated content. The training process involves feeding the MT engine with company-specific bilingual corpora (data sets), often imported in TM format, which includes not just the translations but also metadata such as the translation date, the linguist's identity, and the degree of match (exact or fuzzy) with the source text.

This rich dataset enables the MT engine to learn and predict the company's translation expectations. Rather than generating generic translations based on a broad interpretation of the source material, the trained engine produces tailored outputs that reflect the specific linguistic and stylistic preferences of the company. Like customisation, not all MT platforms are equipped to facilitate training, making platform selection a critical consideration for organisations looking to adopt this approach.

Which Way to Go?

MT Customisation and MT Training can both be combined within the same programme, as MT Customisation can be applied post-training to further refine an engine.

Choosing between MT Customisation and MT Training depends on several factors, including the specific use case, the available bilingual data, the MT platform's capabilities, and the expertise of the team responsible for implementing the MT strategy. By carefully considering these aspects, organisations can select the most appropriate method to enhance their translation processes, ensuring high-quality, brand-consistent, and regionally appropriate content.

In the end, it's about making sure the engine's translations sound right and make sense for your company, whether that means giving it a quick lesson or a full-on course.

How can LanguageLine help you?

Our MT programme experts will work with organisations looking to adopt MT, carefully considering these aspects as well as other factors like automatically estimating the quality of the engines or implementing human quality control mechanisms to enhance their translation processes.

Want to learn more? Contact us to learn about how we can help you create a customised MT programme that aligns with your needs.

Sales Enquiries - Contact Us - LanguageLine UK


Translation & Localization - Get the expertise you need to communicate clearly at any scale in over 290 languages. Download the guide