en

Many thanks!

We have received your enquiry and will be in touch as soon as possible.

SwissGlobal Newsletter

Lost in translation? Not with SwissGlobal. Stay up to date with language industry news, tips, interviews and more – subscribe to our monthly newsletter.

Integrate glossaries into your machine translations for maximum accuracy

Integrating a glossary into your machine translation process is essential for maintaining consistency and accuracy. Whether during training of a machine translation (MT) model, in real-time translation with an online translator, or during post-editing, leveraging a glossary ensures that your preferred terminology is consistently applied.

This article will give you glossary creation tips and guide you through the stages of machine translation to learn where glossaries can be integrated into your workflow to optimise results.

Where does glossary integration fit into your MT workflows?

Key stages where terminology can be integrated include:

  • In the training phase, the MT model learns preferred terms from the start.
  • Real-time translation, where glossaries ensure that terminology is applied consistently as the translation is generated.
  • Post-editing of MT output, where human translators use a glossary to refine and perfect the output.

Glossary integration at any point ensures that the final translation aligns with the original intent and maintains high quality throughout. This structured approach improves translation accuracy and enhances overall efficiency.

Why should you integrate terminology in MT training?

Advantages:Key considerations:
Lasting impact: Embeds your preferred terminology into the core of the MT model.
Consistency: Ensures consistent terminology use from the start.
Efficiency: Reduces the need for extensive post-editing.
Extensive clean training material: Requires a well-prepared and consistent source language dataset.
One-time effort: Although it’s a one-time effort, it demands thorough preparation to be effective.
Limited flexibility: Changes in terminology require retraining the engine, making it inherently less adaptable to updates.

Incorporating terminology during the training phase of an MT model can be a game-changer. The model learns and prioritises your preferred terms from the outset by embedding your terminology directly into the training material. This one-time training effort ensures consistency is built into the very foundation of your translation engine.

This approach has several advantages: it embeds your preferred terminology into the core of the MT model, ensures consistent terminology use right from the start, and significantly reduces the need for extensive post-editing.

However, there are important prerequisites and considerations to keep in mind. Effective implementation requires extensive, clean training material, which means having a well-prepared and consistent dataset. While it is a one-time effort, thorough preparation is crucial to its effectiveness.

Another important consideration is the static nature of MT models. An MT model trained with custom data is designed to recognise and use specific terms as they were during the training process. However, glossaries naturally evolve as language and industry terms shift over time. When terminology is updated – whether terms are added, removed, or modified – the model may struggle to adapt, using outdated terms or failing to recognise new ones.

Due to the structured nature of MT model training, incorporating these changes would require retraining the model, which can be resource-intensive and time-consuming. While the initial integration of specific terminology helps ensure consistency, it’s important to know the model’s limited ability to adapt to ongoing changes without retraining.

How does real-time glossary integration work?

Advantages:Key Considerations:
Flexibility: Offers real-time application of your preferred terminology across various contexts.
User-friendly:
Ideal for internal teams, ensuring consistent terminology use.
On-the-fly adjustments:
Easily adaptable to different projects and languages.
Terminology list: Requires a well-structured list or database of terms.
1:1 mapping of terms: Needs positive 1:1 mapping to avoid translation ambiguity.


Modern MT engines, like DeepL, allow injecting glossaries during translation. By uploading your terminology list, you can ensure real-time adherence to your guidelines through plugins or built-in glossary functions. This integration offers several advantages, including the flexibility to apply your preferred terminology and allowing for on-the-fly adjustments. It’s user-friendly, making it ideal for internal teams to maintain consistent terminology across the board.

That said, there are significant considerations to account for. When building a DeepL glossary, it’s essential to prioritise terms critical to your content that significantly impact translation accuracy. Overly generic terms may not add much value and should be left out. Focus on industry-specific terms, technical jargon, proper nouns, and key phrases where consistency is crucial. Another important aspect to consider is avoiding duplicates. DeepL applies only the first equivalent it encounters, ignoring any subsequent variations. This can lead to unexpected results if similar or identical terms are listed multiple times with different translations. Reviewing frequently used terms and phrases helps prioritise what should be included and ensure that each term is listed only once with the most appropriate translation.

Proper formatting, using a simple source-to-target structure in a CSV file, aids in seamless integration. Use a simple format with one column for source terms and one for the target translation, ensuring no extra characters.

The value of using post-editing for consistency

Advantages:Key considerations:
Human-in-the-loop: This adds a human layer to refine MT output, ensuring high-quality translations. 
Terminological accuracy: This reinforces the correct use of specialised or complex terminology. 
Consistent quality: This maintains consistency and accuracy, particularly in specialised fields.
Clear specifications: Requires guidelines and integration into the post-editing workflow.
Effort: Demands thorough effort from translators to match the output with glossary terms.


A glossary is a valuable tool for human translators in the post-editing phase. Post-editing is the human element in the machine translation process that refines the machine output, enhancing the overall quality of translation. After the MT engine generates the initial output, post-editors can use the glossary to ensure the final product aligns with your terminology standards. This approach helps reinforce using specialised or complex terminology correctly, ensuring terminological accuracy.

Choose SwissGlobal for your glossary integration

Integrating a glossary at various stages of your machine translation workflow can be very helpful in achieving high-quality translation. Independently of which part of your workflow a glossary is integrated at, whether during custom model training, integration in DeepL or on the post-editing stage, a well-maintained glossary ensures that your preferred terminology is consistently applied, enhancing accuracy and efficiency.

Ready to optimise your translation process?Contact us to learn how our glossary management services can help you achieve greater consistency and accuracy in your machine translation process.