Optimizing the translation process: AI, agencies and countries in interaction

Requirements and recommendations for an optimal translation process

For internationally active companies that offer products that require explanation, the translation quality and the translation process are crucial for their marketing in the respective countries. In this blog, I would like to use my project experience to describe various aspects that need to be taken into account and make recommendations to make the process as efficient as possible while ensuring the best possible translation quality.

Organization and tools for the translation process

High-quality translations are achieved above all when technical terms are used correctly and consistently and the translations have been checked by native speakers. An efficient process can be achieved through defined personnel responsibilities, the use of tools and the standardization of processes:

• The deliberate definition of personnel responsibility seems self-evident, but in practice it is not always so clearly identifiable. It means, on the one hand, designating employees who continuously monitor the translation process and manage the translation orders and, on the other hand, designating employees from agencies or the countries who create and check the translations.
• Tools used in the translation process include TMS (translation memory system, such as Trados or Across), which are always used by translation agencies (even if clients are sometimes unaware of this), AI (such as DeepL, Google or Microsoft Translator) and tools that support content creation (such as Congree). Yes, and of course the system in which the texts are originally created, such as PIM, editorial or CMS systems.
• The standardization of processes begins with the creation (and modification) of texts in the original language. The creation itself is crucial for the subsequent translation quality. For example, if technical terms are not used correctly and consistently in the source language, even the best translation will not help. Other aspects include text management and the text workflow, which are described below.

Recommendations for content creation - terminology

Of the many points that immediately spring to mind, I'll start with the most "stressful" point: terminology. As mentioned above, this is about the correct and consistent use of technical terms in the original texts. These should be laid down in a glossary, not only the terms to be used, but also prohibited terms ("negative terms").

Glossaries typically contain hundreds to thousands of technical terms. Why is this exhausting? Because employees have very different ways of speaking, and such standardizations always lead to friction as to which term should prevail and which terms should no longer be used. Here's a little anecdote: I once experienced one employee getting his way on the catalog page ("junction box") and another in the table of contents ("junction box").

How do you arrive at the technical terms? There are various ways that should be followed in equal measure. On the one hand, I like to ask whether there are already Excel lists of technical terms. You can often find them there. Term extraction" is more time-consuming: terms are first extracted from the translation orders using software and listed in Excel. These lists are checked for "incorrect terms" and consolidated. The more terms that already exist, the fewer are extracted; in the best case, everything matches. The consolidated technical terms are then translated into all languages and used for the translation tools (TMS or AI). For example, with TMS you can save the translations of terms in a small translation database or upload them directly to the translation order with AI tools. When translating, the translation is then immediately suggested for the relevant technical term, which reduces translation times and also minimizes correction costs.

Ideally, the correct use of terms should already be checked during content creation, for example with Congree. The tool can be integrated directly into the text editor. For example, we have integrated Congree's authoring service into our PIM system via an API interface.

Recommendations for text management

We have now talked a lot about terminology. There is another point that is far less strenuous and has a very positive effect on the translation process. This involves breaking down enumerated texts into individual, granular text modules. The text modules can often be reused in other bulleted texts, especially if you have products with many variants. This has two effects on cost reduction: firstly, reuse saves translations, and secondly, changes only have a minimal effect, as only the individual text module needs to be "touched" and not the longer bulleted text. It also has a positive effect on translation quality, as the translations are always consistent. Of course, the PIM or CMS system must offer the functionality described above, so you should take this into account when making your selection.

Another point that is often neglected concerns graphics that contain text: usually in the form of individual technical terms, for example captions in diagrams, drawings or functional graphics. These captions are often found in other graphics. Their translations always have to be laboriously set by hand in the graphic and saved as a file for each language.

The following procedure is recommended here, for example with the use of Adobe Illustrator:

1.) Extracting the texts from the graphic and replacing them with placeholders.

2) Save texts as text modules, translate and reuse in other graphic designs

3) Automatic insertion of text modules or translations into the graphic instead of placeholders

4) Automatic saving of the graphic designers per language as a file.

Recommendations for the standardization of processes and the use of tools

The simplest recommendation is to appoint one or two central contacts to manage the translation process. All translation orders should be managed centrally.

This promotes my golden rule of "do the same thing the same way".

There are two possible ways of standardizing processes in relation to translation:

1.) Pre-translation with AI (using the glossary as described above) and review by native speakers, for example in-country.

2) Translation by agency with TMS (using the glossary), with or without subsequent checking.

Of course, a mixture of both variants is also possible, for example specifically by language, as not all languages can be optimally translated with AI.

A note on variant 1: In order to be AI Act-compliant, it is mandatory to mark AI-translated texts and to store when and with which tool the AI-generated text was created.

Use of TMS and AI

Tools will be used in both variants. It is important to ensure a seamless, direct technical connection. State-of-the-art are API interfaces and especially for TMS the COTI Level 2 interface (https://de.wikipedia.org/wiki/COTI).

The workflow must meet the following functional requirements:

• It should only be possible to determine the translation requirements that are really necessary, i.e. where translations are missing or where texts have changed.
• Texts that are currently being translated must be flagged to take account of any text changes during the translation process.
• Changes to the original text ("master text") must trigger a translation of existing languages ("revision").
• Versioning of texts - and synchronization of the appropriate translation must be possible
• Formatting of texts, such as headings, paragraphs, lists, bold/italics, must be implicitly retained during translation; ideally, they should even be formatted displayed. Nothing is more unnecessary than formatting each text again after translation!

Checking and release of translations

Checking translations is actually mandatory in the case of pre-translation with AI; corrections are very likely necessary and useful.

Even when working with agencies, which typically also use proofreaders, proofreading can still lead to corrections.

It makes sense for examinations to be carried out by native speakers, either at the head office or in the countries. The check should therefore be possible online. The result is a correction and release.

Conclusion

The recommendations described are intended to achieve the following goals:

• Make new languages available as quickly as possible (weeks instead of months!)
• Reduce translation and correction costs by up to 80%
• Increase the quality of content through the uniform use of technical terms and translations as well as a standardized approval process

Here is a summary of the aspects that should be considered when setting up an optimal translation process:

• Terminology
• Process for creating texts in the source language
• Management of texts (including bulleted texts, texts in graphics)
• Text workflow with change and approval process
• Automatic determination of translation requirements
• Use of tools, including TMS (with COTI2 interface) and AI (with API interface)
• Control of translation orders by central contact persons and checking by native speakers with a web interface