An in-depth guide to survey translation

Market researchers are well aware of the survey translation minefield of bias and speeder answers. And yes, we even explored in 5 Top Market Research Trends how AI can take much of that headache away. 

But potential pitfalls become a whole lot deeper when surveys go global. Put bluntly, much can be lost or unwittingly gained in a botched survey translation.

In this blog post, we will look at how survey translations can contribute effectively to global market research rather than to its detriment. Because yes, doing them badly can come back and bite your business. 

We will examine current best practices in translating surveys, and strip down the jargon of survey overlay and XLATE files. And we will be able to conclude if we can deploy tools and survey platforms to conquer the world ourselves, or perhaps need a little expert help along the way.

Survey translation: Getting it right

To put us in a receptive frame of mind, let’s remind ourselves why survey translations matter. The world’s 1.3 billion English speakers cover just 17% of the global population. European giants like Cint have seen a doubling of non-English surveys from 5 million in 2016 to 10 million in 2020. Languages featuring more in survey research include Arabic, Korean, Vietnamese and Polish. Surveys translated into the world’s top ten most spoken languages can expand reach to 80% of the world.

A good multilingual survey will be localised. Native speakers refine the content to address cultural background and belief systems. Respondents will understand the questions better and give more accurate answers as a result. The fact is, translation and good localisation can make or break a project.  Getting it wrong is not an option. Interestingly, the pilot testing of a 2023 Danish study surveying 500,000 adults identified more frequent misinterpretation of their survey questions due to cultural issues than linguistic translation errors. 

Bad or no localisation can result in:

• Higher dropout rates when content becomes incomprehensible or appears ridiculous.
• Inaccurate data from the respondents who battle on and make assumptions about meaning. Costly data scrubbing will result.
• Being potentially offensive to that particular culture.
• Making the business appear inept and unprofessional. 

Boost translated survey success:

• Keep survey questions simple and concise minus the jargon. Response rates will be higher.
• Stick to closed multiple-choice or yes/no questions if possible. They reduce translation costs and avoid potential data analysis mistakes when coding open-ended responses. 
• Use graphics and visual aids sparingly if unsure, as these will all have to be localised. Colours, designs and images have different meanings to different cultures. Studies show, however, that emoji-based sad to happy scales work better universally than numbers. For example, 1 or 10 is high, depending on the country. Use of faces instead of words has been known to bring variance between countries down by a third. 
• Be aware of local data and privacy laws.

Survey translation methodology

Studies show that no basic survey translation workflow is foolproof, so no surprises there. But traditional methodologies continue to be tried and tested, and new workflows are improving accuracy. 

1. Traditional back translation 

• Translator 1 forward translates the source language questions into the target language.
• Translator 2 back translates the target language back into the source language.
• Researchers check the back translation against the original, and remediate any potential errors made by Translator 1.

Additional phases can be added based on World Health Organization (WHO) guidelines:

• Feedback from small pre-test groups followed by further remediation.
• Cognitive interviewing of small groups for further check testing before the final version.

Pitfalls: A bad or awkward first translation is not always detected if Translator 2 smooths it over upon back translating it to the source language. Similarly, a bad back translation might falsely signal that remediation is needed for the first one when none is necessary, costing time and money. Proper process is lacking as it does not highlight where and what translators have fixed. There has been a growing consensus that back translation can provide limited or misleading insights, and sectors such as health in particular have called out for more accurate data as a result.

2. Survey translation collaboration

Best practices now lean towards the TRAPD or Committee method for multinational, multicultural or multiregional surveys (3MC). The collaboration of translators and market researchers working together has proved more reliable in diagnosing multiple survey translation issues and identifying exactly where and how to fix them. 

• Translation: Two or more native speakers each produce independent draft translations.
• Review: The two translators and bilingual experts compare the translations and agree on the final version. Market researchers who speak nothing other than the source language can collaborate at this stage. They are the most aware of the objectives of the survey and the intent behind each question.
• Adjudication: A fluent adjudicator who understands the research design and subject matter signs off the final translation. 
• Pretest: The target language questionnaire is tested in a small study to refine.
• Documentation: All translations and edits are documented throughout. This is useful for future projects in the same language as reviewed and tested text can be reused.

The collaborative model can be complex with multiple stages, and has many variations according to budget. But simply not working in silos and sharing information throughout the process can go a long way to cutting costs and major mistakes. Market researchers should brief translators on how the survey will be delivered, which demographics are being targeted, and the required reading level for understanding. Translators in turn should identify terms or concepts that might be difficult to translate, for additional clarification or simplification. Final discrepancies can be settled together by a committee of translators, expert bilingual reviewers and market researchers. 

3. Kicking off survey translations with artificial intelligence 

Ever-improving Artificial Intelligence (AI) can be incorporated into the translation workflow to save time and drive down costs, but only as the initial starting point. Some argue that Neural Machine Translation such as Google Translate and DeepL are tools that can give a consistent level of quality across all languages. A 2021 extensive study on the accuracy of Google Translate in analysing open-ended survey responses indeed found that Google Translate performed similarly to the native-language coders of the professional translation company, but only in certain languages, Greek and Korean did not fare so well. Translation Memory (TM) is an AI tool can store past translations, unlike machine translation. It streamlines the localisation process by matching source and destination segments or phrases, so saving resources and time. It provides consistency for future translations. Translation Glossary and Translation Style Guides can also provide useful shortcuts.

And you might hate chatbots, but some chatbot surveys have achieved good results when surveying immigrants, and particularly in countries where the default communication is mobile. Response rates were also high when used to conduct quick and concise surveys on hotel stays.   

Multilingual survey launch: Time to tool up 

Going right back to basics, I guess we can always start with a Word document. Or can we? Are you seriously going to copy and paste each and every translated question into your survey platform in ten different languages? Not only that, but standard error messages might pop up to your responder in your source language, accompanied by the odd piece of random code. Just one easy lesson on how to look completely unprofessional. 

So if you’re still on a learning curve, it’s good to know that all the best known survey platforms work with multilingual translations. Most give detailed instructions and video demos, and allow manual or auto translation at the click of a button. Auto translation, however (some platforms use Google Translate) should act as a base to be checked as outlined earlier in best practices. The survey is transferred into the format encoding the survey and its translations, here referred to as working files. They are onboarded onto your survey platform.

Survey translation platforms

Alchemer (SurveryGizmo), Forsta (Confirmit and Decipher), Qualtrics, Voxco and SurveyMonkey

Working files

XML (eXtensible Markup Language) to share structured information or tables across the internet, and XLATE (translate) and XLIFF (XML Localization Interchange File Format) which contain the original source text as well as its translation. 

Success tips

Things can get complicated with changes. It’s best practice not to translate until you have completely finished building your survey. For example, Qualtrics advises that continuing to edit the original survey after using the auto-translate feature entails deleting all the previous auto-translations before generating a whole new set. Text changes are not automatically translated. 

Finally, test, test, test all translation versions of your survey. You can use various browser language settings to ensure that there are no blank answer options.

Translated survey launch: The expert’s role

Even some of the biggest market research companies hand over the management of their multilingual survey translations to the experts. Let’s look at why. 

Technologies

A good multi-language service provider will strike the perfect balance between human expertise and taking advantage of the most advanced Computer Assisted Translation (CAT) tools. They naturally research the best day-in-day-out to stay ahead of the curve. Discovering all your technical options can save both time and money. 

Professional translation providers also provide a collaborative team combining both industry-specialist engineers and linguists. That combination of the complete overlay services package will ensure your surveys are accurate and able to engage the highest number of respondents.  The vendor will deploy your TM bank for any future projects, and can advise on using XML, XLATE, XLIFF formats or others as the most compatible with your preferred survey platform. 

Web codes and piped text, when a respondent’s answer determines the follow-on question or links to another survey, will all be taken care of. Local professionals can also conduct rigorous link-checking. Correct survey links will result in high-quality survey navigation and contextual relevance, both factors that can otherwise result in high dropout rates.

The company’s APIs can also remove any further stress from the process by offering seamless content transfer, enabling your software to communicate with theirs.  

They will be your go-to expert for all those technical questions.  

Localisation

It is difficult to localise surveys to high standards without taking advantage of those multi-language vendors who have offices and local linguists at the ready all over the world, effectively operating 24/7.

Certified quality

The provider’s network of professional tried and tested linguists will have an in-depth knowledge of the target languages, and be familiar with international market research requirements. Information security will be assured under ISO27001, and ISO9001 will indicate that the company have a monitored Quality Management System (QSM) in place. That ensures your project will go through a rigorous quality assurance (QA) review phase as part of the survey translation process, and that all their linguists have been properly vetted.

Collaboration

You will be assigned an account and project manager as your point of contact, enabling close collaboration throughout the project. That personalised service will allow you as much or as little control as you need. 

Results

High quality and approved translations managed by one experienced source will ensure consistency of data, and produce more accurate results for analysis.  

One of Our Survey Translation Case studies

Kadence International faced quality and process issues around attempted localisation work on their multi-country quantitative and qualitative market research projects. They sought a partner with global offices offering a wide range of language services. Global Lingo fit the bill offering translations in150 languages from its offices in EMEA, the Americas and APAC. Kadence were allocated a dedicated Singapore-based project manager. We immediately introduced TM technology to the multilingual localisation of the survey across 12 markets, as well as a secondary QA review phase to raise standards. Further open-ended question responses were translated for coding and successful analysis as part of the large international study.   

Survey translation: The ultimate guide summary

So there you have it. The DIY and expert solutions. Which to choose? Much obviously depends on budget restrictions, but conducting an international survey is fraught with risks, lack of appropriate localisation and subsequent loss of reputation being just one. And in the current climate, organisations are increasingly challenging to be shown fully transparent methodology when it comes to ensuring that their survey data is accurate. Launching an international product or service is an expensive business. No one can afford to get that market research wrong.    

Next steps for your survey translation journey

To explore your survey translation options further, contact Global Lingo for a free consultation via telephone, email or online.