By Marie Hense, Vice President of Data Quality, Toluna
Today’s world is becoming increasingly multilingual. Countries’ populations consist of people from diverse backgrounds and with various languages, whether due to regional differences (in large countries such as India) or immigration and the subsequent formation of subpopulations (such as Spanish-speaking communities in the US).
To facilitate day-to-day living in these multilingual societies, technology has stepped up to the plate with the development of translation tools, including web-based translation services, voice-based translators, and translation add-ons that automatically translate everything shown on a device’s web browser in real time. While these tools are brilliant for everyday life and enabling communication in diverse communities, the blurring of linguistic lines poses a challenge for research.
There are three main areas of consideration surrounding this topic: data accuracy, inclusivity, and fraudulent activity. All three need to be carefully evaluated to decide whether respondents using an auto-translate function while taking a survey should be permitted to participate or be excluded from the data collection.
In research, we need to ensure that questions are understood in exactly the same way from one respondent to the next, and the nuanced wording of questions makes a real difference in this. If you’ve ever used an auto-translator, you will know that the quality of the translations is still far from perfect.
For those of us working in research, the potential inaccuracies introduced by auto-translations can seriously challenge our ability to compare responses. Auto-translations can also introduce skew by using ‘loaded’ language in place of the original text’s carefully crafted, neutral wording. This can bias respondents and cause them to reply in ways they wouldn’t have done in the survey’s original language.
Some people have an auto-translate function activated on their device because they need it to navigate their day-to-day life. By excluding these respondents, we may risk the representativeness of our sample. This would exclude respondents from certain backgrounds, whose demographics, socioeconomic factors, behaviours, and attitudes may be different than those who speak the survey’s language – and thus we would fall short in providing representation for key subpopulations in the market we’re researching.
Some ill-intentioned respondents hack into surveys from foreign countries and use auto-translators to understand them, even if they don’t speak the language. In doing so, they’re often able to fly under the radar by selecting semi-logical answers that pass basic in-survey quality checks, such as red herring or trap questions.
Excluding respondents from a survey needs to be done with care and consciousness. In a world where we are competing for respondents’ attention and time, we cannot afford to exclude any respondents unless their inclusion would risk the quality or integrity of our data. In this case, the risks of misunderstood questions and fraudulent respondents outweigh the risk of reduced representativity and inclusion.
Respondents who have an auto-translate tool activated on their device should be prevented from entering surveys in order to ensure data accuracy and the genuine nature of the participants. In cases where there is a desire for full population representation, surveys should be scripted in several languages for the same market to ensure the inclusion of key subpopulations.
While technology is a fantastic enabler that has transformed the speed, agility, and scalability of insights to provide businesses with intelligence that helps inform and shape their strategies, we must be conscious of how advancements in technology may impact the quality of our insights. Research should never lose its fundamental benefit to organisations: providing information for direction and decision-making – but only good-quality data will be able to do so effectively.
This article was first published in the Q1 2023 edition of Asia Research Media