Like Buttons on a Vending Machine: Language Issues in Survey Creation

by | October 13, 2011
Category: Blog

From Surveying 101, we all know that how you ask a question can radically change the answer you get. We work in a bilingual office (English and Kiswahili) and are researching a bilingual population (Kiswahili and Kimaasai), so this truth is even more poignant.

It was important for us that our survey was delivered to the enumerators in Kiswahili because it lessens the risk that enumerators inconsistently translate the questions that they ask farmers. We also needed an English copy of the survey for a variety of reasons. Consequently, we had a few translation issues to work through.

Benefits of ODK
Open Data Kit is a great tool for conducting surveys in several languages. The surveys are set up in such a way that the words and buttons that appear on the screen act as a skin. The enumerator can press a button that says poultry in Swahili (kuku), but ODK can still save the answer into the database as “poultry.” With ODK, we can easily conduct a survey in many languages across many countries, but manage a database of responses that are all in one language[1].

You can also think of it as the label on your vending machine button. No matter what the label says, that button will dispense a drink from the same part of the machine. In this way, both ODK and vending machines can be easily utilized around the world. For ODK, it is just a matter of creating “labels” in the language that is most useful for your respondents.

When to translate
Our survey was initially discussed and written in English. We have partners in New York City and Haiti that were contributing questions and ideas, so this was a necessity. The survey went through many drafts before it reached the final version, and we had to decide at what point to start the Kiswahili translation. On one hand, we needed time to translate, discuss, and vet the Kiswahili translation and word choice. On the other hand, every change to the English version meant additional changes to the Kiswahili version. Furthermore, we were creating a paper-based version and a digital version in both languages, so any change meant a change to four documents. Eventually we decided to start translation when we were about 80% finished with the English version. In the future we might wait a little longer. When you have to make every change in four places, you increase your risk of entering something incorrectly.

Vetting the language
Kiswahili, like most broadly spoken languages, enjoys different spellings, expressions, and vocabularies, so it is very hard to reach a consensus on a “correct” translation. We had five or six native Kiswahili speakers review the translation, and each person requested at least a few corrections. Ultimately, we deferred most contentious issues to the Kiswahili speaker with the greatest knowledge of the people in our research area. Our goal was not to develop the most grammatically correct set of questions, but rather it was to develop the most easily understood set of questions. Even if we use this survey again with another Kiswahili-speaking population, we will likely adjust the translation for that population.

[1] CAVEAT: This is not necessarily true for questions that require a text response. The information will appear in the data exactly how it was entered into the Android device. If your enumerators across several countries all speak English, they could enter all text responses in English. Otherwise, someone will still need to translate those responses at some point.

Submit Comment

Your email address will not be published. Required fields are marked *