Does reading in a second language cost citizens understanding? A seven-country experiment

The question

MultiPoD is building an online space where citizens from across Europe can deliberate on policy across language borders. Before designing how the platform handles language, the consortium needed to know whether language matters: does reading political information in a second language reduce what people take from it, and does offering people a choice of language help?

What we did

We embedded a three-arm experiment in a survey of 8,195 adults in Austria, Belgium, France, Greece, Portugal, Spain and the United Kingdom, recruited through online panels (Bilendi, and Prolific in the United Kingdom) in January and February 2026. Multilingual respondents, people who speak at least one language beyond their first, were randomly assigned to read the same 250-word article on the EU AI Act in one of three ways: in their first language, in their second language with no choice, or in a language of their choosing.

The estimates use the 4,630 multilingual respondents who could be assigned to any of the three versions. Another 231, most of them in Spain, named a second language with no article version, such as Catalan or Basque, so they could not be given the second-language version. Including them changes the estimates very little.

Six factual questions about the article followed immediately. We also asked how difficult the text felt and recorded how long people spent reading it.

The study was not pre-registered. The hypotheses came from earlier research, and we report every specification we estimated: unadjusted, with demographics and country fixed effects, and with full controls.

What we found

Reading in a forced second language cost about half a correct answer: −0.51 on the six-question test (95% CI −0.64 to −0.39), roughly 14% of the control group’s average score of 3.7. The estimate barely moves across specifications, from −0.51 to −0.52. People in that condition also rated the text as harder, by 0.25 points on a five-point scale, and spent roughly 13% longer on it. More effort, less understanding.

Offering a choice avoided the penalty. Outside the United Kingdom, more than nine in ten people offered a choice picked their national language, so most read what they would have read in the control condition. They scored slightly above the control group (+0.15 with full controls, 95% CI 0.03 to 0.27), a difference that is statistically significant only once covariates are included. We found no detectable change in how difficult they found the text. Their reading time was somewhat longer, by about 10% with full controls, so choosing a language is not entirely free of cost.

The penalty is statistically significant in Austria, Belgium, France, Greece and the United Kingdom, and not distinguishable from zero in Portugal and Spain. Country estimates are much less precise than the pooled result and are best read descriptively.

What this cannot tell us

Online panel respondents tend to be younger, more digitally literate and more politically attentive than the general population, so effects in the wider population could be larger. In every country, roughly three quarters or more of the multilingual respondents speak English: the study speaks mainly to citizens with a foreign language alongside the national one.

Speakers of regional, minority and heritage languages with no article version, such as Catalan, Galician or Basque in Spain, could not be assigned to the second-language version. The study therefore does not speak to them, although they are among the groups that matter most for MultiPoD.

The test covers one short, factual article; longer or more technical texts could produce different effects. Completion was about two percentage points lower in the forced second-language group, and bounds that allow for this leave the penalty clearly negative.

What it means for the platform

The study’s recommendations for MultiPoD follow directly from the three findings: default to each user’s preferred language rather than to English, make language choice simple, visible and reversible, label translated or summarised content clearly, and monitor comprehension and effort by language so that no group carries a heavier burden.

Ethos Lab led the study within MultiPoD Work Package 1. MultiPoD is funded by the European Union’s Horizon Europe programme. Views and opinions expressed are those of the authors only and do not necessarily reflect those of the European Union.