Assessing soft skills in a second language: what the research shows

The question employers face

Should a candidate be assessed in the language the job is done in, or in the language they know best? Both answers have a point. If the job is done in Finnish, Finnish is part of the job. But if the aim is to see how someone handles a difficult customer or a stressed colleague, a second language can hide the skill you are trying to see. The research below helps separate the two.

What a second language changes

It uses up working memory. Speaking or writing in a second language takes attention that would otherwise go to the problem. In a 2024 study of 857 applicants and students in health-care programmes in Germany, non-native speakers of German scored lower than native speakers on a written situational judgment test, and the gap was smaller on a test that used animation instead of text; the authors point to the extra mental effort of working in a second language. Under time pressure, this can show as shorter answers and fewer things handled at once.

Reading between the lines is slower and less reliable. Understanding what a colleague means but does not say (a hint, a complaint wrapped in politeness, irony) is harder in a second language. In a 2022 study of second-language speakers of English, those with lower proficiency drew the implied meaning less often than those with higher proficiency, and working it out slowed them down more.

Emotion is flatter, in both directions. People show less emotion in a second language and also react less to emotional content in it. The "foreign language effect" has been reported in many experiments: people often make more detached, less emotional choices when a problem is presented in their second language (the effect is small, and not every replication finds it). In a hiring simulation, this can read as calm, or as cold, depending on the assessor.

The same answer gets a lower rating. A 2024 meta-analysis of hiring studies found that candidates with a non-standard accent or dialect were rated lower for the same content. AI systems can carry the same pattern: a 2024 study in Nature found that large language models judged people more negatively when the same meaning was written in a dialect (African American English) rather than in standard American English. Any assessment that uses AI scoring has to test for this directly.

Self-report shifts with language. Personality questionnaires give slightly different answers depending on the language they are taken in. This matters for questionnaires; it matters less for methods that observe behaviour.

Which soft skills are most affected

The findings above point mostly at reading people: noticing what a colleague feels, catching an indirect request, choosing the right tone with an upset customer. Skills such as social perceptiveness, service orientation, empathy, negotiation and persuasion depend on exactly the things that a second language slows down. Skills that are mostly about structure and decisions (planning a shift, prioritising, deciding what to do with limited information) are likely touched less, though time pressure still costs some of the candidate's attention.

A practical consequence: a second-language candidate who scores lower on people-reading skills but normally on decision skills may be showing a language effect, not a skill gap.

The counter-case: sometimes the language is the job

None of this means the job's language should be ignored. If a support agent will take Finnish calls from Finnish customers, reading an upset Finnish customer in Finnish is the skill, and assessing it in English would be the unfair choice. EU equality law allows a requirement that puts people of some ethnic origins at a disadvantage, such as a language requirement, only when the work justifies it and it is appropriate and necessary; if a candidate shows facts that suggest discrimination, the employer has to prove there was none. The honest rule is: assess in the job's language when the language is part of the job, and let the candidate choose when it is not.

What test standards say

Professional standards for adapting tests across languages are clear and strict. A translated test is not the same test until it is shown to be: the International Test Commission's guidelines ask for careful adaptation, a check that scores mean the same thing in each language (measurement equivalence), and documentation of both. A score from a version that has not been checked should not be compared with a score from another version. This is the standard any multilingual assessment, human or AI-scored, should be held to.

How to assess fairly in practice

1. Decide whether the language is a requirement. Write down why. "Our customers speak Finnish" is a reason; "the office language is English" may or may not be, depending on whether the role actually works in English.
2. If it is not, let the candidate choose. Offer the languages the job can be done in, and let the candidate pick the one they work best in. Do not ask for their first language; ask which language they want to be assessed in.
3. Use plain language where you can. Shorter sentences and common words lower the load without changing the task. Keep the situation and its difficulty the same for everyone.
4. Look at the pattern, not one score. A dip on people-reading skills with normal decision skills deserves a second look in the interview, in the candidate's stronger language if the job allows.
5. Compare only within a checked language. If candidates were assessed in different languages, treat the scores as comparable only if the assessment has been shown to score the same behaviour the same way in both. Ask the vendor for that evidence; if it does not exist, compare within language.
6. Tell candidates which language they will be assessed in before they start, and record which one they used.