Failure of asylum investigation by Swedish authority calls for quality

As reported on Swedish Radio, SR Ekot, on Wednesday, an asylum investigation does not supposedly meet basic standards.

Yet again has a case where linguistics has been used with disputable success in an asylum investigation surfaced.  The particular shortcomings may vary from case to case.  Together they tell the story of a system malfunction.

Each derailed case, whether discovered or not, is important.  Sadly, there are no signs they will cease to crop up any time soon.  The real number of failures is not known and is not estimated with any accuracy, even by the authority, the Swedish Migration Board.

Defined method

One would expect a strictly defined method to be followed by the provider of the language analysis, but authority requirements are few and lax.  While a patent office would scrutinize the mechanism of an invention to assess its utility, the authority is content with the language method being a black box.

Not knowing whether the black box is likely to yield consistent and reliable results, it is still possible to look at a random sample.  The more items you inspect without encountering a failure, the more certain you can be that the proportion of failures is below a certain threshold.

The Swedish Migration Board was satisfied after 10 items and concluded the quality was good.  Now, if there is 1 failure in 20, the likelihood of not encountering an error in the random sample is about 60 %. In other words if the test was designed to detect a failure rate of 5 %, 6 random samples out of 10 would fail to do so.

Three criteria

An impressionistic approach to language testing is a poor substitute for scientific rigour.  To promote transparency and intersubjectivity, three criteria should be met by any dialect attribution:

Firstly, the ear of a speaker of the suggested dialect, which is narrowly defined, should be employed.  This speaker should not be used for judgments about other dialects.  Secondly, the observations by the native speaker should exemplify linguistic patterns and each such patterns used in the argument should be supported by several examples, including any deviant usages.  Thirdly, the patterns referred to should be described in the literature.

Just like an expert ornithologist can describe what traits he would be looking for to classify a bird in a certain way before he sets about examining, so must the linguist be able to do with a dialect. The only honest alternative would be to humbly refrain from making any scientific claims.

Jurisprudence is accumulating slowly but steadily in this area throughout the European Community.  Linguistic theory and novel research is being increasingly deployed to enable safer and more clearly defined practices to develop.  Verified welcomes all initiatives to raise the standard of quality in this area.  This effort will likely benefit from contributions from all stakeholders: authorities and courts, NGOs and demographic experts, linguists and statisticians.

Only then can we hope to see fewer of these disturbing cases.

Lars Johan Lundberg
Director Research & Development
Verified