It's correct according to the current algorithm. It's just not very useful ;)

The likely reason for that is that "de" appears in names of many languages in many languages, for example "amazighe de l’Atlas central" in French. See https://www.unicode.org/cldr/charts/44/by_type/locale_display_names.languages__a-d_.html#346e71a8ed4de6f8

Perhaps we could make a special case for "de", given that these are the first letter of a language that very frequently search for, and that they appear in many names of unrelated languages in a way that makes results not great. We already made some simplistic customizations and shortcuts in the search index, such as adding a "Castilian" alias (this name doesn't appear anywhere in the CLDR, but lots of people use it to search for Spanish), so making a special case for "de" sounds possible, too.

You can see the matches in the API request: https://meta.wikimedia.org/w/api.php?action=languagesearch&format=jsonfm&formatversion=2&search=de

I'd suggest avoid creating substring matches in the database for "de".

Change 1009466 had a related patch set uploaded (by Nikerabbit; author: Nikerabbit):

[mediawiki/extensions/UniversalLanguageSelector@master] Language name search: don't match on short infix terms

https://gerrit.wikimedia.org/r/1009466

Changes in terms of the suggested languages with the latest patch in place can be reviewed here: https://jsoneditoronline.org/#right=cloud.f98285f2f46c4c06bf556c30d32d7929&left=cloud.c7b00d2cb04041b3923d4ea8fb4e6ab5

Change #1009466 merged by jenkins-bot:

[mediawiki/extensions/UniversalLanguageSelector@master] Language name search: don't match on short infix terms

https://gerrit.wikimedia.org/r/1009466

In T358862#9590551, @Amire80 wrote:

It's correct according to the current algorithm. It's just not very useful ;)

The likely reason for that is that "de" appears in names of many languages in many languages, for example "amazighe de l’Atlas central" in French. See https://www.unicode.org/cldr/charts/44/by_type/locale_display_names.languages__a-d_.html#346e71a8ed4de6f8

It's searching all other languages too? That seems bizarre. The interface isn't in those languages, nor is the browser, they're not fallbacks for the current language or the target language either, they're not the autonyms, and those names aren't even being shown.

I could maybe understand searching all languages if less likely matches were further down and the matched text were shown, but right now it apparently thinks I'm more likely to be searching for the Catalan name of Minnan with the interface in English and the browser in German, than I would be searching for German by its language code or its autonym.

In T358862#9644826, @abi_ wrote:

Changes with the latest patch can be reviewed here: https://jsoneditoronline.org/#right=cloud.f98285f2f46c4c06bf556c30d32d7929&left=cloud.c7b00d2cb04041b3923d4ea8fb4e6ab5

The link isn't working for me any more, but when I originally looked at it, it was a slight improvement (there were fewer matches), but the matches still didn't make much sense overall.

In T358862#9682031, @Nikki wrote:

In T358862#9644826, @abi_ wrote:

Changes with the latest patch can be reviewed here: https://jsoneditoronline.org/#right=cloud.f98285f2f46c4c06bf556c30d32d7929&left=cloud.c7b00d2cb04041b3923d4ea8fb4e6ab5

The link isn't working for me any more, but when I originally looked at it, it was a slight improvement (there were fewer matches), but the matches still didn't make much sense overall.

Fixed. There were 86 matches before, and after Niklas' patch there are 55 matches.

In T358862#9682031, @Nikki wrote:

It's searching all other languages too? That seems bizarre. The interface isn't in those languages, nor is the browser, they're not fallbacks for the current language or the target language either, they're not the autonyms, and those names aren't even being shown.

With the language selection we wanted to provide flexibility when searching for your language. For example a user landing on Wikipedia article from a language they do not speak (e.g., following a link on social media or search results) may want to check if the content is available in a language they know. As we explore the next iteration for the language selector (T287860) we can consider to polish the approach but some level of cross-language search may still be needed.