Stemmer is not working properly on indexing or on search

I am setting up Search API for a german language website. The default stemmer won't be much help for german. Anyway, I guess it shouldn't hurt. But it does. It seems that stemming isn't used on indexing, but on search. For example I have the title "Eine neue WEBinar-Reihe". The following keywords will be indexed:

select * from search_api_db_default_index_text;
+------------------+---------------+----------------------------+-------+
| item_id          | field_name    | word                       | score |
+------------------+---------------+----------------------------+-------+
| entity:node/1:de | title         | eine                       |  5000 |
| entity:node/1:de | title         | neue                       |  5000 |
| entity:node/1:de | title         | webinarreihe               |  5000 |
...

But on search the stemmer transform the search keywords to "ein", "neue" and "webinarreih". Its cutting the final e for eine and webinarreihe, due to Porter2::step5(). Sofar that seems to be correct, but if it happens on the search process it should happen also on the index process.

I disabled stemming, as it's not useful. But maybe the findings are still helpful. Maybe it's simply related to the language. I don't know, if the default stemmer only works on english content, what would be correct, but than something needs to tell the search, that it needs to be running in german without invoking stemmer too (the current website has only german enabled, english is disabled).