When "language" is selected, then internal prediction module do not allow to have different "min word length" for different data source. The text expander is what it is, it expands text, it do not have prediction feature builtin, as opposed to "language" module. I wouldn't like to change that :)
When "textExpander" only mode is used(in addition to above), then you could potentially select "min word length", but it will be applied to all languages. Per language settings are not planed as this make configuration unintuitive and confusing for users.
could there be an option to set minimum word length separately for the text expander?