Currently, the keywords tool requires that Freqlist 1 and Freqlist 2, if they are in different corpora, be based on p-attributes with the same handle.
So, if pone selects "LEmma" as "the things to compare", then the handle used will be the handle of Lemma in corpus 1. (LEt's say, "lem").
If corpus 2 has its lemma in an attribute with a different handle (let's say, dic_head_word), then the analysis simply won't work. Either nonse will come back, or there'll just be an errror.
To get around this: allow the user to set a different attribute as the one to use from the "foreign" corpus.
This will be tricky, UI wise, because the list of atts to choose among will change dynamically depending on exactly what annotations the corpus that has been selected in the dropdown has.
This would also be an expert option - one very much not for peopllewho don't undestand CWB attribute identifiers.