Lemmatization of analogical/creative morphological forms

[ellepannitto]

Hello!

I'm following on issue #1179 as I have a related problem.

While lemmatizing data for spoken Italian (with @caterinamauri and @EleonoraZucchini), we found this example:

SP1  era tra i migliori germanisti che ci fossero in italia eh negli anni sessanta 
    e c'è tutto il carteggio che siccome erano amici è anche pieno di battute di ironie no? 
    eh cases era una persona molto mh

SP2  a cases?

SP3  tutto cases e chiesas

SP1  brillante eh e chieses eh mh

SP2  di chiesas

The intended meaning is that the speaker is referring to a person whose surname is Cases, and is playfully reusing the Italian expressions "casa e chiesa" and "uomo di chiesa" (i.e. a pious, church-going person), reshaping them into "cases e chiesas".

The morphological distortion here is intentional: the speaker is producing a kind of analogical formation, loosely calquing Spanish-like plural morphology onto Italian forms. How should we treat these cases?

It's not a learner case but does it make sense to distinguish the two aspects? Or can we analyze these cases of "creative" lemmas (creative for very different reasons, but still not random) all alike?

[nschneid]

With wordplay (e.g. puns) there can probably be multiple layers of interpretation and therefore multiple justifiable analyses.

I would probably go with whatever lemma seems simplest based on the superficial patterns of the language, and maybe add a special MISC feature like Wordplay=Yes to be able to find such cases in the future. (Relatedly, EWT has some instances of metalinguistic mentions that are marked Mentioned=Yes.)