Notes from Karin Michelson's talk

For anyone who missed Karin's talk on Tuesday, here are a few notes!

---

Every verb has a prenominal prefix, if you have agents and patients, there are 58 that occur. Each has up to 10 allomorphs, and if you count all the forms, you have 326 options. How do you analyse this?

You can separate vowel and stem, this lets you reduce the number of prefixes, but the number of stem classes increases.

Synchronic analysis: posit a single underlying form and derive the other 4. BUT! Some of the phonological rules are completely unphonological.

Possible solution?

One rule for each cell. You still have a lot of rules, but there isn't anything wrong with having a bunch of different rules having as their output the same allomorph, then you refer to the different stem type. Input: semantic features and stem class. Output: realization. Shift the role of phonology.

For a lot of linguistics, we assume the 1 meaning:1 form assumption. How much can we rely on this for all processes? It's especially difficult when we have idiosyncratic data.

What if people just remember a stem class or a form that identifies a stem class, then are productive with that? The one form allows you to predict all the other forms.