I always get a bit confused by the argument optionality choices on the morphology page.
If I have a language where subject agreement is required when a subject is dropped and not permitted if the subject is overt, I would say that “overt-arg” is “not-permitted” on the subject?
But then how do I enforce that there isn’t agreement when the subject is present? Do I need a non-inflecting rule that says overt-arg permitted?
Yes, you’ll need a zero-marking rule in that same PC that effectively says that the subject must be overt. (This will end up as [ OPT - ] on the subject, but I don’t know the choices terminology for it off the top of my head.) Customize a grammar like that and test it, I guess? Or better: Look at Safiyyah’s regression tests.
I took a look at Safiyyah’s regression tests. For subj drop she has 8 tests for the 9 combinations of required/optional/not-possible subject marking for overt and dropped subjects (excluding not-possible for overt, not-possible for dropped). The subject marking rules in these choices files are specified for either dropped-arg or overt-arg, permitted or not-permitted and that’s it. Her regression tests don’t posit any 0-rules. However, in Saleem (2010) says: “In some of the nine logical possibilities, enforcing these constraints requires sending the verb through “nomarker” lexical rules so that constraints associated with markerless verbs can be enforced.” But the paper doesn’t go into any more detail regarding which of the combinations this is relevant for.
So I’ll need to look at the customization script or customize a grammar to figure out how the constraints work.
A side note: her regression tests don’t use the OPT feature (which is available in the questionnaire drop-down). Maybe it’s for lexically licensed arg-opt or maybe it’s a pseudo feature that’s relevant for customizing lexical rules, but not for the choices file?
I suspect it’s that it’s for the lexically licensed arg-opt.
It’s hard to tell from the customization script and I can’t find any discussion of this in Saleem’s thesis (maybe it’s there but I can’t find it). But logically, I think the only two cases where a 0-rule is necessary are 1) when the argument marker is required for overt arguments and not possible for dropped arguments and 2) when the argument marker is not possible for overt arguments and required for dropped arguments. For any where argument marking is optional in one of the two cases, we shouldn’t need to do any extra work to anything enforce agreement, right?
In that case, if subj marking is required for dropped subjects and not permitted for overt objects, the argument marking lrt should be overt-arg: permitted. This pc should be obligatory and another lrt (with no argument marking) in the pc should dropped-arg: not-permitted.
Yes, this sounds correct to me.