Unaccusativity in HPSG

[aranovch at sprynet.com]

Hello,  Stefan, and thanks for the comments. I think for the molst part we agree, and I am glad to see that there is so much work on the fine-grained grammatical aspects of unaccusativity in HPSG.

About my paper: yes, what I show is that when you look beyond the few examples everyone talks about, the generalization that unaccusativity is responsible for the availability of postverbal subjects does not hold. Rather, the test seems to be sensitive to information structure, which in turns is connectd to lexical semantics.

The most interesting piece of data is that change of state verbs (i.e. get sick, rot, grow, etc.) do not seem to behave like other 'unaccusatives.' I have seen this unexpected behavior in many other tests, and I am curious to know whether in the German tests you talk about  (prenominal participles and passive) aren't also sensitive to this group.

For me, the question is this: do we need a syntactic representation for the split in the class of intransitives, or can we by-pass that, attributing the contrasting behacior of 'unergatives' and 'unaccusatives' tho differences in semantics/pragmatics directly? Maybe in the cases you are talking about the generalization has to do with theta-roles, and not with a syntactic feature that distinguished internal from external arguments.

If one still needs to do so, however, I think an alternative is to play around with the configurations that we already have. In my paper I suggests that one class of postverbal subject in Spanish stays inside the VP. This is what in GP approaches was taken to be a manifestation of an internal argument: these are unaccusative internal arguments that never make it to SPEC-IP. I suggested that these are arguments that stay in the COMPS list, not in the SUBJ value. In these verbs, SUBJ is empty, and a non-canonical NP occupies the first slot in COMPS. So, one way to represent unaccusative verbs is to have an identity between SUBJ and one member of ARGST, but not the first one: the first ARGS member is a non-canonical NP. In other words: unaccusative subjects are subjects, but not a-subjects.

SUBJ       <NP1>
COMPS     <    >
ARGST     <NP-noncanon, NP1>

I don't know if there are any advantages of this approach over the one you suggest. I avoids having to introduce additional features for the purpose of distinguishing internal from external arguments. But the price is another mismatch between the valence lists and ARGST.

Cheers,

-RA