This is to follow up on the discussion in the EMB students meeting on whether or not a head-compositional version of head-spec is needed to model some possessive strategies. The ERG uses a normal (non-head-compositional) head-spec rule to model possession, and so I am working to see if I can adapt their analysis for my library.
After looking at it today, I think my answer is still cautiously ‘no.’ Here’s the reason: the overall INDEX of the possessive phrase must be the INDEX of the head of the possessive phrase. In English, this can be built into the lexical entry for 's – it can simply identify its INDEX with the INDEX of the head of the possessive phrase.
How would this work in the case where the possessive marker is an affix on the possessor? Presumably, you’d have the lexical rule that adds the affix do the same thing that 's does – identify its INDEX with the INDEX of the possessum. As long as the lexical rule also puts the right indices into the poss_rel, it should work fine.
Here’s where things go wrong: the ERG analysis only works because in English, the possessor noun combines with its determiner before it combines with the possessive marking 's. So, in the sentence the dog’s cat, you get all the usual semantics for the NP the dog, where the ARG0 of dog is identified with the ARG0 of its determiner. In this analysis, the possessor affix must apply to the possessor noun before it combines with its determiner, meaning, it’s the INDEX value that it has by the time you get around to attaching the determiner is no longer its own. So you get this, where the determiner for dog has as its ARG0 the ARG0 for cat:
So, that’s where I’m at. I’m not convinced there’s absolutely no way around this, but the head-compositional head-spec rule certainly does the trick.
Thanks for following up … it is indeed tricky. So we’re talking about something like “the dog-s cat” where “-s” is an affix and not a clitic, right?
ordinary spec-head-rule in both cases
[ POSS + ] (or [ CASE poss ] or whatever makes sense in your library)
a non-branching rule that takes the [ POSS + ] NP “the dog-s” and builds a DP that points the INDEX at the right thing in the poss_rel?
To do that last little bit of magic, you’d need a pointer to the right argument in the poss_rel on the [ POSS + ] NP. I see two ways to do this:
(1) The affix-adding lex rule is putting in poss_rel and it links the noun’s XARG to the relevant argument position.
(2) The non-branching DP -> NP rule puts in the poss_rel through its C-CONT and thus has its hands on it anyway.
Does that makes sense?
I think that all makes sense to me, yes. I have tried implementing option #1 and ran into this issue: the NP ‘the dog-s’ is a non-head-compositional head-spec phrase, so its HOOK value comes from the determiner – meaning the XARG of ‘dog-s’ is not accessible to the unary rule. There might be a workaround to that, but I didn’t see one that would work without modifying the lexical type definition for determiners, which I am loath to do. So I am working on implementing option #2 and having issues with diff-lists, but I’ll let you know what happens there.
So, option #2 works (with lots of debugging help from you )! Here’s the parse tree:
So, it sounds like my claim for the HPSG abstract should instead be “Look how many patterns we can generate using just the phrase types in the matrix.”
As for my library, I’m not sure which approach is actually going to make for a better implementation. Here are the arguments for each version, as I see them:
Unary phrase rule:
- Avoids adding a phrase rule (head-compositional head-spec) (Although you may have to add a poss-specific head-spec rule anyway if possessive phrase order doesn’t match general word order in the lg (don’t have examples to hand)
- My library would match the argument I’m making in the HPSG abstract better, without extra explaining.
Head-compositional head-spec rule:
- Avoids adding a phrase rule (possessor unary rule)
- In the rest of my library, lexical rules that add semantic or syntactic info have some sort of phonological contribution as well – poss_rel is always on inflecting lexical rules, not zero-inflecting rules. Adding the unary rule that carries poss_rel would break this trend. This is more of an aesthetic argument than a theoretical one, admittedly.
- I have this analysis already implemented in my library, pretty deeply ingrained. This is really not any kind of argument
Settling the implementation is not quite as urgent as getting my analysis settled for the HPSG abstract, so this might be better settled in an actual advisement meeting.
Yes, let’s talk when we next meet in person. I’m glad the HPSG abstract story is clear, though!