Constraining poss-unary-rule


Right now, I’m working on implementing the analysis for specifier-like possessors that uses the normal head-spec-phrase, rather than head-compositional head-spec. In order to make it work, there’s a unary phrase rule that the possessor must go through that adds in the necessary semantics having to do with possession. This works just fine in cases where the possessor is marked (either by affix or adposition) as being a possessor. The problem is that there are cases where only the possessum is marked. In those cases, I can’t constrain the unary phrase rule to only target noun phrases with a POSSESSOR feature (because the possessor isn’t marked), and so it goes nuts and applies that unary rule to all noun phrases. I’m going to try to think of ways to constrain it, but right now, I haven’t come up with anything yet and wanted to throw it out there to see if anyone can think of possible solutions. Thanks!

Here’s the unary rule as presently constituted:

poss-unary-phrase := basic-unary-phrase &                                                                                        
              [ SYNSEM.LOCAL [ CONT.HOOK #hook,
                               CAT [ HEAD.POSSESSOR nonpossessive,
                                  VAL [ SPR #spr,
                                         COMPS #comps,                                                                                                    
                                         SUBJ #subj,                                                                                                
                                         SPEC < [ LOCAL [ CAT [ VAL.COMPS < > ,
                                                                HEAD +np & [ PRON - ] ],                                                               
                                                          CONT.HOOK #hook & [ INDEX #possessum & [ COG-ST uniq-id ], 
                                                                              LTOP #lbl ] ] ] > ] ] ],      
             C-CONT [ RELS <! arg12-ev-relation & [ PRED "poss_rel",                                                                             
                                                      LBL #lbl,                                                             
                                                      ARG1 #possessum,
                                                      ARG2 #possessor ],                                                                                  
                                quant-relation & [ PRED "exist_q_rel",                                                                                   
                                                   ARG0 #possessum,                                                                                  
                                                   RSTR #harg ] !>,                                                                                   
                        HCONS <! qeq & [ HARG #harg, LARG #lbl ]!>,
                        ICONS <! !>   ],                                                                     
                ARGS < [ SYNSEM.LOCAL [ CAT [ HEAD.POSSESSOR possessor,
                                                                     VAL [ SPR #spr & olist,
                                                    COMPS #comps & olist,                                                                  
                                                    SUBJ #subj,                                                                           
                                                    SPEC < > ],                                                                                           
                                              HEAD +np ],                                                                                             
                                        CONT.HOOK.INDEX #possessor ] ] > ].


Could you instead add the semantics to the possessum?


So, possessor isn’t marked, but it’s head-spec? Why not treat this as juxtaposition? Also – is there a real language that you’ve got in mind here?


Guy – So, I tried a version of this, and couldn’t get it to work, unfortunately. The reason the unary rule is needed is that head-spec is a non-head-compositional rule, but the index of the whole possessive phrase should be the same as the head (possessum)'s index. So the unary rule that the possessor must go through does the work of identifying the index of the head (possessum) with it’s own index. Then the head-spec rule can pick up the index from the non-head daughter and everything works. I tried having the lex rule that the possessum goes through do this same work by identifying its HOOK with the HOOK of its SPR, but without the unary rule on the possessor side as buffer, this actually ends up identifying the index values of possessor and possessum, which leads to a broken MRS.

Emily – The possessum is marked in these cases (and usually agrees with the possessor), so juxtaposition won’t work – particularly because the juxtaposition rules can’t handle agreement.

As far as attestation goes, it looks like Ainu is an example of this (unfortunately, determining something is a spec-like possessor is hard without required determiners, because you basically need a negative example of ‘*det det possessor possessum’ being ungrammatical). In general though, I have already ruled out the head-mod analysis from ever occurring when the possessum is marked by an adposition/clitic (because that analysis didn’t work), so if I want to cover languages like Ewe, I’d need to make head-spec a possibility when the possessum is the only marked constituent.


If the possessors aren’t marked, then yes, any old NP is going to go through that non-branching rule. I’d say this is leading to overgeneration because the mother of that rule is showing up in places other than the spec daughter of head-spec? Or is there something beyond that? (Because if not, then it sounds like you’ll probably have this problem even with marking on the daughter.)


The context I’m seeing it showing up in is on the SUBJ daughter of the head-subj phrase – I can keep it from attaching to (say) the possessum by barring from attaching to things marked as possessums, but when the possessor isn’t marked, it just attaches to all unmarked NPs.


So I think you’ll see this same problem with NPs marked as possessors when you need this non-branching rule. Something on the mother of the rule has to say “This is not an ordinary NP, do not use in ordinary NP positions”. That something could be CASE, or it could be the head value is actually det, not noun.


I’ll go ahead and try the det option, so I can avoid “hallucinating case” for now :slight_smile:


Update: this seems to have worked just fine :slight_smile: