Ensuring agreement of past participles with their complements; LSYNSEM?

In Spanish, in constructions like He has a brother named Brian, the past participle must agree with the complement of the original verb, in this case, to name. So, Someone (doesn’t matter who) named the brother Brian, so Brian is the complement of name, and ultimately named and brother must agree in number and gender:

(1) Tiene un hermano         llamado      Brian
     Has  a  brother:masc    named.masc  Brian
    "He has a brother named Brian." [spa] 

The grammar currently overgenerates and will admit any combination of genders here. I am trying to decide where the right place is to add constraints.

I believe the desired tree is this one:

Screenshot from 2024-02-06 16-29-26

MRS:

Everything is properly linked and the gender of hermano (brother) is properly specified.

But unfortunately, the grammar also happily produces the same kind of tree for the ungrammatical (2):

(2) Tiene una hermana         llamado      Brian
     Has  a  sister:fem    named.masc  Brian
    Intended: "He has a sister named Brian." [spa] 

Screenshot from 2024-02-07 13-46-00

In the tree, the third V in the unary chain is the past participle rule, and in that structure, the AGR value is still easily accessible:

Screenshot from 2024-02-06 16-40-14

But then the first complement of the verb gets dropped, the top V in that chain is the optcomp rule. I am not sure whether this makes sense, or maybe that is the problem. If the complement needs to be realized through something other than the head-complement rule, then it should either be dropped or slashed. In the ERG, in the similar construction, it will be dropped, and the tree will look similar, with an AdjPart phrase formed from the VP named Brian and the resulting AP then modifying brother.

But if the complement is dropped, then is it a reasonable thing to try and hold on to the agreement information of that dropped complement? One issue is, the PNG values are appropriate for AGR structures of type _individual but verbs have events and those do not have PNG. So I am not quite sure how to propagate the agreement features starting at that top V node which is the optcomp.

The optcomp phrase has a feature called LSYNSEM which I haven’t seen before. Maybe I could use that? But like I said, I don’t know what it is for, originally. Below first a very large picture showing the geometry and then the LSYNSEM more closely. It is a HEAD feature:

I can constrain the values of this LSYNSEM thing to the head daughter’s of the optcomp rule:

Screenshot from 2024-02-07 13-48-11

I can then constrain the adj-participle phrase to identify these agreement values with its own and this way I get rid of the tree. The question is, what is this LSYNSEM and am I trying to use it properly? Does anyone know anything about LSYNSEM?

I would guess that llamado here is a passive participle, used as a modifier. So the agreement is with the subject. Is that consistent with the morphology?

Isn’t the subject the person who performs the naming, not the person being named?

It’s true that there is agreement with the subject already implemented there but since it does not result in bad sentences being ruled out, I assumed something is missing. “A person named Brian” – is “person” really a subject of the verb “to name”? It’s its first complement I think? The subject (whoever chose the name) is not expressed, so the agreement with the subject shouldn’t matter? Or am I mixing everything up.

In “a person named Brian”, named is the passive participle, so the subject is the modifiee is person. I’m assuming the same is true in Spanish.

In French, the passive participles agree with the subject even when they are predicates (not modifiers): Elle a été poursuivie par un lapin / Il a été poursuivi par un lapin

What happens in Spanish?

Though this isn’t going to automatically work for passive participles used as modifiers — you’ll have to enforce agreement with MOD, not (just) SUBJ.

Yes, it is the same in Spanish (see examples (1)-(2) in the first post), the participle agrees with the noun it modifies.

So maybe the issue is that the passivization is not carried out correctly (though that would be surprising to me given how old and big the grammar is).

The verb llamar (to name) has two complements: the thing being named and the name. So, in this case, we need to link the original first complement of the verb to the thing that the participle is going to modify. Not sure why this isn’t happening, or rather, why the bad sentences aren’t ruled out.

Here’s the adj-part-phrase. The agreement between the MOD and the SUBJ is there but it does not have the effect of enforcing agreement between “brother/sister” and “named”:

adjpart_phrase := unary-phrase & 
  [ SYNSEM [ LIGHT -,
             SLSHD #slshd,
             LOCAL [ AGR #ind,
                     COORD #coord & -,
                     COORD-STRAT #coord-strat & zero,
                     CAT [ HEAD adj &
                               [ MOD < synsem &
                                       [ LOCAL intersective-mod &
                                               [ AGR #ind,
                                                 CAT [ LASTNMOD -,
                                                       HEAD noun & [ KEYS.KEY nom_rel ],
                                                       VAL [ SUBJ < >,
                                                             SPEC < > ] ] ] ] >,
                                 KEYS.KEY v_event_rel ],
                           VAL [ SPR < >,
                                 SUBJ < >,
                                 COMPS #comps,
                                 CLTS #clts,
                                 SPEC < > ] ],
                    CONT #cont,
                    CTXT #ctxt ],
	    NON-LOCAL #non-local & [ QUE 0-dlist ] ],
    ARGS < [ INFLECTED +,
             SYNSEM [ SLSHD #slshd,
                      LOCAL [ COORD #coord,
                              COORD-STRAT #coord-strat,
                              CAT [ HEAD verb &
                                         [ AUX -,
                                           VFORM part,
                                           VOICE passive,
                                           MOD < >,
                                           KEYS.KEY v_event_rel ],
			   	    VAL [ SUBJ < [ LOCAL.CONT.HOOK.INDEX #ind ] >,
                                          COMPS #comps & < >,
                                          CLTS #clts & < > ] ],
                              CONT #cont & [ HOOK #hook ],
                              CTXT #ctxt ],
                      NON-LOCAL #non-local ] ] >,
    C-CONT [ HOOK #hook,
             RELS <! !>,
             HCONS <! !> ] ]
"""
; --- unary phrase type for adj-pastpart
; removed "PRD non-prd" to deal with esos tíos son muy pesados
""".

Here’s the past participle rule. Neither this rule nor its parent concern themselves with any agreement, which perhaps is the problem. (I have just found this rule. Before, I was mistakenly looking at a similarly named inflectional rule in another file).

v_pastpart-trans_dlr := ppart-lex-rule &
  [ SYNSEM.LOCAL [ CAT [ HEAD [ INV -,
                                PRD prd & [ COPV ser ] ],
                         VAL [ SUBJ < unexpressed & [ OPT +,
                                                      LOCAL np_nom_local & [ CONT #objcont & [ HOOK.INDEX #arg2 ] ],
                                                      NON-LOCAL #ononloc ] >,
                               COMPS < [ OPT +,
                                         LOCAL mrkd_np_local & [ CAT.HEAD.KEYS.KEY _por_p_sel_rel,
                                                                 CONT.HOOK.INDEX #subjind ],
                                         NON-LOCAL #snonloc ] . #comps >,
                               CLTS #clts ] ],
                   CONT.RELS.LIST < [ ARG2 #arg2 ], ... > ],
    DTR [ INFLECTED +,
          SYNSEM.LOCAL.CAT.VAL [ SUBJ < [ LOCAL np_nom_local & [ CONT.HOOK.INDEX #subjind ],
                                          NON-LOCAL #snonloc ] >,
                                 COMPS [ FIRST [ LOCAL np_acc_local & [ CONT #objcont ],
                                                 NON-LOCAL #ononloc ],
                                         REST #comps ],
                                 CLTS #clts ] ] ]
  """
  -- with transitive verbs
  """.

ppart-lex-rule := const-ltol-rule &
  [ ALTS.PASS -,
    SYNSEM [ PUNCT #punct,
             SLSHD #slshd,
             LOCAL [ CAT [ MC #mc,
                           HEAD verb & [ AUX -,
                                         VFORM part,
                                         VOICE passive,
                                         TAM #tam & [ MOOD #mood ],
                                         KEYS.KEY #key,
                                         MOD < >,
                                         LSYNSEM #lsynsem ],
                           VAL.SPR < [ OPT +,
                                       LOCAL [ CAT [ HEAD.KEYS.KEY degree_rel,
                                                     MC na ],
                                               CONT.HOOK [ LTOP #hand,
                                                           XARG #index ] ],
                                       NON-LOCAL [ SLASH 0-dlist,
                                                   QUE 0-dlist ] ] > ],
                     CONT [ HOOK [ INDEX #index,
                                   LTOP #hand,
                                   XARG #xarg ],
                            HCONS #hcons,
                            RELS #rels & [ LIST < [ ARG0 #index & [ E.MOOD #mood ] ], ... > ] ],
                     CTXT #ctxt ],
             LKEYS #lkeys,
             NON-LOCAL #nonloc ],
    DTR [ INFLECTED +,
          ALTS.PASS +,
          SYNSEM [ PUNCT #punct & [ RPUNCT no_punct ],
                   SLSHD #slshd,
                   LOCAL [ CAT [ MC #mc,
                                 HEAD verb & [ PRD non-prd,
                                               VFORM part,
                                               VOICE active,
                                               TAM #tam,
                                               KEYS.KEY #key,
                                               LSYNSEM #lsynsem ],
                                 VAL.SPR < > ],
                           CONT [ HOOK [ LTOP #hand,
                                         XARG #xarg ],
                                  HCONS #hcons,
                                  RELS #rels ],
                           CTXT #ctxt ],
                   LKEYS #lkeys,
                   NON-LOCAL #nonloc ] ],
    C-CONT.RELS <! !> ]
  """
  --- LR types for participles
  -- LR type building pastparts for passive constructions
  """.

However, adding agreement to the past participle lexical rules still doesn’t rule out the bad sentence. It looks like somehow, I am loosing the agreement after the optcomp rule applies (the top V in the V chain), and I don’t understand why because the optcomp obviously identifies the subject of the mother and the subject of the daughter.

Here’s the daughter of optcomp from the tree, in the second screenshot the highlighted tag is “5”:

Screenshot from 2024-02-07 17-26-57

Here’s the mother of the optcomp from the same tree, underspecified agreement:

Here’s the optcomp phrase and the specific subtype that is used:

optcomp-phrase := head-valence-phrase & head-only & head-compositional &
  [ INFLECTED #infl,
    SYNSEM canonical-synsem &
           [ MODIFIED #mod,
             SLSHD #slshd,
             LIGHT -,
             LOCAL [ COORD #coord,
                     COORD-STRAT #cstrat,
                     CAT [ MC #mc,
                           VAL [ SUBJ #subj,
                                 COMPS #comps,
                                 SPR #spr,
                                 SPEC #spec,
                                 CLTS #clit ] ] ],
             NON-LOCAL #nlocal ],
    HEAD-DTR [ INFLECTED #infl & +,
               SYNSEM [ MODIFIED #mod, 
                        SLSHD #slshd,
                        LOCAL [ COORD #coord & -,
                                COORD-STRAT #cstrat & zero,
                                CAT [ MC #mc,
                                      VAL [ SUBJ #subj,
                                            SPR #spr,
                                            SPEC #spec,
                                            COMPS < [ OPT +,
                                                      NON-LOCAL [ QUE 0-dlist,
                                                                  REL 0-dlist ] ] 
                                                    . #comps >,
                                            CLTS #clit ] ] ],
                        NON-LOCAL #nlocal ] ],
    C-CONT [ RELS  <! !>,
             HCONS <! !> ] ].

optcomp-v_phrase := optcomp-phrase &
  [ SYNSEM.LOCAL.CAT [ ;HEAD.LSYNSEM.LOCAL.AGR #agr,
                       POSTHEAD #ph,
                       VAL [ COMPS #comps ] ],
    HEAD-DTR.SYNSEM.LOCAL [ ;AGR #agr,
                            STR.HEADED solely,
                            CAT [ POSTHEAD #ph, 
                                  HEAD verb & [ KEYS.KEY v_event_rel ],
                                  VAL [ SUBJ < [ ] >,
                                        COMPS < [ NON-LOCAL.SLASH 0-dlist ]. #comps > ] ] ] ].

…in the structure for the optcomp construction, the gender-number features can still be found in CONT|HOOK. Which doesn’t surprise me because I knew they weren’t entirely lost (they are in the MRS!). It’s just that that doesn’t preclude a masculine participle from modifying a feminine noun, for example.

Hold on! :slight_smile: Editing a wrong file :slight_smile:

OK, after adding agreement identity between the first complement of the verb being passivized and the resulting participle, the optcomp phrase has the proper agreement values on the subject:

These of course persist after the application of the head-complement rule which attaches the second complement (the name, Brian or whatever).

But now the adj-participle phrase is still happy to have the participle agree with a noun of another gender. Here is the adj-participle phrase again. I can get rid of the unwanted tree I think by adding AGR #ind to the daughter’s SUBJ (right now there is only a contraint on the HOOK). Does that sound like the right thing? The relationship between HOOK and AGR is still not fully clear to me.

adjpart_phrase := unary-phrase & 
  [ SYNSEM [ LIGHT -,
             SLSHD #slshd,
             LOCAL [ AGR #ind,
                     COORD #coord & -,
                     COORD-STRAT #coord-strat & zero,
                     CAT [ HEAD adj &
                               [ MOD < synsem &
                                       [ LOCAL intersective-mod &
                                               [ AGR #ind,
                                                 CAT [ LASTNMOD -,
                                                       HEAD noun & [ KEYS.KEY nom_rel ],
                                                       VAL [ SUBJ < >,
                                                             SPEC < > ] ] ] ] >,
                                 KEYS.KEY v_event_rel ],
                           VAL [ SPR < >,
                                 SUBJ < >,
                                 COMPS #comps,
                                 CLTS #clts,
                                 SPEC < > ] ],
                    CONT #cont,
                    CTXT #ctxt ],
	    NON-LOCAL #non-local & [ QUE 0-dlist ] ],
    ARGS < [ INFLECTED +,
             SYNSEM [ SLSHD #slshd,
                      LOCAL [ COORD #coord,
                              COORD-STRAT #coord-strat,
                              CAT [ HEAD verb &
                                         [ AUX -,
                                           VFORM part,
                                           VOICE passive,
                                           MOD < >,
                                           KEYS.KEY v_event_rel ],
                                           ;LSYNSEM.LOCAL.AGR #ind ],
			   	    VAL [ SUBJ < [ LOCAL.CONT.HOOK.INDEX #ind ] >,
                                          COMPS #comps & < >,
                                          CLTS #clts & < > ] ],
                              CONT #cont & [ HOOK #hook ],
                              CTXT #ctxt ],
                      NON-LOCAL #non-local ] ] >,
    C-CONT [ HOOK #hook,
             RELS <! !>,
             HCONS <! !> ] ]
"""
; --- unary phrase type for adj-pastpart
; removed "PRD non-prd" to deal with esos tíos son muy pesados
""".

What I find surprising in that rule is that the HOOK.INDEX of the MOD isn’t linked to anything. I bet identifying it with #ind would work — and would probably also lead to better semantics.

Do you mean, in the adjpart phrase above? If I add a CONT.HOOK.INDEX #ind constraint to the mother, nothing breaks (at least wrt one test sentence) but the bad tree is still not ruled out unless I add LOCAL.AGR #ind constraint to the SUBJ of the daughter.

Yes, in adjpart-phrase. I don’t have a good theory of how AGR is used in this grammar, but if it works and doesn’t break anything else on regression testing, why not?

1 Like