Affix Coordination

I’m trying to describe how affix coordination is handled in the Grammar Matrix and I don’t quite have a clear picture. As far as I know, the coordination library predates the morphotactics library, so I’m a little unclear on how (if?) these were integrated.

The questionnaire asks for the orthography of the affix and if it’s a prefix or suffix. coordination.py creates an irule for that affix. No coordination feature becomes available on the morphology page after affix coordination is specified (like it does for eg. negation).

Is the irule created by coordination.py able to attach just anywhere? Is there any potential interaction between this and the morphotactics inputs/outputs I should anticipate? This is the relevant code in coordination.py:

if pre or suf:
    # first the rule in mylang
    mylang.add(pn + '-bottom-coord-rule :=\
           ' + bot + 'bottom-coord-rule &' + passup + '\
           [ SYNSEM.LOCAL.COORD-STRAT "' + num + '",\
             SYNSEM.LOCAL.COORD-REL.PRED "_and_coord_rel",\
             DTR.SYNSEM.LOCAL.CAT.HEAD ' + headtype + ' ].')

    # now the spelling change rule in irules.tdl
    rule = pn + '-bottom :=\n'
    if pre:
        rule += '  %prefix (* ' + pre + ')\n'
    else:
        rule += '  %suffix (* ' + suf + ')\n'
    rule += '  ' + pn + '-bottom-coord-rule.'
    irules.add_literal(rule)

I rather suspect it was never retrofitted …

Makes sense- it would be an easy thing to overlook. What’s the expected behavior? In theory, do you know if a rule like that would just attach to anything. In other words would it generate:
swim-PST-COORD and swim-COORD-PST, but at least it would parse both?
Or would the graph/inputs get messed up somewhere so it wouldn’t parse one or both of those?

I suspect that what’s happening is that the coord rule is entirely a thing apart, i.e. unable to attach to or feed any of the other rules. You might look at the grammar from this paper:

Fokkens, Antske, Laurie Poulson and Emily M. Bender. 2009. Inflectional Morphology in Turkish VP Coordination. In Müller, Stefan (ed), Proceedings of the HPSG09 Conference . Stanford: CSLI. pp.110-130. [.bib]

… available here: http://www.delph-in.net/matrix/turkish/

That’s post-morphotactics v1, so maybe we managed it somehow?

It looks like you managed by using examples where the word marked with coordination has no other inflection… so that’s a dead end.

But- I do have some inferred grammars that exhibit this phenomenon, so I found an example from Lezgi:

за-н-ни
1sg-erg-and

за-н-ни (which is attested) does seem to parse properly. It goes through the ergative lex rule and then the bottom coord rule. So I think that’s what we want.

However, за-ни-н does not parse. I don’t get the bottom-coord-rule when I “show parse chart”… I can do it with interactive unification though (that seems suspicious- why would that be?). But then if I drag the output of the bottom-coord-rule through into the ergative lexical rule via interactive unification, I get this unification error:

image

I think what this means is that the affix coord rule as it is now, looks for a word that is infl satisfied and then attaches. But it’s not a suitable input to other rules. Am I interpreting that correctly?

Can you give me the tdl for the bottom-coord-rule?

If за-н-ни is attested but за-ни-н isn’t, I wouldn’t expect the inferred grammar to parse за-ни-н anyway…

n11-bottom-coord-rule := infl-bottom-coord-rule &
[ SYNSEM.LOCAL.COORD-STRAT “11”,
SYNSEM.LOCAL.COORD-REL.PRED “_and_coord_rel”,
DTR.SYNSEM.LOCAL.CAT.HEAD noun ].

Does it matter that it’s not attested? It doesn’t look like MOM’s graph of attested affix orders is relevant here because the coordination library is adding the affix completely separately.

I see: the inferred morphotactics use the -rule-dtr types to control what attaches to what. Since n11-bottom-coord-rule doesn’t inherit from noun-pc17-rule-dtr, it’s not a possible input to that rule. This is what is ruling out за-ни-н, and is in fact related to that being unattested.

Looking at the definition of infl-bottom-coord-rule, it looks like it assumes it attaches ‘last’ that is, it drops any forbids/requires constraints. That wouldn’t stop other things from attaching to it, but since it doesn’t get appropriate supertypes, they don’t.

infl-bottom-coord-rule := add-only-rule &
			  inflecting-lex-rule &
  [ INFLECTED infl-satisfied,
    DTR.INFLECTED inflected,
    SYNSEM.LOCAL [ COORD +,
                   COORD-REL #crel ],
    C-CONT [ RELS <! #crel !>,
             HCONS <! !> ]].

Are you saying that if за-ни-н was attested, it would parse? I don’t understand how if infl-bottom-coord-rule, which is added by the coordination library and not the morphotactics library, assumes that it attaches last.

No, I don’t think it would parse with the infl-bottom-coord-rule, even if attested, because the coordination library is not fully integrated into the morphotactics.