Lab 6
Language: Miyako
Adam Ledyard and Sanae Sato


Embedded Clauses

Ryukyuan make embedded clauses (both declaractive and interrogative) with complementizer "=ru(u)"(or s alternative "=su(u)"). 


Example Sentence in the testsuit

	cftar=ru=u sstar
	cf-tar=ru=u ss-tar
	arrive-PAST=COMP=ACC know-PAST
	I realized [the fact that] (she) arrived.


Here are the definition of complementizer(s). We only have =ru as our example in the testsuit.

	;;; Complementizers
	   
	 ru_comp := declarative-comp-lex-item &
	    [ STEM < "ru" >].
	  
	 ruu_comp := declarative-comp-lex-item &
	    [ STEM < "ruu" >].
	  
	 su_comp := declarative-comp-lex-item &
	    [ STEM < "su" >].
	  
	 suu_comp := declarative-comp-lex-item &
	    [ STEM < "suu" >].

In order to case-marker "=u" to accept comp(lementizer) as its argument, we added following  changes in miyako.tdl.

	+npc :+ [ CASE case ].


	case-marking-lex := basic-one-arg & raise-sem-lex-item &
	  [ SYNSEM.LOCAL.CAT [ HEAD adp &
		                   [ CASE #case,
		                      MOD < > ],
		               VAL [ SPR < >,
		                     SUBJ < >,
		                     COMPS < #comps >,
		                     SPEC < > ] ],
	    ARG-ST < #comps &
		     [ LOCAL.CAT [ HEAD +nc &
		                        [ CASE #case ],
		                   VAL.SPR < > ] ] > ].


	 interrogative-comp-lex-item := complementizer-lex-item &
	    [ SYNSEM.LOCAL.CONT.HOOK.INDEX.SF ques ].
	  
	 declarative-comp-lex-item := complementizer-lex-item &
	    [ SYNSEM.LOCAL.CONT.HOOK.INDEX.SF prop ].



Non-verbal Predicates
  
Bare-NPs

Bare-NPs show up in Miyako in a very specific environment.  They only show up with it is in nonpast.

		kari=a sinsii
		kari=a sinsii
		3SG=TOP teacher
		He (is) a teacher

From what we can tell, determiners are allowed in the NP predicate.  I adapted the n-bar rule from
class with a few modifications that were suggested:

	 np-bar-predicate-rule := unary-phrase &
	  [ SYNSEM.LOCAL.CAT [ HEAD verb,
	 		      VAL [ COMPS < >,
	 			    SUBJ < [ LOCAL [ CONT.HOOK.INDEX #arg1,
	 					     CAT [ HEAD adp,
	 					     	   VAL.SPR < > ] ] ] > ] ],
	    C-CONT [ HOOK [ LTOP #ltop,
	 		   INDEX #index & [ E.TENSE nonpast ],
	 		   XARG #arg1 ],
	 	    RELS <! arg12-ev-relation &
	 		  [ PRED "_be_v_id_rel",
	  		    LBL #ltop,
	 		    ARG0 #index,
	 		    ARG1 #arg1,
	 		    ARG2 #arg2 ] !>,
	 	    HCONS <! qeq & [ LARG #larg ] !> ],
	   ARGS < [ SYNSEM.LOCAL [ CAT [ HEAD noun,
	 				VAL.SPR < > ],
	 			  CONT.HOOK [ INDEX #arg2,
	 				      LTOP #larg ]]] > ].



Copula

The copula and existential verbs are needed for the non-verbal predicate so this needed to be addressed. Also note that the existential verb is the only verb that requires agreement with the subject so a new feature was introduced, animacy since a different verb is needed depending on the subject.  


Ryukyuan have two forms of copula verb (ar & jar). We have implemented as transitive verb which takes +njrp  (disjunctive type for nouns, adjectives, adverbs and adpositions). A little more restriction is necessary not to parse ungrammatical sentences.



This is what we add as new type:

	transitive-copula-ext-verb-lex := main-verb-lex & transitive-lex-item &
	  [ SYNSEM.LOCAL.CAT.VAL.COMPS < #comps >,
	    ARG-ST < [ LOCAL.CAT.HEAD adp ],
		     #comps &
		     [ LOCAL.CAT [ VAL [ SPR < >,
		                         COMPS < > ],
		                   HEAD +njrp ] ] > ].

Also copula verb needed to gothough TENSE rule in order to get correct analysis. Therefore, we added following lexical rule.

	trans-copula-verb-lex := transitive-copula-ext-verb-lex & ASPECT-or-converb-rule-dtr & MOOD-rule-dtr & TENSE-rule-dtr & negation-rule-dtr &
	  [ INFLECTED.TENSE-FLAG -,
		SYNSEM.LOCAL.CAT.VAL.COMPS.FIRST.OPT-CS type-id ].


Since the only difference between copula verbs and existential verbs are the animacy, theoretically speaking existential verbs (such as "u", "ur", "ar", and "jar") should work as same way for the non-verbal predicate. However, our grammar is not fully working with existential verbs yet.


Sentences getting correct analysis:

	# nominal predicate (NP)   
	#Example 72
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	kari=a sinsii jar0
	kari=a sinsii jar-0
	3SG=TOP teacher COP-NPST
	He is a teacher


	# nominal predicate (NP)
	#Example 73
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	kari=a sinsii
	kari=a sinsii
	3SG=TOP teacher
	He (is) a teacher

	# nominal predicate (NP)
	#Example 74
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	kari=a sinsii a-tar
	kari=a sinsii a-tar
	3SG=TOP teacher COP-PAST
	He was a teacher

	# nominal predicate (NP)
	#Example 75
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	kari=a sinsii=du atar
	kari=a sinsii=du a-tar
	3SG=TOP teacher=FOC COP-PAST
	He was a teacher


	# nominal predicate (NP)
	#Example 76
	Source: author
	Vetted: f
	Judgment: u
	Phenomena: cop
	kari=a sinsii
	kari=a sinsii
	3SG=TOP teacher 
	He (was) a teacher



	#(35) Adj/Adv
	#Example 77
	Source: author
	Vetted: 	f
	Judgment: g
	Phenomena: cop
	kari=a takaf=du ar0
	kari=a taka-f=du ar-0
	3SG=TOP tall-ADV=FOC COP-NPST
	He is tall


	#Example 80 Animacy is important in the copula 
	Source: author
	Vetted: f
	Judgment: u
	Phenomena: cop
	pztu=nu cc=n jar0
	pztu=nu cc=n jar-0
	man=NOM moon=DAT COP-NPST
	The man is on the moon 

Sentences getting incorrect analysis:

	# nominal locative (PP)
	#Example 78
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	pztu=nu cc=n ur0
	pztu=nu cc=n ur-0
	man=NOM moon=DAT COP-NPST
	The man is on the moon 

	# nominal locative (PP)
	#Example 79
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: cop
	pztu=nu cc ur0
	pztu=nu cc ur-0
	man=NOM moon COP-NPST
	The man is the moon 




Polar Questions

In Miyako, Polar questions can be labeled with a sentence final question-clitic which the customization
grammar handled quite nicely.  Unfortunately, an internal focus-clitic can be used to make it an
interrogative so the following sentences are all grammatical:

		#Example 53 -ru Focus particle on subject, makes it an interrogative 
		Source: a
		Vetted: f
		Judgment: g
		Phenomena: wo
		vva=ga=ru uri=u aztar
		vva=ga=ru uri=u az-tar
		2SG=NOM=FOC that=ACC say-PAST
		Did you say that?

		#Example 54 -ru Question particle on verb, makes it an interrogative 
		Source: a
		Vetted: f
		Judgment: g
		Phenomena: wo
		vva=a uri=u aztar=ru
		vva=a uri=u az-tar=ru
		2SG=TOP that=ACC say-PAST=Q
		Did you say that?

		#Example 55 -ru Q
		Source: a
		Vetted: f
		Judgment: g
		Phenomena: wo
		vva=ga=ru uri=u aztar=ru
		vva=ga=ru uri=u az-tar=ru
		2SG=NOM=FOC that=ACC say-PAST=Q
		Did you say that?

As you can see, the 'ru' q-clitic is optional when a 'ru' focus-clitic is present so a new feature was
added and a major overhaul of declaratives and interrogatives was done based on an analysis used for
Russian in 2010:

	basic-zero-arg :+ [ SYNSEM.NON-LOCAL.YNQ 0-dlist ].
	basic-one-arg :+ [ SYNSEM.NON-LOCAL.YNQ 0-dlist ].
	basic-two-arg :+ [ SYNSEM.NON-LOCAL.YNQ 0-dlist ].
	basic-three-arg :+ [ SYNSEM.NON-LOCAL.YNQ 0-dlist ].
	 
	basic-head-opt-comp-phrase :+
	 [ SYNSEM.L-PERIPH #periph,
	   ARGS < [ SYNSEM.L-PERIPH #periph ] > ] .

	non-local :+ [ YNQ 0-1-dlist ].

	binary-headed-phrase :+ mc-na-headed-phrase.

	basic-binary-phrase :+
	 [ SYNSEM.NON-LOCAL.YNQ [ LIST #list,
	   			  LAST #last ],
	   ARGS < [ SYNSEM.NON-LOCAL.YNQ [ LIST #list,
	   	    			   LAST #middle ] ],
		  [ SYNSEM.NON-LOCAL.YNQ [ LIST #middle,
		    			   LAST #last ] ] > ].

	same-ynq-unary-phrase := unary-phrase &
	  [ SYNSEM.NON-LOCAL.YNQ #ynq,
	    ARGS < [ SYNSEM.NON-LOCAL.YNQ #ynq ] > ].

	 non-ynq-word := lex-item &
	  [ SYNSEM.NON-LOCAL.YNQ 0-dlist ].
	 
	 question-clitic-lex := no-hcons-lex-item &
	  [ SYNSEM [ LOCAL [ CAT [ POSTHEAD +, 
	 			  VAL [ SPR < >,
	    	    	    	      	COMPS < >,
	 				SUBJ < >,
	 				SPEC < > ],
	 			  HEAD adp &
	 			    [ MOD < [ LIGHT +,
	 			      	      L-PERIPH + ] > ] ] ,
	 		    CONT.RELS <! !> ],
	 	    NON-LOCAL.YNQ <! *top* !> ] ].
	   
	 int-cl := head-only & interrogative-clause &
	  [ SYNSEM [ LOCAL.CAT [ VAL #val,
	    	    	      	MC bool ],
	 	    NON-LOCAL.YNQ <! !> ],
	    HEAD-DTR.SYNSEM [ LOCAL.CAT [ MC na,
	    		     	       	 VAL #val ],
	 		     NON-LOCAL.YNQ <! *top* !> ] ].
	 
	 decl-cl := head-only & declarative-clause & same-ynq-unary-phrase &
	  [ SYNSEM [ LOCAL.CAT [ VAL #val,
	    	    	      	MC bool ],
	 	    NON-LOCAL.YNQ <! !> ],
	    HEAD-DTR.SYNSEM [ LOCAL.CAT [ MC na,
	    		     	       	 VAL #val ],
	 		     NON-LOCAL.YNQ 0-dlist ] ].
	 
	 mc-na-headed-phrase := headed-phrase &
	  [ SYNSEM.LOCAL.CAT.MC na,
	    HEAD-DTR.SYNSEM.LOCAL.CAT.MC na ].
 
	   [ SYNSEM [ LOCAL [ COORD -,
		              CAT [ VAL [ SUBJ < >,
		                          COMPS < > ],
		                    MC +,
		                    HEAD +vc &
		                         [ FORM finite ] ] ],
	     	     NON-LOCAL.YNQ 0-dlist ] ].


At this point, we are overgenerating since the int-cl and decl-cl is not working correctly so there 
are interrogatives being parsed when the 'ru' particle is not present.  This will need to be addressed
next week since we ran out of time.



Trigger Rules

To fix the lack of generation due to hand editing of the lexicon file we tweaked the rules so that they
pointed to the correct TRIGGER values found in the lexicon file.  We also added a trigger rule for 
adjectives and embedded clauses:

	nu-adj_gr := arg0e_gtr &
	[ CONTEXT [ RELS <! [ PRED "~.*_adj_rel"] !> ],
	  FLAGS.TRIGGER "nu-adj" ].


	comp_gtr := generator_rule &
	   [ CONTEXT [ RELS <! [ ARG2 #h & handle ],
		               [ ARG0 #e & event  ] !>],
	     FLAGS [ EQUAL <#e, #h>,
		     TRIGGER "ru_comp" ]].

	nom-marker1_gr_1 := arg0e_gtr &
	  [ CONTEXT.RELS <! [ ARG1 #i & individual ] !>,
	    FLAGS [ SUBSUME < #i >,
		    TRIGGER "nom-marker2" ] ].

	nom-marker1_gr_2 := arg0e_gtr &
	  [ CONTEXT.RELS <! [ ARG2 #i & individual ] !>,
	    FLAGS [ SUBSUME < #i >,
		    TRIGGER "nom-marker2" ] ].

	nom-marker1_gr_3 := arg0e_gtr &
	  [ CONTEXT.RELS <! [ ARG3 #i & individual ] !>,
	    FLAGS [ SUBSUME < #i >,
		    TRIGGER "nom-marker2" ] ].

Generation has started to work, but are getting quite a few generated sentences since the combination
of morphemes quickly causes a huge volume of possible sentences.

Double clitics

Double clitics for information structure are possible in Mikayo:

		#Example 81 Focus particle
		Source: a
		Vetted: s
		Judgment: g
		Phenomena: info
		pztu=nu=du ff0
		pztu=nu=du ff-0
		man=NOM=FOC come-NPST
		A man comes over.

The focus marker attaches to a case marked NP but it can attach to other words such as adverbs, 
adjectives and verbs:


		#Example 77
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: cop
		kari=a takaf=du ar0
		kari=a taka-f=du ar-0
		3SG=TOP tall-ADV=FOC COP-NPST
		He is tall

Topic can either attach to a bare-np, replacing the case marking for NOM or a special topic marker
doublely attaches to the ACC.  You only see 'baa' attached to ACC-marked NPs.

		#Example 83 There can be more than one TOP tagged element
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: info
		ba=a mm=mu=baa faan0
		ba=a mm=mu=baa faa-n-0
		1SG=TOP potato=ACC=TOP eat-NEG-NPST
		I don't eat potatoes.

To make sure that 'du' does not attach to the left of the verb, a special 'no-case' was needed:

	no-case := case. ; focus marker

A new lex-type was created for the focus-marking adposition so it could attach to a case marked phrase
or a verb or adverb.  We are treating it as modifier to accomplish this:

	focus-marking-adp-lex := basic-one-arg & no-hcons-lex-item & no-icons-lex-item &
	 [ SYNSEM.LOCAL [ CONT.RELS <! !>,
		   CAT [ POSTHEAD +,
	   			 MKG fc,
				 HEAD adp &
				    [ MOD < [ LOCAL [ CONT.HOOK.ICONS-KEY focus,
	 					      CAT [ HEAD +vrp,
							    VAL.COMPS < > ]]] > ],
			       VAL [ SPR < >,
				     SUBJ < >,
				     COMPS < >,
				     SPEC < > ] ] ] ].

We wrote a similar type for the acc-topic marker and restricted it to an accusatively marked NP:

	acc-topic-marking-adp-lex := basic-one-arg & no-hcons-lex-item & no-icons-lex-item &
	  [ SYNSEM.LOCAL [ CONT.RELS <! !>,
			   CAT [ POSTHEAD +,
	   			 MKG tp,
				 HEAD adp &
				    [ MOD < [ LOCAL [ CONT.HOOK.ICONS-KEY topic,
	 					      CAT [ HEAD.CASE acc,
							    VAL.COMPS < > ]]] > ],
			       VAL [ SPR < >,
				     SUBJ < >,
				     COMPS < >,
				     SPEC < > ] ] ] ].




Rules

The rules file was edited to allow for new syntactic structures:

		 head-adj-int := head-adj-int-phrase.
		 
		 npbar-pred := np-bar-predicate-rule.
		 
		 int-cl := int-cl.
		 
		 decl-cl := decl-cl.


Results:


We have started to get better coverage, from 50 to 56.  We are getting parses
where none were seen before.  There is overgeneration since the int-cl/decl-cl
isn't working properly so if it parses, then there are 2 parses where one is
an interrogative and one is a declarative.

There is one case where there is massive overgeneration:

	kari=a sinsii
	3SG=TOP teacher
	He is a teacher

The interaction of the bare-np predicate, the messed up int-cl/decl-cl and 
optional arguments have caused an overgeneration of 176 parses.  It is 
grammatical and it wasn't parsing before so it's an overstep in the right 
direction. Another sentence that's ungrammatical and wasn't parsing before is 
now parsing due to this interaction as well:

	pztu=nu naratar aagu=u
	man=NOM learn-PAST song=ACC
	A man learned a song

The word order is all wrong but there is some crazy parsing going on that
will need to be looked at.

The embedded clauses are not parsing but we are close.  The complementizer
does attach to the complement and the case marker does as well but there is a 
conflict that we don't understand at the moment and will be looking at it more
closely through go-post.  Basically PP is a bit ill-formed so there is a 
unfication error when the verb and embedded clausal argument try to unify.

	cf-tar=ru=u ss-tar
	arrive-PAST=COMP=ACC know-PAST
	(I) know that (she) arrived.

'that she arrived' parses but when attaching to the verb 'know' errors occur
between event and ref-ind.

Double acclitics are parsing and the mrs's are looking correct so a sentence 
like:

	vva=ga=ru uri=u aztar=ru.
	you=NOM=FOC that=ACC say-PAST=Q
	You said that?

	pztu=nu=du ff0
	man=NOM=FOC comes.over-NPST
	A man comes over.

They are over generating due to the int-cl/decl-cl but other than that, they
are working wonderfully.

The basic TOPIC marker that was working before is no broken.  We only noticed
this when we ran the test suite, so we don't know exactly what's going on there.
It might have to do with the new focus-marker/acc-topic-marker types that were
introduced in this lab.

Other than the two sentences that massively overgenerate, we aren't seeing
any other parsing problems other than the int-cl/decl-cl problem.

This means, we are getting better coverage but overgeneration is being driven up by a few analyses from the Lab6 grammar.

