Lab 7
Language: Miyako (Irabu Ryukyan)
Adam Ledyard and Sanae Sato


The following objectives are addressed in this Lab 7 write up.

  From Lab6:

  1. Ambiguity elimination
  2. Topic Marker bug
  3. Bug with transitive type
  4. Complementizer

  Lab 7 Phenomena: 

  5. Potential/ Modal "Can" 
  6. Negation  / Non-verbal Negation
  7. Evidentials / Hearsay
  8. Overall Summary


==========================
1. Ambiguity elimination
==========================

	We had some major over generation when we first implemented the interrogative/
	declarative clause fix for the 'ru' q-clitic.  To combat this we restricted
	the types so many of the overgenerated sentences were eliminated:

		mvi-clause := head-only &
		 [ SYNSEM.LOCAL [ COORD -,
				  CAT [ HEAD +vc,
						VAL [ SUBJ < >,
						  COMPS < > ] ] ] ].
		  
		int-cl := mvi-clause & 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 := mvi-clause & 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 ] ].
					 
	Coordination is blocked, the clause has to be fully saturated before applying
	this rule and the YNQ diff-list was appropriately applied where needed in other
	unary-phrase types.

==========================	
2. Topic Marker bug:
==========================

	There was a small bug where:
	
		kari=a sinsii
		3SG=TOP teacher
		He is a teacher
		
	was not parsing.  This was due to a name conflict in the lexicon and it was 
	producing a '=a' clitic that was equivalent to the 'baa'-acc topic marker.
	That was rectified and now it parses correctly.

==========================	
3. Bug with transitive type:
==========================

	There was another sentence that called for a change:
	
		#Example 85 Topic marking
		Source: a
		Vetted: s
		Judgment: g
		Phenomena: info
		vva=a sinsii=n nari0
		vva=a sinsii=n nari-0
		2SG=TOP teacher=DAT become-NPST
		You become a teacher
	
	where become is a nom-acc transitive verb only, so the dative caused problems.
	We underspecified the case of the second argument but made sure that it had
	an adposition.  This allowed the sentence to be parsed and caused no ill effects
	on the other parses.


==========================
4. Embedded Clause / Complementizer
==========================

	Embedded Clause ( using complementizer) is still not working properly as it should be. currently we are getting 98 parses (including correct parse). We will continue work on this proble,, and will update the status.


Following materials are for the LAB 7 Phenomena.


==========================
5. Potential/ Modal "Can" 
==========================


  
  - Description
	Ryukyuan's potential expression is derived by affixiation using the passive suffix -(r)ai. The /r/ is deleted when attaching to a class 2 verb. (i.e. the verb does not end with 'i'.) Thus, faa- 'eat' + -(r)ai results in faa-ai (eat-POT). We could not find a lexical item for 'glass' so we subsituted the word for 'sand' (nnagu) instead. Therefore, our new example sentence for "I can eat glasses" would be 

  - IGT Example (Test Suite)

    Example 86 
    Source: a
    Vetted: f
    Judgment: g
    Phenomena: m
    ba=ga  nnagu=u faa-ai-0
    1SG=NOM sand=ACC eat-POT-NPST
    I can eat glass


  - Implementation

	The potential affix is a morpheme that attaches to the stem of the verb
	and appears to fit into the ASPECT slot that was previously defined.
	We renamed the supertype as PREinflection since that is how it's labeled in
	the grammar:

	
		PRE-lex-rule-super := MOOD-rule-dtr & TENSE-rule-dtr & infl-lex-rule & negation-rule-dtr &
		  [ INFLECTED #infl,
		    DTR PRE-or-converb-rule-dtr &
						[ INFLECTED #infl ] ].

						

	We wrote a potential-lex-rule that inherits from the PRE-lex-rule 
	and the cont-change-only-lex-rule:
	
		potential-lex-rule := PRE-lex-rule-super & cont-change-only-lex-rule &
		  [ C-CONT [ RELS <! arg1-ev-relation &
					[  PRED "_can_v_rel",
					   LBL #ltop,
					   ARG0 #index,
					   ARG1 #harg ] !>,
				  HOOK [ LTOP #ltop,
				  INDEX #index ],
			  HCONS <! qeq & [ HARG #harg,
					   LARG #larg ] !> ],
			DTR.SYNSEM.LOCAL.CONT.HOOK.LTOP #larg ].
	

	The difference in the other aspect morphemes is that they inherit from 
	add-only-no-ccont-rule instead so now the sentence below parses:
	
		#Example 86 
		Source: a
		Vetted: f
		Judgment: g
		Phenomena: m
		ba=ga  nnagu=u fau-ai-0
		1SG=NOM sand=ACC eat-POT-NPST
		I can eat glass
		
	

	- Analysis
	The sentence parses, and by looking at Simple MRS, it seems to be correctly indexed for the relationship between sand (nnagu: _glass_n_rel), and pronoun (ba: _pronoun_n_rel) as  eaten/eater relations. But we are not fully sure that our potential affix is correctly been indexed because the ARG0 is indexed with INDEX of the whole AVM. (It can be interpret as that it gives potential modal to the "eating" event. Correct us on this if we are wrong.)

==========================
6. Negation / Non-verbal Negation
==========================
  - Description

	To express "it does not hurt me" in Ryukyuan,  we came up with a sentence which treat "hurt" as adverbialized-adjective simply because we thought that will sound more natural in Ryukyuan. Therefore, we needed to deal with non-verval negation in order to implement this sentence. 

In general (in terms of the grammar book), both non-verval negation and verb negation was treated as same affixation phenomina. For non-verval negation, a copula-esque verb (it was called as 'state-verb' in the book to dintinguish from the copula). However, we decided to treat this phenomina in two different types.


  - IGT Example (Test Suite)

	Here are the new sentences we added for this lab.


	    Example 87
	    Source: a
	    Vetted: f
	    Judgment: g
	    Phenomena: neg
	    ba=ga agaf=du njaan-0
	    ba=ga agaf=du njaan-0
	    1st=TOP hurtadv=FOC  NEG-NPST
	    I am not (in a) hurt (state)  (=It does not hurt me.) 

	    Example 88 
	    Source: a
	    Vetted: f
	    Judgment: g
	    henomena: neg
	    agaf=fa njaa-n-0
	    agaf=fa njaa-n-0
	    hurtadv=FOC  negstate-NEG-NPST
	    (I am) not (in a) hurt (state) (=It does not hurt me.)



  - Implementation

	There are two types of negation, verbal morphemes and a copula-esque verb:
	---------------------	
	Type 1. Verb Negation
	---------------------

	The negation morpheme provided by the choices file gave us the types below:
	
		negation-lex-rule-super := MOOD-rule-dtr & TENSE-rule-dtr & cont-change-only-lex-rule & infl-lex-rule &
		  [ INFLECTED #infl,
			DTR negation-rule-dtr &
				[ INFLECTED #infl ] ].

				
		negative-lex-rule := negation-lex-rule-super &
		  [ C-CONT [ HOOK [ XARG #xarg,
							LTOP #ltop,
							INDEX #ind ],
					 RELS <! event-relation &
							 [ PRED "neg_rel",
							   LBL #ltop,
							   ARG1 #harg ] !>,
					 HCONS <! qeq &
							  [ HARG #harg,
								LARG #larg ] !> ],
			SYNSEM.LKEYS #lkeys,
			DTR.SYNSEM [ LKEYS #lkeys,
						 LOCAL [ CONT.HOOK [ XARG #xarg,
											 INDEX #ind,
											 LTOP #larg ],
								 CAT.HEAD verb ] ] ].

		For the most part it works correctly.  There seems to be a
		slight bug in the semantics that still needs to be addressed.
								 
	---------------------------
	Type 2. Non-verbal negation
	---------------------------

		There is also a copula like verb that attaches to non-verbal
		predicates to show negation:
		
			ba=ga    takaf=du     njaan0
			ba=ga    takaf=du     njaan-0
			1st=TOP tall.ADVLZ=FOC  NEG-NPST
			I am not (in a) tall (state). (= I'm not tall. )
			
		It works the same as the copula as far as taking the adverbialized
		adjective as its predicate plus the focus marker so at this point
		we have it inheriting from the copula-verb-lex:
		
		
			njaan := copula-verb-lex &
			 [ STEM < "njaan">,
			   SYNSEM.LKEYS.KEYREL.PRED "neg_rel" ].
	   

	- Analysis

		We get the correct parse chart but there are some problems with the
		semantics.  This means we need to develop another type, but at this
		point we are not sure how to proceed.




==========================
7. Evidentials / Hearsay
==========================

	The goal of parsing one of our test corpus sentences, we decided to look at
	evidentials so we could parse this sentence:
	
	#Sentence 10 of test corpus
	Source: author
	Vetted: f
	Judgment: g
	Phenomena: wo
	tooriike=n nartar=ca
	tooriike=n nar-tar=ca.
	Tooriike=DAT beome-PAST=HS
	(It) became Tooriike.
	
	To do this we added a EV feature to the event environment:

		event :+ [ EV evid ].

		evid := predsort.
		reportative := evid.
		quotative := evid.
		
	We then based the clitic of of the questiong particle lex-item found in
	our grammar and set the EV value to reportative:
		
		reportative-lex-item := complementizer-lex-item &
		  [ SYNSEM.LOCAL [ CONT.HOOK.INDEX.EV reportative,
						   CAT.VAL.COMPS.FIRST.LOCAL.CAT.HEAD.FORM finite ] ].



==========================
8. Overall Summary
==========================


Overall, we successfully have increased coverage and decreased (compare to Lab6) overgeneration by following:

	                   In (Coverage)   Analyses (Overgeneration)
	Lab 7 (88 items) :     69.8          3.33
	Lab 6 (86 items) :     56.0         12.43
	Lab 5 (86 items) :     50.0          1.00

In addition to the fix for topic-marker and transitive type bugs, we fixed minor errors such as spelling errors and wrong separation markers in the testsuit. That also contributed to the overall coverage. What we left off from Lab 7 is not too much, but the affix negation (-njaan) needs a little more work (in terms of semantics), and will try to implement as an auxiliary.

Also, we still have problems in generation direction. Hopefully, we will be able to address the problems and enhance our grammar through the next lab.
