LING 567 Grammar Engineering
Lab5 Write Up- Miyako Ryukyuan

Adam Ledyard
Sanae Sato



=====================                                                          
1. Addressing Lab4 remarks
=====================


- TENSE

We switched our tense back to "obligatory" (it is non-irrealis moods). Here is what we have in choices file:

    has-aux=yes
    aux-comp-order=after
    aux-comp=vp
    multiple-aux=no


- Copula/Existential verb

Our grammar still needs to be worked on these verbs (Animacy feature is halfway embedded but not fully incorporated yet to give correct syntactical analysis.) 

We have reordered the verb-lex hierarchy in order to distinguish existential (and animacy distinction) from non-verbal predicates (involving the copula, one form of which is homophonous with the existential verb).


- Embedded Clause

Our grammar have not worked with the embedded clause yet; however, we found some examples of the clitic noun =su(u) or =ru working as a complementizer. Here is the example:

    cf-tar=[ru]=u  cnu=du ss-tar
    arrive-PAST=COMP=TOP yesterday+FOC know-PAST
    'Yesterday did (I) realize [the fact that] (she) arrived'

Is this the kind of example should we look for? Since double clitics (mentioned in later) is not working in our grammar yet, and =ru is also the question marker, so it will increase ambiguity if we add this to our grammar.


=====================
2. Ryukyuan in general for Lab5
=====================

In Lab5, we have covered adjectives (with NP), adverbs, demonstratives, and argument optionality both in subject dropping and object dropping. Miyako Ryukyuan does not have constrain on definiteness markers and agreement between adjectives and head noun.

Also, we revisited what we have left off from previous labs such as unification of pronoun and its markers by hand-editing tdl.



=====================
3. Modifiers - Adjectives
=====================


An adjective of Ryukyuan is derived from reduplication of PC-root or derived PC-stem (PC: property concept) such as imii+imi "small" or ssii+ssu "white".

In syntactical point of view, adjectives in Irabu Ryukyuan does not construct adjective phrases. Rather, they are 'parasitic' on NP and VP structure, and can appear in both NP and VP phrase. Lexical Adjective take Genitive marker in order to act as a modifier. (Genitive marker as clitic in our grammar.) This is another kind of add-position and different from English Adjective cases. It fills the modifier slot of a larger NP by taking =nu (GEN marker) as following example:

     ujakii+ujaki=nu pztu=tu kiban:+kiban=nu pztu
     RDP+rich+GEN    man=ASC RFP+poor=GEN    man
     'A rich man and a poor man'

When appearing in a VP, an adjective fills the modifier slot by taking =du (FOC marker, instead of GEN marker.) Also, the verb that follows the lexical verb must be a progressive auxiliary:


     kama=a [mii+imi]=du ur-0=ri
     that.place=TOP RDP+small=FOC PROG-NPST=eh
     "That place is small, eh?" (Lit. That place is small-ing.)

     kantja=a [jarabii+jarabi]=du ur-0=dara
     3SG=TOP RDP+child=FOC PROG-NPST=EMPH
     "They are childish, you see?" (Lit. They are childish-ing, you see.)


In Lab5, we modeled constitution with NP and got it work, but not with VP. It is because the auxilary (copula) verb is not functioning properly yet. Also, we do not deal with the reduplication process, and instead we directly defined derived words as lexicons. Here is the example sentence in our test suite.


     #(41)
     #Example 61 The adj must go before the noun in the NP and takes GEN case
     Source: author
     Vetted: f     
     Judgment: g
     Phenomena: adj
     ujakiiujaki=nu pztu=nu niv0
     ujakiiujaki=nu pztu=nu niv-0
     rich=GEN man=NOM sleep-NPST
     A rich man sleeps


Unfortunately, adjectives constituting with VP is not working in our grammar because of copula verb. But I am assuming it can be modeled with similar approach as adjective taking genitive marker. Instead of specifying "nouny" head, =du can be implemented as adj-clitic looking for "verby" head.


=====================
4. Modifiers - Adverbs
=====================


Similar to adjectives, adverbs in Miyako Ryukyuan is derived from PC stem (Property Concept stem) by affixation with -f as in taka-f (high-ADVLZ) 'highly', japa-f (soft-ADVLZ) 'softly'. Those derived adverbs may serve as a predicate adjunct or as a predicate complement. 

Except affixation, modeling adverb in our grammar was pretty straight forward, just specifying modified constituent to be verbal. For example, 


    kari=a taka-f=du ar-0
    3SG=TOP tall-ADVLZ=FOC exit-NPST
    "He is tall" (Lit. he is in a tall state.)

    kari=a taka-f=du nar-tar.
    3SG=TOP tall-ADVLZ=FOC become-PAST
    "He became tall" (Lit. he became in a tall state.)


We aim to model the PC-stem & suffixation into our grammar but not working propery yet. So instead, specified derived adverbs as our lexical entries as a workaround for now.


This is the example sentence from our test suite for adverbs.

    ### Adverbs ###

    #Example 66 Adverbs take a stem and add -f to adverbialize it.    
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: adv
    ba=ga vva=u japa-f kurusadi
    ba=ga vva=u japa-f kurusa-di 
    1SG=NOM 2SG=ACC soft-ADV kill-INTENT
    I kill you softly

It is not parsing in our result because we specified as japa-f as two morphemes (which we are aiming to model) but our current grammar is parsing following sentence as we expected.

    ba=ga vva=u japaf kurusadi
    ba=ga vva=u japaf kurusa-di 
    1SG=NOM 2SG=ACC softADV kill-INTENT
    I kill you softly



=====================
5. Demonstratives
=====================
=====================
5. Demonstratives
=====================

Miyako has a three way distinction, proximal, remote and close to hearer so the this hierarchy
was inserted into the language file:

		demonstrative_a_rel := predsort.
		proximal+dem_a_rel := demonstrative_a_rel. ; close to speaker
		distal+dem_a_rel := demonstrative_a_rel.   ; away from speaker
		remote+dem_a_rel := distal+dem_a_rel.      ; away from speaker and hearer
		hearer+dem_a_rel := distal+dem_a_rel.      ; near hearer


Since there are no other determiners, only one super-type is needed so we merged the supertype
and the demonstrative determiner subtype:

	determiner-lex-supertype := norm-hook-lex-item & basic-zero-arg &
	  [ SYNSEM [ LOCAL [ CAT [ HEAD det,
				   VAL [ SPEC.FIRST.LOCAL.CONT.HOOK [ INDEX #ind & [ COG-ST activ+fam ],
									 LTOP #larg ],
									SPR < >,
									SUBJ < >,
									COMPS < >]],
				 CONT [ HCONS < ! qeq &
					 [ HARG #harg,
					   LARG #larg ] ! > ,
					  RELS < ! [ PRED "exist_q_rel" ], 
							   #altkey & 
					   arg1-ev-relation &
					   [ LBL #larg,
						ARG1 #ind ] ! > ]],  
			 LKEYS [ KEYREL quant-relation &
				 [ ARG0 #ind,
				   RSTR #harg ], 
				 ALTKEYREL #altkey ] ] ].

and of course the lexical entries were updated to use this:


		kunu := determiner-lex-supertype &
		  [ STEM < "kunu" >,
			SYNSEM.LKEYS.ALTKEYREL.PRED "proximal+dem_a_rel" ].

		unu := determiner-lex-supertype &
		  [ STEM < "unu" >,
			SYNSEM.LKEYS.KEYREL.PRED "hearer+dem_a_rel" ].

		kanu := determiner-lex-supertype &
		  [ STEM < "kanu" >,
			SYNSEM.LKEYS.KEYREL.PRED "remote+dem_a_rel" ].
			
Each demonstrative has an updated PRED value according to the new hierarchy and it inherits
determiner-lex-supertype instead of the standard one found in matrix.tdl. The pronoun-noun-lex
type was also updated so all pronouns have COG-ST value of activ-or-more and SPECI + :

	pronoun-noun-lex := no-spr-noun-lex &
	  [ SYNSEM.LOCAL [ CAT.HEAD.ANIMATE +,
					   CONT.HOOK.INDEX [ COG-ST activ-or-more,
									     SPECI + ] ] ].

=====================
6. Argument Optionality
=====================

Unfortunately, our grammar document (or any other sources) did not mention anything about argument optionality for both subject and object dropping. However, there were some subject dropping example sentences in our grammar document, then we assumed that it is valid phenomena. There was no subject dropping examples in the document.


In Lab 3, when we first dealt with argument optionality through customization system, we decided to assume that dropping object is not allowed since we could not find any evidence. However, considering similarity between Miyako Ryukyuan and standard Japanese, we decided to assume object dropping is legitimate. (At least chances are high.)

Our examples in the test suite are follows:

    #Example 19 grammatical arugument optionality on subject
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: pro-d
    tigami=u tur0
    tigami=u tur-0
    letter=ACC take-NPST
    (I) take a letter.

    #Example 20 ungrammatical arugument optionality on object
    Source: author
    Vetted: f
    Judgment: g                                                                     Phenomena: pro-d
    ba=ga tur0
    ba=ga tur-0
    1SG=NOM take-NPST
    I take (something).


In order to expose and pass up right semantics for argument optionality (in this case, we assumed default as OPT-CS type-id), we edited miyako.tdl as follows: 

BEFORE:

trans-verb-lex := nom-acc-transitive-verb-lex & ASPECT-or-converb-rule-dtr & MOOD-rule-dtr & TENSE-rule-dtr & negation-rule-dtr &
  [ INFLECTED.TENSE-FLAG - ].

AFTER:

trans-verb-lex := nom-acc-transitive-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 ].

By looking at the result, it seems to be that our grammar is doing the right thing. Sample sentences in testsuite is as follows:


=====================
7. Other TDL editing - Unification Problem
=====================

We finally incorporated pronoun unification features into our grammar and working as we expected. (i.e. PRONOUN feature is doing the right work.) This fix leads us to reduce our error drastically. 


    #Example 15 1PL inclusive Pronoun
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: c
    duu nu iditar
    duu=nu idi-tar
    1PL.incl=NOM exit-PAST
    We exited.


=====================
8. Test Courpus
=====================


=====================
9. Testing Grammar
=====================

Here are the major errors observed in Lab 5.


- Coordination (Asociative) Errors

    #Example 57 both NPs are marked with ASC label
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: crd
    ujakiiujaki=nu pztu=tu kibannkiban=nu pztu=tu niv0
    ujakiiujaki=nu pztu=tu kibannkiban=nu pztu=tu niv0-0
    rich=GEN man=ASC poor=GEN man=ASC sleep-NPST
    The rich man and the poor man sleep.


- Dual Clitics Errors

We are still confused how to implement dual clitics such as "ai=sii=du".

    #Example 59
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: crd
    kuzmi=u burii ai=sii=du fau-tar
    kuzmi=u bur-i-i, ai=sii=du fau-tar
    small.claw=ACC break-THM-MED, that.way=INS=FOC eat-PST
    (I) broke the small claws and ate it in that way.


- Lack of Lexicon.

    #Example 60
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: emb-d
    saki=u numtigaa, ba=a sugu=du niv0
    saki=u num-tigaa, ba=a sugu=du niv-0
    Sake=ACC drink-CVB.COND 1SG=TOP right.away=FOC sleep-NPST
    When (I) drink Sake, I sleep easily.


- Adverbs 

As I mentioned above, Adverb example sentences are not parsing right now because our testsuite define adverb as the word consisted from two morphemes such as japa-f. This is what we would like to implement, so we kept our test suite as it is. The temporal remedyfor our current grammar is defineing 'japaf' and 'takaf' as adverbs. (And it gives us correct analysis.) 


    #Example 66 Adverbs take a stem and add -f to adverbialize it.
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: adv
    ba=ga vva=u japa-f kurusadi
    ba=ga vva=u japa-f kurusa-di 
    1SG=NOM 2SG=ACC soft-ADV kill-INTENT
    I kill you softly


    #Example 68 The adverb must come before the verb
    #(fixed as ungramatical example by sanaes 5/2/2013)
    Source: author
    Vetted: f
    Judgment: u
    Phenomena: adv, wo
    ba=ga vva=u kurusadi japa-f 
    ba=ga vva=u kurusa-di japa-f  
    1SG=NOM 2SG=ACC kill-INTENT soft-ADV
    I kill you softly


Note: Example 67 is giving us false-positive with same reason.
In order to have proper adverbial form, 'japa' has to be followed by -f.
    
    #Example 67 The suffix is required for adverbialization
    Source: author
    Vetted: f
    Judgment: u
    Phenomena: adv
    ba=ga vva=u japa kurusadi
    ba=ga vva=u japa kurusa-di 
    1SG=NOM 2SG=ACC soft kill-INTENT
    I kill you softly


- Embedded Clause

Following sentence does not parse because of  (and also for becauselack of lexicon too.)

    #Example 69
    Source: author
    Vetted: f
    Judgment: g
    Phenomena: emb-d
    ba=a saki=u numtigaa, sugu=du niv0
    ba=a [saki=u num-tigaa], sugu=du niv-0
    1SG=TOP [Sake=ACC drink-CVB.COND] right.away=FOC sleep-NPST
    [When I drink Sake], I sleep easily.


=====================
10. Summary                
=====================


The grammar for Adjective needs to be working with VP constituent. (In order to get this analysis right, we also need to work on copula/auxiliary verb.) The grammar model for Adverbs also needs to be enhanced to recognize stem and -f adverbial marker.

Demonstrative is working properly, and so does argument optionality (both in subject dropping and object dropping). 


