1. "Can" modal

In Russian, the can modal (mochj) is a separate word which takes a VP complement and a subject. Mochj is inflected for tense, person and number. The VP complement it takes is in infinitive form.

Example:
Ja       mogu         estj     steklo.

Ja       mo-gu        estj     steklo

1.SG     can-1SG.PRS  eat.INF  glass.ACC

I can eat glass



To implement this, we added a verb type:

mozhno-modal-lex  := verb-lex & single-rel-lex-item & trans-first-arg-raising-lex-item-1 & mozhno-lex-rule-dtr &
  [ SYNSEM.LOCAL [ CAT.VAL [ SUBJ < #subj >,
                             COMPS < #comps >,
                             SPR < >,
                             SPEC < > ],
                   CONT.HOOK.XARG #xarg ],
    ARG-ST < #subj &
             [ LOCAL [ CONT.HOOK.INDEX #xarg,
                       CAT [ VAL [ SPR < >,
                                   COMPS < > ],
                             HEAD noun &
                                  [ CASE nom ] ] ] ],
             #comps &
             [ LOCAL.CAT [ VAL [ SUBJ < [  ] >,
                                 COMPS < > ],
                           HEAD verb &
                                [ FORM infinitive ] ] ] > ].


We also added a lex-rule supertype and a set of lexical rules to inflect appropriately. For example:

mozhno-modal-lex  := verb-lex & single-rel-lex-item & trans-first-arg-raising-lex-item-1 & mozhno-lex-rule-dtr &
  [ SYNSEM.LOCAL [ CAT.VAL [ SUBJ < #subj >,
                             COMPS < #comps >,
                             SPR < >,
                             SPEC < > ],
                   CONT.HOOK.XARG #xarg ],
    ARG-ST < #subj &
             [ LOCAL [ CONT.HOOK.INDEX #xarg,
                       CAT [ VAL [ SPR < >,
                                   COMPS < > ],
                             HEAD noun &
                                  [ CASE nom ] ] ] ],
             #comps &
             [ LOCAL.CAT [ VAL [ SUBJ < [  ] >,
                                 COMPS < > ],
                           HEAD verb &
                                [ FORM infinitive ] ] ] > ].

Finally we added a lexical entry:


mozhno-modal-lex  := verb-lex & single-rel-lex-item & trans-first-arg-raising-lex-item-1 & mozhno-lex-rule-dtr &
  [ SYNSEM.LOCAL [ CAT.VAL [ SUBJ < #subj >,
                             COMPS < #comps >,
                             SPR < >,
                             SPEC < > ],
                   CONT.HOOK.XARG #xarg ],
    ARG-ST < #subj &
             [ LOCAL [ CONT.HOOK.INDEX #xarg,
                       CAT [ VAL [ SPR < >,
                                   COMPS < > ],
                             HEAD noun &
                                  [ CASE nom ] ] ] ],
             #comps &
             [ LOCAL.CAT [ VAL [ SUBJ < [  ] >,
                                 COMPS < > ],
                           HEAD verb &
                                [ FORM infinitive ] ] ] > ].

The above example is parsing with exactly one tree. However, we are not getting the _can_V_rel in the "I can eat glass" example sentence.

2. Negation

Russian uses an adverbial strategy (i.e. sentential negation is expressed via an adverb
which modifies the V, VP, or S), therefore it has been covered by the customization
script.  

For our negation example, we added "rezh" verb to our lexicon.tdl (which is roughly 
translated as "cut" or "hurt" in some instances):


	rezh := tr-nom-acc-1-conj-imp-verb-lex &
  		[ STEM < "rezh" >,
    		  SYNSEM.LKEYS.KEYREL.PRED "_hurt_v_rel" ].


We are now able to parse sentence #239, but we are getting 22 parses right now (before
we were getting even more, i.e. 64, until we constrained the YNQ value in the int-cl
and decl-cl).  

	#239   
	Source: a, e
	Vetted: t
	Judgment: g
	Phenomena: negation
	Ono        menja        ne      rezhet.
	Ono 	   menja	ne	rezh-et.
	3.SG.NEUT  me.ACC       not     cut.3SG.PRS
	It doesn't hurt me.

The main reason for that appears to be the coordination rules (for example, object "menja" is 
attaching to subject "ono" instead of attaching to VP "ne rezhet" and that seems to be licensed 
by our current coordination rules. We would need two things to fix that: 

   a). to constraint the case of N coordinands to be the same:

	n4-top-coord-rule := basic-n-top-coord-rule & apoly-top-coord-rule &
  		[ SYNSEM.LOCAL.COORD-STRAT "4",
    		  LCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case,
    		  RCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case ].

	np4-top-coord-rule := basic-np-top-coord-rule & apoly-top-coord-rule &
 		 [ SYNSEM.LOCAL.COORD-STRAT "4",
  		   LCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case,
  		   RCOORD-DTR.SYNSEM.LOCAL.CAT.HEAD.CASE #case ].

        and since that appears not to be enough (we didn't get a decrease in ambiguity),

   b).  to allow the object to attach to VP even if it precedes the VP  (which happens only in
        2 parses out of 22).  Not sure how to do that (don't think we can do it through specifying
        the position in the sentence like we did for "li" and it actually looks like the problem is
	in the coord rules and constrainig them.  Another possibility would be to constrain the 
	-bottom-coord-rules below, but it is not exactly clear how:

           n4-bottom-coord-rule := unary-bottom-coord-rule & n-bottom-coord-phrase &
            [ SYNSEM.LOCAL [ COORD-STRAT "4",
                             COORD-REL.PRED "_and_coord_rel" ] ].

           np4-bottom-coord-rule := unary-bottom-coord-rule & np-bottom-coord-phrase &
            [ SYNSEM.LOCAL [ COORD-STRAT "4",
                             COORD-REL.PRED "_and_coord_rel" ] ].


Sentence #240 is less ambiguous because of object being placed after the verb
(i.e. we are only getting 2 parses for this sentence):

	#240   
	Source: a, e
	Vetted: t
	Judgment: g
	Phenomena: negation
	Ono        ne	  rezhet	menja.
	Ono 	   ne	  rezh-et	menja.
	3.SG.NEUT  not	  cut-3SG.PRS	me.ACC
	It doesn't hurt me.

CLEAN UP SECTION


1).  MATRIX YES-NO QUESTIONS

We explicitly identified the L-PERIPH values on the SYNSEM of the head daughter and the
modifier daughter in basic-head-mod-phrase-simple:  

	basic-head-mod-phrase-simple :+
		[ HEAD-DTR.SYNSEM.L-PERIPH #periph,
		NON-HEAD-DTR.SYNSEM.LOCAL.CAT.HEAD.MOD < [ L-PERIPH #periph ] > ].



We also constrained the bare-np-phrase, basic-head-opt-comp-phrase and 
basic-head-opt-subj phrase to inherit from the same-ynq-unary phrase instead of 
same-periph-unary-phrase as we had at first:

	bare-np-phrase := basic-bare-np-phrase & same-ynq-unary-phrase & mc-na-headed-phrase &
  	  [ C-CONT.RELS <! [ PRED "exist_q_rel" ] !>,
  	   SYNSEM.LIGHT - ].

	basic-head-opt-comp-phrase :+ same-ynq-unary-phrase & mc-na-headed-phrase &
 	  [ SYNSEM.L-PERIPH #periph,
  	   ARGS < [ SYNSEM.L-PERIPH #periph ] > ].

	basic-head-opt-subj-phrase :+ same-ynq-unary-phrase & mc-na-headed-phrase & 
          [ SYNSEM.L-PERIPH #periph,
           ARGS < [ SYNSEM.L-PERIPH #periph ] > ].


We instantiated int-cl and decl-cl in rules.tdl:

	int-cl := int-cl.

	decl-cl := decl-cl.


And made the root symbol to have an empty YNQ value in roots.tdl:

	root := phrase &
 		 [ SYNSEM [ LOCAL [ COORD -,
                   	CAT [ VAL [ SUBJ < >,
                               	    COMPS < > ],
                         	MC +,
                         	HEAD +vc &
                              [ FORM finite ] ] ],
	     NON-LOCAL.YNQ 0-dlist ] ].


After we did all this, this was not causing any problems any longer:

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

and we can parse the "li" sentences now.  

The only problem was getting extra ambiguity, i.e. for sentence #163 below we get 
two parses now, lisenced by Int-cl and Decl-cl (with the ADV licensed by Int-cl):

	#163   (subj   li   verb   obj)	Source: a, e	Vetted: t	Judgment: g	Phenomena: Matrix yes-no questions	Ivan li 	     kupil        sobaku?	Ivan li 	     kupi-l       sobak-u	Ivan INTERROGATIVE   buy-3SG.PST  dog-ACC.SG	Did Ivan buy a dog?

but it was resolved by constraining int-cl to not take "li" as its daughter, i.e. we
constrained the HEAD value of the daughter to +vj in int-cl and decl-cl.  And now we
are getting only one parse for #163.  

  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,
				    HEAD +vj ],
                                  NON-LOCAL.YNQ <! *top* !> ]].

  decl-cl := head-only & declarative-clause & same-ynq-unary-phrase &
    [ SYNSEM.LOCAL.CAT [ VAL #val,
                         MC bool ],
      HEAD-DTR.SYNSEM [ LOCAL.CAT [ MC na,
                                    VAL #val,
				    HEAD +vj ],
                        NON-LOCAL.YNQ 0-dlist ]].



2).  EMBEDDED SENTENCES

instead of:

     embed-clausal-verb-lex := main-verb-lex & clausal-second-arg-trans-lex-item &
         [ ARG-ST < [ LOCAL.CAT.HEAD noun ],
                  #comps &
                  [ LOCAL.CAT [ VAL [ SPR < >,
                                      COMPS < > ],
                       HEAD comp ] ] >,
          SYNSEM.LOCAL [ CAT.VAL.COMPS < #comps >,
                         CONT.HOOK.INDEX.SF prop-or-ques ] & 
		         [ CAT.MC - ] ].


we now have this to constrain the MC and SF on the verb's complement
(an alternative way to write this also suggested in your email was giving errors,
so we went with this one):

    embed-clausal-verb-lex := main-verb-lex & clausal-second-arg-trans-lex-item &
          [ ARG-ST < [ LOCAL.CAT.HEAD noun ],
                   #comps &
                   [ LOCAL.CAT [ VAL [ SPR < >,
                                       COMPS < > ],
                                 HEAD comp ] ] >,
            SYNSEM.LOCAL [ CAT [ VAL.COMPS < #comps &
                         [ LOCAL [ CAT.MC -,
                           CONT.HOOK.INDEX.SF prop-or-ques ]] > ]]].


Before we could not parse the interrogative sentences:

You suggested to make the question-embedding verbs select for
[HEAD verb] complements.  Since our interrogative-embed-clausal-verb-lex
inherited from embed-clausal-verb-lex (which selected [HEAD comp] complement,
our interrogative-embed-clausal-verb-lex no longer inherits from it, but instead
inherits from its supertypes and constraits its complement to be [HEAD verb].  

so now instead of: 

     interrogative-embed-clausal-verb-lex := embed-clausal-verb-lex &
           [ SYNSEM.LOCAL [ CAT.VAL.COMPS < [ LOCAL [ CAT [ HEAD verb,
          		    		      	      	    MC - ], 
     		    		              CONT.HOOK.INDEX.SF ques ] ] > ] ].

we have this:

     interrogative-embed-clausal-verb-lex := main-verb-lex & clausal-second-arg-trans-lex-item &
           [ ARG-ST < [ LOCAL.CAT.HEAD noun ],
                    #comps &
                    [ LOCAL.CAT [ VAL [ SPR < >,
                                        COMPS < > ],
                                  HEAD verb ] ] >,
             SYNSEM.LOCAL [ CAT [ VAL.COMPS < #comps &
                          [ LOCAL [ CAT.MC -,
                            CONT.HOOK.INDEX.SF ques ]] > ]]].

This allows us to parse now the sentence #241 with just one parse:

	#241   interrogative-embedded clause: subj  li  verb obj
	Source: a, e
	Vetted: t
	Judgment: g
	Phenomena: embedded questions
	Ja       sprosil           Ivan		li	kupil        	sobaku.
	Ja       sprosi-l          Ivan		li	kupi-l       	sobak-u
	1.SG     ask-PERF.1SG.PST  Ivan.NOM	Q	buy-3SG.PST  	dog-ACC	
	I asked whether Ivan bought a dog.

(although #234 is still with 25 parses, but that's due to coordination issues):

	#234   interrogative-embedded clause: v  li  subj obj
	Source: a, e
	Vetted: t
	Judgment: g
	Phenomena: embedded questions
	Ja       sprosil           kupil        li   Ivan        sobaku.
	Ja       sprosi-l          kupi-l       li   Ivan        sobak-u
	1.SG     ask-PERF.1SG.PST  buy-3SG.PST  Q    Ivan.NOM    dog-ACC	
	I asked whether Ivan bought a dog.

The problem that we have now, however, is that because interrogative clause does no longer
require the [HEAD comp] complement, the ungrammatical sentence #235 (ungrammaticality due
to the interrogative particle missing) is now parsed as grammatical because its complement
is [HEAD verb].  

	#235   interrogative-embedded clause: *v subj obj
	Source: a, e
	Vetted: t
	Judgment: u
	Phenomena: embedded questions
	Ja       sprosil            kupil         Ivan        sobaku.
	Ja       sprosi-l           kupi-l        Ivan        sobak-u
	1.SG     ask-PERF.1SG.PST   buy-3SG.PST   Ivan.NOM    dog-ACC	
	intended: I asked whether Ivan bought a dog.


3). Locative adposition not parsing

Test sentence:

Kniga			na	stole. 
Knig-a			na	stol-e  
book-FEM.SG.NOM  	on	table-SG.MASC.PREP

The above test sentence was not parsing and through unification testing for head-comp rule and the root condition, two errors were found:

a) there was no lexical rule to inflect the noun stol in singular prepositional case
b) the root condition was looking for MOD < > and in the above sentence that was not the case.

Once these errors were fixed, the locative adposition "na" is parsing.


4).  VERBS CLEAN UP

Verb "rezatj" is a first-conjugation verb with consonant stem 2: i.e. it undergoes the consonant
mutation throughout conjugation.  Verb "spatj" is a 2-conj verb also with the mutation of the 
consonant in the first-person singular of the present tense, i.e. "p"-->"pl".  We need to take
care of these mutations and we were hoping to get it done this lab, but we didn't get to it yet.  
It is not causing any problems for our testsuite (i.e. we don't have these examples there).  


5).  COORDINATION RULES

were briefly discussed above for identifying the coordinands' cases, etc. in the Negation Section.  
We need to figure out how to prevent the object from attaching to the subject in the subj-obj-verb 
word order.  

ITSDB

Coverage: 
Lab 6: 66.9
Lab 7: 69.4

Overgeneration:
Lab 6: 2.6
Lab 7: 10.3
