#Modification

#Head-Modifier Rules

##Adjectives 

Yiddish adjectives occur between the determiner and the noun. The following examples show correct and incorrect positioning of adjectives in the noun phrase.

			Source: author
			Vetted: f
			Judgment: g
			Phenomena: {adjectives}
			Dos voyl kind zingt.
			Dos voyl-e kind zing-t.
			DEF.NOM.N nice-NOM.SG.N child sing-3SG.
			The nice child sings.

			Source: author
			Vetted: f
			Judgment: u
			Phenomena: {adjectives}
			Dos kind voyle zingt.
			Dos kind voyl-e zing-t.
			DEF.NOM.N child nice-NOM.SG.N sing-3SG.
			The nice child sings.

			# The adjective is almost an adverb here, but shouldn't be inflected. 
			Source: author
			Vetted: f
			Judgment: u
			Phenomena: {adjectives}
			Dos kind zingt voyle.
			Dos kind zing-t voyle-e.
			DEF.NOM.N nice good-NOM.SG.N sing-3SG.
			The child sings nice.
			
The full set of examples have been omitted as they are present in our lab four write up. Yiddish only has prehead modifiers, so we created a rule called `adj-head-int` as described in the lab directions. Our grammar already contained `+nvcdmo` which was generated automatically by the matrix:

				adj-head-int := adj-head-int-phrase.
				+nvcdmo :+ [ MOD < > ].

We parsed a small sample of sentences selected from our testsuite which did not contain adjectives and found that the `adj-head-int` rule did not fire. This, of course, was expected due to the inclusion of `+nvcdmo` in our grammar. We also observed that sentences with adjectives were parsing correctly and the adjective relations were being predicated of the right indices.


#Adverb

Yiddish adverbs are not subject to inflection. The base form of adjectives in Yiddish can be used as manner adverbs. Manner adverbs can also be formed with derivational affixes out of bases including adjectives, past participles, and nouns, however our grammar not address these. Manner adverbs in Yiddish appear in the final position of unmarked clauses, in which case they attach to the VP. This is illustrated below:


		Source: a:193
		Vetted: t
		Judgment: g
		Phenomena: {adverb}
		Dos kind zingt shayn.
		Dos kind zing-t shayn.
		DEF.ACC.N child sing-1SG nicely.
		The child sings nicely.

Adverbs can also be topicalized by being moved to the first position, resulting in them attaching to the sentence. This is illustrated below:

		Source: author based on a:193
		Vetted: f
		Judgment: g
		Phenomena: {adverb}
		Shayn dos kind zingt.
		Shayn dos kind zing-t.
		nicely DEF.ACC.N child sing-1SG.
		Nicely, the child sings.

The remaining possible positions (at least in a simple declarative  sentence such as this) are ungrammatical. These are addressed in the lab four writeup. 

We created an adverb lex-type using the tdl provided in the lab directions.

	adverb-lex := basic-adverb-lex & intersective-mod-lex &
		[SYNSEM [ LOCAL [ CAT [ HEAD.MOD <[ LOCAL.CAT.HEAD verb]>,
				VAL [ 	SPR < >,
						SUBJ < >,
						COMPS < >,
						SPEC < >
				
		]]]]].
	
We tried parsing the sentence "Dos kind zingt shayn." which produced spurious ambiguity. It yielded three parse trees, all with correct (and identical) MRSs which varied in the order which the `head-adj-int`, `subj-head`, and `basic-head-opt-comp` rules were applied.

In the first two trees, `basic-head-opt-comp` and `head-adj-int` exchanged places. In the final tree, `head-adj-int` was the root instead of `subj-head`. In the case of the first two trees, this ambiguity appears to be analogous to the ambiguity we observed in the argument optionality module and we suspect that we can cull this effect by constraining adverbs in a similar fashion. This ambiguity needs to be eliminated in future versions of our grammar, but we are holding off on fixing it until we completely separate the adjectival and adverbial phenomena.

We plan on adding a rule that allows the base from of an adjective to be used adverbially and inflected forms to be parsed exclusively as adjectives.

#Adjective Agreement

Yiddish adjectives agree with the case, gender and number of the noun they modify. Agreement is marked via a suffix on the adjective. The following examples show adjectival agreement in the nominative case:

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {adjectives, agreement}
		Der guter man zingt.
		Der gut-er man zing-t.
		DEF.SG.NOM.M good-NOM.SG.M man sing;1SG.
		The good man sings.

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {adjectives, agreement}
		Di gute froy zingst.
		Di gut-e froy zing-st.
		DEF.SG.NOM.F good-NOM.SG.F woman sing-2P.
		The good woman sings.

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {adjectives, agreement}
		Dos gute kind zingt.
		Dos gut-e kind zing-t.
		DEF.PL.NOM.N good-NOM.SG.N child sing-3SG.
		The good child sings.
		
All possible grammatical and ungrammatical combinations of adjectives and nouns are illustrated in the lab four write up.  

Below is some general information on writing lexical rules. Please also refer to the lexical rules emitted by the customization system. Adjective agreement lexical rules should be of the "add only" type. Note that if you have an apparently uninflected form (i.e., a zero in the paradigm), you'll need to make sure it goes through a constant lexical rule (no spelling change) which fills in the relevant feature values.

We created rules for each type of inflection and adjective could receive. For specific genders we collapsed case-differentiated forms where they were harmonizable by under-specifying the associated rule. We collapsed all three plural forms into one rule since the inflection is identical for all cases and genders. These represent the matrix equivalent of position classes.

		sg-masc-nom-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE nom,
		                                            CONT.HOOK.INDEX.PNG [ GEND masc,
		                                                                  NUM sg ] ] ].

		sg-masc-acc-or-dat-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE non-nom,
		                                            CONT.HOOK.INDEX.PNG [ GEND masc,
		                                                                  NUM sg ] ] ].

		sg-fem-nom-or-acc-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE non-dat,
		                                            CONT.HOOK.INDEX.PNG [ GEND fem,
		                                                                  NUM sg ] ] ].

		sg-fem-dat-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE dat,
		                                            CONT.HOOK.INDEX.PNG [ GEND fem,
		                                                                  NUM sg ] ] ].

		sg-neut-nom-or-acc-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE non-dat,
		                                            CONT.HOOK.INDEX.PNG [ GEND neut,
		                                                                  NUM sg ] ] ].

		sg-neut-dat-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL [ CAT.HEAD.CASE dat,
		                                            CONT.HOOK.INDEX.PNG [ GEND neut,
		                                                                  NUM sg ] ] ].

		pl-adjective-lex := adjective-lex & infl-lex-rule & add-only-no-ccont-rule &
		  [ SYNSEM.LOCAL.CAT.HEAD.MOD.FIRST.LOCAL.CONT.HOOK.INDEX.PNG.NUM pl ].

Since the PNG constraints for our adjectives corresponds with the distribution of determiner-noun agreement for definite articles in Yiddish, we simply used the PNG constraints for `MOD` from `SPEC.FIRST` in the corresponding determiner rule that was generated by the matrix. We then added the following instances to `irules.tdl` which map onto instances of the aforementioned position classes.

		sg-masc-nom-adjective-suffix :=
		%suffix (* -er)
		sg-masc-nom-adjective-lex.

		sg-masc-acc-or-dat-adjective-suffix :=
		%suffix (* -n)
		sg-masc-acc-or-dat-adjective-lex.

		sg-fem-nom-or-acc-adjective-suffix :=
		%suffix (* -e)
		sg-fem-nom-or-acc-adjective-lex.

		sg-fem-dat-adjective-suffix :=
		%suffix (* -er)
		sg-fem-dat-adjective-lex.

		sg-neut-nom-or-acc-adjective-suffix :=
		%suffix (* -e)
		sg-neut-nom-or-acc-adjective-lex.

		sg-neut-dat-adjective-suffix :=
		%suffix (* -n)
		sg-neut-dat-adjective-lex.

		pl-adjective-suffix :=
		%suffix (* -e)
		pl-adjective-lex.
		
When we ran a small sample of both positive and negative testsuite examples from out suite to check if agreement was working, it seemed to indicate that our implementation was sound.
		

# Demonstratives and definiteness (and more)

## Pronouns

In order to model the cognitive status role of pronouns, we
constrained their INDEX values to be [COG-ST activ-or-more, SPECI + ].
This was done by creating a pronoun supertype that all of our pronouns
inherited from. This also served to clean up the names of the
pronouns, which were still rather nouny in their nomenclature.


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


## Definiteness on determiners

Yiddish marks definiteness through the form of the determiner, having
both definite and indefinite determiners:

# Case agreement on determiners.
Source: author based on a:224
Vetted: f
Judgment: g
Phenomena: {determiner,case,agreement}
Der lerer shraybt di zatsn.
Der lerer shrayb-t di zats-n.
DEF.NOM.M teacher write-1SG  DEF.PL.ACC.F sentence-PL.
The teacher writes the sentences.

Source: author
Vetted: f
Judgment: g
Phenomena: {agreement}
A man zingt.
A man zing-t.
INDF.SG man sing-1SG.
A man sings.


# Demonstrative determiner

In addition to the definite and indefinite determiners, there is also
the one demonstrative determiner yen-, which is inflected for person,
number, gender and case, and is translated as 'that'.

Source: a:248
Vetted: t
Judgment: g
Phenomena: {demonstratives}
Di mame git yenem kind yents bukh.
Di mame git yenem kind yents bukh.
DEF.NOM.F mother give;1SG DEM.DAT.N child DEM-ACC.N book.
The mother gives that child that book.

To implement the handling of determiners, we copied the
determiner-lex-supertype as instructed, and this was then inherited
from by three determiner subtypes which model the different kinds of
determiners in Yiddish: definite determiners (def-determiner-lex),
indefinite determiners (indef-determiner-lex) and demonstrative
determiner (dem-determiner-lex). The first two of these subtypes
inherit from the single-rel-lex-item type, as well as constraining the
value of the COG-ST value belonging to the noun the determiner is the
specifier of to be uniq+fam+act for definite determiners, and type-id for
indefinite determiners. 


def-determiner-lex := determiner-lex-supertype & single-rel-lex-item &
  [ SYNSEM.LOCAL.CAT.VAL.SPEC.FIRST.LOCAL.CONT.HOOK.INDEX.COG-ST uniq+fam+act ].

indef-determiner-lex := determiner-lex-supertype & single-rel-lex-item &
  [ SYNSEM.LOCAL.CAT.VAL.SPEC.FIRST.LOCAL.CONT.HOOK.INDEX [ COG-ST type-id,
                                                            PNG.NUM sg ] ].

We then created the type dem-determiner-lex so as to model the two relations
we are analyzing demonstratives as introducing determiners as per the
instructions, and our lexical entries for determiners then inherited
from these three subtypes as relevant for each entry.


dem-determiner-lex := determiner-lex-supertype &
[ SYNSEM [ LOCAL [ CAT.VAL.SPEC.FIRST.LOCAL.CONT.HOOK [ INDEX #index & [ COG-ST activ+fam ],
                                                        LTOP #ltop ],
                   CONT.RELS <! relation & [ PRED "exist_q_rel" ], 
                               #rel & arg1-ev-relation & [ LBL #ltop,
                                                           ARG1 #index ] !> ],
           LKEYS.ALTKEYREL #rel ] ].


Yiddish makes no mid-far distinction within demonstratives, so we
chose the following hierarchy for the demonstrative relation that
they introduce:
 

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


We had already added the various forms of yen- individually, as the
form does not follow any existing inflection patterns we have
modeled. All we had to do then was make the PRED feature of the
ALTKEYREL of every lexical entry for yen- take the value
distal+dem_a_rel.

After coding this all up, we looked at the indexed MRSs of various
sentences to see if this was all hooked up and working with the cog-st
and speci values as appropriate. Indeed, everything was working
correctly, except for dropped arguments in opt-comp phrases -- which
of course comes up in the next section.

Proximal demonstrative determiners in Yiddish are realized by placing
stress on the definite articles. Since this is not amenable to being
captured via the orthography, we will not handle this. However, there
is a particle/suffix form that modifies determiners to intensify
proximate determiners, which we can implement as a proximal
demonstrative:

Source: author
Vetted: f
Judgment: g
Phenomena: {demonstratives}
Di mame git dos-o kind dos bukh.
Di mame git dos-o kind dos bukh.
DEF.NOM.F mother give;1SG DEM.DAT.N-DEM child DEF.ACC.N book.
The mother gives this child the book.

Source: author
Vetted: f
Judgment: g
Phenomena: {demonstratives}
Di mame git ot dos kind yents bukh.
Di mame git ot dos kind dos bukh.
DEF.NOM.F mother DEM give;1SG DEM.DAT.N child DEF.ACC.N book.
The mother gives this child the book.

Source: author
Vetted: f
Judgment: g
Phenomena: {demonstratives}
Di mame git ot-o dos kind yents bukh.
Di mame git ot-o dos kind dos bukh.
DEF.NOM.F mother DEM-DEM give;1SG DEM.DAT.N child DEF.ACC.N book.
The mother gives this child the book.


This we can model, however we have not yet done so this week, and will
be on the priority number one list for the leftovers pile next week.
(Yes, this information should have been in the write up, two weeks ago!! 
We've added these examples to the test suite)


# Argument optionality

We firstly fixed up the one residual issue with the syntax of argument
optionality. This was the bug that was discussed on Go Post, where
OSV sentences with optionally transitive verbs were failing to parse. The fix,
discovered by Emily, was to remove the 'unexpressed' type constraint as follows:

nocons := nolist & cons &
  [ FIRST unexpressed & [ OPT - ],
    REST  nolist ].

Became:

nocons := nolist & cons &
  [ FIRST unexpressed & [ OPT - ],
    REST  nolist ].

In order to handle the semantics of the lexically licensed object
dropping in Yiddish, we took advantage of the OPT-CS feature, which,
through the magic of the Matrix machinery, populates the desired
COG-ST automatically, but only when the object has actually been
dropped. In order to do this we made the OPT-CS feature of the
complement of the verb take the value type-id, the value that we want
the COG-ST feature of the dropped noun to have.

trans-verb-lex := nom-acc-transitive-verb-lex & 
  [ SYNSEM.LOCAL.CAT.VAL.COMPS < [ OPT-CS type-id ] > ].

Checking this out with relevant parsing sentences shows that the
COG-ST of an unexpressed referent is indeed coming up with the value
type-id.


#Coverage

The modifier rules that we added to the grammar allowed us to cover sentences with adjectives and adverbs that we were not able to previously parse. Please view the example of spurious ambiguity provided under the adverb section.

		# adverb in wrong location
		Source: author based on a:223-224
		Vetted: t
		Judgment: u
		Phenomena: {adverb, word order}
		Der man shayn leyent dos bukh.
		Der man shayn leyen-t dos bukh.
		DEF.NOM.M man nicely read-1SG DEF.ACC.N book.
		The man reads the book nicely.

The above was an ungrammatical sentence since the adverb is in the second position yet it parsed. We believe this is because there is issue with the V2 module in the grammar matrix; we did not have time to analyze this further. This was not an isolated incident, there was one other example of this:
		
		Source: a:223-224
		Vetted: t
		Judgment: u
		Phenomena: {adverb, word order}
		Shayn der man leyent dos bukh.
		Shayn der man leyen-t dos bukh.
		nicely DEF.NOM.M man read-1SG DEF.ACC.N book.
		The man reads the book nicely.
		
When we compared to our baseline testsuite run, we noticed that our ambiguity increased per test item by .05 analyses. Under lexical, the new grammar increased by .20 in comparison with our baseline reflecting our handling of lexical items.
		
These are our coverage numbers:

		Total items:            213
		Positive items:         138
		Negative items:          13
		Parsing positive items:  53%
		Parsing negative items:  17.3%
		Coverage:                38.4%
		Overgeneration:          75%

#Test Corpus

We took the 10 sentences for our corpus from the Corpus of Modern
Yiddish (http://www.corpustechnologies.com:8080/YNC/search/). The
sentences we chose came from a variety of sources, including religious
texts, fiction, and newswire. We had to first overcome some
limitations in the search interface of the corpus which did not allow
us to sort by item length but through a workaround we were able to
extract short sentences. Unfortunately we have not yet had time to put
items from this corpus into the lexicon. Accordingly, we do not yet
parse any of these sentences.

We constructed the corpus in the same way that we made our test suite,
using the make item script and using the translit-seg line as the
target for the output item file. We were also able to use the glosses
provided by the corpus to help us perform the gloss annotation. We
plan to add unknown lexical items from this corpus into our lexicon
next week, as well as look for more items to increase the size of the
corpus.

Some examples:

# Forvets 2008.06.27
Source: a
Vetted: f
Judgment: a
Phenomena: {}
di mayse iz ober a sakh komplitsirter
di mayse iz ober a sakh komplitsirter
DEF.NOM.F story be;3SG but INDF.SG many complicate;PP.

# Forverts 2009.03.13
Source: a
Vetted: f
Judgment: a
Phenomena: {}
vos tuen di iberike fertsn toyznt yidn?
vos tu-en di iberik-e fertsn toyznt yid-n?
what do-3PL DEF.NOM.F extra-FSG 14 thousand Jew-PL?

#Tanakh: Shoftim, Yehoyesh
Source: a
Vetted: f
Judgment: a
Phenomena: {}
hot im zayn yung dershtokh-n, un er iz geshtorbn
hot im zayn yung dershtokh-n, un er iz ge-shtorb-n
have;2PL\3SG ACC.3P.M POSS young stab_to_death-PP and NOM.3P.M be;3SG PP-die-PP.
