#Lab 2: Matrix Customization

##Pronouns

Yiddish has has four major pronoun paradigms: personal, reflexive,
interrogative, and relative. Since all of the paradigms besides the
personal require imposing semantic constraints, we focused primarily
on implementing the personal pronouns for lab one since the the Matrix
Customization system is well-suited for this task.


###Figure 1: Personal Pronouns Grouped by by Case, Person, Number & Gender

+-----------+-------------------+-------------------+-------------------------------------+
|           |1st Person         |2nd Person         |3rd Person                           |
+-----------+-------------------+-------------------+-------------------------------------+
|           |Singular    Plural |Singular    Plural |Masculine  Neuter   Feminine  Plural |
+-----------+-------------------+-------------------+-------------------------------------+
|           |                   |                   |                                     |
|Nominative |ikh         mir    |du          ir     |er         es       zi        zey    |
|           |                   |                   |                                     |
|           |                   |                   |                                     |
|Accusative |mikh        undz   |dikh        aykh   |im         es       zi        zey    |
|           |                   |                   |                                     |
|           |                   |                   |                                     |
|Dative     |mir         undz   |dir         aykh   |im         im       ir        zey    |
|           |                   |                   |                                     |
+-----------+-------------------+-------------------+-------------------------------------+

The Yiddish personal pronouns are marked for person, case, number. The
third person personal pronouns are also marked for gender. These are
some examples of pronouns indicating case, person, and number:
	
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Er zet.
		Er ze-t.
		3SG.NOM.M see-3SG.
		He sees.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zi zet.
		Zi ze-t.
		3SG.NOM.F see-3SG.
		She sees.
		
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Es zet.
		Es ze-t.
		3SG.NOM.N ze-3SG.
		It sees.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zi rufnst dikh.
		Zi rufn-t dikh.
		DEF.3SG.F.NOM call-3SG 2SG.ACC.
		She calls you.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zi runft im.
		Zi runf-t im.
		DEF.3SG.F.NOM call-3SG 3SG.ACC.
		She calls him.


In the matrix we implemented this with a nonhierarchical lex-type
specification. There are some lexemic redundancies that we tried to
address by creating under-specified types. We under-specified them to
capture two usages of a lexeme that disagreed on one morphosyntactic
feature but we did not do this for when they diagreed for multiple
morphosyntactic features since that would be too comlex and would not
encha. For example, 'im' is a lexical entry under the type
`3p-acc-or-dat-sg-masc-pron` which captures but the accusative and
dative usage of the lexeme for the masculine gender. However the
lexeme is also used in the accusative case for the neuter gender so it
has an additional lexical entry under the the type
`3p-dat-sg-neut-pron`.

The cases lexeme usages we were able to capture into a single type
usually involved case-leveling or gender-leveling. In the first and
second person plural pronouns, we see accusative-dative leveling,
since in both cases 'undz' and 'aykh' are used respectively. In their
plural form, the third person pronouns show case-leveling for the
accusative and dative and gender-leveling so the homogenous 'zey' is
used. The third person masculine pronouns also show case-leveling for
the accusative and dative because the pronoun 'im' refers to both. The
redundancies other than 'mir' we could not capture were 'mir' which in
the first person can either refer to the singular dative or the plural
nominative and 'ir' which in the second person refers to the plural
nominative and in the third person refers to the dative singular
feminine.

Yiddish third person pronouns also indicate gender. So, the pronouns
'er' and 'zi' can be assigned to a male or female person as well as
inanimate objects that are masculine or feminine; if these do not
apply, the noun is neutral and receives an 'es'. Some examples of this
include:

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Es zet.
		Es ze-t.
		3SG.NOM.N ze-3SG.
		It sees.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Er zet.
		Er ze-t.
		3SG.NOM.M see-3SG.
		He sees.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zi zet.
		Zi ze-t.
		3SG.NOM.F see-3SG.
		She sees.

##Rest of the Noun Phrase

The full noun phrase in Yiddish comprises at the very least a HEAD,
which in most instances is a noun. The noun phrase may include a
determiner and modifiers which precede the `HEAD` but these are not
required members of the noun phrase. The order of elements in the noun
phrase is mostly fixed to `DET Adj N`; an example of this would be 'Di
gute frage' which translates to 'The good question'. Element order in
the NP can be illustrated using a tree structure such that at each
element appears to the left of the element it is modifying:

###Figure 2: Yiddish NP Word Order

     a                                    all
     +                                    +
     |                                    |
     |                                    |
     +-----> good                         +-------->your
              +                                     +
              |                                     |
              |                                     |
              +-----> question                      +------->good
                                                              +
                                                              |
                                                              |
                                                              +------->questions
                                                              
We focused on only the determiner and noun elements of the noun phrase
since we aren't required to implement adjectives in Lab 2. However,
the adjective element is germane to determiner-noun optionality. The
determiner for most Yiddish nouns is optional unless the noun is a
proper noun which makes a determiner impossible in the base case. If
the proper noun is modified by an adjective though, the determiner
becomes optional for a proper noun. We implemented this by creating
two noun types in our matrix that capture both the optional and
impossible case, however we haven't modeled the case in which a proper
noun takes a determiner only in the presence of an adjective. The
following are some examples of determiner noun-ordering and
optionality extracted from our test-suite:

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Di mener zingen.
		Di mener zing-en.
		DEF.3PL.NOM men-3PL.NOM sing-3PL.
		The men sing.
		
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Mener zingen.
		Mener zing-en.
		Men-3PL.NOM sing-3PL.
		Men sing.
		
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {agreement}
		Mener di zingen.
		Mener di zing-en.
		Men-3PL.NOM DEF.3PL.NOM sing-3PL.
		The men sing.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		A koyekhdike Daytshland iz groys.
		A koyekhdik-e Daytshland iz groys.
		3SG.INDF powerful-DEF.NOM.M Germany be;3SG large.
		A powerful Germany is large.
		
				
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Daytshland iz groys.
		Daytshland iz groys.
		Germany be;3SG large.
		Germany is large.
		

Determiners in Yiddish can be definite or indefinite and vary based on
case, gender and number. The indefinite articles 'a' and 'an' act like
their English counterparts. Below is a table of Yiddish definite
determiners.

###Figure 3: Determiners Grouped by Case, Gender and Number

   +----------+-----------------------------------------+
   |          |  Masculine   Neuter   Feminine   Plural |
   |----------|-----------------------------------------|
   |Nominative|  der	      dos	  di         di     |
   |          |                                         |
   |Accusative|  dem	      dos	  di	     di     |
   |          |                                         |
   |Dative	  |  dem	      dem	  der	     di     |
   |          |                                         |
   +----------+-----------------------------------------+

Determiners must agree with their nouns in case, number and
gender. Here are a few examples from our test suite:
		
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {agreement}
		Di man zingt.
		Di man zing-t.
		DEF.3SG.NOM.F man sing-3SG.
		The man sings.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Di froy zingt.
		Di froy zing-t.
		DEF.3SG.NOM.F woman sing-3SG.
		The woman sings.
		
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {agreement}
		Di froy zingst.
		Di froy zing-st.
		DEF.3PL.NOM.F woman sing-3SG.
		The woman sings.
		
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {agreement}
		Dem froy zingst.
		Dem froy zing-st.
		DEF.3PL.ACC.M woman sing-3SG.
		The woman sings.

Pluralization patterns for nouns in Yiddish are hard to formalize. The
default pattern for pluralizing a noun is suffixing it with '-s' for
nouns that end in an unstressed 'r', 'm', 'n' or a vowel. The second
largest class of nouns take can be pluralized with the suffix '-n'
besides which there is a very large class of irregularly pluralized
nouns that can be suffixed with '-es' and '-er' among many other
forms.

Our test suite also covers adjectives, auxiliaries, and adjective-noun
agreement but we haven't incorporated them into our grammar yet.


##Word Order

Yiddish has V2 word order for declarative matrix clauses. This means
that the finite verb is always the second constituent. The unmarked
ordering of constituents is SVO, however Yiddish makes strong usage of
topicalization, meaning that most constituents can be fronted to
become the topic. When this happens, the finite verb remains in second
position, as per the V2 constraint.

The following examples illustrate grammatical and ungrammatical word
order for intransitive and unmarked transitive sentences:

		Source: author based on a:246.
		Vetted: t
		Judgment: g
		Phenomena: {word order}
		Dos kind zingt.
		Dos kind zing-t.
		DEF.ACC.N child sing-1SG.
		The child sings.
		
		Source: author based on a:246.
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Zingt dos kind.
		Zing-t dos kind.
		sing-1SG DEF.SG.ACC.N child.
		The child sings.
		
		Source: a:223-224
		Vetted: t
		Judgment: g
		Phenomena: {word order}
		Der man leyent dos bukh.
		Der man leyen-t dos bukh.
		DEF.NOM.M man read-1SG DEF.ACC.N book.
		The man reads the book.
		
		Source: author based on a:224
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Leyent dos bukh der man.
		Leyen-t dos bukh der man.
		read-1SG DEF.ACC.N book DEF.NOM.M man.
		The man reads the book.
		
		Source: author based on a:224
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Dos bukh der man leyent.
		Dos bukh der man leyen-t.
		DEF.ACC.N book DEF.NOM.M man read-1SG.
		The man reads the book.

The following example shows topicalization of the object in a
transitive sentence:

		Source: a:224
		Vetted: t
		Judgment: g
		Phenomena: {word order}
		Dos bukh leyent der man.
		Dos bukh leyen-t der man.
		DEF.ACC.N book read-1SG DEF.NOM.M man.
		It's the book that the man is reading. 

The following examples show grammatical and ungrammatical placements
of the verb in ditransitive sentences:

		Source: a:248
		Vetted: t
		Judgment: g
		Phenomena: {word order}
		Di mame git dem kind dos bukh.
		Di mame git dem kind dos bukh.
		DEF.NOM.F mother give;1SG DEF.DAT.N child DEF.ACC.N book.
		The mother gives the child the book.
		
		Source: author based on a:248
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Git di mame dem kind dos bukh.
		Git di mame dem kind dos bukh.
		give:1SG DEF.NOM.F mother DEF.DAT.N child DEF.ACC.N book.
		The mother gives the child the book.
		
		Source: author based on a:248
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Di mame dem kind git dos bukh.
		Di mame dem kind git dos bukh.
		DEF.NOM.F mother DEF.DAT.N child give;1SG DEF.ACC.N book.
		The mother gives the child the book.
		
		Source: author based on a:248
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Di mame dem kind dos bukh git.
		Di mame dem kind dos bukh git.
		DEF.NOM.F mother DEF.DAT.N child DEF.ACC.N book give;1SG.
		The mother gives the child the book.

It is also possible to topicalize both direct and indirect objects of
ditransitive verbs. This is discussed later on in the case section.

In order to implement this, we simply selected the V2 option. Yiddish
differs from other characteristically V2 languages (notably German) in
that infinitive verb forms are not generally in final position and
embedded clauses retain V2 word order. Depending on whether these
constraints have been implemented into the Grammar Matrix's handling
of V2 word order, it will be interesting to see how well it supports
Yiddish word order.


##Case

Yiddish uses a nominative-accusative case system, and has three cases
in total: nominative, accusative and dative. There is no genitive
case, however pronouns are marked for possession and a limited class
of nouns have inflectional morphology marking possession. Case is
expressed on determiners, pronouns, adjectives and a very limited
class of nouns that include kinship terms.

The following examples show case being expressed on pronouns and
determiners. The full grammatical and ungrammatical permutations can
be found in the Pronoun section and The Rest of the NP section.

		Source: a:246
		Vetted: t
		Judgment: g
		Phenomena: {case}
		Du zingst.
		Du zing-st.
		2SG.NOM sing-2SG.
		You sing.
		
		Source: author
		Vetted: t
		Judgment: g
		Phenomena: {case}
		Mir leyent es.
		Mir leyen-t es.
		1PL.NOM read-1SG 3SG.ACC.N.
		We read it.
		
		Source: a:223-224
		Vetted: t
		Judgment: g
		Phenomena: {case}
		Der man leyent dos bukh.
		Der man leyen-t dos bukh.
		DEF.SG.NOM.M man read-1SG DEF.SG.ACC.N book.
		The man reads the book.
		
		Source: author based on a:224
		Vetted: f
		Judgment: g
		Phenomena: {case}
		Der lerer shrajbt di zacn.
		Der lerer shrajb-t di zac-n.
		DEF.NOM.M teacher write-1SG  DEF.PL.ACC.F sentence-PL.
		The teacher writes the sentences.

The idiosyncratic case marking on nouns works as follows: personal
names, close kinship terms and a few particular words such as 'jid'
(Jew), 'mentsch' (person) and 'rebe' (rabbi) can take the suffix -n
the accusative and dative cases. 

		# Accusative case agreement on proper name rivke.
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {case}
		Dos kind zent rivken.
		Dos kind zen-t rivke-n.
		DEF.NOM.N child zen-3SG Rivke-ACC. 
		The child sees Rivke.
		
		# Dative Case agreement on mentsch
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {case}
		Der mentsch git dem mentschn dem epl.
		Der mentsch git dem mentsch-n dem epl.
		DEF.NOM.M person give;1SG DEF.ACC.M person-DAT DEF.ACC.M apple.
		The person gives the person the apple.
		
		# Dative Case agreement on mame
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {case}
		Di mame git der mamen dem epl.
		Di mame git der mame-n dem epl.
		DEF.NOM.F mother give;1SG DEF.DAT.F mother-DAT DEF.ACC.M apple.
		The mother gives the mother the apple.
		
		# Incorrect use of dative case on mame
		Source: author
		Vetted: f
		Judgment: u
		Phenomena: {case}
		Dos kind zent di mamen.
		Dos kind zen-t di mame-n.
		DEF.NOM.N child zen-3SG DEF.ACC.F mame-DAT.
		The child sees the mother.

Since topicalization is possible with almost any constituent, there
are less constraints upon where NPs with specific case may be placed
-- certainly none for subject position and none for object position in
transitive verbs. One context where constraints can be found is in
verbs with multiple complements. In unmarked clauses, the order of
verbal complements is indirect object then direct object. This means
that in ditransitive verbs, a full dative NP comes before a full
accusative NP. We are currently not completely sure whether this is a
soft or hard constraint, but for now will assume it to be hard. The
following is therefore ungrammatical:

		Source: author based on a:248
		Vetted: f
		Judgment: u
		Phenomena: {word order}
		Di mame git dos bukh dem kind.
		Di mame git dos bukh dem kind.
		DEF.SG.NOM.F mother give;1SG DEF.SG.ACC.N book DEF.SG.DAT.N child.
		The mother gives the child the book.

This particular constraint currently remains unimplemented as it is
too fine-grained for the Grammar Matrix to handle. It should also be
noted that the above applies to full NPs, and when one or more
pronouns are involved, the constraints on ordering change. This is
outside the scope of this current report though.

It is possible for both direct and indirect objects of a ditransitive
verb to be topicalized, and since there are constraints on the
ordering of verb complements in unmarked clauses, there are likely to
be such constraints on the ordering of the nominative argument (now to
the right of the verb) and the other complement of the verb, however
we have not yet located data to determine what these constraints are,
if any.

Case was implemented as a feature that we applied to the relevant
pronoun and determiner forms. Possession is currently unhandled but
will be implemented as a lexical rule that requires the -s noun affix
and applies a Boolean valued POSS feature, this way allowing the
constraint to be applied alongside case constraints.

We implemented the idiosyncratic case marking on nouns, by creating
two separate lexical rules, one for feminine nouns and the other for
masculine nouns. Both require the -n affix and in addition, apply only
to lexical types setup specifically for nouns that can have case
marking, with one inheriting from feminine nouns and one from
masculine nouns. The lexical rule that applies to masculine nouns
stipulates the constraint that the case is either accusative or
dative, and for feminine nouns, just that the case is dative.


##Verb Morphology

In Yiddish, verbs are inflected for person and number. The full
paradigm is indicated below:

		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Ikh zen.
		Ikh zen.
		1SG.NOM see;1SG.
		I see.
		
		Source: author
		Vetted: f
		Judgement: g
		Phenomena: {agreement}
		Mir zenen.
		Mir zen-en.
		1PL.NOM zen-1PL.
		We see.
		
		Source: author
		Vetted: f
		Judgement: g
		Phenomena: {agreement}
		Mir zingn.
		Mir zing-n.
		1PL.NOM sing-1PL.
		We sing.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Du zenst.
		Du zen-st.
		2SG.NOM see-2SG.
		You see.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Ir zenst.
		Ir zen-t.
		2PL.NOM see-2SG.
		You see.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Er zent.
		Er zen-t.
		3SG.NOM.M see-3SG.
		He sees.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zey zenen.
		Zey zen-en.
		3PL.ACC see-3PL.
		They see.
		
		Source: author
		Vetted: f
		Judgment: g
		Phenomena: {agreement}
		Zey zingn.
		Zey zing-n.
		3PL.ACC sing-3PL.
		They sing.

We implemented this via a lexical rule which takes the relevant affix
and checks for the appropriate agreement in person and number on the
subject.
 
##Test Suite Coverage

Total Items: 79
Positive: 53
Negative: 26
Parsed Positives: 18
Parsed Negatives: 0
Coverage: 34%
Over-generation: 0%

Our test suite included many items that make use of auxiliaries,
adjectives, and copula which account for nearly all of our unparsed
grammatical examples. We had no over-generation and the tsdb did not
obtain parses for our ungrammatical items. One bug that quickly became
apparent from looking at parse trees is that the singular noun lex
rule (which does not take an affix) is being incorrectly applied to
pronouns as the lexical rule applies to all nouns. Even though this
doesn't currently result in any overgeneration, this should be fixed,
and we plan to do this by creating a basic noun lex type that all
nouns inherit from, which the number lexical rules can then.



##Unimplemented Properties

Please see the notes interleaved in the documentation of our matrix
customization choices.

##Matrix Customization System Improvements

For V2 language order the matrix assumes that word order is free aside
from the second verb position constraint. This not true for Yiddish
which places some constraints on the relative ordering of verbal
complements, with respect to their case, for verbs that have more than
one complement. This kind of fine-grained constraint is not available
to us.
