[lkb] Coordination in LKB

[Nidhi Sharma]

Dear All

I am a research scholar, working on Agreement and Coordination in Hindi. I
have looked at some interesting agreement patterns and I am trying to
impement them using LKB.  I am struggling to get the facts implemented and
going through a difficult stage in my research. I look forward to your
expertise in the field and would be really grateful if you provide some
support and directions. I understand that my mail would require some of your
precious time however, I will be really indebted to you if you could help me
as you all are experts in the field and I'm sure your guidance will get me
through my work.

First, I would like to summarise the coordination facts in Hindi below :

Hindi Coordinate noun phrases differ in their agreement strategies between
the predicate and its argument depending on the nature of the conjunct head
noun i.e. animate/inanimate
With regard to agreement between a predicate and coordinate noun phrase,
although Hindi exhibits the widely attested “resolution agreement strategy”,
it also uses a less familiar “Closest Conjunct Agreement (CCA)”. There are
basically three different agreement patterns listed below:

When two animate nouns are conjoined, the verb takes the masculine default
form is any one of the conjuncts is masculine
When two animate nouns are conjoined, the verb take the feminine plural form
is both the conjuncts are feminine.
When two nonanimate nouns are combined, the verb agrees with the rightmost
conjunct in both its number and gender features.
 Also, agreement between a pronominal adjective and a Coordinate noun phrase
is always CCA for both number and gender.

Next, I propose the following analyses

HPSG based analyses (to be implemented in LKB)

ANALYSIS I

This analysis is a modification of the one proposed in

Villavicencio, Aline, Louisa Sadler, and Doug Arnold. 2005. An HPSG Account
of Closest Conjunct Agreement in NP Coordination in Portuguese. In Stefan
M¨uller (ed.), Proceedings of the HPSG05 Conference,Stanford, CA. CSLI
Publications: http://www-csli.stanford.edu/publications.

This analysis involves the introduction of new features LAGR and RAGR which
contain agreement information about the left and right conjunct. So, all
lexical nouns, adjectives and determiners have the feature LAGR and RAGR in
addition to CONCORD. Also, in a non coordinate structure LAGR, RAGR and
CONCORD will have the same value. The values of LAGR and RAGR will travel up
to the mother i.e. coordinate NP, and the CONCORD will be structure shared
with RAGR in case of CCA and it gets a default masc pl in case of resolved
agreement. Also, when both LAGR and RAGR are fempl (feminine plural) at noun
level, the NP has a CONCORD fempl.

I propose to implement the same with following steps:

Steps:

1. First, we propose to introduce two subtypes of nominal-coordinate-phrase
(which is itself a subtype of coordinate phrase): one for animate coordinate
phrases i.e. animate-n-coord-ph  and one for nonanimate coordinate phrases
i.e. nonanimate-n-coord-ph . As resolved gender agreement is seen only in
the case of animate coordinate phrases, we introduce two subtypes of
animate-n-coord-ph i.e. animate-n-coord-ph-m for phrases that resolve to
masculine and animate-n-coord-ph-f that resolve to feminine.
 The above subtypes will satisfy the constraint that

a coordinate structure is feminine if all the conjuncts are feminine and it
is masculine, if a single conjunct is masculine.

2. Then, we introduce the features LAGR and RAGR, which store agreement
information about the leftmost and rightmost conjuncts. LAGR contains
information about the leftmost conjunct and RAGR contains information about
the rightmost conjunct. The feature CONCORD is used to contain resolved
agreement information.

So, any lexical noun, adjective or determiner will have these head features
and their values will be passed up.

We will mention both NUM and GEND values for LAGR, RAGR and CONCORD as the
rules need to look into the actual values

3. We will have to constrain NUM value of the CONCORD to be plural for
animate-n-coord-ph

4. We need to constrain the rules in such a manner, that if any of the
conjuncts has GEND masculine, the CONCORD| GENDER for the coordinate phrase
is masculine and if both the conjuncts are GEND fem, the CONCORD value for
Coordinate phrase is CONCORD|GEND feminine.

5. For animate nouns, we will have to define both NUM and GEND values for
LAGR, RAGR and CONCORD. Then a rule needs to be written will require the
CONCORD|NUM and CONCORD| GEND for the coordinate phrase to be the same as
(shared with) RAGR| NUM and RAGR| GEND.

6. For agreement between adjectives and nouns, I suppose CONCORD|NUM and
CONCORD| GEND will be shared with the values of RAGR.

 ANALYSIS II

This analysis is focused more on implementing using the LKB system

The Steps to be followed are:

I will create a new HEAD feature FORM with its values as animate and
inanimate.  This feature will be defined for nouns.
I want to rewrite the NP coordination rules in the following manner
 RULE 1
NP coordination rule for inanimate nouns

It basically states that the AGR (Agreement) value of the coordinate NP is
token identical with that of the right conjunct

 NP [ AGR  #agr1 ] ------à   N  [AGR  # agr1 ]   and   N [  AGR    agr2]

(Coordinate np)                    (Left conjunct)                    (Right
conjunct)

The feature AGR gives the NUM(BER) and GEN(DER) information.

The constraint states that when two (inanimate nouns) are combined; the AGR
value of the NP comes from the right conjunct.

RULE 2: NP coordination rule for animate nouns -1

It basically states that if either of the conjuncts is GENDER masculine, the
coordinate NP gets the AGR [masc plural].

NP   [  AGR  [  NUM    pl                       N [AGR[  GEND  masc] ]   &
N[AGR   agr]
                          GEND   masc]  ]    à


                                                      OR

NP   [  AGR  [  NUM    pl                       N [AGR[  agr] ]   &
N[AGR[GEND  masc]
                          GEND   masc]  ]    à


So, if either of the conjuncts is masculine, the NP is by default masculine
plural. NUM can be left unspecified for the conjunct nouns.

RULE 3: NP coordination rule for animate nouns -2

It basically states that if both the conjuncts are GENDER feminine, the
coordinate NP gets the AGR [feminine plural].

NP   [ AGR [  NUM    pl                  N [AGR[GEND  fem] ]  & N[AGR[GEND
fem  ]
                       GEND   fem]  ]    à
Also, the rules will have to be constrained so that they combine only nouns
with the following  features :

SYNSEM. LOCAL. CAT. HEAD [ FORM  animate]

Or

SYNSEM. LOCAL. CAT. HEAD [ FORM  inanimate]

So far we have not looked at a combination of animate and inanimate nouns.

My Questions:

1) In order to implement the first analysis I would have to define LAGR,
RAGR and CONCORD as subtypes of INDEX, or then specify NUM and GEND features
on all of them. Later changes need to be made to the rules.

I’m afraid this would lead to redundancy and also, does the customization
page (the system) allow us to create subtypes of INDEX.

2)  The second analysis requires constraining the rules as stated.  When
writing phrasal rules, can we write values instead of just sharing of the
values because our data needs the rules to look into the NUM and GEND value
and not just structure sharing. How do I constrain animacy/nonanimacy in the
NP coordination rule?

Please share your knowlege and expertise in the field and suggest me with
some directions and/or solutions to the above problems.

Your help is greatly appreciated

Thank you in advance

Awaiting your response

Regards

Nidhi
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.delph-in.net/archives/lkb/attachments/20100820/ff85978b/attachment.html>