<!DOCTYPE article PUBLIC "-//NLM//DTD JATS (Z39.96) Journal Archiving and Interchange DTD v1.0 20120330//EN" "JATS-archivearticle1.dtd">
<article xmlns:xlink="http://www.w3.org/1999/xlink">
  <front>
    <journal-meta />
    <article-meta>
      <title-group>
        <article-title>Dynamic Social Choice for Anaphora Resolution</article-title>
      </title-group>
      <contrib-group>
        <contrib contrib-type="author">
          <string-name>Sumiyo Nishiguchi</string-name>
          <email>nishiguchi@rs.tus.ac.jp</email>
          <xref ref-type="aff" rid="aff0">0</xref>
        </contrib>
        <aff id="aff0">
          <label>0</label>
          <institution>Department of Liberal Arts, Faculty of Science Division I, Tokyo University of Science 1-3 Kagurazaka, Shinjuku</institution>
          ,
          <addr-line>Tokyo 162-8601</addr-line>
          <country country="JP">Japan</country>
        </aff>
      </contrib-group>
      <abstract>
        <p>Disambiguation of pronoun reference has been an important issue for both theoretical and computational linguists. While linguistic theories on binding conditions eliminate impossible readings to a certain extent, many inter-sentential anaphora remain ambiguous. Nishiguchi (2011, 2012a,b, 2014, 2016a,b) consider pronoun resolution as a social choice among discourse participants which obeys Arrow's Impossibility Theorem (Arrow 1963). This paper further discusses discourse update of Social Welfare Function which provides updated variable assignment.</p>
      </abstract>
    </article-meta>
  </front>
  <body>
    <sec id="sec-1">
      <title>-</title>
      <p>(1) Frances: ...Not while Emma’s not here. You
know Emma
Billy: Mm.</p>
      <p>Frances: she’s, she was walking with Lisa
and I weren’t there and her Mum sh– jus– ,
like she muc– , she mucks about a lot and
she told Leigh that if he don’t serve her he’s
gonna die, she’s gonna punch him right!
However, proximity does not always resolve
referential ambiguity of pronouns. Him in (2a)
unambiguously means someone other than the
closest John—some discourse-salient entity. In (2b),
the pronoun is ambiguous.</p>
      <p>b. Johni said hefi=jg likes himselffi=jg.</p>
      <p>
        In linguistic binding theory
        <xref ref-type="bibr" rid="ref15 ref2">(Chomsky 1981,
Reinhart 1983)</xref>
        , antecedents are called binders,
which bind bindees that are anaphoric pronouns,
e.g., him or himself. Condition B is that
pronouns must be free in their local domain,
meaning that they are not bound by the antecedent by
means of coindexing and c-commanding relation.
C-command is roughly equivalent to precedence,
with some restrictions.
      </p>
      <p>However, (3) is ambiguous in four ways and can
have either one of the following interpretations: i)
John broke John’s leg, ii) John broke Bill’s leg, iii)
Bill broke Bill’s leg, or iv) Bill broke John’s leg.
He and his can be bound by either John or Bill.
The binding theories have no way of
disambiguating these pronouns since there is no way of
knowing speaker intention. Proximity does not predict
the different readings in (3) either.
(3) Anna: Billj is a good goalkeeper.</p>
      <p>Kim: Johni said hei=j broke hisi=j leg
recently.
1</p>
    </sec>
    <sec id="sec-2">
      <title>Social Choice Theory</title>
      <p>
        Although Social Choice Theory
        <xref ref-type="bibr" rid="ref1 ref17 ref4 ref7">(Arrow 1963,
Moulin 1988, Taylor 2005, Gaertner 2009)</xref>
        has
only been briefly mentioned in
        <xref ref-type="bibr" rid="ref18">van Rooij (2011</xref>
        ) in
relation with interadjective comparison, Arrow’s
Impossibility Theorem is obeyed in a social choice
of pronominal reference. Typically, social choice
theory explains collective decision making in case
of voting and has solved the problems with
majority decision. Preferences are ordering between
alternatives and should satisfy the following
axioms. When R stands for a knowledge of all pairs
and x, y and z for alternatives,
Axiom 1. For all x and y, either xRy or yRx.
Axiom 2. For all x, y, and z, xRy and yRz imply
xRz.
      </p>
      <p>
        Axiom 1 states that the relation R is
connected—every candidate is related to each
other. Relations that satisfy Axiom 2 are
transitive. In (4), N, a finite set of individuals or voters,
consists of five individuals and , a nonempty set
of alternatives or candidates, has three members.
Let L( ) denote the set of all linear orders on
. A profile R is a vector of linear orders, or
preferences. Ri is a vector of preferences of an
individual i. NxR&gt;y denotes the set of individuals
that prefer the candidate x to y. Supposing R the
profile given in this model, NoR&gt;c is a set of people
who prefers Obama to Clinton, that are, Anna,
Heather and George
        <xref ref-type="bibr" rid="ref3">(cf. Endriss 2016)</xref>
        .
(4) a. N = fa, k, h, g, ng
e. SWF F: L( )N ! L( )
      </p>
      <p>A social welfare function (SWF) F is a function
which takes individual’s preferences and returns
collective preference. Arrow demonstrated that
any SWF for three or more alternatives the
following conditions must be a dictatorship. Condition 2
states that the relative ranking of two candidates
remains unchanged regardless of other candidates.
Theorem 1 (General Possibility Theorem
(Impossibility Theorem)). If there are at least three
alternatives which the members of the society are free
to order in any way, then every social welfare
function satisfying Conditions 1 and 2 and yielding a
social ordering satisfying Axioms 1 and 2 must be
either imposed or dictatorial.</p>
      <p>Condition 1 (Pareto condition). A SWF F satisfies
the Pareto condition if, whenever all individuals
rank x above y, then so does society: NxR&gt;y = N
implies xF(R)y
Condition 2 (Independence of irrelevant
alternatives (IIA)). A SWF F satisfies IIA if the relative
social ranking of two alternatives only depends on
their relative individual rankings: NxR&gt;y = NxR&gt;0y
implies xF(R)y , xF(R’)y
Condition 3 (Nondictatorship). There is no
individual i such that for every element in the domain
of rule f, 8x, y 2 X: xPiy ! xPy (Sen 1979)</p>
    </sec>
    <sec id="sec-3">
      <title>Application to Pronoun Resolution</title>
      <p>SWF for pronoun resolution satisfies Arrow’s
Impossibility Theorem, or General Possibility
Theorem, by satisfying Axioms 1, 2, Pareto
Condition and IIA but demonstrating dictatorship.
Pronoun resolution is compared with voting by
multiple voters, discourse participants. The candidates
or choices would be different interpretation of the
sentence. In (5), the referent of he is
ambiguous. Chris meant he to be Bob, while Naomi
interpreted him to be John. As the disagreement on
pronominal reference is consolidated in the
discourse, pronoun resolution is certainly a social
choice and Social Choice Function (SCF) decides
the antecedent.
(5) Chris: John said he broke his leg.</p>
      <p>Naomi: Did he? John looked fine when I saw
him this morning.</p>
      <p>Chris: It is Bob who broke his leg.</p>
      <p>Naomi: I thought you were talking about
John.</p>
      <p>When individuals I = fc, ng, candidates = fj,
bg, Chris and Naomi’s ordering is jRcb ^ bRnj,
denote the set of linear orders on by L( ).
Preferences (or ballots) are taken to be elements of
L( ). A profile R 2 L( )I is a vector of
preferences. SCF or voting rule is a function F : L( )I
! 2 n; mapping a given profile to a nonempty
set of winners; e.g., a singleton set fbg for (5).
SWF is a function F : L( )I ! L( ) mapping any
given profile to a (single) collective preference
order. Although the preferences between the
candidates vary between the individuals, SWF returns
a single preference order and ambiguities are
resolved during the conversation.</p>
      <p>There are three possible antecedents for she in
(1)—Emma (e), Lisa (l) and Lisa’s mother (m).
Let us say that Billy (b) prefers e to l, and also l
to m to be the antecedent. On the other hand, the
speaker Francis (f) prefers m to l, and l to e
according to the proximity. All three candidates are
ordered in accordance with Axiom 1, i.e., eRbl ^
lRbm and mRf l ^ lRf e. Transitivity also holds
for pronoun antecedent preferences. Each of them
implies eRblRbm and mRf lRf e. SWF for pronoun
resolution also meets Pareto condition. When the
interpretation of the addressees agrees with the
one of the speaker, the decision of the society
follows. It is unlikely that pronouns refer to someone
else other than speaker’s intention and hearer’s
interpretation. A SWF F satisfies IIA if the relative
social ranking of two alternatives only depends
on their relative individual rankings. Let us say
that the preference relations are denoted by R and
R’. Assume that IIA does not hold and consider
a dialogue in (7) where the relative rankings
between Bob and John is affected by irrelevant
candidate Victor’s ranking. The social decision differs
from the relative ranking between John and Bob of
speaker and hearer, which does not happen, in (8).
(6) Chris: Bob is a good skier. But John said he
broke his leg.</p>
      <p>Naomi: Did he? Poor Bob!
(7) Chris: Victor is a good skier and so is Bob.</p>
      <p>But John said he broke his leg.</p>
      <p>Naomi: Did he? Poor Bob!
(8) bR’cvR’cj ^ bR’njR’nv 6! jF(R)b</p>
      <p>Then, NbR&gt;j = NbR&gt;0 j implies bF(R)j , bF(R’)j
The speaker’s decision on pronominal reference
dictates the social preference. Even when there
is disagreement or misunderstanding, the speaker
corrects unifies interpretation in general, as in (9).
Pronoun resolution is dominated, or dictated, by
the speaker’s meaning.
(9) Chris: Bob is a good skier. But John said he
broke his leg.</p>
      <p>Naomi: Did he? Poor Bob!</p>
      <p>Chris: No. I mean John broke his leg.
(10) xPcy ! xPy
Proof. Suppose: xPcy ! xPy, that is, xPcy !
yRx, where R is weak preference. However, the
dialogue normally proceeds jPcb ! jPb as in (10).
Contradiction.</p>
      <p>Lemma 1. The social welfare function for
pronoun resolution is IIA and Pareto but is
dictatorial.
3</p>
    </sec>
    <sec id="sec-4">
      <title>Dynamic Update of SCF</title>
      <p>In linguistic literature, a variable assignment
function g has been assumed to assign the referent to
indices indexed to pronouns. For example, g may
assign John to the variable x: g(x) = John. Now, g
can be considered to be SCF which selects a
referent for a pronoun socially. Let us define g and
the space as in (11). The assignment function g is
updated throughout the discourse as in (12).
(11) a. g = f&lt;x, i&gt;: x refers to ig
b. Information state consists of Social
Welfare Function F, Social Choice
Function g for variable assignment,
individual’s preferences R, individuals in the
discourse X, a set of indices such as i, a set of
discourse participants V, and relation
between decisions B.</p>
      <p>= &lt; F, G, R, X, I, V, B&gt;
(12) 1 There were ooh’s and aah’s when
hex1 finished, and some unbridled
laughter. Aileena was looking dubiously at hery1
husbandh but hex2 was in no mood to
disapprove.</p>
      <p>2 Hex3 winked at the Duked and called
across to himx4, ‘What a grand thing, your
Honour, to have a wedding without a
minister!’ The Duked did hisx5 stately bow at
that and then Donaldm was calling for
another song.</p>
      <p>3 Some of the veteransv were on the
point of giving tongue but young Donald
McCullochm was on hisx6 feet and moving
into the middle of the ring, hex7 was full of
himselfx8, sparkling with mischief but with
an undertow of ardour.</p>
      <p>4 ‘Duncan Ban MacIntyreb wrote a song for
hisx9 wife Maryr.</p>
      <p>5 I do not know if Alexl used it to court
his10 M aryr – hex11 must have used
something —‘The joke was unconscious but
crowing laughter came from the young menn
beside the whisky jar. (BNC A0N1311-1315,
King Cameron)
(13) a. g1= f&lt;y1, a&gt;, &lt;x2, h&gt;g</p>
      <p>I = fa, rg (a: author, r: reader)</p>
      <p>S = fa, hg
b. g2 = f&lt;x3, h&gt;, &lt;x4, d&gt;g</p>
      <p>S = fa, h, d, mg
c. g3 = f &lt;x6, m&gt;, &lt;x7, m&gt;, &lt;x8, m&gt;g</p>
      <p>S = fa, h, d, v, mg
d. g4 = f&lt;x9, b&gt;g</p>
      <p>S = fa, h, d, v, m, b, rg
e. g5 = f&lt;x10, l&gt;, &lt;x11, l&gt;g</p>
      <p>S =fa, h, d, v, m, b, r, l, ng
f. [[hery]]g1 = a</p>
      <p>
        G is regarded as SCF. Also, the set of best
elements S’ can be called its choice set of the whole
set of alternaties, and is denoted g(S’, R)
        <xref ref-type="bibr" rid="ref16">(cf. Sen
1979)</xref>
        R is a sequence of individual’s preferences
where Rx is a preference ordering of x.
(14) g1(S, R) = fa, hg
g2(S, R) = fh, dg
g3(S, R) = fmg
g4(S, R) = fbg
g5(S, R) = flg
      </p>
      <p>As the author’s dynamic preferences change in
the discourse as in (15a), g is updated throughout
the discourse by means of a relation B.
(15) a. 2: hRad for hex3 ^ dRah for hex4 ^
dRahIam for hex5 (aIxb: x is indifferent
between a and b, ^: dynamic
conjunction)
b. Social Decision: hRd ^ dRh ^ dRhIm
c. B(gn,gn+1)
(16) Dynamic Social Welfare Function:</p>
      <p>FnBFn+1BFn+2,...
4</p>
    </sec>
    <sec id="sec-5">
      <title>Comparison with Other Studies</title>
      <p>
        Dynamic Predicate Logic
        <xref ref-type="bibr" rid="ref5">(Groenendijk and
Stokhof 1991)</xref>
        consider update semantics where
two states differ with respect to variable
assignment. When h[x]g, the state g is updated with
respect to the assignment to x. The current
paper consider an abstract function B between two
SCFs.
        <xref ref-type="bibr" rid="ref14">Parkes and Procaccia (2013)</xref>
        model
dynamic decision making under constantly changing
preferences using Markov decision processes, in
which the states coincide with preference profiles
and a policy corresponds to a social choice
function.
5
      </p>
    </sec>
    <sec id="sec-6">
      <title>Detection of Speaker Intention</title>
      <p>In order to implement Dynamic Social Choice
for pronoun disambiguation, speaker’s intention
needs to be detected from the text. The phrases
such as “I mean” are used to resolve ambiguity of
pronominal reference in the discourse. as in (17).
(17) ‘...And Sarah Morgan likes the idea of
Angela marrying someone in the government.’
McLeish considered this cold and rational
assessment. ‘When did you last see her? Miss</p>
    </sec>
    <sec id="sec-7">
      <title>Angela Morgan, I mean.’ (BNC AB9)</title>
      <p>Out of 18 instances of “I mean PNP” (PNP
stands for proper name” ) found with the query
“I mean N” in BNC, 7 instances had a preceding
pronoun, the caraphor.</p>
    </sec>
  </body>
  <back>
    <ref-list>
      <ref id="ref1">
        <mixed-citation>
          <string-name>
            <surname>Arrow</surname>
            ,
            <given-names>K. J.</given-names>
          </string-name>
          :
          <year>1963</year>
          ,
          <string-name>
            <given-names>Social</given-names>
            <surname>Choice</surname>
          </string-name>
          and Individual Values,
          <volume>2</volume>
          <fpage>edn</fpage>
          , Yale University Press, New Haven.
        </mixed-citation>
      </ref>
      <ref id="ref2">
        <mixed-citation>
          <string-name>
            <surname>Chomsky</surname>
          </string-name>
          , N.:
          <year>1981</year>
          , Lectures on Government and Binding, Foris Publications, Dordrecht.
        </mixed-citation>
      </ref>
      <ref id="ref3">
        <mixed-citation>
          <string-name>
            <surname>Endriss</surname>
            ,
            <given-names>U.</given-names>
          </string-name>
          :
          <year>2016</year>
          ,
          <article-title>Judgment aggregation</article-title>
          , in H. Moulin,
          <string-name>
            <given-names>F.</given-names>
            <surname>Brandt</surname>
          </string-name>
          ,
          <string-name>
            <given-names>V.</given-names>
            <surname>Conitzer</surname>
          </string-name>
          ,
          <string-name>
            <given-names>U.</given-names>
            <surname>Endriss</surname>
          </string-name>
          ,
          <string-name>
            <given-names>J.</given-names>
            <surname>Lang</surname>
          </string-name>
          and
          <string-name>
            <surname>A. D.</surname>
          </string-name>
          Procaccia (eds),
          <source>Handbook of Computational Social Choice</source>
          , Cambridge University Press, pp.
          <fpage>399</fpage>
          -
          <lpage>426</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref4">
        <mixed-citation>
          <string-name>
            <surname>Gaertner</surname>
            ,
            <given-names>W.</given-names>
          </string-name>
          :
          <year>2009</year>
          ,
          <string-name>
            <surname>A</surname>
          </string-name>
          <article-title>Primer in Social Choice Theory</article-title>
          .
          <article-title>Revised edition</article-title>
          ,
          <source>LSE Perspectives in Economic Analysis</source>
          , Oxford University Press.
        </mixed-citation>
      </ref>
      <ref id="ref5">
        <mixed-citation>
          <string-name>
            <surname>Groenendijk</surname>
            ,
            <given-names>J.</given-names>
          </string-name>
          and
          <string-name>
            <surname>Stokhof</surname>
            ,
            <given-names>M.</given-names>
          </string-name>
          :
          <year>1991</year>
          ,
          <article-title>Dynamic predicate logic</article-title>
          ,
          <source>Linguistics and Philosophy</source>
          <volume>14</volume>
          ,
          <fpage>39</fpage>
          -
          <lpage>100</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref6">
        <mixed-citation>
          <string-name>
            <surname>Leass</surname>
            ,
            <given-names>H. J.</given-names>
          </string-name>
          :
          <year>1991</year>
          ,
          <article-title>Anaphora resolution for machine translation: A study</article-title>
          ,
          <source>IWBS Report 187.</source>
        </mixed-citation>
      </ref>
      <ref id="ref7">
        <mixed-citation>
          <string-name>
            <surname>Moulin</surname>
          </string-name>
          , H.:
          <year>1988</year>
          ,
          <article-title>Axioms of Cooperative Decision Making, Econometric Society Monographs</article-title>
          , Cambridge University Press, Cambridge.
        </mixed-citation>
      </ref>
      <ref id="ref8">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          :
          <year>2011</year>
          ,
          <article-title>Computational social choice for pronoun resolution, Ipsj sig-nl.</article-title>
        </mixed-citation>
      </ref>
      <ref id="ref9">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          : 2012a,
          <article-title>Social choice for anaphora resolution</article-title>
          ,
          <source>NLP2012 Proceedings</source>
          , pp.
          <fpage>97</fpage>
          -
          <lpage>100</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref10">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          : 2012b,
          <article-title>Social choice for anaphora resolution</article-title>
          ,
          <source>Society for Social Choice</source>
          , New Delhi.
        </mixed-citation>
      </ref>
      <ref id="ref11">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          :
          <year>2014</year>
          ,
          <article-title>Application of social choice theory in pronoun resolution</article-title>
          ,
          <source>Society for Social Choice</source>
          , Boston College.
        </mixed-citation>
      </ref>
      <ref id="ref12">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          : 2016a,
          <article-title>Social choice for anaphora resolution</article-title>
          ,
          <source>Studies in Liberal Arts and Sciences</source>
          , Vol.
          <volume>48</volume>
          , Tokyo University of Science, pp.
          <fpage>147</fpage>
          -
          <lpage>156</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref13">
        <mixed-citation>
          <string-name>
            <surname>Nishiguchi</surname>
            ,
            <given-names>S.</given-names>
          </string-name>
          : 2016b,
          <article-title>Social choice for disambiguation of pronominal reference, Society for Social Choice</article-title>
          , Lund University.
        </mixed-citation>
      </ref>
      <ref id="ref14">
        <mixed-citation>
          <string-name>
            <surname>Parkes</surname>
            ,
            <given-names>D. C.</given-names>
          </string-name>
          and
          <string-name>
            <surname>Procaccia</surname>
          </string-name>
          , A. D.:
          <year>2013</year>
          ,
          <article-title>Dynamic social choice with evolving preferences</article-title>
          ,
          <source>Proceedings of the 27th AAAI Conference on Artificial Intelligence</source>
          , Palo Alto, CA, pp.
          <fpage>767</fpage>
          -
          <lpage>773</lpage>
          .
        </mixed-citation>
      </ref>
      <ref id="ref15">
        <mixed-citation>
          <string-name>
            <surname>Reinhart</surname>
            ,
            <given-names>T.</given-names>
          </string-name>
          :
          <year>1983</year>
          ,
          <article-title>Anaphora</article-title>
          and
          <string-name>
            <given-names>Semantic</given-names>
            <surname>Interpretation</surname>
          </string-name>
          , The University of Chicago Press, Chicago.
        </mixed-citation>
      </ref>
      <ref id="ref16">
        <mixed-citation>
          <string-name>
            <surname>Sen</surname>
            ,
            <given-names>A. K.</given-names>
          </string-name>
          :
          <year>1979</year>
          ,
          <string-name>
            <given-names>Collective</given-names>
            <surname>Choice</surname>
          </string-name>
          and
          <string-name>
            <given-names>Social</given-names>
            <surname>Welfare</surname>
          </string-name>
          , North-Holland, Amsterdam.
        </mixed-citation>
      </ref>
      <ref id="ref17">
        <mixed-citation>
          <string-name>
            <surname>Taylor</surname>
          </string-name>
          , A.:
          <year>2005</year>
          ,
          <string-name>
            <given-names>Social</given-names>
            <surname>Choice</surname>
          </string-name>
          and
          <source>the Mathematics of Manipulation</source>
          , Cambridge University Press, Cambridge.
        </mixed-citation>
      </ref>
      <ref id="ref18">
        <mixed-citation>
          <string-name>
            <surname>van Rooij</surname>
            ,
            <given-names>R.</given-names>
          </string-name>
          :
          <year>2011</year>
          ,
          <article-title>Measurement and interadjective comparisons</article-title>
          ,
          <source>Journal of Semantics</source>
          <volume>28</volume>
          ,
          <fpage>335</fpage>
          -
          <lpage>358</lpage>
          .
        </mixed-citation>
      </ref>
    </ref-list>
  </back>
</article>