<?xml version="1.0" standalone="yes"?>
<Paper uid="C96-2158">
  <Title>Anaphor Resolution and the Scope of Syntactic Constraints</Title>
  <Section position="5" start_page="938" end_page="939" type="metho">
    <SectionTitle>
3 Towards the Algorithm
</SectionTitle>
    <Paragraph position="0"> The lnaill question concerns the adequate implementation of Chomsky's I)inding t)rinciples. Some a l)riori remarks on theoretic subtM;ies and on the eml)loyed ret)resentation are in t)lace.</Paragraph>
    <Section position="1" start_page="938" end_page="938" type="sub_section">
      <SectionTitle>
3.1 lnterdei)endency Sensitiveness
</SectionTitle>
      <Paragraph position="0"> As state(l t)y (Correa, 1988), an immediate imI)lemen|at|on of th(; constraints proposed in Binding Theory is unlbasil)le. Chomsky states, merely as a the()rctical device, a flee, inclexing rule wlfich ran(lomly assigns reference in(lexes to surface structure NP nodes. During inapt)lug to the seinanti(: LF (logical form) representation, the t)inding principles s()~'ve as restrictions tbr filtering out the im dex distributions which are considered valid when intert)re, ted as eorefL'rence markers. A direct iml)leme, ntation of this generate-~md-test 1)ro(:edure yields an exponential time complexity.</Paragraph>
      <Paragraph position="1"> Current approaches avoid gen('rate-and-te~st |)y resorting to different strategies. According to 1;11(; most colnmon tectmiquc, for anat)horic NPs, a separate antecedent search is t)ertbrmed, resulting in a quadratic time complexity (e.g. (Hot)bs, 1978; Strube and Hahn, 1995)). Because, howev(;r, the ante(:edent decisions are performed in isolation, invalid index distributions may m'ise. In examph; (Sa) The barbcri told the elientj a story, while hek shaved himl.</Paragraph>
      <Paragraph position="2"> neither of the t)ronouns is confined structurally to one of the intrasenttmtial antece.(lent eandi(tates in tie matrix clause. But, afl, era first decision, e.g.</Paragraph>
      <Paragraph position="3"> (Sb) The barberi told the clientj a story, while he |shaved himz.</Paragraph>
      <Paragraph position="4"> the situation changes, for one of the antect;dent options of the still unresolved l)ronoun is no longer available. Binding principle B may bc violat(.'d: (8c) * The barberi told the. clientj a story, while he |shaved him|.</Paragraph>
      <Paragraph position="5"> An interdepemlen(:y 1)el;we(;it antecedent; choi(:es may arise as well when choosing/)etween discourse alltece(leiltS~ OF as a COllSeqll(;llc( ~, ()f relative, clause attachment, which 1)redetermines coindexing.</Paragraph>
      <Paragraph position="6"> The at)proach presented below is sealsitive to these, decision interdependencies, while avoiding the exponential time comi)lexity of an immedL ate l)inding constraint implem(mtatioil. This is achieved by supplementing the straightforward se,quential strategy with a dynamic reveritication of the binding restrictions in the antecedent selection stet). To avoid that (te.sirable antecedent options are ruled out l)y interdependency, the choices wil;h highest plausibility is given preference to.</Paragraph>
    </Section>
    <Section position="2" start_page="938" end_page="939" type="sub_section">
      <SectionTitle>
3.2 Representing Surihce Structure
</SectionTitle>
      <Paragraph position="0"> The original statement of Binding Theory forms part of GB Theory, in which a broader set &lt;)1' in: tera&lt;:ting l)rin(:iph~s is f&lt;)rmulated. Because the aim of aimi)hor resolution for a specific language is restricted, the reI)resentation (:an be simplifiexl.</Paragraph>
      <Paragraph position="1"> Complicating details which result fl'om the Gll claim to mfiversality may t)e emil;ted.</Paragraph>
      <Paragraph position="2"> lies|des being efficiently searchable, the simplified surface structure has to represent the stru(:t;ural details wtfich are necessary for th(,' verifica-. tion of the 1)in(ling restrictions. In particular, this  comprises subject-object-asymmetry, the demarcation of local domains, and surf.ace order dependent structural variations 3.</Paragraph>
      <Paragraph position="3"> Because the KONTEXT text analysis system is based on a dependency grammar, a mapping process generates the required representation from a dependency trees which is not suitable for a structural verification of the binding principles, because vital details are not structurally visible.</Paragraph>
      <Paragraph position="4"> The attempt of directly Verifying BT restrictions on dependency structure, as suggested by Strube and Hahn (Strube and Hahn, 1995), does not seem adequate, because important details are ignored.</Paragraph>
      <Paragraph position="5"> The structures which were generated for some of the above examples are as follows: 4</Paragraph>
      <Paragraph position="7"> The marker nodes STHAT and SREL are delimiters of local domains, to which the binding principle verification functions are sensitive.</Paragraph>
      <Paragraph position="8"> Special techniques are employed in representing local NP domains, which are introduced by deverbative NPs and NPs with possessive markers (saxonian genitive, genitivus possessivus, possessive pronoun, or certain attributive PPs), e.g.</Paragraph>
      <Paragraph position="9">  opened. The NP barber and the reflexive pronoun himself may be coindexed only indirectly via the possessive pronoun his, which is of type B, and hence forced to take a nonlocal antecedent. In accordance with intuitive judgement, a local instance of the NP storyj blocks the eoindexing of the possessivc pronoun and its dominating noun.</Paragraph>
      <Paragraph position="10"> Here again, the mechanism which copes with interdependencics is appliedfi Technically, new NP types C' (example (10)) and B' (relative pronoun, 3This concerns certain cases of subject and object clause extraposition as well as, in particular, the object NPs contained in the VP, for which a right branching structure is generated, yielding a base for a structural determination of admissible antecedents for reflexive pronouns, which is mainly governed by subject-object asymmetry and surface order.</Paragraph>
      <Paragraph position="11">  Chomsky's GB theory. Because of its restricted aim, however, it is nmch simpler.</Paragraph>
      <Paragraph position="12"> cf. section 3.1) are introduced for which binding principles C and B are verified, respectively, but for which no antecedent search is performed.</Paragraph>
    </Section>
    <Section position="3" start_page="939" end_page="939" type="sub_section">
      <SectionTitle>
3.3 The Algorithm
</SectionTitle>
      <Paragraph position="0"> The KONTEXT anaphor resolution algorithm, as shown in figure 1, consists of three phases: constraint application, preference criteria application and plausibility sorting, and antecedent selection including reverification of constraints which may be involved in decision interdependencies.</Paragraph>
      <Paragraph position="1"> Two binding constraint verification procedures are employed which differ in the handling of type A NPs. According to binding principle A, a reflexive pronoun requires 'constructively' a local antecedent (step l(b)i). Example (10), however, illustrates that further nonloeal coindexings are admissible. This gives rise to a weak version of binding constraint verification, the usage of which is of vital importance to the fimctioning of the interdependency test step 3b.</Paragraph>
    </Section>
  </Section>
  <Section position="6" start_page="939" end_page="940" type="metho">
    <SectionTitle>
4 Evaluation
</SectionTitle>
    <Paragraph position="0"> As a proper base for comparison, the theoretical analysis is restricted to the contribution of intrasentential antecedent search. Let n be the number of NP nodes in the surface structure representation. Because the number of anaphoric NPs and intrasentential candidates is bounded by n, and the individual a priori verifications of the binding principles contribute costs proportional to the number of nodes in the surface structure tree, the worst case time complexity of step 1 is O(n3). A similar analysis, assuming a clever handling which prevents individual interdependency checks from being done more then once, reveals that the complexity of step 3 is O(n 3) too. Therefore, since the scoring and sorting step 2 does not exceed this limit, the overall worst case complexity is O(n3).</Paragraph>
    <Paragraph position="1"> In tests on architect biographies drawn from (Lampugnani, 1983), the algorithm correctly resolved approximately 90 per cent of type B pronouns (including possessives), and, as expected, all occurrences of reflexives, which occur quite scarcely in the test corpus. The set of possible antecedents tends to be reduced drastically during constraint application. Interdependency collisions did not happen too frequent. This tendency is strongly supported by the case role inertia heuristic, which promotes a complementary distribution of preferred antecedents for type B pronouns cooccurring in a domain of binding.</Paragraph>
    <Paragraph position="2"> The strategy of considering the more plausible antecedent choices first does not eliminate interdependency collisions in general, and, moreover,  1. For each anaphoric NP Y, determine the set of possible antecedents X: (a) Verify morphosyntactic or lexical agreement with X (congruence in person, number, and gender, lexical recurrence ete, depending on the type of Y) (b) If the antecedent candidate X is intru.sentential, check whether the binding l)rinciples of Y and X are satisfied: for the proposed eoindexing, i. verify that the binding principle of Y is satisfied constructively, ii. verify that the binding principle of X is not violated. (c) If Y is a type B 1)ronoun, antecedent candidate X is intr~scntential, and, according to surfa(:e order, X follows Y (i.e. tit(.' resumption would be cataphorie), verify that X is definite. 2. Plausibility scoring and sorting: (a) For each surviving pair (Y/, A~) of anaphor and antecedent candidate: deterinine the munerical plausibility score v(Y/, Xj), which ranks Xj relatively to Y/, based on case role inertia, recency, cataphor penalty, and subject prefe.rence, deI)ending on the type of I~. (b) (local sorting) For each anai)hor Y: sort their individual antecedent cm, didates Xj according to decreasing plausibility v(Y, X a ).</Paragraph>
    <Paragraph position="3"> (c) (global sorting) Sort the anaphors V according to decreasing I)lausibility of their individual best antecedent candidate.</Paragraph>
    <Paragraph position="4"> Antecedent Selection: Consider anaphors Y in the order determined in step 2c. Suggest antecedent candidates X(Y) in the order determined in step 213. Select X(Y) as candidate if there is no interdependency, i.e. if (a) the morphosyntactic featmes of Y and X(Y) are still compatible% (b) for each NP Z whose coindexing with X(Y) has been determined in the current invocation of  the anaphor resolution algorithm: the coindexing of Y anti Z which results as a side effect when chosing X(Y) as antecedent for Y does not violate the binding 1)rincil)les. To allow for m, efficient detection of intex(let)endencies , store the selected antecedent separately fl'om corefercnt occurrences contributed by earlier invocations of the algorithm.  does not guarantee that the global maximum of plausibility is reached. Because of its practical performance, however, it proved to be a satisfactory substitute for the generate-and-test strategy.</Paragraph>
  </Section>
  <Section position="7" start_page="940" end_page="941" type="metho">
    <SectionTitle>
5 Exploring the Limits
</SectionTitle>
    <Paragraph position="0"> The determination of the substructure describing a local domain iv not always easy. Whereas for NPs with possessive markers (of. example (10)) the matter tends to be clear, a common source of difficulties emerges from adjectivally used participles and from deverbative NPs. In the latter case, e.g. a genitival attribute may instantiate, dependlug on the NP, either the subject (.qenitiwts subjectivus) or the object (.qenitiwts objectivus) (for German, cf. (Teubert, 1979)). As the following examl)les demonstrate, it iv insufficient to know merely about the existence of a h)cal domain. In general, it is necessary to determine the instantia= tion of its participants, but this, at least in certain &lt;:ases, involves pragmatic inferencing.</Paragraph>
    <Paragraph position="1"> (lla) Pauli accepts the decision for himi.</Paragraph>
    <Paragraph position="2"> (llb) * Pauli accepts the decision for himselfi.</Paragraph>
    <Paragraph position="3"> According to accet)tability judgements, decision introduces a local binding domain. But a change 6In German, this kind of interdependency may arise, due to lnorphosyntactic ambiguity, in case of multiple occurrences of the pronoun sic.</Paragraph>
    <Paragraph position="4"> of tile matrix clause verb leads to a different judgement, while tile syntactic structure is preserved:  (12a) Pauli revises the decision for himi.</Paragraph>
    <Paragraph position="5"> (12b) Pauli rcviscs the decision for himselfi.</Paragraph>
    <Paragraph position="6">  The clue lies in the observation that a pragmatic restriction is governing the instantiation of the implicit local subject in exmnples (11), but not in examples (12). in (11@, duc to the obvious conclusioil that someone who accepts an action is not the conscious actor of it, .Paul is pragmati(:ally ruled out as the local subject of the decision domain. On the other hand, revise leaves open whether Paul or someone else is the decider. This explanation is confirmed by the following data:  (13a) Paull revises Sarn'sj decision for h, imi.</Paragraph>
    <Paragraph position="7"> (13b) * Pauli revises Sam 'sj decision fi)r himselfi . (13e) *Pauli revises hisi decision for himi.</Paragraph>
    <Paragraph position="8"> (13d) Pauli revises hisi decision for himselfi.</Paragraph>
    <Paragraph position="9">  Current approaches (Strube and Hahn, 1995; Lappin amt Leass, 1994) ignore this subtlety by merely taking into account NP domains Which are established by possessive determiners. As a consequence, wrong results may be obtained, e.g. in case of example (lla), as there is no t)ossessive modifier, Paul will not be considered to be an mttecedent candidate for him. With these difficulties in mind, questionable antecedent decisions may t)e  marked as depending on particular local instantiations, by this means providing a starting point for more comprehensive considerations which take into account the relation between structural restrictions and the resolution of ellipsis.</Paragraph>
  </Section>
class="xml-element"></Paper>