<?xml version="1.0" standalone="yes"?>
<Paper uid="C96-2152">
  <Title>Disambiguation by Prioritized Circumscription</Title>
  <Section position="3" start_page="0" end_page="901" type="intro">
    <SectionTitle>
1 Introduction
</SectionTitle>
    <Paragraph position="0"> '\]'his p;~pcr presents a lll(.q;llod of (tisa.ml/igu;LdOn t;Lsk 1)y a va.rbmt of (;ircunmcription, prioritized ci'rcum, scriptiou, (M(:Car|;hy. 1986; l,ifhchitz, 1985) and discuss its iml)h',nlental:ion by a hierarchical l:onstr;I,int logic progrmnming (HCIA)) bmguagc, sltc\[l a,s (Borning et al., 1989).</Paragraph>
    <Paragraph position="1"> Dis~unl)igua, I;i(/n is a v(',l'y intporl;ant t;ask in mttura.\[ bmguage l)ro(:essing. To reso\]ve aml)iguity, hunt~ms se, enl to use not only syntactic (:onslxa,ints but a,lso wtrious lev(;ls (1t7 heuristics such as grammati(:;d 1)r(~Dren(:es (\[\[ol/1)s, t990) a,nd sema,ntic t)ref(',r(',n(;(',s (Wilks, 1975).</Paragraph>
    <Paragraph position="2"> For t~xamtlh'., SUl)pose thai: we have the following SCllt(HICCS.</Paragraph>
    <Paragraph position="3"> John just saw a man with a telescope. (M He bought the teles('ope yesterday. (1)) Although dmr(', is ;m alnt)iguil;y on mc;milig of the 1)hrase, &amp;quot;'widt a teh'~scope&amp;quot;(ldw, teh;scot)(; ix either used by ,}ohn or (:arricd by the man), we might conclude t:hc preflwred rc~uling a.~ f'olh)ws.</Paragraph>
    <Paragraph position="4"> From dm above sentences. &amp;quot;He&amp;quot; wouhl be equal to .John l)c(:ause tho subject t;ends to be cont.inued t.o the ne.xt sentence }rod .John probably had a telescope at the time of seeing a ma.n fi:om |;he scnteiwe (b) ;u,l ilw.r|;i~ of possession. Therefore.</Paragraph>
    <Paragraph position="5"> from this prefl:rrcd re;~(ting, we conchuh', that the telescope is used as a, (levi(:(; to sec a man.</Paragraph>
    <Paragraph position="6"> However, this re;~ding is not tinal since at least the fl)llowing prcfl'~rences are involved in the above reading and these preferences Call be (teib.~ted 1)y s(;rong(;r hfl'orlm)A;ion.</Paragraph>
    <Paragraph position="7"> Syntactic preferen(:e: The sul)ject 1;(re(Is to I)c col\[\[;illltc(\[.</Paragraph>
    <Paragraph position="8"> Semantic prefer(m('(.': If ;~ person buys something at time i, then he shouhl h;~ve it at time j where i &lt; j.</Paragraph>
    <Paragraph position="9"> In or(tot to dcmons(;ratc defea.sibilil;y of t)rcfer -o.11(:(! l'ltl('.s, :Ul)t)osc the following S(HItCII(:C is added ,,ft(,,. th(, ahoy, s(..,m,n(:(,s (a) ~u.l (b).</Paragraph>
    <Paragraph position="10"> Bu G he gave the lnan the telescope (c) this morning.</Paragraph>
    <Paragraph position="11"> Then, we migh|; oh;rage ;~ preferred reading th~l; (,he mint should have had a telescope, and dmr(&gt; for('., t;he t(~lcs(:ope was carried by the man at the. time of John's see, ing the man. In this re;uting, a(; legist, (;he following preference rule of mmther inertia of possession is used.</Paragraph>
    <Paragraph position="12"> ff a I)(',rson gives some.thing to tim other person at time i, then the other l)erson shouhl btve it a,t time j where i &lt; j.</Paragraph>
    <Paragraph position="13"> This (:onfli(:l,s with (;he 17Orlner semantic preferen(:e of inertia of possession by 1)uying, but the above preference is st, ronger tbm the former since the time of giving is later t;h;m the time of buying.</Paragraph>
    <Paragraph position="14"> rl'hlls, the folJlllCF t)reference b(~C()lll(R-; 1lO hmer 3111)licablc by the new scutcncc.</Paragraph>
    <Paragraph position="15"> This kind of revision of reading cmmot be rcl)res('.nl,ed by infl'.ren(:e in (:l;~ssical logic siu(:e in classi(:;d logic, once wc gel; ~ inferred result, we can no hmger rel:ra.(:t the result (monotonic pr'operty). Therefore, to ml(lerstand tit(,, phenolnen&amp; we need other reasoning met;hods rout in f~(:t,  many researches h;tve been using general reasoninn f~am(:works in Artificial Intelligence. such a,s abduction (ttobbs et al., 1993), prol)abilistic network (Chm'niak and Gohhnan, 1989), truth lnaint(mance system (Zernik and Brown, 1988), default logic (Quantz, 1993) and conditional logic (Lascarides, 1993). In this paper, wc ttropose another alternative, that is, circ,,.m.~cription (McCarthy, 1986: Lifschitz, 1985). Even though circumscrit)tion is one of the most pottular fornlMisn,s in the COllllllllllit, y of llOlllllOllOtOlliC reasoning rcs(,.ar(:h, it is surprishlg that w'.ry few h~ts examined feasibility of (:ircumscrit)tion for disa, mbiguation. Our work of disambiguation by intcrt)retation ordering is originated from (Satoh, 1991) and in a more recent work, Kameymna (Kameyama, 1994) has indcp(mdently propos('.d usage of circumscription for interpretation of pronominal anaphora.</Paragraph>
    <Paragraph position="16"> In this paper, we explore this direction further. In circumscription, we give a prefcrclme of der over logical interpretations and consid('.r the most 1)referable models. This representation naturally corrcst)onds with a disambiguation task sin(w. we can regard a logical interpr(~tation as a possible remling, and disaml)igua, tion as a, task to get the most prcf(wal)h; reading among possil)le readlags. Among variants of circumscription, prioritizcd circu.mscriptio'n, is suitable to ret)rcsent various strength of preference rules. In f)rioritized &lt;fircumscril)tion, we can (livide preference rules into hiernrchy and this hierarchy giv(,,s a t)riority relation ov(,.r pr(~f('.rences. Therefore, we directly represent rules in the hierarchy in prioritized circumsccil)tion. null We believe, that circumscritttion has the following allvantage, s in the (:ask of resolving ambiguity. * Since wc use a tirst-order predicate calculus for a basic language, we (:all rot)resent various kinds of inforln;ttion such as grammaticsfl rules and semantical rules in one Damework.</Paragraph>
    <Paragraph position="17"> * There is only one extra underlying mc(:hanism besides iufi'.rellC(', rules for the first-order predicate cah:ulus, that; is, introducing an order over h)gical interltretations. Therefore, re.asoning process (:an be un(hwstoo(l easily COlnt)ared to other lnechallisln using numerical reasoning or comt)h:x inferca(:(; rules.</Paragraph>
    <Paragraph position="18"> * We (lo not nec(l to assign detailed mlmcri(:al values to t)referellce rules in ord(;r to express t)riority over t)r(~ference rules, but just specify a t)ref(:rence level of the rules. This representation can t)e regar(led as all assignment of qualitative strength for 1)ref(~l'ell(;(~ rules all(\[ reduces a \])ur(letl of tel)resenting a ttriority over preference rules gready. Moreover, this prioritization is general since we can repr(~sent a various kind of priority besides specilicity.</Paragraph>
    <Paragraph position="19"> * It is important to retain 1)ossible readings if we can not w.solve aml)iguity yet. In cir(:umscrip-Lion, we can consider multiple preferable models, not nec(;ssary the single pr@rable model. So, if ther(~ are yet multiple possible readings as a result of disambiguation, we can keep these possit)le readings as multiple i)retb.rat/le lnodels.</Paragraph>
    <Paragraph position="20"> In this l)aper, we also dis(:uss an implenmnration by using hierarchical constraint logic programlning (HCLP) language sltch as (Borning ct al., 1989). HCLP language is similar to constraint h)gic progralnming bmguage except that we (;nil represent a constraint hierm:dly. Thus, there is a corrcst/on(lencc l/ctwecn a solution of an HCLP language and the most t)rcferablc models of prioritized circunmcription. Ill this patter, we use our HCLP language based on a l)oolean constraint solver to get tlJ(', most t)rcferal)lc models from t)ref er(mce rules rot)resented as bo(/h,,mt constraints in tlt(', HC'L1 ) language. Wc demonstrate how the.</Paragraph>
    <Paragraph position="21"> ;fi)ove example of the disalnbiguation is tre;~ted in tlm HCLP language.</Paragraph>
  </Section>
class="xml-element"></Paper>