File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/94/c94-2126_metho.xml
Size: 36,563 bytes
Last Modified: 2025-10-06 14:13:40
<?xml version="1.0" standalone="yes"?> <Paper uid="C94-2126"> <Title>A TREATMENT OF FUNCTIONAL DEFINITE DESCRIPTIONS</Title> <Section position="2" start_page="0" end_page="789" type="metho"> <SectionTitle> 2. RANGE OF FDD </SectionTitle> <Paragraph position="0"> When we process real texts, it is apparent that FDD is a non-trivial part of the use of definite descriptions. FDD constitutes a large portion of l)l) use among so-called first-mention use. In a recent study of a Swedish corpus, Fraurud (1990) reports that 60.9% of toted I)D occurrences me first-mention use, aud such complex 1)1) as the ones in the form of'the X of Y' in corresponding English structures (a typical form of Iq)D) accounts for 41.2% among the first-mention use. Since FDD can be found among simple DD, the percentage of FDD further increases. Although Fraurud's study reveals its importance in Swedish, it is easy to assume a comparable situation in English. In literature, FDD is usually regarded as a limited phenomenon that is difficult to formalize. Itowevcr, such a view toward I,Z)D is shortsighted since it not only undermines the importance of I'\])D with respect to other use of DI), but also misses the important relations to functional anaphoric expressions in other categories. null Hawkins (1978) studies various usage of definite deseriptions and proposes an analysis based on the theory of Familiarity. Two of his usage of the first-mention tel'mites are FDD under consideration here: an associative use such as 'a car' - 'the steering wheel', and a larger situation use such as 'a town' - 'the church', lie claims that in these cases common knowledge shared by a speaker and the hearer is very general and inferable from lexical information without pragmatic information. Hawkins' study signifies the importance of Familiarity presupposition of definite descriptions as well as the range of definite descriptions although his study covers only file major usage and misses some important issues, which we will examine below.</Paragraph> <Paragraph position="1"> Kamp (1984) provides the first but very insightful formal semantic analysis of FDD. In Kmnp's Discourse Reprcsentation theoretic treatment of FDD, a head noun of FDD always introduces a new reference marker for the individual that it denotes and a functional anaphoric referent into a universe of the DRS. tie notes that FDD ranges over types of functional referents. That is, if the functional referent of FDD is pronominal, the l~7)D needs to utilize pronominal resolution mechanism, and if demonstrative, then demonstrative resolution mechanism. Kamp's work suggests the range of FDD distribution is wider than that suggested by ltawkins mid shows us a point of departure.</Paragraph> <Paragraph position="2"> Lobner(1985) proposes an analysis of DD based on a lexical distinction among nouns; sortal nouns, functional nouns, and non-functional relational nouns. Sortal nouns are typical nouns that denotes individuals. Functional nouns are relational nouns with situational arguments. Ile studies a wide range of FDDs and classifies many of them into a bag of functional nouns. As a result, functional nouns include 'weather', 'time', 'sun', 'speaker', 'hearer', 'president', 'referee', 'bride', 'head', 'top', 'surlace', 'height', 'weight', 'birth', 'death', 'beginning', 'end', etc. lie points out that timctionai nouns allow modifications with pp, adjectives and adverbs, but non-functional nouns do not. Thus, the fact that 'the present wife' is acceptable but 'the present son' is not indicates the difference between functional nouns and non: functional relational nouns. However, even seemingly non-controversial relational noun such as 'daughter' can fred certain situation where it is modified by a pp. For example, a man has two daughters, one studying in L.A. and the other working in Seattle. We may refer to his daughters 'the daughter in L.A. and the daughter in Seattle'. This difficulty in distinguishing functional and non-functional nouns is one of the major problems in Lobner's treatment. Another major problem is the lack of constraining mechanism on linking.</Paragraph> <Paragraph position="3"> For example, (2) John's friend got married.</Paragraph> <Paragraph position="4"> Usually John takes pictures of a wedding.</Paragraph> <Paragraph position="5"> The bride hesitated to be photographed.</Paragraph> <Paragraph position="6"> It is not difficult for normal English speakers to identify the function of the second sentence as a background information.</Paragraph> <Paragraph position="7"> Therefore, the intended functional link from 'the bride' in the third sentence should be to the first sentence, tlowevcr, Lobner's analysis incorrectly allows a link between 'the bride' in the third sentence and 'a wedding' in the second sentence.</Paragraph> <Paragraph position="8"> In a recent study, Chierchia (1993) proposes a treatment of FDD based on his theory of Dynamic Binding.</Paragraph> <Paragraph position="9"> Chierchia represents functional anaphoric links by co-indexing a functional antecedent with a superscript and FDD with a subscript as shown below. IIe calls this a-indexation.</Paragraph> <Paragraph position="10"> (1') Every book i about Picasso made the author i rich.</Paragraph> <Paragraph position="11"> lie claims a-indexed anaphoric links are comparable to pronominal anaphoric links. This means that FDD should follow the same constraints on the pronominal anaphoric links. The idea behind this claim is very similar to Kamp's idea for elliptic DI)D. qhat is, both studies note certain similarity between FDD and pronominal possessives. In effect, both studies can explain why the following FDD link is not felicitous in terms of accessibility of pronominal anaphora.</Paragraph> <Paragraph position="12"> (3) Every book i about Picasso was published by Mr.King.</Paragraph> <Paragraph position="13"> #The author i became rich.</Paragraph> <Paragraph position="14"> However, Chierchia's study does not consider other FDD, those that Kamp analyzes. Also, it fails to explain a pair of sentences such as follows: (4) a. Usually it&quot; John meets every pastor who a&ninisters a wedding i, he writes to the bride i.</Paragraph> <Paragraph position="15"> b. #Usually if John meets every boy who has an interesting book i, he writes to the author i.</Paragraph> <Paragraph position="16"> Notice that (4-a) and (4-b) are structurally identical while FDD is felicitous in (4-a) but infelicitous in (4-b).</Paragraph> <Paragraph position="17"> In what follows, I will present a classification of I'T)Ds and their analyses based on DRT.</Paragraph> </Section> <Section position="3" start_page="789" end_page="789" type="metho"> <SectionTitle> 3. TWO CLASSES OF FDD </SectionTitle> <Paragraph position="0"> In this paper, I propose that FDD should be classified into two basic types according to their semantic characteristics. The first class is called relational 1~73D and the second non-relational FDI). Relational FDD denotes relations between objects and these relations are lexical properties of head nouns of FDD. Kinship terms such as 'son', 'mother', 'sister', etc. are typical examples of relational nouns. Non-relational FDD denote functions from sets of individuals to individuals. A typical example is a superlative noun phrase such as 'the tallest tree'. Ordinal number modification such as 'the third man' and identifying adjective modification such as 'the identical book' are other possible examples of this type. The distinction between relational and non-relational 1,7)D is characterized below: l-i) relational FDD takes an object in a given discourse as its functional antecedent and forms a functional anaphoric link, 1-ii) such a functional anaphofic link is licensed by lexical characteristics of the functional description of the head nouns, and l-ili) relational FDD and their functional antecedents can form a construction of &quot;FDD of (anteceden0&quot;, while 2-i) non-relational lq)D takes a selection set instead of all object as a fimctional antecedent, and 2-ii) the link is licensed by a modifier such as superlative, ordinal number, or identifying adjective rather than a head uoun of FDD.</Paragraph> </Section> <Section position="4" start_page="789" end_page="793" type="metho"> <SectionTitle> 4. RELATIONAL FDD </SectionTitle> <Paragraph position="0"> Functional anaphoricity of Relational FDD is marked by a head noun's lexical property. That is, only relational nouns can be heads of relational FDD. There are certain subclasses of relational FDDs. I propose three subclasses: FDD based on i) lexical relations, ii) temporal/locational, and iii) situational roles. Lexical relational FDD is characterized by their paraphrasability wilh true possessives (that is, both forms of &quot;X's Y&quot; and &quot;Y of X&quot;). In effect, this subclass constitutes the largest group mnong l-7)Ds. Situational Role FI)D is marked by social roles and professions such as 'judge' and 'bride'. Another subclass, temporal/ locational I71)1) is marked by relations with time and location. null 4.1. FDD based on Lexical RelatiorLs</Paragraph> <Section position="1" start_page="789" end_page="792" type="sub_section"> <SectionTitle> Possessives and Relations </SectionTitle> <Paragraph position="0"> This class is marked by the paraphrasability to possessive constructions. That is, I assume that FDD with lexical relations must be paraphrased by both forms of&quot;X' s Y&quot; and &quot;the Y of X&quot;. In other words, if English lexicon includes concepts such as relations denoted by certain class of nouns, this information must be shared by many constructions in English, and I assume that it is the case that possessives, compound nouns, and FDD utilize this information. Furthermore, possessive paraphrasability of this type of FDD captares their similarity to the pronominal anaphor that appear inpossessiveconstructions aspossessorpronominals. If this is con'ect, then it predicts that this type of FDD obeys the constraints on anaphoric links that prouominals obey.</Paragraph> <Paragraph position="1"> Barker (1991) proposes a semantic analysis of possessive constructions based on an ,assumption that all posses- null sives are base generated. Following Abney's (1987) DI' hypothesis, Barker proposes the following syntactic analysis of possessives.</Paragraph> <Paragraph position="2"> (5) Johu's mother DP ~ss\] D' /,,,..</Paragraph> <Paragraph position="3"> lie claims that a relation from a possessor to a possessee is lexically detenniuedifthepossessee is arclational noun such as 'sou', 'mother', etc. Such relations are represented by non-monadic predicates. For example, 'mother' will be translated by a dyadic predicate 'mother' such as follows: (6) \[\[motherl\] = ~.x~y \[mother(x,y)\] When a possessee is not a relational noun, then the possessive denotes a l~lation based on general ownership or some sort of closeness relations, aud Barker calls such relations an extrinsic relation. Ileproposes the following two semat~tic tr~mslations lbr possessive determiners corresponding to the above distinction.</Paragraph> <Paragraph position="5"> qhe la'anslation in (7-a) is used in the case ofa lexical relation, while p in (7-b) denotes ml extrinsic relation such as ownership. When they are used in the ~malyses of&quot;Johu' s mother&quot; and &quot;John's human&quot;, resulting trmlslations are as lollows: (8) a. \[\[John's mother\]l = ~y\[mother(j,y)\] b. \[\[John's hunum\]\] = kYlPO,Y) & hunmn(y)\] Note that 'mother' is a relational noun while'humau' is not. The above translations explain why the &quot;bureau of John&quot; is ungrmmnatical as opposed to the gnunmatical counterpart, &quot;the mother of John&quot;. A noun, 'Y' in the toml of &quot;X's Y&quot;, cau be either relational or non-relational, but 'Y' in file form of &quot;the Y of X&quot; must be relational. Therefore, a uon~ relational llOUU such as 'hnn|au' C~UlUOt form all expression &quot;the human of X&quot;. Barker's analysis provides a g~d tbundation lot our insight on relations in functional descriptions in general. Now, we say a noun X is relational if ~md only if it allows both &quot;Y's X&quot; and &quot;X of Y&quot;. All relational nouns are tr,-mslated into nou-nlonadie predicates.</Paragraph> <Paragraph position="6"> ill the above discussion, we did not inquire on the status of definite articles in the lonn of&quot;the Y of X&quot;. One may ask whether or not &quot;,ill relation~d nouns in the lbnn of &quot;Y of X&quot; require definite articles, and file auswer is obviously NO. It seems that only a certain sct of relations possess the uniquedeg uess presupposition on the arguments of the relations. For example, arelalion 'mother~of(X,Y)' that means that Y is the mother of X possesses the uniqueness presupposition on the second argument but not on the first argmneut. In conuast, a relation 'sou-of(X,Y)' thin means Y is a son of X h~ts the uniqueness pl~esuppositiou on the first argument but not tin tile secoud. Relations generally deseribe property/characteristic of one individual that occupies ORe argument of the relatiou. Let us call this argument the primary argument of a rclatilai as oppose to the referential argument that links to a rctcrent of the functioual antecedent. In 'mothcr-of(X,Y)' Y is the prhnary argmneut and in 'son-of(X,Y)' Y is the prhnary argument, q\]ms, ifaprhnary argument ofarclation Y is presupposed as unique, a definite ,'u'ticle is required in the Ibnn of &quot;Y of X&quot;. This is the reason why 'mother' in 'mothcrof' rcquircs adefinite articlebut not 'son' in 'sonof'.</Paragraph> <Paragraph position="7"> Uuiqucncss presuppositi(ms on certain arguments of relations are clearly lexical in natnre. Definite articles in 1&quot;1)1) rellcct this lexically marked presupposition. Cases without uuiqueness presupposition such as 'son-of should be called Functional Indefinite l)escriptions (FID) (See Wada (fotfllcoming) lot further discussion on this type of Functional Anaphora).</Paragraph> <Paragraph position="8"> Now, we extend the scope of ore' examiuation fl'oln possessive-based sliuctures to other cases such as 'a book' 'the author'. Siucc 'author' is a relalion',d noun, we anlicio pate that tile same kind of analysis is possible to the analysis of 'the author'. From file previous discussion, we know that the relational fuuctional auaphotic link between 'a book' ,'rod ' the author' is possible provided that&quot;the author of file book&quot; ~md &quot;the tx~k's author&quot; are both legitimate expressions. In other words, 'auti~of is a relational noun that denotes a dyadic lexical 1&quot;clarion 'author'. The two arguments of the relation are a relcreut for a salient book in a diseour~ aud a referent that is the nuiquely identifiable author of tile book.</Paragraph> <Paragraph position="9"> However, the expression &quot;the author&quot; needs to be linked anaphoricaly to its functional antecedent, n~mlely 'a book'.</Paragraph> <Paragraph position="10"> Kamp assumes timt there is a selection set lbr this definite description and the most salient individual hi the set will be selected as its antecedent. This is no different from regul~tt definite description s~zsolution. Nonetheless, we need to add sonic details It this.</Paragraph> <Paragraph position="11"> As poiuted out by Chiefchia (1993), tile functional ,'umphoric lille must be conslaained. Both KampaudChierchia assume that lq)l) can be ~malyzable as pronominal posses~ sives. The axmphorie links that are interpreted li'om paraphr,'tsed pronomiuals and their antecedents mnst lollow general constraints on pronominals. In I)RT, such a constraint is called tile accessibility condition based on weak subordin~Uioa relation (_>) between DRSs. Keanp and Reyle (1993:120) define it as follows: (9) Accessibility Condition l~et K be a DRS, x a discourse referent and ~ a l)RS~ couditioLL We say that x is accessible froin 3t ill K iff there are K _> K 1 and K 1 > K 2 such that x belongs to UK1 and 3t belongs to COnK2.</Paragraph> <Paragraph position="12"> The above cxmdition roughly tells that wheu ,'in expression can be interpreted as anaphoric to a certain entity, that auaphoric expression must reside within some extention of tile DRS in which the antecedent entity resides. FDD is no exception to this condition. Let us repeat the example of(I) and its variation here.</Paragraph> <Paragraph position="13"> (10 a. Every book about Picasso made the author rich.</Paragraph> <Paragraph position="14"> b. Every book about Picasso was published by Mr.</Paragraph> <Paragraph position="15"> King.</Paragraph> <Paragraph position="16"> #The author became rich.</Paragraph> <Paragraph position="17"> The contrast shown in the above sentences is comparable to the following pair.</Paragraph> <Paragraph position="18"> (11) a. Every book about Picasso made its author rich.</Paragraph> <Paragraph position="19"> b. Every book about Picasso was published by Mr.King.</Paragraph> <Paragraph position="20"> #Its author became rich.</Paragraph> <Paragraph position="21"> The pronominal possessive, 'its', appears in exactly the same location in the above sentences as the FDD s in (10). It seems unarguable to assume that the two obey the same constraint. Indeed, it is more consistent to treat FDD as pronominal anaphora than to treat it as definite anaphora when we consider that referential arguments introduce regular discourse referents such as pronominals.</Paragraph> <Paragraph position="22"> In sum, we observed that lexical relational FDD is licensed by lexical relations of the head nouns. The relations force uniqueness presupposition on the primary arguments of the relations. Furthermore, like pronominal anaphoric links, functional anaphoric links obey accessibility condition. In the following section, we examine other relational FDDs and see whether the above observed characteristics hold.</Paragraph> <Paragraph position="23"> 4.2. FDD based on Situational Roles Certain relations do not keep regular sense of relations that would typically be held between two individuals. That is, those that we consider here usually do not allow paraphrasing &quot;X's Y&quot; and &quot;the Y of X&quot; interchangeably. In this section, we consider a group of nouns that denote a relation between a situation and its unique clement. Let us examine an example first.</Paragraph> <Paragraph position="24"> (12) John attended a wedding last week.</Paragraph> <Paragraph position="25"> The bride was his ex-giflfriend.</Paragraph> <Paragraph position="26"> Hawkins pointed out that both a speaker and a hearer must have a shared knowledge about a common situational setting. In (12), it is non-controversial to assume general knowledge that 'the bride' is a unique and necessary role in the situation of 'a wedding'. I claim that role nouns such as 'bride' are closely related to certain situations and that due to this close relationships, functional anaphoric links are possible. Nouns of typical social roles and professions such as 'president', 'referee', 'judge', 'lawyer', 'driver', 'victim', 'murderer' satisfy for this use ofFDD (Lobner (1985:294)). I assume a situational role noun is non-monadic predicate whose first argument (referential argument) holds a referent for a situation. For example, 'the bride' is repre~ sented in the following way.</Paragraph> <Paragraph position="27"> bride(S,x) 'S' in (13) is a discourse referent for a situational role referential argument of&quot;bride&quot;. Like functional antecedents of relational nouns, 'S' will be resolved with a functional antecedent. Some examples of situation ,setting nouns are 'wedding', 'court', 'case', 'incident', 'accident', 'class room', 'restaurant', etc. I call these words situation triggers. I assume that typical situation triggers and their FDDs must be available in the lexicon as part of common-sense knowledge of English. For example, 'bride' should mark its situation trigger 'wedding' in the lexicon. Of course, this is a trivial solution and we need to determine formal characterization of situation and situation triggers as well as more general solution based on common sense reasoning.</Paragraph> <Paragraph position="28"> Subordination and Situational Role FDD Because Situational Role FDD always appear with a certain situational setting, it is often the case that we see the following kind of contrast.</Paragraph> <Paragraph position="29"> (14) a. Usually if John meets every pastor who administers a wedding, he writes to the bride.</Paragraph> <Paragraph position="30"> b. #Usually if John meets every boy who has an interesting book, he writes to the author.</Paragraph> <Paragraph position="31"> Notice that since (14-a) and (14-b) are structurally identical, both 'wedding' and 'book' should not be accessible to 'bride' and 'author', respectively. As we see in (14), it is not the case. Presumably, (14-a) is represented in the following</Paragraph> <Paragraph position="33"> that 'y' is not accessible to S in UK3. The question hem is why seemingly impossible link is allowed in (14-a) but not in (14-b).</Paragraph> <Paragraph position="34"> It has been known that there arc several cascs in which the accessibility condition violation does not result in infelicitous anaphoric links. Roberts (1987) provides a DRT based analysis for a similar phenomenon with pronominal anaphora. Consider the following example.</Paragraph> <Paragraph position="35"> (15) I larvey courts a girl at every convention.</Paragraph> <Paragraph position="36"> She always comes to the banquet with him.</Paragraph> <Paragraph position="37"> She claims that a pronominal 'she' is linked to 'a girl' since tile second sentence is modally subordinated in the conse- null quent of the DR conditional in the first sentence. In other words, the second sentence is under the quantification of the situation in the first sentence due to the lact that the modalily appears in the second sentence. Notice that even in (15), an incidence of FI)D is apparent: 'the banquet' is functionally linked to 'convention'.</Paragraph> <Paragraph position="38"> In (14-a), we c,'m safely assume that the consequent sentence is subordinate to the DR condilional' s consequent, while in (14-b) such explanation does not seem to be available. Once again, we see a certain simih-uity bctwecn pronominal anaphora and situational role lq)D.</Paragraph> <Paragraph position="39"> 4.3. FDD Imsed on Temporal/Lncational Relations The third group to consider here consists of the following kinds of expressions.</Paragraph> <Paragraph position="40"> (16) a. the mottling of December 31 the spring of 1988 b. the midst of Rmnadan the beginning of thc war (17) a. the top of the house the edge of the bridge the side of the cat' the bottom of the bottle b. the north of London c. the middle of the bridge Note that all of the head nouns in the above FDI) denote either temporal as in (16) or locational points as in (17) with respect to oilier temporal ~w locative lXfints. What makes these FI)D distinctive from file two other relational Fl)l)s examiucd so lar is 1) that Ihey caimot be p,'uaphrascd by possessivc conslnlclion of the form &quot;X's Y&quot; although &quot;the Y of X&quot; form is acceptable as shown at)eve; that is, they are no| lexical relations, and 2) that they arc not situaliomd roles. Furthcrmorc, as a group, these nouns typically link to functional antecedents that arc anchored expressions. Notc that this obscrvation allows us It consider certain close relationship bctwccn this subclass and expressions such as l~)llows: (18) the city of New York the port of Los Angeles Exprcssions in (18) arc usually considcrcd as proper names, i.e., anchored cxprcssions. Noncthclcss, it is possible to considcr 'lhc city o1&quot; as a fnnclion.</Paragraph> </Section> <Section position="2" start_page="792" end_page="793" type="sub_section"> <SectionTitle> 4.4. DRT Treatment of Relational FI)D </SectionTitle> <Paragraph position="0"> All of the relational FDI)s are translated into non-monadic relational predicates. The number of argmnents depend upon relations toxically sp(y.:ificd in the lexicon for relational nonns, l:or example, most ()1' kinship terms are dyadic predicates but some derived nominals will have the same numhcr of arguments as the mmlber of arguments that Iheir verbal countcrpmls lX)SSeSS.</Paragraph> <Paragraph position="1"> Nonetheless, at the time of traqslating a telatiomd noun, whether or not the noun is used anaphorically, functional anaphorically, or non-anaphorically is not known. &quot;l'herefore, we cannot select an appropriate DRS construction principle at the time ofWanslation of FDD. What I would like to propose is that we lranslate FDD into non-monadie relation predicates but do nothing more than the translation at this time. I hypothesize that any un-instantiated referential argument introduces all anaphoric type reference marker.</Paragraph> <Paragraph position="2"> This rcfcrencc marker can be processed further in three ways: finding its antecedent, linding its functional antecedcnt, and finding neithcr its m~teeedent nor its functional antecedent. The lbllowing FDD construction rule states the above scnario.</Paragraph> <Paragraph position="3"> (19) FDD CR Given a relational FDD phrase 'the N', 1) Introduce a relational condition, R, with an appropriate argument structure in Con K.</Paragraph> <Paragraph position="4"> 2) Introduce a new reference marker, n, for a principle m'gument of R in U K.</Paragraph> <Paragraph position="5"> 3) Introduce a set of new reference markers for the rest of the argnments of R in U K.</Paragraph> <Paragraph position="6"> 4) Substitute u for 'the N' in p.</Paragraph> <Paragraph position="7"> Let us take an cx,'unple of relational IT)l) and see how the abeve CR will be applied.</Paragraph> <Paragraph position="8"> (20)/lsmdly if John buys an interesting book, he writes to the author.</Paragraph> <Paragraph position="9"> At the time of translating the phrase, 'to the author', we have the following DRS under constx~ction.</Paragraph> <Paragraph position="11"> book(M) I v / write(e2,x,Y) interesting(al) \] \[ 'to the author' Since 'author' is a relational noun, the lexicon provides information coneenting its lexical denotation of the relation, n,'unely a predicate 'author' with two argument position. The above lq)D conslrnction rule produces a DRS as follows:</Paragraph> <Paragraph position="13"> At Ihis moincnt, three things must be done: 1) resolve 'x', which is lk)r the pronoun, 'he', 2) resolve 'thel', which strands lot the entity of'authod, and 3) resolve 'y', that stands for the fimctional antecedent lot 'author'.</Paragraph> <Paragraph position="14"> Resolving 'x' with 'j' is trivial. 'j' is the only entity that is accessible to 'x' with gender, number satisfaction. }low- null ever, 'thel' cannot be resolved with any antecedent since there are no previously mentioned 'author' or deictically salient antecedent available from the above DRS. What we have to do is to accommodate it. Thus, we leave 'thel' in UK2 without doing anything. Now, we resolve 'y' with 'al', an entity that stands for 'book'. 'al' is accessible to 'y' due to the extension ofDRS K 1 to K 2. Note that this functional link is only possible when the lexicon provides common sense information that specifies the kind of relation that is held between 'book' and 'author'. After these resolution operations, we have the following completed DRS.</Paragraph> <Paragraph position="15"> What happens when there is a previously mentioned phrase? Consider the following example.</Paragraph> <Paragraph position="16"> (22) Usually if John meets the author of a book, he praises the author.</Paragraph> <Paragraph position="17"> The first incident of 'author' is the case of explicit FDD in Kmnp (1984). Its functional antecedent is provided by 'of' phrase. The second occurrence of 'author' is non-functionally linked to the initial mention of 'author'. I assume that the second incidence of 'author' introduces the same relational condition. And, due to the non-functional link to the initial incidence of 'author', the arguments will be filled with the exact copies of the arguments of the initial 'author'. Therefore, we have the following DRS.</Paragraph> <Paragraph position="19"> Note that when 'y' is linked to 'al' via a regular anaphoric link of 'the2' to 'thel', it is not nccessury to invoke another anapbora resolution procedure for 'y'. The situation is just the s,'une ,'ks in the case of explicit FDD such as 'the mother of John'. The first argument of a relation 'mother(A,thel)' is syntactically connected to entity denoted by the 'og phrase.</Paragraph> </Section> </Section> <Section position="5" start_page="793" end_page="794" type="metho"> <SectionTitle> 5. NON-RELATIONAL FDD </SectionTitle> <Paragraph position="0"> A group of FDD that we call non-relational FDD are IT)Ds such as &quot;the tallest man&quot;, &quot;the third book&quot;, and &quot;the same girl&quot;. They all lake obligatory definite articles but are lirst mentions without having regular sense of antecedents.</Paragraph> <Paragraph position="1"> However, the references of the three exmnples are determined with respect to some sets of individuals in the discourse. This cetlainly satisfies our definition of FDD.</Paragraph> <Paragraph position="2"> Nonetheless, the function and the mechanism underlying the function differ substantively from relational FDD discussed so far. In this section, we will see how they differ and I will sketch a treatment for this class of FDD briefly. I would like to refer readers to the work (Wada (forthcoming)) for further and detailed discussion on non-relational FDD.</Paragraph> <Paragraph position="3"> First of all, relational H)D and non-relation FDD differ structurally. A typical syntactic structure of relational FDD is an NP of the form \[np \[dot the\], \[nbar \[n A\]\]\]. N (lexically 'A' in the form) must be a relational noun and it can take a complement 'of' phrase. Non-relational FDD, on the other hand, must contain an adjunct phrase headed by one of elements such as superlative adjectives, ordinal numerals, and identifying adjectives like 'very' and 'same'. Thus, its typical form is \[np \[det the\], \[nbar \[x, 'A'\], \[n 'B'\]\]\]. 'A' is the FDD licensing lexical element and 'B' is any noun.</Paragraph> <Paragraph position="4"> Second, the functional link of non-relational FDD is licensed by a head of the adjunct phrase such as a superlative adjective, an ordinal number, or an identifying adjective rather than ahead noun (relationalnoun) ofFDD. I call these heads of adjunct phrases functional modifiers. Functional modifiers denote functions that introduce discourse referents which will be resolved with functional m~tecedents in a discourse.</Paragraph> <Paragraph position="5"> Third, non-relational FDD takes a selection set instead of an object as their functional antecedents. A selection set is a maximal set of objects in a given discourse that satisfies descriptions in the head nouns of the non-relational FDD.</Paragraph> <Paragraph position="6"> The phenomenon is reminiscent to plural anaphora where the antecedents are usually maximal sets.</Paragraph> <Paragraph position="7"> Finally, the uniqueness presupposition of the referent of non-relational FDD is provided by the lexical/semantic characteristics of the functional modifiers while it is due to the lexical relation of the head noun in relational l~7)D.</Paragraph> <Paragraph position="8"> In this paper, we examine three subclasses of this FDD: superlatives such as 'the strongest man', ordinal ntmaber modification such as 'the third book', and identifying adjective modification such as 'the stone car'.</Paragraph> <Section position="1" start_page="793" end_page="794" type="sub_section"> <SectionTitle> Basic Analysis of Non-relational FDD </SectionTitle> <Paragraph position="0"> Kamp (1984) provides an analysis of superlative constructions. In that, he treats superlatives as a set of comparatives under a universal quantifier that introduces a DRT-conditional. The set of comparatives are distributed over members of the selection set given in the discourse.</Paragraph> <Paragraph position="1"> Kamp shows this selection sot as analogous to 'among them'.</Paragraph> <Paragraph position="2"> Consider the following.</Paragraph> <Paragraph position="3"> (24) Three men came to harvest rice.</Paragraph> <Paragraph position="4"> The weakest man operated a combine machine.</Paragraph> <Paragraph position="5"> The superlative 'the weakest man' takes maximal set of 'the three men who came to harvest rice' as its functional antecedent and returns the unique individual thatsatisfies a condition that this individual is weaker than any member of the set except himself. The following DRS represents Kamp's In our treatment, we add a dyadic predicate 'weake st(X, the 1 )' to the above representation to indicate that the the set of three men is the functional antecedent of this lq)D and the set is dislxibuted over members of the set.</Paragraph> <Paragraph position="6"> Both ordinal numeral phrases and identifying adjectives a~e analyzed to have the s,'une logical structures as the superlative adjectives discussed above. The ordinal numeral case is based on enumeration operations via DRT-conditional instead of a set of comparisons under an universal quantifier. Once enumeration operation takes place, each of the members of a set can be refen'ed with an index, the ordinal number. The identifying adjective case is treated exactly like the case of superlatives.</Paragraph> </Section> </Section> <Section position="6" start_page="794" end_page="794" type="metho"> <SectionTitle> 6. IMPI,EMENTATION dE FDD RESOLUTION </SectionTitle> <Paragraph position="0"> 1TP's Interprctext natural language understanding system has been under development ill the past few years. We reported some early results in MUC3 Conference and elsewhere (Dahlgren, et at. 1991). The system includes a large Naive Semantic lexicon, a principle-based wide coverage parser with a sense disambiguation mechanism, a DRS construction module, an anaphora resolver, mid lexical and discourse database handlers. FDI) resolution was implemented as part of a large anaphora resolution mechanism.</Paragraph> <Paragraph position="1"> hi file implementation, searching order antong hypotheses is very hnportmit. Certain ordering eliminates possibilities of available resolution, mid other cases cause increase ill processing load of the resolution. Thus, our goal is to reduce the processing load as much as possible by reducing search space and to reduce resolution errors by setting item specific hypotheses ordering at the stone time.</Paragraph> <Paragraph position="2"> All of tim FDI) arc potential instances of shnple subsequent-mention I)D. For exmnplc, a DI) with a lexical relational noun such as 'the mother' may have a regular antecedent such as 'a mother' in the discourse. Although first-mention DDs that include l:Dl)s arc statistically more common than typical subsequent-mention l)Ds, a possibility of being subsequent-mcntion DD should be tested first, qhis is because of tile fact that the range of description satisfying antecedents are more constrained than the range of functionally satisfying antecedents.</Paragraph> <Paragraph position="3"> Generally, we should hypothesize more restricted assumption before general ones. In our implementation, each FDI) subckLss possesses its own set of ordered hypotheses.</Paragraph> <Paragraph position="4"> For cxmnple, temporal/Ideational relation FDD typically have anchored expressions as their lunctitmal antecedents and our resolution module searches in this restricted area (an anchored object Its0 before it searches in the previous discourse. For another exmnple, situational role FDD checks simple subsequent-mention case first; if it falls, then it tries to find a situation-trigger in the accessible universe of discourse. Cun'ently, we have been conducting a large scale evaluation on anaphora resolution.</Paragraph> </Section> class="xml-element"></Paper>