<?xml version="1.0" standalone="yes"?>
<Paper uid="J98-2001">
  <Title>A Corpus-based Investigation of Definite Description Use</Title>
  <Section position="2" start_page="0" end_page="190" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> The work presented in this paper was inspired by the growing realization in the field of computational linguistics of the need for experimental evaluation of linguistic theories--semantic theories, in our case. The evaluation we are considering typically takes the form of experiments in which human subjects are asked to annotate texts from a corpus (or recordings of spoken conversations) according to a given classification scheme, and the agreement among their annotations is measured (see, for example, Passonneau and Litman 1993 or the papers in Moore and Walker 1997). These attempts at evaluation are, in part, motivated by the desire to put these theories on a more &amp;quot;scientific&amp;quot; footing by ensuring that the semantic judgments on which they are based reflect the intuitions of a large number of speakers; 1 but experimental evaluation is also seen as a necessary precondition for the kind of system evaluation done, for example, in the Message Understanding initiative (MUC), where the performance of a system is evaluated by comparing its output on a collection of texts with a standardized annotation of those texts produced by humans (Chinchor and Sundheim 1995). Clearly, 1 For example, recent work in linguistics shows that agreement with a theory's predictions may be a matter of how well the actual behavior distributes around the predicted behavior, rather than an all-or-nothing affair (Bard, Robertson, and Sorace 1996).</Paragraph>
    <Paragraph position="1"> (~) 1998 Association for Computational Linguistics Computational Linguistics Volume 24, Number 2 a MUC-style evaluation presupposes an annotation scheme on which all participants agree.</Paragraph>
    <Paragraph position="2"> Our own concern are semantic judgments concerning the interpretation of noun phrases with the definite article the, that we will call definite descriptions, following (Russell 1919). 2 These noun phrases are one of the most common constructs in English, 3 and have been extensively studied by linguists, philosophers, psychologists, and computational linguists (Russell 1905; Christophersen 1939; Strawson 1950; Clark 1977; Grosz 1977; Cohen 1978; Hawkins 1978; Sidner 1979; Webber 1979; Clark and Marshall 1981; Prince 1981; Heim 1982; Appelt 1985; L6bner 1985; Kadmon 1987; Carter 1987; Bosch and Geurts 1989; Neale 1990; Kronfeld 1990; Fraurud 1990; Barker 1991; Dale 1992; Cooper 1993; Kamp and Reyle 1993; Poesio 1993).</Paragraph>
    <Paragraph position="3"> Theories of definite descriptions such as (Christophersen 1939; Hawkins 1978; Webber 1979; Prince 1981; Helm 1982) identify two subtasks involved in the interpretation of a definite description: deciding whether the definite description is related to an antecedent in the text4--which in turn may involve recognizing fairly fine-grained distinctions-and, if so, identifying this antecedent. Some of these theories have been cast in the form of classification schemes (Hawkins 1978; Prince 1992), and have been used for corpus analysis (Prince 1981, 1992; Fraurud 1990); 5 yet, we are aware of no attempt at verifying whether subjects not trained in linguistics are capable of recognizing the proposed distinctions, which is a precondition for using these schemes for the kind of large-scale text annotation exercises necessary to evaluate a system's performance, as done in MUC.</Paragraph>
    <Paragraph position="4"> In the past two or three years, this kind of verification has been attempted for other aspects of semantic interpretation: by Passonneau and Litman (1993) for segmentation and by Kowtko, Isard, and Doherty (1992) and Carletta et al. (1997) for dialogue act annotation. Our intention was to do the same for definite descriptions.</Paragraph>
    <Paragraph position="5"> We ran two experiments to determine how good naive subjects are at doing the form of linguistic analysis presupposed by current schemes for classifying definite descriptions. (By &amp;quot;how good&amp;quot; here we mean &amp;quot;how much do they agree among themselves?&amp;quot; as commonly assumed in work of this kind.) Our subjects were asked to classify the definite descriptions found in a corpus of natural language texts according to classification schemes that we developed starting from the taxonomies proposed by Hawkins {'.1978) and Prince (1981, 1992), but which took into account our intention of having naive speakers perform the classification. Our experiments were also designed to assess the feasibility of a system to process definite descriptions on unrestricted text and to collect data that could be used for this implementation. For both of these reasons, the classification schemes that we tried differ in several respects from those adopted in prior corpus-based studies such as Prince (1981) and Fraurud (1990). Our study is also different from these previous ones in that measuring the agreement among annotators became an issue (Carletta 1996).</Paragraph>
    <Paragraph position="6"> For the experiments, we used a set of randomly selected articles from the Wall Street Journal contained in the ACL/DCI CD-ROM, rather than a corpus of transcripts of spoken language corpora such as the HCRC MapTask corpus (Anderson et al. 1991) or the TRAINS corpus (Heeman and Allen 1995). The main reason for this choice was 2 We will not be concerned with other cases of definite noun phrases such as pronouns, or possessive descriptions; hence the term definite description rather than the more general term definite NP. 3 The word the is by far the most common word in the Brown corpus (Francis and Kucera 1982), the LOB corpus 0ohansson and Hofland 1989), and the TRAINS corpus (Heeman and Allen 1995).</Paragraph>
    <Paragraph position="7"> 4 We concentrated on written texts in this study. See discussion below. 5 Both Prince's and Fraurud's studies are analyses of the use of the whole range of definite NPs, not just of definite descriptions.</Paragraph>
    <Paragraph position="8">  Poesio and Vieira A Corpus-based Investigation of Definite Description Use to avoid dealing with deictic uses of definite descriptions and with phenomena such as reference failure and repair. A second reason was that we intended to use computer simulations of the classification task to supplement the results of our experiments, and we needed a parsed corpus for this purpose; the articles we chose were all part of the Penn Treebank (Marcus, Santorini, and Marcinkiewicz 1993).</Paragraph>
    <Paragraph position="9"> In the remainder of the paper, we review two existing classification schemes in Section 2 and then discuss our two classification experiments in Sections 3 and 4.</Paragraph>
    <Paragraph position="10"> 2. Towards a Classification Scheme: Linguistic Theories of Definite Descriptions When looking for an annotation scheme for definite descriptions, one is faced with a wide range of options. At one end of the spectrum there are mostly descriptive lists of definite description uses, such as those in Christophersen (1939) and Hawkins (1978), whose only goal is to assign a classification to all uses of definite descriptions. At the other end, there are highly developed formal analyses, such as Russell (1905), Heim (1982), L6bner (1985), Kadmon (1987), Neale (1990), Barker (1991), and Kamp and Reyle (1993), in which the compositional contribution of definite descriptions to the meaning of an utterance, as well as their truth-conditional properties, are spelled out in detail. These more formal analyses are concerned with questions such as the quantificational or nonquantificational status of definite descriptions and the proper treatment of presuppositions, but tend to concentrate on a subset of the full range of definite description use. Among the more developed semantic analyses, some identify uniqueness as the defining property of definite descriptions (Russell 1905; Neale 1990), whereas others take familiarity as the basis for the analysis (Christophersen 1939; Hawkins 1978; Heim 1982; Prince 1981; Kamp and Reyle 1993). We will say more about some of these analyses below.</Paragraph>
    <Paragraph position="11"> Our choice of a classification scheme was dictated in part by the intended use of the annotation, in part by methodological considerations. An annotation used to evaluate the performance of a system ought to identify the anaphoric connections between discourse entities; this makes familiarity-based analyses more attractive. From a methodological point of view, it was important to choose an annotation scheme that (i) would make the classification task doable by subjects not trained in linguistics, and (ii) had already been applied to the task of corpus analysis. We felt that we could ask naive subjects to assign each definite description to one of a few classes and to identify its antecedent when appropriate; we also wanted an annotation scheme that would characterize the whole range of definite description use, so that we would not need to worry about eliminating definite .descriptions from our texts because they were unclassifiable.</Paragraph>
    <Paragraph position="12"> For these reasons, we chose Hawkins's list of definite description uses (Hawkins 1978) and Prince's taxonomy (Prince 1981, 1992) as our starting point, and developed from there two slightly different annotation schemes, which allowed us to see whether it was better to describe the classes to our annotators in a surface-oriented or a semantic fashion, and to evaluate the seriousness of the problems with these schemes identified in the literature (see Fraurud 1990).</Paragraph>
    <Section position="1" start_page="184" end_page="187" type="sub_section">
      <SectionTitle>
2.1 The Christophersen/Hawkins List of Definite Description Uses
</SectionTitle>
      <Paragraph position="0"> The wide range of uses of definite descriptions was already highlighted in Christophersen (1939). In the third chapter of his book, Hawkins (1978) further develops and extends Christophersen's list. He identifies the following classes, or uses, of definite descriptions:  Computational Linguistics Volume 24, Number 2 Anaphoric Use. These are definite descriptions that cospecify with a discourse entity already introduced in the discourse. 6 The definite description may use the same descriptive predicate as its antecedent, or any other capable of indicating the same  antecedent (e.g., a synonym, a hyponym, etc.).</Paragraph>
      <Paragraph position="1"> (1) a. Fred was discussing an interesting book in his class. I went to discuss the book with him afterwards.</Paragraph>
      <Paragraph position="2"> b. Bill was working at a lathe the other day. All of a sudden the machine stopped turning.</Paragraph>
      <Paragraph position="3"> c. Fred was wearing trousers. The pants had a big patch on them. d. Mary travelled to Paris. The journey lasted six hours.</Paragraph>
      <Paragraph position="4"> e. A man and a woman entered a restaurant. The couple was received by a waiter.</Paragraph>
      <Paragraph position="5"> Immediate Situation Uses. The next two uses of definite descriptions identified by  Hawkins are occurrences used to refer to an object in the situation of utterance. The referent may be visible, or its presence may be inferred. The visible situation use occurs when the object referred to is visible to both speaker and hearer, as in the following examples:  (2) a. Please, pass me the salt.</Paragraph>
      <Paragraph position="6"> b. Don't break the vase.</Paragraph>
      <Paragraph position="7"> Hawkins classifies as immediate situation uses those definite descriptions whose referent is a constituent of the immediate situation in which the use of the definite description is located, without necessarily being visible: (3) a. Beware of the dog.</Paragraph>
      <Paragraph position="8"> b. Don't feed the pony.</Paragraph>
      <Paragraph position="9"> c. You can put your coat on the clothes peg.</Paragraph>
      <Paragraph position="10"> d. Mind the step.</Paragraph>
      <Paragraph position="11"> 6 There are some complex terminological problems when discussing anaphoric expressions. Following  standard terminology, we will use the term referent to indicate the object in the world that is contributed to the meaning of an utterance by a definite description---e.g., we will say that Bill Clinton is the referent of a referential use of the definite description the president of the USA in 1997. We will then say, following Sidner's terminology (Sidner 1979), that a definite description cospecifies with its antecedent in a text, when such antecedent exists, if the definite description and its antecedent denote the same object. This is probably the most precise way of referring to the relation between an anaphoric expression and its antecedent; note that two discourse entities can cospecify without referring to any object in the world--e.g., in The (current) king of France is bald. He has a double chin, as well., he cospecifies with the (current) king of France, but this latter expression does not refer to anything. However, since we will mostly be concerned with referential discourse entities, we will often use the term corefer instead of cospecify. Apart from this, we have tried to avoid more complex issues of reference insofar as possible (Donnellan 1972; Kripke 1977; Barwise and Perry 1983; Neale 1990; Kronfeld 1990).  Poesio and Vieira A Corpus-based Investigation of Definite Description Use Larger Situation Uses. Hawkins lists two uses of definite descriptions characteristic of situations in which the speaker appeals to the hearer's knowledge of entities that exist in the nonimmediate or larger situation of utterance--knowledge they share by being members of the same community, for instance.</Paragraph>
      <Paragraph position="12"> A definite description may rely on specific knowledge about the larger situation: this is the case in which both the speaker and the hearer know about the existence of the referent, as in the example below, in which it is assumed that speaker and hearer are both inhabitants of Halifax, a town which has a gibbet at the top of Gibbet Street: (4) The Gibbet no longer stands.</Paragraph>
      <Paragraph position="13"> Specific knowledge is not, however, a necessary part of the meaning of larger situation uses of definite descriptions. While some hearers may have specific knowledge about the actual individuals referred to by a definite description, others may not. General knowledge about the existence of certain types of objects in certain types of situations is sufficient. Hawkins classifies those definite descriptions that depend on this knowledge as instances of general knowledge in the larger situation use. An example is the following utterance in the context of a wedding: (5) Have you seen the bridesmaids? Such a first-mention of the bridesmaids is possible on the basis of the knowledge that weddings typically have bridesmaids. In the same way, a first-mention of the bride, the church service, or the best man would be possible.</Paragraph>
      <Paragraph position="14"> Associative Anaphoric Use. Speaker and hearer may have (shared) knowledge of the relations between certain objects (the triggers) and their components or attributes (the associates): associative anaphoric uses of definite descriptions exploit this knowledge. Whereas in larger situation uses the trigger is the situation itself, in the associative anaphoric use the trigger is an NP introduced in the discourse.</Paragraph>
      <Paragraph position="15"> (6) a. The man drove past our house in a car. The exhaust fumes were terrible. b. I am reading a book about Italian history. The author claims that Ludovico il Moro wasn't a bad ruler. The content is generally interesting.</Paragraph>
      <Paragraph position="16"> c. I went to a wedding last weekend. The bride was a friend of mine. She baked the cake herself.</Paragraph>
      <Paragraph position="17"> Unfamiliar Uses. Hawkins classifies as unfamiliar those definite descriptions that are not anaphoric, do not rely on information about the situation of utterance, and are not associates of some trigger in the previous discourse. Hawkins groups these definite descriptions in classes according to their syntactic and lexical properties, as follows: NP complements. One form of unfamiliar definite descriptions is characterized by the presence of a complement to the head noun.</Paragraph>
      <Paragraph position="18"> (7) a. Bill is amazed by the fact that there is so much life on Earth.</Paragraph>
      <Paragraph position="19"> b. The philosophical aphasic came to the conclusion that language did not exist.</Paragraph>
      <Paragraph position="20">  Computational Linguistics Volume 24, Number 2 c. Fleet Street has been buzzing with the rumour that the Prime Minister is going to resign.</Paragraph>
      <Paragraph position="21"> d. I remember the time when I was a little girl.</Paragraph>
      <Paragraph position="22"> Nominal modifiers. The distinguishing feature of these phrases, according to Hawkins, is the presence of a nominal modifier that refers to the class to which the head noun belongs.</Paragraph>
      <Paragraph position="23">  (8) a. I don't like the colour red.</Paragraph>
      <Paragraph position="24"> b. The number seven is my lucky number.</Paragraph>
      <Paragraph position="25"> Referent-Establishing Relative Clauses. Relative clauses may establish a referent for the hearer without a previous mention, when the relative clause refers to something mutually known.</Paragraph>
      <Paragraph position="26"> (9) a. What's wrong with Bill? Oh, the woman he went out with last night was nasty to him. (But: ?? Oh, the woman was nasty to him.) b. The box (that is) over there Associative clauses. Some definite descriptions can be seen as cases of bridging references in which both the trigger and the associate are specified. The modifiers of the head noun specify the set of objects with which the referent of the definite description is associated.</Paragraph>
      <Paragraph position="27"> (10) a. I remember the beginning of the war very well.</Paragraph>
      <Paragraph position="28"> b. There was a funny story on the front page of the Guardian this morning. c .... the bottom of the sea.</Paragraph>
      <Paragraph position="29"> d .... thefight during the war.</Paragraph>
      <Paragraph position="30"> Unexplanatory Modifiers Use. Finally, Hawkins lists a small number of modifiers that require the use of the: (11) a. My wife and I share the same secrets.</Paragraph>
      <Paragraph position="31"> b. Thefirst person to sail to America was an Icelander.</Paragraph>
      <Paragraph position="32"> c. The fastest person to sail to America...</Paragraph>
    </Section>
    <Section position="2" start_page="187" end_page="188" type="sub_section">
      <SectionTitle>
2.2 The Semantics of Definite Descriptions
</SectionTitle>
      <Paragraph position="0"> Some of the classes in the Christophersen/Hawkins classification are specified in a semantic fashion; other classes are defined in purely syntactic terms. It is natural to ask what these uses of definite descriptions have in common from a semantic point of view: for example, is there a connection between the unfamiliar and unexplanatory uses of definite descriptions and the other uses? (The unfamiliar uses with associative clauses seem related to the associative anaphoric ones, and both seem related to the uses based on referent-establishing relative clauses.) Many authors, including Hawkins himself, have attempted to go beyond the purely descriptive list just discussed.</Paragraph>
      <Paragraph position="1">  Poesio and Vieira A Corpus-based Investigation of Definite Description Use One group of authors have identified uniqueness as the defining property of definite descriptions. This idea goes back to Russell (1905), and is motivated by larger situation definite descriptions such as the pope and by some cases of unexplanatory modifier use such as thefirst person to sail to America. The hypothesis was developed in recent years to address the problem of uniqueness within small situations (Kadmon 1987; Neale 1990; Cooper 1993). 7 Another line of research is based on the observation that many of the uses of definite descriptions listed by Hawkins have one property in common: the speaker (or writer) is making some assumptions about what the hearer (or reader) already knows.</Paragraph>
      <Paragraph position="2"> Speaking very loosely, we might say that the speaker assumes that the hearer is able to &amp;quot;identify&amp;quot; the referent of the definite description. This is also true of some of the uses Hawkins classified as unfamiliar, such as his nominal modifiers and associative clause classes. Attempts at making this intuition more precise include Christophersen's (1939) familiarity theory, Strawson's (1950) presuppositional theory of definite descriptions, Hawkins's (1978) location theory and its revision, Clark and Marshall's (1981) theory of definite reference and mutual knowledge, as well as more formal proposals such as Heim (1982).</Paragraph>
      <Paragraph position="3"> Neither the uniqueness nor the familiarity approach have yet succeeded in providing a satisfactory account of all uses of definite descriptions (Fraurud 1990; Birner and Ward 1994). However, the theories based on familiarity address more directly the main concern of NLP system designers, which is to identify the connections between discourse entities. Furthermore, the prior corpus-based studies of definite descriptions use that we are aware of (Prince 1981, 1992; Fraurud 1990) are based on theories of this type. For both of these reasons, we adopted semantic notions introduced in familiaritystyle accounts in designing our experiments--in particular, distinctions introduced in Prince's taxonomy.</Paragraph>
    </Section>
    <Section position="3" start_page="188" end_page="189" type="sub_section">
      <SectionTitle>
2.3 Prince's Classification of Noun Phrases
</SectionTitle>
      <Paragraph position="0"> Prince studied in detail the connection between a speaker/writer's assumptions about the hearer/reader and the linguistic realization of noun phrases (Prince 1981, 1992).</Paragraph>
      <Paragraph position="1"> She criticizes as too simplistic the binary distinction between given and new discourse entities that is at the basis of most previous work on familiarity, and proposes a much more detailed taxonomy of &amp;quot;givenness&amp;quot;---or, as she calls it, assumed familiarity--meant to address this problem. Also, Prince's analysis of noun phrases is closer than the Christophersen/Hawkins taxonomy to a classification of definite descriptions on purely semantic terms: for example, she relates unfamiliar definites based on referent-establishing relative clauses with Hawkins's associative clause and associative anaphoric uses. 8 Hearer-New~Hearer-Old. One factor affecting the choice of a noun phrase, according to Prince, is whether a discourse entity is old or new with respect to the hearer's knowledge. A speaker will use a proper name or a definite description when he or she assumes that the addressee already knows the entity whom the speaker is referring to, as in (12) and (13).</Paragraph>
      <Paragraph position="2">  (12) I'm waiting for it to be noon so I can call Sandy Thompson.</Paragraph>
      <Paragraph position="3"> 7 LObner (1985) generalizes this idea, with good results; we will return to this work later. 8 Clark and Marshall (1981) also proposed a revision of Hawkins' theory that merges some of the classes on semantic grounds.</Paragraph>
      <Paragraph position="4">  Computational Linguistics Volume 24, Number 2 (13) Nine hundred people attended the Institute.</Paragraph>
      <Paragraph position="5"> On the other hand, if the speaker believes that the addressee does not know of Sandy Thompson, an indefinite will be used: (14) I'm waiting for it to be noon so I can call someone in California.</Paragraph>
      <Paragraph position="6"> Discourse-New~Discourse-Old. In addition, discourse entities can be new or old with respect to the discourse model: an NP may refer to an entity that has already been evoked in the current discourse, or it may evoke an entity that has not been previously mentioned. Discourse novelty is distinct from hearer novelty: both Sandy Thompson in (12) and the someone in California mentioned in (14) may well be discourse-new even if only the second one will be hearer-new. On the other hand, for an entity, being discourse-old entails being hearer-old. In other words, in Prince's theory, the notion of familiarity is split in two: familiarity with respect to the discourse, and familiarity with respect to the hearer. Either type of familiarity can license the use of definites: Hawkins's anaphoric uses of definite descriptions are cases of noun phrases referring to discourse-old discourse entities, whereas his larger situation and immediate situation uses are cases of noun phrases referring to discourse-new, hearer-old entities. 9 lnferrables. The uses of definite descriptions that Hawkins called associative anaphoric, such as a book ... the author, are not discourse-old or even hearer-old, but they are not entirely new, either; as Hawkins pointed out, the hearer is assumed to be capable of inferring their existence. Prince called these discourse entities inferrables. (This is the class of definite descriptions for which Clark \[1977\] used the term bridging references.) Containing Inferrables. Finally, Prince proposes a category for noun phrases that are like inferrables, but whose connection with previous hearer's knowledge is specified as part of the noun phrase itself--her example is the door of the Bastille in the following example: (15) The door of the Bastille was painted purple.</Paragraph>
      <Paragraph position="7"> At least three of the unfamiliar uses of Hawkins--NP complements, referent-establishing relative clauses, and associative clauses--fall into this category. (See also Clark and Marshall \[1981\].)</Paragraph>
    </Section>
    <Section position="4" start_page="189" end_page="190" type="sub_section">
      <SectionTitle>
2.4 Some Remarks about Coverage
</SectionTitle>
      <Paragraph position="0"> Perhaps the most important question concerning a classification scheme is its coverage.</Paragraph>
      <Paragraph position="1"> The two taxonomies we have just seen are largely satisfactory in this respect, but a couple of issues are worth mentioning.</Paragraph>
      <Paragraph position="2"> First of all, Prince's taxonomy does not give us a complete account of the licensing conditions for definite descriptions. Of the uses mentioned by Hawkins, the unfamiliar definites with unexplanatory modifiers and NP complements need not satisfy any of the conditions that license the use of definites according to Prince: these definites are  9 In Clark and Marshall's (1981) terminology, one would say that different copresence heuristics can be used to establish mutual knowledge.</Paragraph>
      <Paragraph position="3">  Poesio and Vieira A Corpus-based Investigation of Definite Description Use not necessarily discourse-old, hearer-old, inferrables, or containing inferrables. These uses fall outside of Clark and Marshall's classification, as well.</Paragraph>
      <Paragraph position="4"> Secondly, none of the classification schemes just discussed, nor any of the alternatives proposed in the literature, consider so-called generic uses of definite descriptions, such as the use of the tiger in the generic sentence The tiger is a fierce animal that lives in the jungle. The problem with these uses is that the very question of whether the &amp;quot;referent&amp;quot; is familiar or not seems misplaced--these uses are not &amp;quot;referential.&amp;quot; A problem related to the one just mentioned is that certain uses of definite descriptions are ambiguous between a referential and an attributive interpretation (Donnellan 1972). The sentence The first person to sail to America was an Icelander, for example, can have two interpretations: the writer may either refer to a specific person, whose identity may be mutually known to both writer and reader; or he or she may be simply expressing a property that is true of the first person to sail to America, whoever that person happened to be. This ambiguity does not seem to be possible with all uses of definite descriptions: e.g., pass me the salt seems only to have a referential use. Again, the schemes we have presented do not consider this issue. The question of how to annotate generic uses of definite descriptions or uses that are ambiguous between a referential and an attributive use will not be addressed in this paper.</Paragraph>
    </Section>
    <Section position="5" start_page="190" end_page="190" type="sub_section">
      <SectionTitle>
2.5 Fraurud's Study
</SectionTitle>
      <Paragraph position="0"> A second problem with the classification schemes we have discussed was raised by Fraurud in her study of definite NPs in a corpus of Swedish text (Fraurud 1990). Fraurud introduced a drastically simplified classification scheme based on two classes only: subsequent-mention, corresponding to Hawkins's anaphoric definite descriptions and Prince's discourse-old, and first-mention, including all other definite descriptions.</Paragraph>
      <Paragraph position="1"> Fraurud simplified matters in this way because she was primarily interested in verifying the empirical basis for the claim that familiarity is the defining property of definite descriptions; she also observed, however, that some of the distinctions introduced by Hawkins and Prince led to ambiguities of classification. For example, she observed that the reader of a Swedish newspaper can equally well interpret the definite description the king in an article about Sweden by reference to the larger situation or to the content of the article.</Paragraph>
      <Paragraph position="2"> We took into account Fraurud's observations in designing our experiments, and we will compare our results to hers below.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>