File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/96/p96-1035_metho.xml
Size: 23,859 bytes
Last Modified: 2025-10-06 14:14:19
<?xml version="1.0" standalone="yes"?> <Paper uid="P96-1035"> <Title>Resolving Anaphors in Embedded Sentences</Title> <Section position="4" start_page="0" end_page="264" type="metho"> <SectionTitle> 2 Intrasentential Antecedents </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="0" end_page="0" type="sub_section"> <SectionTitle> 2.1 Embedded sentences and elementary </SectionTitle> <Paragraph position="0"> events An embedded sentence contains either more than one verb or a verb and derivations of other verbs (see sentence 1 with verbs said and forming). 1) Three of the world's leading advertising groups, Agence Havas S.A. of France, Young & Rubicam of the U.S. and Dentsu Inc. of Japan, said they are forming a global advertising joint venture.</Paragraph> <Paragraph position="1"> Broadly speaking embedded sentences concern more than one fact. In sentence 1 there is the fact of saying something and that of forming a joint venture. We call such a fact an elementary event (EE hereafter). Thus an embedded sentence will contain several EEs. Factors that influence embedded sentences are mainly semantic features of verbs. For example the verb to say, that takes a sentence complement favours an introduction of a new fact, i.e., &quot;to say something&quot;, and the related fact. There are other classes of verbs such as want to, hope that, and so on. In the following, subordinate phrases, like relative or causal sentences, will be also considered as embedded ones.</Paragraph> </Section> <Section position="2" start_page="0" end_page="263" type="sub_section"> <SectionTitle> 2.2 Embedded sentences with </SectionTitle> <Paragraph position="0"> intrasentential antecedents First of all, we will distinguish the Possessive, Reciprocal and Reflexive pronouns (PRR hereafter) from the other pronouns (non-PRR hereafter).</Paragraph> <Paragraph position="1"> On the basis of 120 articles, of 4 sentences on average, containing 332 pronouns altogether, we made the following assumption (1): Assumption: non-PRR pronouns can have intrasentential antecedents, only if these pronouns occur in an embedded sentence.</Paragraph> <Paragraph position="2"> The statistics below show that of 262 non-PRR pronouns, there are 244 having intrasentential antecedents, all of which occur in embedded sentences and none in a &quot;simple&quot; sentence. The remaining 18 non-PRR pronouns have intersentential antecedents.</Paragraph> <Paragraph position="4"> With intrasentential antecedents 244 in an embedded sentence With intrasentential in a simple 0 sentence With intersentential antecedents 18 Our assumption means that, while the PRR pronouns may find their antecedents in an non embedded sentence (e.g., sentences 2 and 3) the non-PRR pronouns can not.</Paragraph> <Paragraph position="5"> 2) Vulcan made i~s initial Investment in Telescan in May, 1992.</Paragraph> <Paragraph position="6"> 3) The agencies HCM and DYR are ~hemselves joint ventures.</Paragraph> <Paragraph position="7"> Without jumping to conclusions, we cannot avoid making a parallel with the topological relations defined in the binding theory (Chomsky, 1980), between two coreferring phrases in the syntactic tree level. Assumption 1 redefines these relations in an informal and less rigorous way, at the semantic level, i.e., considering semantic parameters such as the type of verbs that introduce embedded sentences.</Paragraph> </Section> <Section position="3" start_page="263" end_page="264" type="sub_section"> <SectionTitle> 2.3 Using Sidner's Focusing Approach </SectionTitle> <Paragraph position="0"> To resolve anaphors one of the most suitable existing approaches when dealing with anaphor issues in a conceptual analysis process is the focusing approach proposed by Sidner. However, this mechanism is not suitable for intrasentential cases. We propose to exploit its main advantages in order to build our anaphora resolution mechanism extending it to deal also with intrasentential antecedents.</Paragraph> <Paragraph position="1"> We describe the main elements of the focusing approach that are necessary to understand our method, without going into great detail, see (Sidner, 1979) (Sidner, 1981) (Sidner, 1983). Sidner proposed a methodology, modelling the way &quot;focus&quot; of attention and anaphor resolution influence one another. Using pronouns reflects what the speaker has focused on in the previous sentence, so that the focus is that phrase which the pronouns refer to. The resolution is organised through the following processes: * The expected focus algorithm that selects an initial focus called the &quot;expected focus&quot;. This selection may be &quot;confirmed&quot; or &quot;rejected&quot; in subsequent sentences. The expected focus is generally chosen on the basis of the verb semantic categories. There is a preference in terms of thematic position: the &quot;theme&quot; (as used by Gruber and Anderson, 1976 for the notion of the object case of a verb) is the first, followed by the goal, the instrument and the location ordered according to their occurrence in the sentence; the final item is the agent that is selected when no other role suits.</Paragraph> <Paragraph position="2"> * The anaphora interpreter uses the state of the focus and a set of algorithms associated with each anaphor type to determine which element of the data structures is the antecedent. Each algorithm is a filter containing several interpretation rules (IR).</Paragraph> <Paragraph position="3"> Each IR in the algorithm appropriate to an anaphor suggests one or several antecedents depending on the focus and on the anaphor type.</Paragraph> <Paragraph position="4"> * An evaluation of the proposed antecedents is performed using different kinds of criteria (syntax, semantics, inferential, etc.) * The focusing algorithm makes use of data structures, i.e., the focus registers that represent the state of the focus: the current focus (CF) representation, alternate focus list (AFL) that contains the other phrases of the sentence and the focus stack (FS). A parallel structure to the CF is also set to deal with the agentive pronouns.</Paragraph> <Paragraph position="5"> The focusing algorithm updates the state of the focus after each sentence anaphor (except the first sentence). After the first sentence, it confirms or rejects the predicted focus taking into account the results of anaphor interpretation.</Paragraph> <Paragraph position="6"> In the case of rejection, it determines which phrase is to move into focus.</Paragraph> <Paragraph position="7"> This is a brief example (Sidner 1983) : a Alfred and Zohar liked to play baseball.</Paragraph> <Paragraph position="8"> b They played it every day after school before dinner. null c After lheir game, Alfred and Zohar had ice cream cones.</Paragraph> <Paragraph position="9"> d They tasted really good.</Paragraph> <Paragraph position="10"> * In a) the expected focus is &quot;baseball&quot; (the theme) * In b) &quot;it&quot; refers to &quot;baseball&quot; (CF). &quot;they&quot; refers to Alfred and Zohar (AF) The focusing approach always prefers the previous sentences' entities as antecedents to the current sentences. In fact only previous sentence entities are present in the focus registers. Thus phrases of the current sentence can not be proposed as antecedents. This problem has already been underlined, see (Carter, 1987) in particular who proposed augmenting the focus registers with the entities of the current sentence. For example in sentence 4 while the focus algorithm would propose only &quot;John&quot; as an antecedent for &quot;him&quot;, in Carter's method &quot;Bill&quot; will also be proposed.</Paragraph> <Paragraph position="11"> 4) John walked into the room. He told Bill someone wanted to see him.</Paragraph> <Paragraph position="12"> The focusing mechanism fails in the expected focus algorithm when encountering anaphors occurring in the first sentence of a text, which we call initial anaphors, such as They in sentence (1). The problem with initial anaphors is that the focus registers cannot be initialised or may be wrongly filled if there are anaphors inside the first sentence of the text. It is clear that taking the sentence in its classical meaning as the unit of processing in the focusing approach, is not suitable when sentences are embedded. null We will focus on the mechanisms and algorithmic aspects of the resolution (how to fill the registers, how to structure algorithms, etc.) and not on the rule aspects, like how IRs decide to choose Bill and not John (sentence 4).</Paragraph> </Section> </Section> <Section position="5" start_page="264" end_page="266" type="metho"> <SectionTitle> 3 Our Solution </SectionTitle> <Paragraph position="0"> As stated above, embedded sentences include several elementary events (EEs). EEs are represented as conceptual entities in our work. We consider that such successive EEs involve the same context that is introduced by several successive short sentences. Moreover, our assumption states that when non-PRR anaphors have intrasentential antecedents, they occur in embedded sentences. Starting with these considerations, the algorithm is governed by the hypotheses expanded below.</Paragraph> <Section position="1" start_page="264" end_page="264" type="sub_section"> <SectionTitle> 3.1 Main hypotheses </SectionTitle> <Paragraph position="0"> First hypothesis : EE is the unit of processing in the basic focusing cycle.</Paragraph> <Paragraph position="1"> An EE is the unit of processing in our resolution algorithm instead of the sentence. The basic focusing cycle is applied on each EE in turn and not sentence by sentence. Notice that a simple sentence coincides with its EE.</Paragraph> <Paragraph position="2"> Second hypothesis : The &quot;initial&quot; EE of a well formed first sentence does not contain non-PRR pronouns just as an initial simple sentence cannot. null For example, in the splitting of sentence 1 into two EEs (see below), EEl does not contain non-PRR pronouns because it is the initial EE of the whole discourse.</Paragraph> <Paragraph position="3"> EEl) 'C/rhree of the world's leading advertising groups, Agence I-Iavas S.A. of France, Young & Rubicam of the U.S. and Dentsu Inc. of Japan, said&quot; EE2) &quot;they are forming a global advertising joint venture.&quot; Third hypothesis : PRR pronouns require special treatment.</Paragraph> <Paragraph position="4"> PRR could refer to intrasentential antecedents in simple sentences (such as in those of sentences 3 and 4). An initial EE could then contain an anaphor of the PRR type. Our approach is to add a special phase that resolves first the PRRs occurring in the initial EE before applying the expected focusing algorithm on the same initial EE. In all other cases, PRRs are treated equally to other pronouns.</Paragraph> <Paragraph position="5"> This early resolution relies on the fact that the PRR pronouns may refer to the agent, as in sentence 3, as well as to the complement phrases. However the ambiguity will not be huge at this first level of the treatment. Syntactic and semantic features can easily be used to resolve these anaphors. This relies also on the fact that the subject of the initial EE cannot be a pronoun (second hypothesis).</Paragraph> <Paragraph position="6"> Having mentioned this particular case of PRR in initial EE, we now expand on the whole algorithm of resolution.</Paragraph> </Section> <Section position="2" start_page="264" end_page="265" type="sub_section"> <SectionTitle> 3.2 The Algorithm </SectionTitle> <Paragraph position="0"> In the following, remember that what we called the basic focusing cycle is the following successive steps * applying the resolution rules, * applying the focusing algorithm, i.e., updating the focus registers * the evaluation of the proposed antecedents for each anaphor.</Paragraph> <Paragraph position="1"> The algorithm is based on the decomposition of the sentence into EEs and the application of the basic focusing cycle on each EE in turn and not sentence by sentence.</Paragraph> <Paragraph position="2"> The complete steps are given below (see also figure 1): Step 1 Split the sentence, i.e., its semantic representation, into EEs.</Paragraph> <Paragraph position="3"> Step 2 Apply the expected focus algorithm to the first EE.</Paragraph> <Paragraph position="4"> Step 3 Perform the basic focusing cycle for every anaphor of all the EEs of the current sentence. Step 4 Perform a collective evaluation (i.e., evaluation that involves all the anaphors of the sentence), when all the anaphors of the current sentence are processed.</Paragraph> <Paragraph position="5"> Step 5 Process the next sentence until all the sentences are processed: * split the sentence into EEs count when applying the focusing algorithm. For example, in sentence 1, the intrasentential antecedent Bill will be taken into account, because EEl would be processed beforehand by the expected focusing algorithm.</Paragraph> <Paragraph position="6"> 2. The problem of initial anaphors is then resolved. The expected focusing algorithm is applied only on the initial EE which must not contain anaphors.</Paragraph> </Section> <Section position="3" start_page="265" end_page="266" type="sub_section"> <SectionTitle> 3.3 Examples and results </SectionTitle> <Paragraph position="0"> To illustrate the algorithm, let's consider the following sentence : Lafarge Coppee said it would buy 10 percent in National Gypsum, the number two plasterboard company in the US, a purchase which allows it to be present on the world's biggest plasterboard market.</Paragraph> <Paragraph position="1"> At the conceptual level, there are 3 EEs. They are involved respectively by the said, buy, and allows verbs. They correspond respectively to the following surface sentences: EEl &quot;Lafarge Coppee said&quot; EE2 &quot;it would buy 10 percent in National Gypsum, the number two plasterboard company in the US&quot; EE3 &quot;a purchase which allows it to be present on the world's biggest plasterboard market.&quot; Consider the algorithm : * the expected focusing algorithm is applied to the first EE, EEl, which contains non-PRR anaphors.</Paragraph> <Paragraph position="2"> * the other phases of the algorithm, i.e., the basic focusing cycle, are applied to the subsequent EEs : - EE2 contains only one pronoun it, which is resolved by the basic focusing cycle - it in EE3 will be resolved in the same way. The anaphora resolution has been implemented as a part of a conceptual analyser (Azzam, 1995a). It dealt particularly with pronouns. It has been tested on a set of 120 news reports. We made two kinds of algorithm evaluations: the evaluation of the implemented procedure and an evaluation by hand. For the implementation the success rate of resolution was 70%. The main cases of failure are related to the non implemented aspects like the treatment of coordination ambiguities and the appositions, or other anaphoric phenomena, like ellipsis.</Paragraph> <Paragraph position="3"> For the second evaluation which concerns the real evaluation of the approach,i.e., without going into the practical issues concerning implementation, the success rate was 95%. The main cases of failure were due to the cases that were not considered by the algorithm, like for example the pronouns occurring before their antecedents , i.e., cataphors. Such cases occur for example in sentences 5 and 6 pointed out by Hobbs (IIobbs, 1985) to discuss the cases that are not handled easily in the literature.</Paragraph> <Paragraph position="4"> 5) Mary sacked out in his apartment before Sam could kick her out.</Paragraph> <Paragraph position="5"> 6) Girls who he has dated say that Sam is charming.</Paragraph> <Paragraph position="6"> Our algorithm fails in resolving his in 5, because the algorithm searches only for the entities that precede the anaphor in the text. The same applies for he in 6. However improving our algorithm to process classical cases of cataphors, such as that in sentence 6, should not require major modifications, only a change in the order in which the EEs are searched. For example, to process pronouns of the sentence 6 split into two EES (see below), the algorithm must consider EE2 before EEl. This means applying the step 2 of the algorithm to EE2, then step 3 to EEl. The sentence 5 should require specific treatment, though.</Paragraph> <Paragraph position="7"> EEl) &quot;that Sam is charming&quot; EE2) &quot;Girls who he has dated say&quot; IIobbs also pointed out the cases of &quot;picture noun&quot; examples, as in sentences 7 and 8: 7) John saw a picture of him.</Paragraph> <Paragraph position="8"> 8) John's father's portrait of him.</Paragraph> <Paragraph position="9"> In 7 our algorithm is successful, i.e., it will not identify him with John because of our previous assumption (section 2.2). However our algorithm would fail in 8 because the non-PRR pronoun him could refer to John which occurs in the same EE.</Paragraph> <Paragraph position="10"> Notice that Hobbs' (I-Iobbs, 1985) remark that &quot;the more deeply the pronoun is embedded and the more elaborate the construction it occurs in, the more acceptable the non reflexive&quot; is consistent with our assumption.</Paragraph> <Paragraph position="11"> For example in the embedded sentence 9 where either the reflexive (himself) or non reflexive pronouns (him) may be used, it is more natural to make use of him.</Paragraph> <Paragraph position="12"> 9) John claimed that the picture of him hanging in the post office was a fraud.</Paragraph> </Section> </Section> <Section position="6" start_page="266" end_page="267" type="metho"> <SectionTitle> 4 The Conceptual Level </SectionTitle> <Paragraph position="0"> We comment here on the main aspects of the conceptual analysis that are related to the anaphora resolution process. They concern mainly the way of splitting embedded sentences and the problems of determining the theme and of managing the other ambiguities and the several readings.</Paragraph> <Paragraph position="1"> The conceptual analyser's strategy consists of a continuous step-by-step translation of the original natural language sentences into conceptual structures (CS hereafter). This translation uses the results of the syntactic analysis (syntactic tree). It is a progressive substitution of the NL terms located in the syntactic tree with concepts and templates of the conceptual representation language. Triggering rules are evoked by words of the sentence and allow the activation of well-formed CS templates when the syntactico-semantic filter is unified with the syntactic tree. The values caught by the filter variables are the arguments of the CS roles, i.e., they fill the CS roles. If they are anaphors, they are considered to be unbound variables and result in unfilled roles in the CS. The anaphora resolution aims therefore at filling the unfilled roles with the corresponding antecedents.</Paragraph> <Section position="1" start_page="266" end_page="266" type="sub_section"> <SectionTitle> 4.1 Splitting into EEs </SectionTitle> <Paragraph position="0"> The splitting of a sentence in EE is done on the corresponding CS. A minimal CS is a template comprising a predicate that identifies the basic type of the represented event and a set of roles or predicate cases.</Paragraph> <Paragraph position="1"> For example, the sentence &quot;to say that they agree to form a joint venture&quot; is represented, in a simplified way, with three templates, corresponding to the predicates: * move information (from &quot;to say&quot;), * produce an agreement (from &quot;to agree&quot;), * produce a joint venture (from &quot;to form&quot;). Given that one template at the semantic level represents an elementary event, the splitting is implicitly already done when these templates are created in the triggering phase. Indeed, the syntactico-semantic filter of the triggering rules takes into account the semantic features of words (mainly verbs) for recognising in the surface sentence those that are able to trigger an elementary event.</Paragraph> </Section> <Section position="2" start_page="266" end_page="266" type="sub_section"> <SectionTitle> 4.2 Determining the theme </SectionTitle> <Paragraph position="0"> Gruber and Anderson characterise the theme as follows: if a verb describes a change to some entity, whether of position, activity, class or possession, then the theme is the changed entity, (Gruber, 1976) and (Anderson, 1977). As Carter (Carter, 1987) demonstrated, this definition of Gruber and Anderson is sufficient to apply the focusing mechanism.</Paragraph> <Paragraph position="1"> This assumption is particularly apt when we dispose of a conceptual representation. Indeed, to determine the thematic roles, we established a set of thematic rules that affect for a given predicative occurrence, its thematic functions according to the predicate type, the role type and the argument's semantic class.</Paragraph> </Section> <Section position="3" start_page="266" end_page="267" type="sub_section"> <SectionTitle> 4.3 Managing other ambiguities </SectionTitle> <Paragraph position="0"> An important aspect appears when one designs a concrete system, namely how to make other disambiguation processes cohabit. In the conceptual analyser, the general disambiguation module (GDM) deals with other ambiguities, like prepositional phrase attachment. It coordinates the treat- null ment of different kinds of ambiguities. This is necessary because the conceptual structures (CS) on which the rules are performed could be incomplete because of other types of ambiguities not being resolved. For example, if the CF of the sentence is a PP object that is not attached yet in the CS the thematic rules fail to fill the CF. The GDM ensures that every disambiguation module intervenes only if previous ambiguities have already been resolved.</Paragraph> <Paragraph position="1"> The process of co-ordinating ambiguity processing is fully expanded in (Azzam, 1995b).</Paragraph> </Section> <Section position="4" start_page="267" end_page="267" type="sub_section"> <SectionTitle> 4.4 Multiple readings </SectionTitle> <Paragraph position="0"> When dealing with ambiguities, another important aspect is managing multiple readings. At a certain point when the GDM calls the anaphora module to deal with a given anaphor, the status of the conceptual analysis could be charaeterised by the following parameters : * The set of conceptual structures for the current reading Ri on which the resolution is performed, given that several readings could arise from previous ambiguity processing.</Paragraph> <Paragraph position="1"> * The set of conceptual structures of the current sentence Si where the anaphor occurs; * The set of conceptual structures of the current elementary event EEi where the anaphor occurs after the Si splitting.</Paragraph> <Paragraph position="2"> * The state of the focus (content of the registers), SFi The main assumption is that the anaphora resolution algorithm always applies to a single state, (Ri, Si , EEi, SFi) when resolving a given anaphor (Step 3): a If several antecedents are still possible after the individual evaluation of the anaphor, Ri is then duplicated, in Rij, as many times as there are possibilities.</Paragraph> <Paragraph position="3"> b When performing the collective evaluation of all Si anaphors, every inconsistent Rij is suppressed. null c The result is a set of readings (Rij, Sj , EEj, SFi).</Paragraph> </Section> </Section> class="xml-element"></Paper>