File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/96/p96-1036_metho.xml
Size: 26,820 bytes
Last Modified: 2025-10-06 14:14:19
<?xml version="1.0" standalone="yes"?> <Paper uid="P96-1036"> <Title>Functional Centering</Title> <Section position="3" start_page="270" end_page="270" type="metho"> <SectionTitle> 2 Types of Anaphora Considered </SectionTitle> <Paragraph position="0"> Text phenomena, e.g., textual forms of ellipsis and anaphora, are a challenging issue for the design of parsers for text understanding systems, since imperfect recognition facilities either result in referentially incoherent or invalid text knowledge representations.</Paragraph> <Paragraph position="1"> At the conceptual level, textual ellipsis relates a quasianaphoric expression to its extrasentential antecedent by conceptual attributes (or roles) associated with that antecedent (see, e.g., the relation between &quot;Akkus&quot; (accumulator) and &quot;316LT&quot;, a particular notebook, in (lb) and (la)). Thus, it complements the phenomenon of nominal anaphora, where an anaphoric expression is related to its antecedent in terms of conceptual generalization (as, e.g., &quot;Rechner&quot; (computer) in (lc) refers to &quot;316LT' in (la) mediated by the textual ellipsis in (lb)). The resolution of text-level nominal (and pronominal) anaphora contributes to the construction of referentially valid text knowledge bases, while the resolution of textual ellipsis yields referentially coherent text knowledge bases.</Paragraph> <Paragraph position="2"> (1) a. Ein Reserve-Batteriepaek versorgt den 316LT ca.</Paragraph> <Paragraph position="3"> 2 Minuten mit Strom.</Paragraph> <Paragraph position="4"> (A reserve battery pack - supplies - the 316LT for approximately 2 minutes - with power.) b. Der Status des Akkus wird dem Anwender angezeigt. null (The status of the accumulator - is - to the user indicated.) null c. Ca. 30 Minuten vor der Entleerung beginnt der Rechner 5 Sekunden zu beepen.</Paragraph> <Paragraph position="5"> (Approximately 30 minutes - before the discharge - starts - the computer - for 5 seconds - to beep.) d. 5 Minuten bevor er sich ausschaltet, f'angt die Low-Battery-LED an zu blinken.</Paragraph> <Paragraph position="6"> (5 minutes - before - it - itself- turns off- begins - the low-battery-LED - to flash.) In the case of textual ellipsis, the missing conceptual link between two discourse elements occurring in adjacent utterances must be inferred in order to establish the local coherence of the discourse (for an early statement of that idea, cf. Clark (1975)). In the surface form of utterance (lb) the information is missing that &quot;Akkus'&quot; (accumulator) links up with &quot;316LT&quot;. This relation can only be made explicit if conceptual knowledge about the domain, viz. the relation part-of between the concepts ACCUMULATOR and 316LT, is available (see Hahn et al. (1996) for a more detailed treatment of text ellipsis resolution).</Paragraph> </Section> <Section position="4" start_page="270" end_page="272" type="metho"> <SectionTitle> 3 Principles of Functional Centering </SectionTitle> <Paragraph position="0"> Within the framework of the centering model (Grosz et al., 1995), we distinguish each utterance's backward-looking center (Gb(U,~)) and its forward-looking centers (G! (Un)). The ranking imposed on the elements of the G I reflects the assumption that the most highly ranked element of G I (tin) - the preferred center Cp(Un) - is the most preferred antecedent of an anaphoric or elliptical expression in Un+l, while the remaining elements are partially ordered according to decreasing preference for establishing referential links. Hence, the most important single construct of the centering model is the ordering of the list of forward-looking centers (Walker et al., 1994).</Paragraph> <Paragraph position="1"> The main difference between Grosz et al.'s work and our proposal concerns the criteria for ranking the forward-looking centers. While Grosz et al. assume that grammatical roles are the major determinant for the ranking on the C'y, we claim that for languages with relatively free word order (such as German), it is the functional information structure (IS) of the utterance in terms of the context-boundedness or unboundedness of discourse elements. The centering data structures and the notion of context-boundedness can be used to redefine Dane~' (1974a) trichotomy between given information, theme and new information (rheme). The Cb(U,), the most highly ranked element of C.t(Un-i) realized in \[In, corresponds to the element which represents the given information. The theme of U, is represented by the preferred center C'p(U,), the most highly ranked element of C! (Un).</Paragraph> <Paragraph position="2"> The theme/rheme hierarchy of \[In is represented by CI(U,~ ) which - in our approach - is partly determined by the C! (Un-i): the rhematic elements of Un are the ones not contained in C! (U,_ i ) (unbound discourse dements); they express the new information in Un. The ones contained in Cl(U,_i ) and Cy(U,) (bound discourse elements) are thematic, with the theme/rheme hierarchy corresponding to the ranking in the Cls. The distinction between context-bound and unbound elements is important for the ranking on the C I, since bound elements are generally ranked higher than any other non-anaphoric elements (cf. also Haji~ov~i et al. (1992)).</Paragraph> <Paragraph position="3"> An alternative definition of theme and rheme in the context of the centering approach is proposed by Rambow (1993). In his approach the theme corresponds to the Cb and the theme/rheme hierarchy can be derived from those elements of C! (U,-i) that are realized in \[In. Rambow does not distinguish, however, between the information structure and the thematic structure of utterances, which leads to problems when a change of the criteria for recognizing the thematic structure is envisaged, Our approach is flexible enough to accomodate other conceptions of theme/rheme as defined, e.g., by Haji6ov~i et al. (1995), since this change affects only the thematic but not the information structure of utterances.</Paragraph> <Paragraph position="4"> bound element(s) >~sb~.. unbound element(s) anaphora >X Sbo,,a (possessive pronoun xor elliptical antecedent) >,Sbo,,a (elliptical expression xor head of anaphoric expression) nom head, >pr,c nom head2 >p~,o ... >~, nom headn The rules holding for the ranking on the C' I, derived from a German language corpus, are summarized in Table 2. They are organized into three layers 2. At the top level, >,sb,,~ denotes the basic relation for the overall ranking of information structure (IS) patterns. Accordingly, any context-bound expression in the utterance U,_ i is given the highest preference as a potential antecedent of an anaphoric or elliptical expression in \[In while any unbound expression is ranked next to context-bound expressions.</Paragraph> <Paragraph position="5"> The second relation depicted in Table 2, >iSbou~u, denotes preference relations dealing exclusively with multiple occurrences of (resolved) anaphora, i.e., bound elements, in the preceding utterance. >'Sbo,,d distinguishes among different forms of context-bound elements (viz., anaphora, possessive pronouns and textual ellipses) and their associated preference order. The final element of >,Sbou~u is either the elliptical expression or the head of an anaphoric expression which is used as a possessive determiner, a Saxon genitive, a prepositional or a genitival attribute (cf. the ellipsis in (2c): &quot;die Ladezeit&quot; (the charge time) vs. &quot;seine Ladezeit&quot; (its charge time) or &quot;die Ladezeit des Akkus&quot; (the accumulator's charge time)).</Paragraph> <Paragraph position="6"> For illustration purposes, consider text fragment (1) and the corresponding Oh~C! data in Table 33: In (ld) the pronoun &quot;er&quot; (it) might be resolved to &quot;Akku&quot; (accumulator) or &quot;Rechner&quot; (computer), since both fulfill the agreement condition for pronoun resolution. Now, &quot;der Rechner&quot; (computer) figures as a nominal anaphor, already resolved to DELL-3 16LT, while &quot;Akku&quot; (accumulator) is only the antecedent of the elliptical expression &quot;der Entleerung&quot; (discharge). Therefore, the preferred antecedent of &quot;er&quot; (it) is determined as Rechner (computer).</Paragraph> <Paragraph position="7"> The bottom level of Table 2 specifies >~rco which covers the preference order for multiple occurrences of the same type of any information structure pattern, e.g., the occurrence of two anaphora or two unbound elements (all heads in an utterance are ordered by linear precedence relative to their text position). In sentence (2b), two nominal anaphors occur, &quot;Akku&quot; (accumulator) and &quot;Rechner&quot; (computer). The textual ellipsis &quot;Ladezeit&quot; (charge time) in (2c) has to be resolved to the most preferred dement of the C' I of (2b), viz. the entity denoted by &quot;Akku&quot; (accumulator) (cf. Table 4). Note that &quot;Rechner&quot; (computer) is the subject of the sentence, though it is not the preferred antecedent, since &quot;Akku&quot; (accumulator) precedes &quot;Rechner&quot; (computer) and is anaphoric as well. (la) Cb: DELL-3 16LT: 316LT Cf.&quot; \[DELL-316LT: 316LT, RESERVE-BATTERY-PAcK: Reserve-Batteriepack, TIME-UNIT-PAIR: 2 Minuten, POWER: Strom\] (lb) Cb: DELL-316LT:-Cf: \[DELL-316LT:--, Accu: Akku, STATUS: Status, USER: Anwender\] (lc) Cb: DELL-3 16LT: Rechner Cf: \[DELL-316LT: Rechner, Accu: --, DISCHARGE: Enfleerung, TIME-UNIT-PAIR: 30 Minuten, TIME-UNIT-PAIR: 5 Sekunden\] (ld) Cb: DELL-3 16LT: er Cf: \[DELL-316LT: er, LoW-BATTERY-LED: Low-Battery-LED (2) Der316LTw~dmiteinemNiMH-Akku bestllckt.</Paragraph> <Paragraph position="9"> b. Durch diesen neuartigen Akku wird der Rechner ffir ca. 4 Stunden mit Strom versorgt.</Paragraph> <Paragraph position="10"> (Because of this new type of accumulator - is the computer - for approximately 4 hours - with power - provided.) c. Dartiberhinaus ist die Ladezeit mit 1,5 Stunden sehr kurz.</Paragraph> <Paragraph position="11"> (Also - is - the charge time of 1.5 hours - quite short.) Given these basic relations, we may formulate the composite relation :>,s (Table 5). It states the conditions for the comprehensive ordering of items on C! (x and y denote lexical heads).</Paragraph> <Paragraph position="13"> ~fx and y both represent the same type of IS pattern then the relation >~,,c applies to x and y else/fx and y both represent different forms of bound elements then the relation >rSbo,,a applies to x and y</Paragraph> </Section> <Section position="5" start_page="272" end_page="323" type="metho"> <SectionTitle> 4 Evaluation </SectionTitle> <Paragraph position="0"> In this section, we first describe the empirical and methodological framework in which our evaluation experiments were embedded, and then turn to a discussion of evaluation results and the conclusions we draw from the data.</Paragraph> <Section position="1" start_page="272" end_page="273" type="sub_section"> <SectionTitle> 4.1 Evaluation Framework </SectionTitle> <Paragraph position="0"> The test set for our evaluation experiment consisted of three different text sorts: 15 product reviews from the information technology (IT) domain (one of the two main corpora at our lab), one article from the German news magazine Der Spiegel, and the first two chapters of a short story by the German writer Heiner Miiller 4.</Paragraph> <Paragraph position="1"> The evaluation was carried out manually in order to circumvent error chaining 5. Table 6 summarizes the total numbers of anaphors, textual ellipses, utterances and words in the test set.</Paragraph> <Paragraph position="2"> ag. apho~ ellipses utterances words Given this test set, we compared three major approaches to centering, viz. the original model whose ordering principles are based on grammatical role indicators only (the so-called canonical model) as characterized by Table 1, an &quot;intermediate&quot; model which can be considered a naive approach to free word order languages, and, of course, the functional model based on information structure constraints as stated in Table 2. For reasons discussed below, augmented versions of the naive and the canonical approaches will also be considered. They are characterized by the additional 4Liebesgeschichte. In Heiner Mflller. Geschichten aus der Produktion 2. Berlin: Rotbuch Verlag, pp. 57-63.</Paragraph> <Paragraph position="3"> SA performance evaluation of the current anaphora and ellipsis resolution capacities of our system is reported in Hahn et al. (1996).</Paragraph> <Paragraph position="4"> constraint that elliptical antecedents are ranked higher than elliptical expressions (short: &quot;ante > express&quot;). For the evaluation of a centering algorithm on naturally occurring text it is necessary to specify how to deal with complex sentences. In particular, methods for the interaction between intra- and intersentential anaphora resolution have to be defined, since the centering model is concerned only with the latter case (see Suri & McCoy (1994)). We use an approach as described by Strube (1996) for the evaluation.</Paragraph> <Paragraph position="5"> Since most of the anaphors in these texts are nominal anaphors, the resolution of which is much more restricted than that of pronominal anaphors, the rate of success for the whole anaphora resolution process is not significant enough for a proper evaluation of the functional constraints. The reason for this lies in the fact that nominal anaphors are far more constrained by conceptual criteria than pronominal anaphors. So the chance to properly resolve a nominal anaphor, even at lower ranked positions in the center lists, is greater than for pronominal anaphors. While we shift our evaluation criteria away from simple anaphora resolution success data to structural conditions based on the proper ordering of center lists (in particular, we focus on the most highly ranked item of the forward-looking centers) these criteria compensate for the high proportion of nominal anaphora that occur in our test sets. The types of centering transitions we make use of (cf.</Paragraph> </Section> <Section position="2" start_page="273" end_page="323" type="sub_section"> <SectionTitle> 4.2 Evaluation Results </SectionTitle> <Paragraph position="0"> In Table 8 we give the numbers of centering transitions between the utterances in the three test sets. The first column contains those which are generated by the naive approach (such a proposal was made by Gordon et al. (1993) as well as by Rambow (1993) who, nevertheless, restricts it to the German middlefield only).</Paragraph> <Paragraph position="1"> We simply ranked the elements of C! according to their text position. While it is usually assumed that the elliptical expression ranks above its antecedent (Grosz et al., 1995, p.217), we assume the contrary. The second column contains the results of this modification with respect to the naive approach. In the third column of Table 8 we give the numbers of transitions which are generated by the canonical constraints as stated by Grosz et al. (1995, p.214, 217). The fourth column supplies the results of the same modification as was used for the naive approach, viz. elliptical antecedents are ranked higher than elliptical expressions. The fifth column shows the results which are generated by the functional constraints from Table 2.</Paragraph> <Paragraph position="2"> First, we examine the error data for anaphora resolution for the five cases. All approaches have 99 errors in common. These are due to underspecifications at different levels, e.g., the failure to account for prepositional anaphors (16), plural anaphors (8), anaphors which refer to a member of a set (14), sentence anaphors (21), and anaphors which refer to the global focus (12). Only 6 errors of the functional approach are directly caused by an inappropriate ordering of the C I, while the naive approach leads to 10 errors and the canonical to 7. When the antecedent of an elliptical expression is ranked above the elliptical expression itself the error rate of these two augmented approaches increases to 12 and 9, respectively.</Paragraph> <Paragraph position="3"> We now turn to the distribution of transition types for the different approaches. The centering model assumes a preference order among these transitions, e.g., CONTINUE ranks above RETAIN and RETAIN ranks above SHIFT. This preference order reflects the presumed inference load put on the hearer or speaker to coherently decode or encode a discourse. Since the functional approach generates a larger amount of CONTINUE transitions, we interpret this as a first rough indication that this approach provides for more efficient processing than its competitors.</Paragraph> <Paragraph position="4"> But this reasoning is not entirely conclusive. Counting single occurrences of transition types, in general, does not reveal the entire validity of the center lists.</Paragraph> <Paragraph position="5"> Instead, considering adjacent transition pairs gives a more reliable picture, since depending on the text sort considered (e.g., technical vs. news magazine vs. literary texts) certain sequences of transition types may be entirely plausible, though they include transitions which, when viewed in isolation, seem to imply considerable inferencing load (cf. Table 8). For instance, a CONTINUE transition which follows a CONTINUE transition is a sequence which requires the lowest processing costs. But a CONTINUE transition which follows a RETAIN transition implies higher processing costs than a SMOOTH-SHIFT transition following a RETAIN transition. This is due to the fact that a RETAIN transition ideally predicts a SMOOTH-SHIFT in the following utterance. In this case the SMOOTH-SHIFT is the &quot;least effort&quot; transition, because only the first element of the C! of the preceding utterance has to be checked to perform the SMOOTH-SHIFT transition, while in the case of CONTINUE at least one more check has to be performed. Hence, we claim that no one particular centering transition is preferred over another. Instead, we postulate that some centering transition pairs are preferred over others. Following this line of argumentation, we here propose to classify all occurrences of centering transition pairs with respect to the costs they imply. The cost-based evaluation of different C! orderings refers to evaluation criteria which form an intrinsic part of the centering model 6.</Paragraph> <Paragraph position="6"> Transition pairs hold for two immediately successive utterances. We distinguish between two types of transition pairs, cheap ones and expensive ones. We call a transition pair cheap if the backward-looking center of the current utterance is correctly predicted by the preferred center of the immediately preceding utterance, i.e., Cb(Ui) = Gp(Ui_l),i = 2...n.</Paragraph> <Paragraph position="7"> Transition pairs are called expensive if the backward-looking center of the current utterance is not correctly predicted by the preferred center of the immediately preceding utterance, i.e., Cb(Ui) # Gp(Ui_l),i = 2... n. Table 9 contains a detailed synopsis of cheap and expensive transition pairs. In particular, chains of the RETAIN transition in passages where the Cb does not change (passages with constant theme) show that the canonical ordering constraints for the forward-looking centers are not appropriate, The numbers of centering transition pairs generated by the different approaches are shown in Table 10, In general, the functional approach shows the best re6As a consequence of this postulate, we have to redefine Rule 2 of the Centering Constraints (Grosz et al., 1995, p.215) appropriately, which gives an informal characterization of a preference for sequences of CONTINUE over sequences of RETAIN arid, similarly, sequences of RETAIN over sequences of SHIFT. Our specification for the case of text interpretation says that cheap transitions are preferred over expensive ones, with cheap and expensive transitions as defined in Table 9.</Paragraph> <Paragraph position="8"> Suits, while the naive and the canonical approaches work reasonably well for the literary text, but exhibit a poor performance for the texts from the IT domain and the news magazine. The results for the latter approaches become only slightly more positive with the modification of ranking the antecedent of a textual ellipsis above the elliptical expression, but they do not compare to the results of the functional approach.</Paragraph> <Paragraph position="9"> We were also interested in finding out whether the functional ordering we propose possibly &quot;includes&quot; the grammatical role based criteria discussed so far.</Paragraph> <Paragraph position="10"> We, therefore, re-evaluated the examples already annotated with Gb/C! data available in the literature (for the English language, we considered all exampies from Grosz et al. (1995) and Brennan et al.</Paragraph> <Paragraph position="11"> (1987); for Japanese we took the data from Walker et al. (1994)). Surprisingly enough, all examples of Grosz et al. (1995) passed the test successfully. Only with respect to the troublesome Alfa Romeo driving scenario (cf. Brennan et al. (1987, p.157)) our constraints fail to properly rank the elements of the third sentence C! of that example. 7 Note also that these results were achieved without having recourse to extra constraints, e.g., the shared property constraint to account for anaphora parallelism (Kameyama, 1986).</Paragraph> <Paragraph position="12"> We applied our constraints to Japanese examples in the same way. Again we abandoned all extra constraints set up in these studies, e.g., the Zero Topic Assignment (ZTA) rule and the special role of empathy 7In essence, the very specific problem addressed by that example seems to be that Friedman has not been previously introduced in the local discourse segment and is only accessible via the global focus.</Paragraph> <Paragraph position="13"> functional ante > express constraints generate are the same as those generated by Walker et al. including these model extensions. Only a single problematic case remains, viz. example (30) of Walker et al. (1994, p.214) causes the same problems they described (discourse-initial utterance, semantic or world knowledge should be available). Even for the crucial examples (32)-(36) of Walker et al. (1994, p.216-221) our constraints generate the same Cls as Walker et al.' s constraints with ZTA.</Paragraph> <Paragraph position="14"> To summarize the results of our empirical evaluation, we first claim that our proposal based on functional criteria leads to substantially better and -- with respect to the inference load placed on the text understander, whether human or machine -- more plausible results for languages with free word order than the structural constraints given by Grosz et al. (1995) and those underlying a naive approach. We base these observations on an evaluation approach which considers transition pairs in terms of the inference load specific pairs imply. Second, we have gathered some evidence, still far from being conclusive, that the functional constraints on centering seem to incorporate the structural constraints for English and the modified structural constraints for Japanese. Hence, we hypothesize that functional constraints on centering might constitute a general mechanism for treating free an___dd fixed word order languages by the same descriptive mechanism. This claim, however, has to be further substantiated by additional cross-linguistic empirical studies.</Paragraph> </Section> </Section> <Section position="6" start_page="323" end_page="323" type="metho"> <SectionTitle> 5 Comparison with Related Approaches </SectionTitle> <Paragraph position="0"> The centering model (Grosz et al., 1983; 1995) is concerned with the interactions between the local coherence of discourse and the choices of referring expressions. Crucial for the centering model is the way the forward-looking centers are organized. Despite several cross-linguistic studies a kind of &quot;standard&quot; has emerged based on the study of English (cf. Table 1 in Section 1). Only few of these cross-linguistic studies have led to changes in the basic order of discourse entities, the work of Walker et al. (1990; 1994) being the most far reaching exception. They consider the role of expressive means in Japanese to indicate topic status and the speaker's perspective, thus introducing functional notions, viz. ToPIc and EMPATHY, into the discussion. German, the object language we deal with, is also a free word order language like Japanese (possibly even more constrained).</Paragraph> <Paragraph position="1"> Our basic revision of the ordering scheme completely abandons grammatical role information and replaces it with entirely functional notions reflecting the information structure of the utterances in the discourse. Interestingly enough, several extra assumptions introduced to account, e.g., for anaphora parallelism (e.g., the shared property constraint formulated by Kameyama (1986)) can be eliminated without affecting the correctness of anaphora resolutions. Rambow (1993) has presented a theme/rheme distinction within the centering model to which we fully subscribe. His proposal concerning the centering analysis of German (already referred to as the &quot;naive&quot; approach; cf. Section 4) is limited, however, to the German middlefield and, hence, incomplete.</Paragraph> <Paragraph position="2"> A common topic of criticism relating to focusing approaches to anaphora resolution has been the diversity of data structures they require, which are likely to hide the underlying linguistic regularities. Focusing algorithms prefer the discourse element already in focus for anaphora resolution, thus considering context-boundedness, too. But the items of the focus lists are either ordered by thematic roles (Sidner, 1983) or grammatical roles (Suri & McCoy, 1994; Dahl & Ball, 1990)). Dahl & Ball (1990) improve the focusing mechanism by simplifying its data structures and, thus, their proposal is more closely related to the centering model than any other focusing mechanism. But their approach still relies upon grammatical information for the ordering of the centering list, while we use only the functional information structure as the guiding principle.</Paragraph> </Section> class="xml-element"></Paper>