File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/00/w00-1006_metho.xml
Size: 26,446 bytes
Last Modified: 2025-10-06 14:07:25
<?xml version="1.0" standalone="yes"?> <Paper uid="W00-1006"> <Title>From Elementary Discourse Units to Complex Ones</Title> <Section position="3" start_page="0" end_page="47" type="metho"> <SectionTitle> 2 Adjuncts as Discourse Units </SectionTitle> <Paragraph position="0"> The question at what size of textual expressions one should start looking for discourse units has not been suftlciently answered yet.</Paragraph> <Paragraph position="1"> (Mann and Thompson, 1988), for example, state that &quot;Unit size is arbitrary, but the division of the text should be based on some theory-neutral classification. That is, for interesting results, the units should have independent functional integrity. In our analyses, units are basically clauses \[...\]&quot; \[p.248\]. It has been noted previously that sometimes phrases already seem to have enough &quot;functional integrity&quot; to give some &quot;interesting results&quot;, however, until recently there has been no detailed proposal what kind of phrases should be considered.</Paragraph> <Paragraph position="2"> Consider the following (German) example: 2 (1) a. Mit dem P6LXZ-A versucht Elitegroup neue Kunden zu gewinnen.</Paragraph> <Paragraph position="3"> \[With the P6LXZ-A -- Elitegroup tries to attract new customers.\] b. Mit einem PCI-Slot bietet das Motherboard wenig Platz fiir Erweiterungen.</Paragraph> <Paragraph position="4"> \[With one PCI slot (only) -- the motherboard provides only little space for extensions.\] A straightforward coherence analysis with relations from RST (Mann and Thompson, 1988) takes (l-b) as a single unit and links it to (l-a), probably via an Evaluation relation. Paraphrasing sentence (l-b) reveals, however, a plausible decomposition into two interdependent discourse units: (2) a. The motherboard has one PCI slot, b. so it provides only little space for extensions. null Clause (2-a) gives an Explanation for the subordinate clause (2-b). This might be attributed to the impact of the cue word &quot;so&quot; in (2-b). More generally, it has been claimed that whenever an implicit coherence relation can be made explicit by a paraphrase incorporating a specific cue word, then this coherence relation can always be assumed to hold. 3 So, from a methodological point of view it 2In the following, I will summarize the arguments from (Schauer and Hahn, 2000) that focus on prepositional phrases.</Paragraph> <Paragraph position="5"> S&quot;As a test for the presence of an implicit connection it can be required that the connection could have been explicit... &quot;(Martin, 1992, p.184) cannot be justified to analyze Sentence (2) as being composed of two elementary units, while the prepositional phrase &quot;with one PCI slot&quot; should be an indistinguishable part of the whole Sentence (l-b).</Paragraph> <Paragraph position="6"> Obviously, not all kind of phrases may serve as elementary units. As a first criterion to determine whether a phrase is a discourse unit, we propose to consider only those phrases that are not syntactically mandatory complements of their governing syntactic head.</Paragraph> <Paragraph position="7"> Complements are assumed to be specified in terms of subcategorization frames, valency hsts, etc. Adjuncts are phrases which do not match such a complement specification of their governing syntactic head.</Paragraph> <Paragraph position="8"> In Example (l-b), while it is both possible to omit the optional prepositional phrase &quot;with one PCI slot&quot; and to decompose the sentence such that the prepositional phrase turns into a proper clause, this is only hardly possible for any other phrase in this example, say the subject &quot;the motherboard&quot;.</Paragraph> <Paragraph position="9"> The second major criterion operates at the semantic level. Semantic specifications of lexemes, independent of the specific semantic theory one subscribes to, are confined to &quot;typical&quot; properties, e.g., events are characterized by agents, patients, instruments, locations, time frames, etc. The interpretation of phrases referring to these attributes can be taken care of by straightforward semantic interpretation mechanisms. As long as the interpretation makes reference only to such &quot;typical&quot; properties it is not necessary to consider such phrases as discourse units. But when an analyst thinks that the interpretation involves (additional) non-typical, unpredictable information, coherence relations may capture their value-adding meaning. 4 E.g., the prepositional phrase of (l-a) only refers to a such typical property, namely the instrument of an action. This typicality consideration does not carry over to an &quot;explanation&quot; of events, which is the interpretation of &quot;with tion of adjuncts, in contrast, may refer to both typical and non-typical information.</Paragraph> <Paragraph position="10"> one PCI slot&quot; from Sentence (l-b).</Paragraph> <Paragraph position="11"> Ultimately, however, there is also a third criterion that needs to be considered: namely, that a coherence relation can be found by which the phrase may be related to the remainder of the clause it occurs in (Marcu et al., 1999). We propose that this search for an applicable coherence relation may be performed by trying to paraphrase the phrase and its matrix clause in such a way that it results in two clauses that are related by a cue-phrase.</Paragraph> <Paragraph position="12"> (Schauer and H~.hn, 2000) evaluated the applicability of the criteria by' hmnan annotators using two corpora. We found that 11.0% of the prepositional phrases in a German text corpus can be seen as discourse units and 8.0% in an English control set. These PPs are responsible for 14.8% and 12.1%, respectively, of the relations fomad in the texts. The discourse units triggered by prepositional phrases always occur in satellite (i.e. subordinate) position, which is not surprising, given their status of optional information.</Paragraph> </Section> <Section position="4" start_page="47" end_page="48" type="metho"> <SectionTitle> 3 From Prepositional Phrases to </SectionTitle> <Paragraph position="0"/> <Section position="1" start_page="47" end_page="47" type="sub_section"> <SectionTitle> Coherence Relations </SectionTitle> <Paragraph position="0"> For an automated coherence analysis, we will basically take the preposition as a trigger for computing a coherence relation. To make this discussion more concrete, it is embedded in the framework of SYNDIKATE, a text analysis system under development in our lab (Hahn and Romacker, 1999).</Paragraph> </Section> <Section position="2" start_page="47" end_page="47" type="sub_section"> <SectionTitle> 3.1 Semantic Interpretation </SectionTitle> <Paragraph position="0"> Crucial for the discussion here is that the syntactic analysis provided by the system allows to distinguish betweenadjuncts and mandatory complements: for Sentence (l-b), a dependency relation ppadj between the prepositional phrase and its governing syntactic head is derived, that results from the valency specification for the main verb &quot;bietet&quot; (provide). For computing a semantic interpretation s , the corresponding conceptual cormlares of the content words are then checked SWe assume the framework of description logics (Woods and Sclmaolze, 1992).</Paragraph> <Paragraph position="1"> for role compatibility.</Paragraph> <Paragraph position="2"> In Example (l-b), the major interpretation constraints derive fzom the main verb &quot;bietet&quot; (provide) which is represented by the concept PROVIDE. It has three major conceptual roles, PROVIDE-PATIENT, PROVIDE-CO-PATIENT, and INSTRUMENT. The PROVIDE-PATIENT and PROVIDE-CO-PATIENT roles can be filled by some instance of MOTHERBOARD and SPACE, respectively.</Paragraph> <Paragraph position="3"> Focusing on the analysis of the PP, each preposition specifies allowed semantic relations (Romacker et al., 1999). In the case of &quot;mit&quot; (with) they allow an interpretation in terms of the conceptual INSTRUMENT role, so the corresponding role of PROVIDE is frilled with PCI-SLOT. This causes additional conceptual interpretation processes: a production rule checks whether the filler of the INSTRUMENT role may be a PART-OF the conceptual correlate of the syntactic subject, linking SPACE and MOTHERBOARD. Along the line, a HAS-PART-STATE is established that corresponds to the situation that &quot;the motherboard has one PCI slot&quot;. 6</Paragraph> </Section> <Section position="3" start_page="47" end_page="48" type="sub_section"> <SectionTitle> 3.2 Discourse Interpretation </SectionTitle> <Paragraph position="0"> As discussed above, a preposition does not always signal a coherence relation. Also, a preposition may give rise to different coherence relations depending on context, so one can not simply paste a label of &quot;the&quot; coherence relation into the representation. Instead, the computation of coherence relations is performed by testing relation-specific constraints on the configuration of the text-knowledge base resulting from semantic interpretation.</Paragraph> <Paragraph position="1"> Following standard practice, coherence relations are assumed to connect situations. So, we try to derive the relation between the conceptual correlate PROVIDE of the main verb and the inferred state HAS-PART-STATE, which stems from the conceptual interpretation of the prepositional phrase.</Paragraph> <Paragraph position="2"> The preposition &quot;mit&quot; (with) may signal an Explanation. The basis for computing this relation consists in recognizing that the matrix clause gives some kind of evaluation: namely, that the &quot;motherboard provides only little space.&quot;. The degree expression SPACE, with its associated negative POL--MARKER, (Staab and Hahn, 1997) is the trigger for recognizing the evaluative status of the matrix clause. We can now start looking for the signalled Explanation for this negative judgment.</Paragraph> <Paragraph position="3"> Explanation can be found quite easily: usually, a MOTHERBOARD has up to four PCI-SLOTS (see the default value that is denoted at the PART-OF relation in Figure 1). But actually the situation HAS-PART-STATE is a state in which only one is present, which is obviously &quot;little&quot;. The computations that derive the Explanation-relation are all performed by means of production rules.</Paragraph> <Paragraph position="4"> As a preposition might give rise to different coherence relations, the constraints of all those relations have to be checked. None of the constraints that are checked for other coherence relations signaled by 'knit&quot; (e.g. Condition) are satisfiable in this example. Similarly, for Example (l-a) no coherence relation would be derived, because the constraints would not be appicable. So, the computational approach is complementary to the semantic criterion of typicallity discussed in Section 2.</Paragraph> </Section> </Section> <Section position="5" start_page="48" end_page="50" type="metho"> <SectionTitle> 4 Interclausal Coherence </SectionTitle> <Paragraph position="0"> The proposed mechanism essentially treats the preposition as a kind of cue-phrase that determines a set of possible coherence relations, combining it with a inference mechanism. The main advantage of using such a combination is that it is possible to disambiguate cases in which cue-phrases may give rise to different relations without being confined to testing the complete range of coherence relations. This problem is not restricted to prepositions: (Knott and Sanders, 1998) actually build a multi-taxonomy of cue-phrases in which elements may give rise to several relations, depending on context. E.g, &quot;as&quot; has three connotations: &quot;as&quot; may signal temporal, causal or slml\]arity relationship.</Paragraph> <Paragraph position="1"> The presented approach can'ies over neatly to such cue-phrases with multiple relations: indeed, the process described in Section 3 can be seen as a formal reconstruction of the paraphrase. The conceptual representations that is derived for Sentence (2) is basically the same as for (l-b): in both cases a HAS-PART-STATE and a PART-OF relation are computed that connects the MOTHERBOARD and its PCI-SLOT. Finally, the computation of the coherence relation for Sentence (2) differs only with respect to the trigger: in Sentence (2), the cue-phrase &quot;so&quot; triggers the computation instead of the preposition ~mit&quot; in (l-b). The presented approach is thus able to handle coherence relations that operate on the intra-sentential level. Still, this is only a first step to account for the discourse structure of entire texts: cue-phrases are by no means always present. And even if there is a cue-phrase in a new unit to attach, it is still often not clear from the cue alone to which unit the new one should connect: (3) a. Apple presented a new member of the MessagePad organ/zer thmily.</Paragraph> <Paragraph position="2"> b. A new display and a new version of the operating system are introduced.</Paragraph> <Paragraph position="3"> c. However, the old MessagePad 120 is still available.</Paragraph> <Paragraph position="4"> The cue &quot;however&quot; alone does not give enough information to decide whether Sentence (3-c) should connect to (3-b) or to (3-a): further information is needed, like that there is a referential relation between the old MessagePad 120 and the MessagePad family. As a step towards connecting larger units, a recent study (Schauer, 2000) examined the interaction of discourse structure and referentim relations among nominM expressions.</Paragraph> <Section position="1" start_page="49" end_page="50" type="sub_section"> <SectionTitle> 4.1 Structural Dependencies </SectionTitle> <Paragraph position="0"> What seems intuitively clear, is that a theory of discourse structure should reflect the structure of the text under consideration. Rhetorical Structure Theory, however, concentrates on the effects intended by the author of a text.</Paragraph> <Paragraph position="1"> RST focuses on a '~pre-ree~|izational&quot; structure and is not primarily concerned with text phenomena. The results are analyses that do not account for the structural constraints in a text, here referential constraints: the dependency of an anaphoric expression on its antecedent - that it is resolvable and hence interpretable - can hardly be captured in RST.</Paragraph> <Paragraph position="2"> While this is of course not an issue for RST as a theory, it is a prerequisite for any system that wants to account automatically for a text's discourse structure. As an example consider the following fragment: (4) a. The Vaio F190 comes with convincing equipment.</Paragraph> <Paragraph position="3"> b. It features a DVD-ROM and a 6 GB harddisk.</Paragraph> <Paragraph position="4"> In classical RST, example (4-a) could be said to stand in an Evaluation relation to (4-b). The definition of Evaluation requires that the satellite evaluates the nucleus, see Figure 2. r However, this would not capture a structural dependency: the pronominal anaphora &quot;it&quot; ca.nnot be interpreted correctly without the In RST, one could reflect this referential dependency by analyzing (4-b) as giving Evidence for (4-a) (see Figure 3), as an Evidence relation may be applied when the hypothesis is the nucleus. Such an analysis would neglect the &quot;convincing&quot; in (4-a) which is quite an explicit evaluation.</Paragraph> <Paragraph position="5"> In order to account for the structural dependency and the appropriate semantic linkage, we propose to use a new kind of Evaluation-N(ucleus) relation. More generally, this loosens the relation between nucleus-satellite assignment and semantic content of coherence relations.</Paragraph> </Section> <Section position="2" start_page="50" end_page="50" type="sub_section"> <SectionTitle> 4.2 Structural Configurations </SectionTitle> <Paragraph position="0"> Analyzing texts using the outlined notion of structural dependencies as a basic guideline, we determined the structural configurations involving coreference relations. Basically, when two units contain coreferentiai expressions, they are usually connected by a coherence relation which subordinates the anaphoric unit. In the simplest case, the relation is one of Elaboration. However, further linguistic cues or inferences might give rise to semantically &quot;richer&quot; relations.</Paragraph> <Paragraph position="1"> When more than one coreference relation is involved, issues become more complex. For instance, consider the following example: (5) a. The Panasonic LC90S is a 19&quot;-display. b. Although its screen size of 482 ram corresponds to a conventional 21&quot;monitor, null c. considerably less space is required.</Paragraph> <Paragraph position="2"> d. The device can be attached to a videocard via a USB-comaector.</Paragraph> <Paragraph position="3"> Obviously, the nominal anaphor '2he device&quot; in (5-d) requires an antecedent. One possibility would be the pronominal &quot;it&quot; in (5-b), leading to a resolution to &quot;the LC90S&quot;. However, this is not reflected in the discourse structure that seems most appropriate, of.</Paragraph> <Paragraph position="4"> size) is not further elaborated on in (5-d), so one might say there is a mini-segment boundary between these two sentences. Hence, it would also be not correct to analyze (5-d) as an Elaboration of (5a-c), because (5-d) elaborates only (5-a). So, unless further connections (stemming from cue-phrases, inferences or whatever) license a topic continuation, a new unit containing anaphora will connect to the upmost unit bearing their antecedents.</Paragraph> </Section> </Section> <Section position="6" start_page="50" end_page="52" type="metho"> <SectionTitle> 5 From References to Coherence </SectionTitle> <Paragraph position="0"> The configurations just described naturally lead to a combined account of deriving a text's discourse structure and resolving its referring expressions, repeated from (Schauer, 2000) in Figure 5. Basically, the algorithm uses the successful resolution of anaphoric expressions as a guideline for determining the target unit to which a new unit should be connected which in turn restricts the set of units which should be searched for resolving further referring expressions.</Paragraph> <Paragraph position="1"> The algorithm assumes the availability of several capabilities for anaphora resolution.</Paragraph> <Paragraph position="2"> First, for a given unit candidate s a set of noun phrases needs to be identified that may be anaphoric expressions. Second, some resolution process is necessary that checks whether an anaphoric expression can be resolved in a given list of possible antecedents. This process makes heavy use of underlying (domain) knowledge (Hahn et al., 1996): in Example (1), &quot;the P6LXZ-A&quot; is an instance of a MOTHERBOARD. Since ~the motherboard&quot; in (l-a) is a definite noun phrase and syntactic as well as conceptual information match with the plausible antecedent &quot;the P6LXZ-A', a referential link can be established, see the IS-COREFERENTIAL relation in Figure 6.</Paragraph> <Paragraph position="3"> Abstracting away from the actual implementation, the algorithm operates over a tree structure that manages lists of accessible an-SFor the sake of argument and because the algorithm is taken from (Schauer, 2000), clauses will be taken as units for the moment. The issue will be discussed below in more detail.</Paragraph> <Paragraph position="5"> /, found at least one antecedent node ,/ if target_node t:hen /, connect new unit to old node ,/ tecedent candidates and reflects the discourse structure in terms of coherence relations that has been derived so far. 9 The algorithm now loops through all clauses of a text, building up both the tree and the antecedent lists incrementally. Whenever a new clause has to be considered, its left context consisting of the right frontier of the tree is checked for plausible antecedents. When all accessible antecedent nodes are determined, the highest node that provides antecedents for all resolvable anaphoric expressions in the new unit is taken as the target node, in accordance with the discussion in Section 4.2.</Paragraph> <Paragraph position="6"> If a new unit contains no referential expression then the algorithm makes no prediction. If the target node has been found, the new unit is connected to it, i.e. the new unit is established as a satellite to the target unit. This means that the new unit opens a new right-most branch and hence becomes the lowestright node of the tree. So, the new right frontier consists of the newly attached unit, the structure is shown in Figure 6, while the storage of accessible referential objects is managed elsewhere in the system as it interacts with the syntactic parser. modified node and its predecessors.</Paragraph> <Paragraph position="7"> The evaluation in (Schaner, 2000) showed that the predictions made by the algorithm depend crucially on the size of the elementary units considered. When clauses are considered to be the elementary unit size, the predictions are correct in up to 81% of the cases in which the algorithm makes a prediction -under the pre-condition that intra-sentential units axe handled first. Linking units by referentiai makes surely no sense when the units considered axe phrases: of course, syntactic constraints always mandate that intra-clausal units are related with each other first.</Paragraph> <Paragraph position="8"> This algorithm is only a useful approximation towards a complete account of a text's discourse structure. Relating to the highest unit providing antecedents works only when there is a mini segment boundary every time an anaphoric expression is used. Although the algorithm ignores other sources of relations between units -- which are exactly the cases where the algorithm fails -- the results are surprisingly good. As such, enhancing the cue-phrase-only approach by additionally considering coreference relations is in the middie of cheap surface-only approaches and very</Paragraph> <Paragraph position="10"> expensive inference-only methods. A complete approach will also try to account for inter-sentential cue-phrases and also further intercounections, based on the computed representation, see Figure 6) 0</Paragraph> </Section> <Section position="7" start_page="52" end_page="53" type="metho"> <SectionTitle> 6 Related Work </SectionTitle> <Paragraph position="0"> Rhetorical Structure Theory (Mann and Thompson, 1988) was basically used as the theory on which the presented arguments were grounded. Unfortunately, its definitions are intended to be applied by human annotators and have long resisted approaches to automation because of their vagueness. Previous approaches to (automatically) deriving a text's discourse structure either tried to rely on purely surface-based criteria (cuephrases, (Marcu, 1998)) or concentrated on inferences about representations of discourse units (Asher, 1993; Hobbs et al., 1993). However, neither of these approaches tries to account for phrases as discourse units. Two re- null evaluative nature of Sentence (l-b), see the discussion in Section 3.</Paragraph> <Paragraph position="1"> cent studies mention the role of PPs as carriers of coherence relations (Vander Linden and Martin, 1995; Grote et al., 1997), but only for the purpose of text generation.</Paragraph> <Paragraph position="2"> As prepositions (and cue-phrases in general) can signal different coherence relations, the presented computational approach couples a cue-phrase approach like (Marcu, 1998) with inferences using the computed semantic representation. Only recently, there have been some approaches (Webber et al., 1999) that acknowledge the need to incorporate both a treatment of cue-phrase approaches and inferences. However, how these inferences take place is not addressed, neither is the level of intraclansal coherence, that, as we have argued, substantially relies on an integration of beth veins. Purely inferential approaches like (Asher, 1993; Hobbs et al., 1993) have also been criticized for being not really applicable because of computational costs. The presented approach tries to minimize these costs by relying on surface-based criteria that reduce the set of coherence relations that have to be considered.</Paragraph> <Paragraph position="3"> Cue-phrases are not necessarily alone responsible for the discourse structure of texts. Hence, referential relations bel~ween nominal expressions were suggested as means for finding attachment points of new traits that need to be incorporated. (Corston, Oliver, 1998) enlarges the cue-phrase approach of (Marcu, 1998) in a vein sJmi\]ar to the one presented in Section 5. However, how several coreference relations interact with the resulting discourse structure, is not spelled out. Also, in his approach, it remains uncle~ how the correct target node to attach to is identified. Segmented Discourse Representation Theory (SDRT), as described in (Asher, 1993), provides a framework of discourse structure which interacts with referential accessibility constraints. Asher does not rely on coreferences for establishing target units; instead the derivation of a coherence relation (and thereby of the target unit to connect a new unit to) relies on rather abstract connections between &quot;events&quot;. While recognizing coreference relations certainly also requires domain knowledge and inference capabilities, recognizing connections between events seems an even more impossible task. The presented approach is hence more light-weight in nature.</Paragraph> <Paragraph position="4"> (Webber et al., 1999) apply operations on trees for incorporating cue-phrases in a grammar-driven approach to discourse structure. The presented algorithm could be integrated with their approach, to account for the cases of units that are not connected by referential relations but via cue-phrases.</Paragraph> </Section> class="xml-element"></Paper>