File Information
File: 05-lr/acl_arc_1_sum/cleansed_text/xml_by_section/metho/95/j95-1002_metho.xml
Size: 64,367 bytes
Last Modified: 2025-10-06 14:13:59
<?xml version="1.0" standalone="yes"?> <Paper uid="J95-1002"> <Title>Expressing Rhetorical Relations in Instructional Text: A Case Study of the Purpose Relation</Title> <Section position="4" start_page="31" end_page="33" type="metho"> <SectionTitle> 3. Corpus Collection and Representation </SectionTitle> <Paragraph position="0"> The corpus developed for this study was taken from various types of instructional text, including instruction booklets, recipes, and auto-repair manuals. It contains approximately 1000 clauses (6000 words) of instructions, taken from 17 different sources representing a diverse array of process types. These sources include instructions for electronic devices like cordless telephones and clock radios, manipulative processes like auto repair and first aid, and creative processes like cooking and craft making.</Paragraph> <Paragraph position="1"> The one common feature is that they all involve the expression of actions and of the procedural relations between them. 1 As an example of the nature of this text, consider the following excerpt from the instructions for the GTE Airfone (Airfone 1991), which will be called the Remove-Phone text: (2) When instructed (approx. 10 sec.) remove phone by firmly grasping top of handset and pulling out. Return to seat to place calls. (Airfone 1991) 2 1 It should be noted that this corpus is much smaller than the language corpora used in larger statistical studies (Church and Mercer 1993). The deep semantic and pragmatic knowledge that was required for the current study has necessitated this.</Paragraph> <Paragraph position="2"> 2 This paper will add a reference to the end of all examples that have come directly from the corpus, indicating the manual from which they were taken. Examples of actual IMAGENE output will be fully italicized. All other examples are contrived for explanatory purposes.</Paragraph> <Paragraph position="3"> Keith Vander Linden and James H. Martin Expressing Rhetorical Relations This passage gives an example of the variation of expressional form that is common in instructional text. It contains, among other things, two expressions of purpose: &quot;remove phone&quot; and &quot;to place calls.&quot; This notion of purpose, which will be detailed in the next section, is one of actions that are to be realized through the execution of expressed sub-actions. The first is stated as an imperative (&quot;remove the phone&quot;), with the sub-actions expressed in participial form within a by prepositional phrase (&quot;by firmly grasping top of handset and pulling out&quot;). The second (&quot;to place calls&quot;), on the other hand, is expressed in final position as a to infinitive, with its sub-action stated as an imperative (&quot;return to seat&quot;). The problem to be addressed by the corpus analysis in step 2 is to determine the contextual features used to choose these forms as opposed to the alternate forms that could have expressed the &quot;same basic information.&quot; A relational-style database is used to represent the rhetorical, grammatical, and lexical aspects of the corpus. The representation of the grammatical form of the clauses and phrases is based on traditional principles of syntax and semantics. Clauses and phrases are represented in separate tables. Links within the clause table are used to indicate subordinate relations, and links between the clause and phrase tables are used to represent relative clauses and predicate-argument relations. It also includes semantic information such as the agent of a particular action (e.g., the reader or the device) and the semantic type of the predication (e.g., material or relational process).</Paragraph> <Paragraph position="4"> More detail on the database can be found elsewhere (Vander Linden 1993c). The goal was to allow for the representation of any element of the pragmatic, semantic, or syntactic context that might be relevant in the analysis.</Paragraph> <Paragraph position="5"> Mann and Thompson's RST has been used to encode the rhetorical relations between expressions in the corpus (Mann and Thompson 1987, 1988). It was developed as a framework for describing text structure, viewed in terms of the semantic and pragmatic relations that hold between text spans at all levels. The current study has made use of five such relations: Purpose, Precondition, Result, Sequence, and Concurrent. This section will now make some general observations concerning RST, present an example RST analysis of the Remove-Phone text, and conclude with definitions of the relations used in the study.</Paragraph> <Paragraph position="6"> RST distinguishes between what are called nucleus-satellite and multi-nuclear schemata. The nucleus-satellite schema relates two spans of text: one designating a more central span, called the nucleus, and a more peripheral one, called the satellite. This relation is represented graphically with a directed arrow from the satellite to the nucleus. Definitions of relations of this sort specify constraints that apply to the nucleus (N), the satellite (S), and the combination of the two and specify the effects of the expression.</Paragraph> <Paragraph position="7"> Purpose, Precondition, and Result are examples of such relations. The multi-nuclear schema relates one or more spans, designating no span as superordinate or subordinate to any other. Definitions of relations of this sort include specification of the constraints on the nuclei and the combination of nuclei, as well as a specification of the effect of the expression. The Sequence and Concurrent relations are such relations.</Paragraph> <Paragraph position="8"> RST was attractive for the IMAGENE project because of its ability to represent the hierarchical structure of text with rhetorical structures that matched the level of analysis required for the study of expressions of procedural relations. There is considerable debate in the field of discourse analysis concerning the relative importance of intentional structure and rhetorical relations (e.g., Grosz and Sidner 1986; Moore and Pollack 1992), most systems focusing on one or the other. The current study has conflated them, as the instructional texts have not tended to display the complex intentional structure common to persuasive texts and interactive discourses (Vander Linden 1993b).</Paragraph> <Paragraph position="9"> Finally, RST has been used by many researchers for the purpose of text generation (e.g., Moore and Paris 1988; Hovy and McCoy 1989; Scott and Souza 1990; R6sner The RST representation of the Remove-Phone text.</Paragraph> <Paragraph position="10"> and Stede 1992a). This testifies not only to RST's usefulness, but also to the direct applicability of the results of the current study to the field of natural language generation. Because of the common use of RST, the results can be more easily applied to other work in this area. In particular, the focus on the precise forms of expression of rhetorical relations in instructional text fills an important gap in current work (see the work of Scott and Souza \[1990\] on expressing rhetorical relations).</Paragraph> <Paragraph position="11"> Consider the application of RST to the Remove-Phone text. The first problem that must be addressed in any RST analysis is the segmentation of the text into spans that will serve as the atomic units of description. In RST, these spans have typically been clauses, as is the case in the Remove-Phone passage, but certain phrases with propositional content may be considered as well. The spans used in our analysis are propositional units that express single actions. In the Remove-Phone text, there are six such action expressions, listed here in segmented form: by firmly grasping top of handset and pulling out.</Paragraph> <Paragraph position="12"> Return to seat to place calls.</Paragraph> <Paragraph position="13"> The second problem is one of relating these segments in the appropriate rhetorical structure. The current study has used the two aspects of the RST specification that can be mapped onto the procedural structure of the process being expressed, namely, the hierarchical structure of RST and the subset of RST relations that correspond to procedural relations. Each of these two aspects will be discussed with reference to the RST representation for the Remove-Phone text, shown in Figure 1. 3 The first aspect of this structure is its hierarchical nature. The procedural sequence schema at the top of the text hierarchy, for example, indicates that there are two</Paragraph> </Section> <Section position="5" start_page="33" end_page="36" type="metho"> <SectionTitle> 3 The actual RST and grammatical analyses of the text were performed by one of the authors, and the </SectionTitle> <Paragraph position="0"> examples crucial to the formalization of the results were reviewed by both authors. This approach would be difficult in the analysis of certain more complex texts such as persuasive texts, but proved to be adequate in the study of local structure in instructional text.</Paragraph> <Paragraph position="1"> Keith Vander Linden and James H. Martin Expressing Rhetorical Relations spans of text that express a sequence of two actions. The spans themselves can be expressed as single propositional units or as more complex spans, the latter being the case with the two spans in this sequence. This is a representational manifestation of the hierarchical nature of the processes themselves and is displayed graphically by extending the horizontal line of a span to cover all of its subordinate spans. The RST representation of the Remove-Phone text is a small portion of the full hierarchy that represents the entire manual. The current study has focused on the expression of just such local sub-trees; the problems of expression of macro-structure are beyond the scope of the analysis (see Mooney, Carberry, and McCoy 1991).</Paragraph> <Paragraph position="2"> The second aspect of this structure is the nature of the rhetorical relations themselves. The representation makes use of five relations: Purpose, Precondition, Result, Sequence, and Concurrent, which are used as abstractions to identify the lexical and grammatical manifestations of the procedural relations inherent in the process. They are termed informational by Moore and Pollack (1992) and subject matter by Mann and Thompson (1987) because they are based on semantic content rather than on intentional or presentational content. 4 This section will now provide specific definitions of these relations.</Paragraph> <Paragraph position="3"> PURPOSE (taken from Mann and Thompson 1987) constraints on N: presents an activity constraints on S: presents a situation that is unrealized constraints on the N+S combination: S presents a situation to be realized through the activity in N the effect: R recognizes that the activity in N is initiated in order to realize S The Purpose relation is taken directly from the RST specification. In this paper, it refers to a situation in which a higher level activity is realized through the execution of lower level sub-steps. An example can be found in the Remove-Phone text cited above: &quot;... remove phone by firmly grasping top of handset and pulling out.&quot; Here the activity of removing the phone is realized by the execution of the sub-steps of grasping and pulling the phone.</Paragraph> <Paragraph position="4"> PRECONDITION (taken from R6sner and Stede 1992a) constraints on N: presents an action constraints on S: presents an unrealized situation constraints on the N+S combination: S must be realized in order to make it possible or sensible to carry out N the effect: R recognizes that situation S must be realized in order to successfully carry out action N The Precondition relation is a simple amalgam of the standard RST relations Circumstance and Condition. It has been taken from R6sner and Stede's work on generating multilingual instructions (R6sner and Stede 1992a). This particular combination has proven useful in analyzing various kinds of conditions and circumstances that 4 There is some question as to whether these semantically based relations should be termed rhetorical in the classic sense at all (Dale 1993). Because of the prevalence of this use of the term, however, it will be retained in this paper.</Paragraph> <Paragraph position="5"> Computational Linguistics Volume 21, Number 1 frequently arise in instructions, such as the precondition found in the Remove-Phone text: &quot;When instructed (approx. 10 sec.) remove phone .... &quot; Here, the removal of the phone must not be attempted until after the device has instructed the user to do so. RESULT (adapted from constraints on N: constraints on S: constraints on the N+S the effect: Mann and Thompson 1987) none presents either a volitional or non-volitional action or the situation that could have arisen from one combination: N presents a situation that could have caused the situation presented in S; presentation of N is more central to W's purposes in putting forth the N-S combination than is the presentation of S.</Paragraph> <Paragraph position="6"> R recognizes that the situation presented in N could be a cause for the action or situation presented in S The Result relation is a simple amalgam of RST's volitional and non-volitional results. It was useful for analyzing expressions of actions or situations that are expressed as being the result of other actions, as in &quot;Place the handset in the base. The BATTERY CHARGE INDICATOR will light.&quot; (Excursion 1989). Here, the device's action of lighting the indicator is a result of the reader's action of placing the handset in the base. SEQUENCE (taken from Mann and Thompson 1987) constraints on N: multi-nuclear constraints on the combination of nuclei: A succession relationship between the situations is presented in the nuclei the effect: R recognizes the succession relations among the nuclei The Sequence relation is taken directly from the RST specification and refers to actions that are in temporal sequence, as in the following excerpt from the Remove-Phone text: &quot;... by firmly grasping top of handset and pulling out.&quot; CONCURRENT (adapted from Mann and Thompson 1987) constraints on N: multi-nuclear constraints on the combination of nuclei: A simultaneous relation between distinct situations is presented in the nuclei the effect: R recognizes the simultaneous relations among the nuclei Finally, the Concurrent relation is a simple extension of Sequence, referring to actions that are distinct but simultaneous. An example can be found in &quot;Press and hold the mouse button while you move the mouse.&quot; (Macintosh 1988). Here, holding the mouse button and moving the mouse must be done simultaneously. Keith Vander Linden and James H. Martin Expressing Rhetorical Relations</Paragraph> </Section> <Section position="6" start_page="36" end_page="38" type="metho"> <SectionTitle> 4. Corpus Analysis </SectionTitle> <Paragraph position="0"> Two related issues must be addressed in the corpus analysis: .</Paragraph> <Paragraph position="1"> .</Paragraph> <Paragraph position="2"> Determining the range of expressional forms commonly used by instructional text writers.</Paragraph> <Paragraph position="3"> Determining the precise communicative context in which each of these forms is used.</Paragraph> <Paragraph position="4"> With a couple of minor exceptions, the study was performed exclusively on three instruction manuals for cordless telephones (approximately one-third of our corpus), and the results were applied to the remainder of the corpus. The exceptions involved so that and until expressions and expressions of concurrency, which were not well represented in the telephone manuals. Examples of these expressions were taken from the remainder of the corpus. The method employed therefore has much in common with the method proposed by Quinlan and implemented in ID3 (Quinlan 1986) in that the training set is expanded in cases where there are insufficient examples on which to base a full analysis. So far, the testing set has not been expanded to include examples on which to test these so that, until, and concurrent expressions.</Paragraph> <Section position="1" start_page="36" end_page="37" type="sub_section"> <SectionTitle> 4.1 The Range of Expressions </SectionTitle> <Paragraph position="0"> The first task is that of determining the range of lexical and grammatical forms used to express each particular rhetorical relation. The full corpus contains 119 purpose expressions, all but four of which occur in one of the following seven forms (the purpose, i.e., the satellite span of the rhetorical relation, is italicized): (3a) To end a previous call, hold down FLASH \[6\] for about two seconds, then release it. (Code-a-phone 1989) (3b) Follow the steps in the illustration below, for desk installation. (Code-a-phone 1989) (3c) The OFF position is primarily used for charging the batteries. (Code-a-phone 1989) (3d) For frequently busy numbers, you'll want to use REDIAL \[7\], and the pause will have to be in Redial memory. (Code-a-phone 1989) (3e) When instructed (approx. 10 sec.) remove phone by firmly grasping top of handset and pulling out. (Airfone 1991) (3f) Return handset to wall unit from which it was taken. Insert heel first as shown, then push top in firmly. (Airfone 1991) (3g) Tilt pan down slightly at the rear so that the fluid drains out. (Reader's Digest 1981) All four of the issues of lexical and grammatical choice we addressed are displayed here. The purpose expressions can be textually placed in the slot either before or after the expression of their sub-actions. Furthermore, there are seven combinations of grammatical form, linker, and clause combining to choose from, the relative frequencies and percentages of which are given in Table 1 (where the letters in example (3) correspond to the letters in the table). Example (3a) uses a to infinitive form (TNF). (b) For-Nominalization 2 7 9 7.5% (c) For-Gerund 0 3 3 2.5% (d) For-Goal-Metonymy 1 5 6 5.0% (e) By-Purpose 11 1 12 10.0% (f) Adjoined-Purpose 4 0 4 3.3% (g) So-That-Purpose 0 10 10 8.4% Other 4 4 3.3% Example (3b) uses a for prepositional phrase with a nominalization (&quot;installation&quot;) as the complement. Example (3d) uses a for preposition with a gerund phrase as the complement. Example (3c) uses a for preposition with a noun phrase that refers to the object (or goal) of the corresponding action as the complement. This is termed Goal Metonymy. Example (3e) uses a simple imperative for the purpose with by conjoining participial forms of the intended actions. Example (3f) uses a simple imperative for the purpose, with the intended actions in a separate sentence following the purpose. Example (3g) uses a simple imperative for the intended actions, with a so that conjoining a present tense action form of the purpose. 5</Paragraph> </Section> <Section position="2" start_page="37" end_page="38" type="sub_section"> <SectionTitle> 4.2 The Context of Expression </SectionTitle> <Paragraph position="0"> The second task, that of determining the functional context in which each of the forms is used is more difficult. The IMAGENE project employs a hypothesis generation and test cycle, such as the one advocated by Cumming (1990) in an attempt to identify correlations between the contextual features of communicative context on the one hand, and the lexical and grammatical forms on the other.</Paragraph> <Paragraph position="1"> This methodology starts with the range of lexical and grammatical forms corresponding to each of the rhetorical relations considered. In the hypothesis phase, the analyst hypothesizes a feature of the communicative context that appears to correlate with the variation of some aspect of the lexical and grammatical forms. These hypotheses may come from an intuitive analysis of the texts, as well as from the current literature on the subject. The features themselves pertain to any of three aspects of the communicative context (termed metafunctions in systemic linguistics): Ideational--the propositional meaning of the material being expressed (associated with the traditional notion of semantics); Textual--the flow and structure of the text (associated with discourse analysis); and Interpersonal--the human relationships between interlocutors (associated with socio-linguistics) (Halliday 1985). All of these types of features have proven relevant in the analysis. In the test phase, the analyst attempts to validate the hypothesis by querying the database for the relevant information. These two phases are repeated until a good match is achieved or until a relevant hypothesis cannot be found.</Paragraph> <Paragraph position="2"> As an example of this methodology, consider the issue of slot, that is, the determination of which span in a rhetorical relation should be expressed first. The slot of 5 This study has temporarily characterized these so that expressions as action/sub-action expressions. A more detailed analysis of how the situations that give rise to them differ from those for other purposes is yet to be performed. Keith Vander Linden and James H. Martin Expressing Rhetorical Relations purpose expressions in our corpus is split fairly evenly between initial and final purpose expressions. Of the 119 purpose expressions, approximately 48% are fronted and 52% are not fronted. These values are similar to the percentages of initial and final purpose clauses found by Thompson (1985) for procedural texts. Although she reports just over 18% initial purpose clauses for English text in general, she reports 49% initial purpose clauses for a book of recipes and 34% for an auto-repair manual.</Paragraph> <Paragraph position="3"> Thompson's study indicated that one common feature of fronted purpose clauses is that their scope is global, that is, there is more than one expressed proposition that is directly related to the fulfillment of the purpose (e.g., &quot;To achieve purpose A, do B and do C.&quot; where there are two sub-actions, B and C). Such a purpose clause is expressing a context in which the prescribed sub-actions are to be interpreted and thus should be fronted. This provides a good starting hypothesis for determining the slot of purposes, namely, that global purpose clauses are fronted. Upon inspection, we find that only three cases of global purpose expressions in our corpus (7.9%) are not fronted. This yields strong support for the hypothesis and allows us to go on and discern what factors motivated the counter-examples. This process continues either until no distinctions can be found, or until there are not enough examples on which to base the distinctions.</Paragraph> <Paragraph position="4"> The next question to be asked is whether this is the only explanation of fronted purpose expressions, or whether there are other fronted purpose expressions that are not global. This can be addressed by querying all the fronted, non-global purpose clauses in the corpus. In the current corpus, this set is non-empty, which leads to the conclusion that there are other factors at work in the question of purpose slot. The iterative process of hypothesis generation and testing can then be conducted on these other cases in a similar manner.</Paragraph> <Paragraph position="5"> This analysis technique is designed to identify covariation between elements of the communicative context on the one hand and grammatical form on the other. The purpose slot analysis just discussed, for example, identifies the existence of a covariation between purpose scope and slot. This covariation, however, does not constitute proof that the technical writer actually considers the issue of slot during the generation process, nor that the prescribed form is actually more effective than any other. Proof of either of these issues would require psycholinguistic testing. This study provides detailed prescriptions concerning how such testing should be performed, i.e., what forms should be tested and what contexts controlled for, but it does not actually perform the tests. A discussion of this may be found elsewhere (Vander Linden 1993a).</Paragraph> </Section> </Section> <Section position="7" start_page="38" end_page="42" type="metho"> <SectionTitle> 5. IMAGENE </SectionTitle> <Paragraph position="0"> This section will discuss the theoretical framework of the implementation and then detail the treatment for purpose expressions. The full IMAGENE architecture, as depicted in Figure 2, consists of a System Network and a Sentence Building routine and is built on top of Penman. It transforms inputs (shown on the left in Figure 2) into instructional text (shown on the right).</Paragraph> <Section position="1" start_page="38" end_page="42" type="sub_section"> <SectionTitle> 5.1 IMAGENE's Architecture </SectionTitle> <Paragraph position="0"> Penman, a sentence-level generator developed at the USC Information Sciences Institute (Mann 1985; Penman 1989), was employed not only because of its broad coverage of English syntax, but also because it is based on a Systemic-Functional view of language (Halliday 1976). The Systemic view is distinctly functional, that is, it is particularly interested in mapping elements of the communicative context onto the appropriate grammatical forms. As a by-product of this view of language, Penman contains a well-developed implementation of the System Network, the Systemic formalism for representing grammar.</Paragraph> <Paragraph position="1"> The system network is basically a decision network in which each choice point distinguishes between alternative features of the communicative context. It has been used extensively in Systemic Linguistics to address both sentence-level and text-level issues (e.g., Berry 1981; Patten 1988; Fawcett 1990). Such networks are traversed based on the appropriate features of the communicative context, and as a side effect of this traversal, linguistic structures are constructed by realization statements that are associated with each feature of the network. Penman's networks are specifically designed to construct English sentences.</Paragraph> <Paragraph position="2"> IMAGENE's system network is built in a similar manner, but because it constructs text structures rather than sentences, its realization statements have a flavor significantly different from their counterparts in the grammar developed for Penman. 6 We now give a short discussion of how IMAGENE'S realization statements can manipulate the evolving text structure, making reference to the text structure produced by IMAGENE for the portion of the Remove-Phone text shown in Figure 3 (which corresponds to the text span &quot;... remove phone by firmly grasping top of handset and pulling out&quot;). The full analysis of the Remove-Phone text will be given later; here we intend to illustrate only the types of manipulations made by the realization statements.</Paragraph> <Paragraph position="3"> * Inserting nodes into the text structure (iterative-insert, insert, copy)--IMAGENE starts with an empty text structure and uses these statements to insert action nodes as appropriate. In Figure 3, for example, the Remove-Action, Grasp-Action, and Pull-Action nodes refer to the actions of removing, grasping, and pulling the handset.</Paragraph> <Paragraph position="4"> * Ordering the surface expression of the nodes (order, reorder, insert-order, combine)--IMAGENE uses these statements to specify the 6 Penman's sentence-level realization statements work with a single, prespecified list of features of the sentence, called grammatical functions, such as ACTOR, PROCESS, GOAL, and THEME. At the text-level, there is no definitive list of what might be called text functions. Rather, IMAGENE'S realization statements allow the insertion of subscripted elements that can correspond to lists of sequential commands, multiple preconditions, etc.</Paragraph> <Paragraph position="5"> A segment of the Text Structure for the Remove-Phone example.</Paragraph> <Paragraph position="6"> order of expression of the action nodes. The choice of clause-combining strategies is made here as well. In Figure 3, the links are shown as lightfaced directed arrows, marked with either New-Sentence or Continue-Sentence (in this segment, only the latter is used). Reorder is used to change an ordering made earlier in the processing, whereas the others establish the order of newly inserted nodes.</Paragraph> <Paragraph position="7"> Building the RST structure between nodes (structure, unlink)--IMAGENE uses structure to create hierarchical text structure links between nodes and unlink to remove them. Figure 3 contains a purpose relation and a sequential relation. As with reorder, unlink is used to &quot;un-structure&quot; a default structuring.</Paragraph> <Paragraph position="8"> Marking the lexical and grammatical forms of expression of the nodes (mark, iterative-mark)--The realization statements also determine the grammatical form of the expression of each of the nodes in the structure.</Paragraph> <Paragraph position="9"> In Figure 3, the Remove-Phone node, for example, is marked as an imperative, and the Grasp-Action as an ing form with the linker by.</Paragraph> <Paragraph position="10"> IMAGENE'S network consists of approximately 70 systems. It maps those features of the communicative context deemed relevant in the corpus analysis performed in step 2 onto the appropriate lexical and grammatical forms for expressing each action. The network, having the basic high-level structure shown in Figure 4, performs the two basic functions of Content and Rhetorical Status Selection and Grammatical Form Selection. We view Content Selection as the process of choosing the appropriate actions from the process plan to express and Rhetorical Status Selection as the process of choosing the appropriate rhetorical relation to be used in expressing each of these actions. IMAGENE contains a sub-network implementing these two processes that is currently very preliminary (Vander Linden 1993c). 7 This paper focuses on the Grammatical Form Selection portion of the network, that is, the choice, given an action to be expressed and its rhetorical status, of the appropriate lexical and grammatical forms of expression. We will present a detailed discussion of the purpose sub-network below, which is representative of the other grammatical form sub-networks.</Paragraph> <Paragraph position="11"> 7 Paris (1988, 1993) discusses this issue in more detail, particularly as it pertains to user modeling. A high-level view of the systems in the network.</Paragraph> <Paragraph position="12"> The input to IMAGENE is the set of features of the functional context that affect the form of expression of the plan, called the text-level inquiries in Figure 2. This input is implemented as a set of responses to the inquiries made by the IMAGENE system network pursuant to determining the appropriate path to be taken through the IMAGENE system network. They are analogous to Penman's sentence-level inquiries. Currently, the data structures and code necessary to respond to the inquiries automatically have not been implemented. Rather, the inquiries are answered manually, allowing us to focus on determining the appropriate set of inquiries and the precise lexical and grammatical consequences of the responses of these inquiries. The dashed lines in Figure 2 indicate some of the information sources that the inquiry implementations will access (i.e., the Process Structure, the Penman lexicon and grammar, and the evolving text structure). As a side effect of traversing the network, IMAGENE'S realization statements automatically realize these consequences in a text structure (also shown in Figure 2). The Text Structure, to be discussed more fully in Section 5.3, is represented in IMAGENE'S Text Representation Language (TRL). TRL itself is implemented in LOOM (MacGregor and Bates 1987; Loom 1993).</Paragraph> <Paragraph position="13"> A second input shown in Figure 2, the Process Structure, is a representation of the process to be expressed. It is built in IMAGENE'S Process Representation Language (PRL), which is also implemented in LOOM and will also be discussed in Section 5.3. It is a representation like that produced by a procedural planner, containing the procedural hierarchy of the process being expressed as well as some information about the lexical items used to express each action and its arguments. It is currently built by hand, which allows us to focus on the problem of expression rather than on planning. As previously mentioned, the Process Structure will eventually become a fundamental source of procedural information for the text inquiries. Currently, however, it is simply used by the final component of the architecture, the Sentence Builder, to specify the appropriate lexical items and case structures for the action input to the text-level inquiries. The Sentence Builder uses the lexical information given in the Process Structure just described, to translate the Text Structure, described above, into the appropriate sentence specification to be passed to Penman for surface realization. This specification is coded in terms of Penman's Sentence Planning Language (SPL) (Kasper 1989), a language that allows the specification of the lexical items and gram- null Keith Vander Linden and James H. Martin Expressing Rhetorical Relations matical structures to be generated by Penman. The translation process is performed by a recursive descent of the Text Structure hierarchy. One SPL command is produced for each sentence in the Text Structure.</Paragraph> </Section> <Section position="2" start_page="42" end_page="42" type="sub_section"> <SectionTitle> 5.2 Purpose Relations in Instructional Text </SectionTitle> <Paragraph position="0"> Purpose expressions arise in the context in which actions are viewed as being related hierarchically, that is, in which one higher level action is realized by the execution of a set of lower level actions. 8 As Section 4 indicated, there are a large number of lexical and grammatical forms in which such procedural relations are typically expressed, each used in a particular functional context. This section discusses the systems that have been included in IMAGENE to distinguish among these contexts.</Paragraph> <Paragraph position="1"> There are other studies of purpose expressions from the point of view of representation and understanding that are of use here (Di Eugenio 1992; Balkanski 1992).</Paragraph> <Paragraph position="2"> Di Eugenio, for example, has worked with by purposes and to infinitive (TNF) purposes in the context of understanding, but does not appear to have distinguished the two forms in her analysis of the procedural relations between actions. The current study is critically interested in discerning principled reasons for choosing between these sorts of expressions.</Paragraph> <Paragraph position="3"> The issues of slot and form of purpose expressions are treated largely independently by IMAGENE. The slot is determined by the sub-network shown in Figure 5. The form is determined by the sub-networks shown later in Figures 7 and 8. This portion of the system network is capable of generating a greater range of purpose expressions than is typical in generation systems and of identifying the functional reasons for choosing one form over the other. It should be noted that while the determinations made by the systems are based solely on the results of the corpus analysis conducted in step 2, the following sections will include intuitive motivations for the realizations the systems make.</Paragraph> <Paragraph position="4"> in Figure 5, formalizes Thompson's notion of the &quot;vastly different functions&quot; for initial and final purpose clauses (Thompson 1985) in the context of instructional text. It typically places the purpose expression in the final position. The exceptions to this are when the scope of the purpose is global, the purpose is considered optional, or the purpose is considered contrastive. These three exceptions are handled by the three systems depicted in Figure 5. 9 The first exception, handled by the Scope system, concerns the number of actions pertained to by the purpose. This correspondence between global purposes and fronted purpose expressions was already discussed in the corpus analysis section, but to give an intuitive feel for this empirical result, consider the awkwardness of restating example (3a) as &quot;?? Hold down FLASH \[6\] for about two seconds, then release it to end a previous call. &quot;1deg The restatement seems to imply, incorrectly, that the purpose applies 8 Goldman has termed these hierarchical relations generation relations (Goldman 1970). A detailed discussion of them can be found elsewhere (Balkanski 1993; Di Eugenio 1993).</Paragraph> <Paragraph position="5"> 9 In the system network notation, vertical lines indicate decision points. The boldfaced names are systems, the normal font names are features, and the italicized names are realization statements. The ordering realization statements are denoted with the operators > and \], meaning order the clauses in the same sentence and order the clauses in separate sentences, respectively. More detail on this notation can be found elsewhere (Winograd 1983, ch. 6).</Paragraph> </Section> </Section> <Section position="8" start_page="42" end_page="51" type="metho"> <SectionTitle> 10 The &quot;??&quot; notation is used to denote a possible form of expression that is not typically found in our </SectionTitle> <Paragraph position="0"> corpus; it does not indicate ungrammaticality.</Paragraph> <Paragraph position="1"> A structural view of purpose demotion. to the last action alone rather than to the sequence of actions. The fronted form, in example (3a), makes no such implication.</Paragraph> <Paragraph position="2"> As can be seen in Figure 5, the Global feature of the Scope system contains five realization statements, all making changes to the evolving text structure. In the Remove-Phone text, for example, these statements restructure or demote the Remove action in the hierarchical structure shown in Figure 6A into a satellite node in the RST structure shown in Figure 6B. They do not, however, actually contain the realization statement to set the textual order of the purpose expression (which would read Purpose > Nucleus); later systems in this branch of the decision network execute this statement. The remaining exceptions occur when the purpose is considered optional or contrastive and are handled by Optionality and Contrastiveness, respectively. Here are examples of them from the corpus: (4) For more information and wall installation instructions, see the Installation Notes on page 3. (Code-a-phone 1989) (5) To place call, dial AREA CODE and NUMBER. To end call, press red HANG UP button. (Airfone 1991) * Keith Vander Linden and James H. Martin Expressing Rhetorical Relations In example (4), the action of getting more information is optional, that is, the reader may or may not want more information at this point in the text. 11 The purpose expression is therefore stated first to set the appropriate context for interpreting the prescribed sub-action. In example (5), the purpose of ending a call is stated in contrast to placing a call in the previous sentence. It is thus fronted to set the appropriate context for the prescribed action. This fronting of contrastive purposes occurred in our corpus in the context of three oppositional semantic situations: (1) initiating/ending; (2) allowing/preventing; and (3) activating/deactivating.</Paragraph> <Paragraph position="3"> The results of this study predict a number of cases in which purposes should not be fronted, which is in contrast to the general claim made by Dixon (1987). He claimed that purposes should always be fronted because this facilitates the top~lown construction of a procedural plan by readers as they progress through the text. Our results show cases in which this rule is not followed by technical writers, that is, when the purpose is neither global, optional, nor contrastive.</Paragraph> <Paragraph position="4"> determines the grammatical form of purpose expressions. The first element of the form selection sub-network is Conditional-Status, which determines whether the high-level purpose being expressed has special conditions pertaining to it, such as the expressed precondition in example (6a) or other conditions that restrict the applicability of the purpose, as in example (7a) (&quot;to wall unit from which it was taken&quot;). If so, either a by purpose or an adjoined purpose expression is used, depending upon the complexity of the resulting sentence as determined by Sentence-Complexity. The slot of these forms is always initial and is determined here, rather than in the slot selection sub-network just discussed. As was the case in our corpus, IMAGENE expresses purposes that involve five or more propositions using the adjoined form and otherwise with the by form. Consider the following examples of these situations: (6a) When instructed (approx. 10 sec.) remove phone by firmly grasping top of handset and pulling out. (Airfone 1991) (6b) ?? To remove phone when instructed (approx. 10 sec.), firmly grasp top of handset and pull out.</Paragraph> <Paragraph position="5"> (7a) Return handset to wall unit from which it was taken. Insert heel first as shown, then push top in firmly. (Airfone 1991) (7b) ?? Return handset to wall unit from which it was taken by inserting heel first as shown, then pushing top in firmly.</Paragraph> <Paragraph position="6"> In example (6a), there is a precondition on the high-level purpose of removing the phone, a feature that correlates well with the use of the by form. Example (6b) seems to make the incorrect implication that the prescribed actions work only &quot;when instructed.&quot; In the second example, the by form would similarly be prescribed by IMAGENE (because of the condition that the handset be returned &quot;to wall unit from which 11 The distinction between conditions and optional purposes is under the purview of rhetorical status selection and is yet to be addressed.</Paragraph> <Paragraph position="7"> 12 Curly braces indicate that all sub-networks on the right should be entered. Square brackets indicate that all inputs must be true before entering the system on the right. The fact that the Global-Purpose feature is required for entry to the Purpose-TNF system, as well as the input conditions represented normally in the figure, is indicated with an arrow pointing to the additional input conditions. These determinations are made by the Scope system that is not repeated here.</Paragraph> <Paragraph position="8"> High-level systems for the purpose form system network.</Paragraph> <Paragraph position="9"> it was taken&quot;), but the number of propositions in the resulting sentence, (7b), appears to be too great (return, taken, insert, shown, push), forcing the use of the adjoined form, (7a). The adjoined purpose form is an example of a case in which the rhetorical structure of a text need not be explicitly signaled with a lexical or grammatical cue (except textual order), called an &quot;inferred connective&quot; by Crothers (1979). RST allows the representation of this situation because its relations are not defined in terms of lexical and grammatical forms (Mann and Thompson 1987).</Paragraph> <Paragraph position="10"> When a purpose does not have conditions upon it and the scope is global, Purpose-TNF marks the purpose as a to infinitive (TNF). Example (3a) illustrated this. These sorts of context-setting purposes are not demoted to phrase status. This reflects the fact that global purposes are not expressed in phrasal form in our corpus.</Paragraph> <Paragraph position="11"> The Volitionality system determines whether the purpose expresses the desire of the reader to get some inanimate substance to perform in some volitional way. This context usually leads to the use of the so that purpose, as shown in example (3g). Quite often these substances are liquids, but may also include other inanimates. What distinguishes liquids appears to be their ability to drip or drain over a period of time. Consider the following alternate forms for expressing purpose: (8a) Sit the person up leaning slightly forward so that blood and saliva can drain from his mouth. (Rosenberg 1985) (8b) ?? Sit the person up leaning slightly forward in order to allow blood and saliva to drain from his mouth.</Paragraph> <Paragraph position="12"> The form in example (8a) is more commonly used in our corpus in this context. Goal-Status determines whether the use of Goal Metonymy is warranted. The term Goal is used here as a case relation, corresponding to what is also called theme (Allen 1987). This metonymy occurs in purposes in which the direct object (or goal) of the purpose clause is more important than the action, as in (9) For frequently busy numbers, you'll want to use REDIAL \[7\], and the pause will have to be in Redial memory. (Code-a-phone 1989) The corpus study revealed that situations in which the full purpose would be something like &quot;to handle frequently busy numbers&quot; or &quot;for dealing with frequently busy numbers,&quot; tend to be expressed using this sort of ellipsis. The goal of the verb, in this case the busy numbers, metonymically refers to the action as a whole. The remainder of the form selection sub-network, shown in Figure 8, is capable of generating three discrete points along the continuum from fully nominal to fully verbal forms (Quirk et al. 1985), namely the nominalization, the gerund, and to infinitive. These are the forms that were present in our corpus. Nominal-Availability will realize a prepositional phrase with a nominalization as the complement whenever the appropriate nominalization exists, as in example (9a). I3 13 This analysis of nominalizations is an example of the descriptive nature of the current study of instructional text. The descriptive observation has been made that when nominalized forms of a verb exist in the lexicon, they tend to be used. A full explanatory account, in the spirit of current Discourse-Functional studies (e.g., Matthiessen and Thompson 1987; Thompson 1987), would attempt to identify the precise aspect of the action or the context of its expression that would dictate the use of a nominalization, thus resulting in the development of a nominalized form in the English language. Such an account is beyond the scope of the current study.</Paragraph> <Paragraph position="13"> Computational Linguistics Volume 21, Number 1 (9a) Follow the steps in the illustration below, for desk installation.</Paragraph> <Paragraph position="14"> (Code-a-phone 1989) This use of phrases with nominalizations as propositional units is common in instructional text as well as in academic text (Cumming 1991) and formal text in general (Hovy 1987). IMAGENE'S architecture implements a particular interpretation of Cumming's proposal (1991) that nominalizations be dealt with at two levels, one at which the actions are not specified for nominal or clausal expression, and another in which they are. IMAGENE'S Process Structure can be seen as the former level, its Text Structure as the latter.</Paragraph> <Paragraph position="15"> Even if a nominalization exists, however, it still may not be used depending upon the determination of Nominal-Arguments and Nominal-Complexity. These systems, based on the examples in our corpus, restrict nominalizations to single, non-complex arguments. Consider the following examples:</Paragraph> <Paragraph position="17"> Use the VOL LO/HI \[2\] switch to adjust volume to your preferred listening level. (Code-a-phone 1989) ?? Use the VOL LO/HI \[2\] switch for volume adjustment to your preferred listening level.</Paragraph> <Paragraph position="18"> FLASH uses proper timing to avoid an accidental hangup. (Code-a-phone 1989) ?? FLASH uses proper timing for accidental hangup avoidance. In cases (10a) and (11a), taken from our corpus, there were nominalizations available, namely &quot;adjustment&quot; and &quot;avoidance,&quot; but neither was used. The adjustment nominalization in (10b) was apparently not used because it required more than one argument. The avoidance nominalization in (11b) appears to have been rejected because the argument &quot;accidental hangup&quot; was itself a nominalization and thus too complex. In both cases, the to infinitive form was preferred.</Paragraph> <Paragraph position="19"> If no nominalization is available, TNF-Arguments will produce the to infinitive (TNF), unless the infinitive form requires the expression of redundant arguments. Here is an example of this case: (12a) The BATT LOW Light \[9\] comes ON when the battery is weak. The handset must be returned to the base for recharging. (Code-a-phone 1989) (12b) ?? The BATT LOW Light \[9\] comes ON when the battery is weak. The handset must be returned to the base to recharge (the battery?).</Paragraph> <Paragraph position="20"> Examples similar to (12a) were found in the corpus, whereas those similar to the alternative to infinitive expression, (12b), were not.</Paragraph> <Section position="1" start_page="47" end_page="48" type="sub_section"> <SectionTitle> 5.3 The Remove-Phone Example </SectionTitle> <Paragraph position="0"> As an example of the data structures used by IMAGENE, consider the PRL representation of the actions from the Remove-Phone text, depicted graphically in Figure 9. '4 Note 14 Return-Action is a child of Place-Action because we have viewed it as the first of the sub-actions of &quot;placing a call.&quot; In cases such as this one, the procedural distinction between child and sibling actions is a tricky one (see Di Eugenio 1993). We have routinely classified actions expressed with to infinitive constructions as parent nodes rather than as sibling nodes. We leave a more complete treatment of this distinction to future work.</Paragraph> <Paragraph position="1"> The Process Structure for the Remove-Phone text.</Paragraph> <Paragraph position="2"> that this structure is currently built by hand. It is assumed that it could be constructed using artificial intelligence planning methodologies. Note also that the PRL structure, in the various slots for each node, specifies the Penman lexical entries for most of the lexical choice issues, thus allowing IMAGENE to concentrate on expressing procedural relations.</Paragraph> <Paragraph position="3"> Given this structural representation of a sequence of actions, the Content and</Paragraph> </Section> <Section position="2" start_page="48" end_page="48" type="sub_section"> <SectionTitle> Rhetorical Status Selection system sub-network can be viewed as using, the inquiry </SectionTitle> <Paragraph position="0"> responses to produce the TRL structure shown in Figure 10. is Again, this process is not the subject of this paper, but is mentioned to provide a more complete discussion of the data structures involved.</Paragraph> <Paragraph position="1"> The Grammatical Form Selection sub-networks can then be seen as operating on the appropriate relations included in this representation and producing the full TRL structure shown in Figure 11. TRL allows the Text Structure to include a representation of the hierarchical structure of the text in terms of RST, including both nucleus-satellite and multi-nuclear schemata. In addition, TRL specifies the textual order and clause combining using additional New-Sentence and Continue-Sentence links. For example, the Instruct, Grasp, Remove, and Pull nodes are all combined into one sentence in Figure 11. Finally, TRL specifies the grammatical form of each action expression using three features that may be attached to expressible nodes in the structure. The Form feature specifies the general grammatical form. For example, the Instruct is marked as Passive, indicating that the agentless passive should be used. The Linker and Tense markers are also used to mark the appropriate linker and tense of the expression.</Paragraph> <Paragraph position="2"> The Sentence Builder then uses a straightforward recursive descent algorithm to produce an SPL command for each of the sentences in the TRL structure. The generated 15 Because the execution of the Content and Rhetorical Status Selection sub-network is interleaved with the execution of the Grammatical Form Selection sub-networks, this structure alone would never exist at any point in the execution of the network. It is, rather, an illustrative view of what the Content and</Paragraph> </Section> <Section position="3" start_page="48" end_page="50" type="sub_section"> <SectionTitle> Rhetorical Status Selection sub-network would realize if it were executed in isolation. </SectionTitle> <Paragraph position="0"> The final Text Structure for the Remove-Phone text.</Paragraph> <Paragraph position="1"> text for this example is shown here: (13) When you are instructed, remove the phone by grasping the top of the handset and pulling it. Return to a seat to place a call.</Paragraph> <Paragraph position="2"> This text is identical to the original text with respect to the four lexical and grammatical issues addressed here. There are, however, a number of other lexical and phrasal differences, including the lexical items chosen for the object references and the use of determiners. These differences arise from the fact that the current study has not specifically addressed the issue of referring expressions. Currently, IMAGENE uses simple algorithms for pronominalization and determiners, which are not based on a detailed corpus study of the forms and functions of the object reference domain. A Keith Vander Linden and James H. Martin Expressing Rhetorical Relations study of referring expressions, similar to our work on expressing rhetorical relations, would allow the development of a more principled solution to this problem.</Paragraph> </Section> <Section position="4" start_page="50" end_page="51" type="sub_section"> <SectionTitle> 5.4 More Examples of IMAGENE's Output </SectionTitle> <Paragraph position="0"> This section includes examples of IMAGENE'S output for the fundamental relations dealt with in the current study, that is, Purpose, Precondition, Result, and action Sequence.</Paragraph> <Paragraph position="1"> It is intended to demonstrate IMAGENE'S breadth of coverage and will not discuss the details of how the forms are motivated.</Paragraph> <Paragraph position="2"> Given the choice to express an action, rhetorically, as a purpose, IMAGENE is capable of producing seven grammatical forms for its expression, most of which can be either fronted or not fronted. Here are the various forms, as generated by IMAGENE according to the distinctions discussed in the previous section: To end a call, hold down the FLASH button for two seconds, then release it. Follow steps in the illustration for desk installation.</Paragraph> <Paragraph position="3"> Use the OFF position for charging the batteries.</Paragraph> <Paragraph position="4"> Use the REDIAL for frequently busy numbers.</Paragraph> <Paragraph position="5"> When you are instructed, remove the phone by grasping the top of the handset and pulling it.</Paragraph> <Paragraph position="6"> (14f) Remove the phone. Grasp the top of the handset, and pull it. (14g) Tilt the pan so that the fluid drains out.</Paragraph> <Paragraph position="7"> Given the choice to express an action, rhetorically, as a precondition, IMAGENE is capable of producing four grammatical forms for its expression, all of which can be either fronted or not fronted and also linked with various lexical items. Here are some representative forms, as generated by IMAGENE: (15a) If light flashes, insert credit card.</Paragraph> <Paragraph position="8"> (15b) The BATTERY LOW INDICATOR will light when the battery is low. (15c) When the phone is installed, and the battery is~harged, move the OFF/STBY/TALK switch to the STBY position.</Paragraph> <Paragraph position="9"> (15d) Return the OFF/STBY/TALK switch to the STBY position after your call. There are two types of results that IMAGENE supports. The first type is non-reader actions that are not the result of an explicit command to monitor a particular device state. IMAGENE expresses this type of result as a future tense clause, as seen in example (16a). The second type is not based on an action in the Process Structure at all, but rather, is a span added by the system networks to signal a state resulting from an expressed action. IMAGENE expresses these as present tense relational expressions, as seen in example (16b). Here are examples of these forms:</Paragraph> <Paragraph position="11"> The BATTERY LOW INDICATOR will light when the battery is low.</Paragraph> <Paragraph position="12"> When the phone is installed, and the battery is charged, move the OFF/STBY/TALK switch to the STBY position. The phone is now ready to use.</Paragraph> <Paragraph position="13"> Computational Linguistics Volume 21, Number 1 Simple sequential actions do not fit into the categories discussed above and are marked as imperative commands. These commands are combined into clauses by the sentence tools system network using and when the concurrency that could be implied is impossible or inconsequential, as in example (140, or then when there is possible unwanted concurrency, as in example (14a).</Paragraph> </Section> </Section> <Section position="9" start_page="51" end_page="53" type="metho"> <SectionTitle> 6. Verifying IMAGENE's Prescriptions </SectionTitle> <Paragraph position="0"> Finally, we compare the output of the text generator with the text in the corpus. For this purpose, IMAGENE'S system network was re-run for all of the approximately 600 action expressions, both those from the training set and those from the testing set. Statistics were kept on how well its realizations matched the expressions in the corpus. 16 These tests were performed without the Penman realization component engaged, comparing the TRL output of the system network with the corpus text. This way, the extensive lexicon that would have been necessary for the surface realization was not required.</Paragraph> <Paragraph position="1"> IMAGENE currently includes a domain model and lexical entries for cordless telephones and a few other specific examples.</Paragraph> <Paragraph position="2"> The match was judged on four separate lexical and grammatical issues: linker, form, slot, and clause combining. The resulting TRL structure had to specify the identical linker (either preposition or conjunction), form (tense, aspect, mood, and voice, or non-finite verb or nominalization), slot (textual order), and combining (if the expression was combined with the following one). An example of this verification process can be found in Section 5.3, in which the IMAGENE-produced Remove-Phone text is shown to match the original text on all four of these issues.</Paragraph> <Paragraph position="3"> Note that the match must be exact. For example, if IMAGENE specifies the conjunction and for a sequence expression when then occurs in the text, the choice of linker would be counted as incorrect, in spite of the fact that the resulting text might be quite understandable. Note also that IMAGENE'S realizations may even be better in some cases than the text in the corpus. Although the general philosophy of the approach taken in the current study is to assume that the choices made by the writers of the corpus are correct, there are isolated cases ~in which the forms in the corpus are probably inappropriate. IMAGENE embodies choices that are consistently made over a range of instructions and thus does not reflect isolated examples.</Paragraph> <Paragraph position="4"> The analysis conducted in step 2 has been based primarily on a small subset of the full corpus, namely on the instructions for a set of three cordless telephone manuals.</Paragraph> <Paragraph position="5"> This training set constitutes approximately 35% of our corpus. The results of this analysis were then implemented in IMAGENE and applied to the full corpus, providing a detailed characterization of the instructions found in the original telephone manuals and a quantitative analysis of how well this characterization applies to the other forms of instructions. IMAGENE's realizations correctly match all four lexical and grammatical issues in 71% of the expressions in the training set and 52% in the testing set. The specific levels of match for the four most common rhetorical relations are detailed in Figures 12 and 13. There is one table for each of the major rhetorical relations, Purpose, Precondition, Result, and Sequence. 17 These tables show the percentage of 16 The training corpus included some non-procedural text that was included for a pilot study done before the focus on procedural text had been determined. It is not handled particularly well by IMAGENE, and the results given in this section will not include it. The testing set is exclusively procedural and is included in full.</Paragraph> <Paragraph position="6"> 17 Because there are relatively few concurrent expressions in our corpus, only 33, those results are not included in this section.</Paragraph> <Paragraph position="7"> The accuracy of IMAGENE's realizations for Purpose and Precondition expressions.</Paragraph> <Paragraph position="8"> IMAGENE'S realizations for linker, form, slot, and clause combining that matched those in the corpus, differentiating between the training set and the testing set. As can be seen in all of the charts, the level of match is better for the training set, but still good for the testing set.</Paragraph> <Paragraph position="9"> For purpose expressions, IMAGENE makes use of four different linkers (by, for, so that, and no linker) and six different forms (to infinitive, imperative, nominalization, gerund, goal metonymy, and simple present tense action) and produces a match on all four lexical and grammatical issues for 81% of the purpose expressions in the training set and 59% in the testing set. Figure 12 gives a breakdown of IMAGENE's accuracy for the four lexical and grammatical issues. To judge these results more fully, consider an alternative system that always generates the single most common purpose form.</Paragraph> <Paragraph position="10"> In our corpus, this is the fronted to infinitive, which occurred in 34% of the purpose expressions. TM Such a system would score 34% under the verification criteria used here. For precondition expressions, the most common form in our corpus is the fronted if present tense clause, which occurred in 19% of the 98 precondition expressions in the corpus. IMAGENE, which produces five linkers and nine forms, produces a match for 67% of the precondition expressions in the training set and 35% in the testing set.</Paragraph> <Paragraph position="11"> As can be seen in the precondition chart in Figure 12, IMAGENE's accuracy is lower for preconditions than for purposes, particularly in the testing set. This reflects the greater diversity of procedural contexts in which preconditions arise and the corresponding diversity of the forms used to express them (see Vander Linden 1994). Certainly, a larger training set is required here, but it is not clear at this point how much larger it 18 Table 1 indicates 32%, but that would be for our corpus with the non-procedural portions of text included. They have been removed here to remain consistent with the statistics shown in this section. The accuracy of IMAGENE's realizations for Result and Sequence expressions.</Paragraph> <Paragraph position="12"> should be. IMAGENE'S accuracy for results and sequence expressions is similar to that presented for purposes and preconditions. It is detailed in Figure 13.</Paragraph> </Section> class="xml-element"></Paper>