<?xml version="1.0" standalone="yes"?>
<Paper uid="J93-3002">
  <Title>A Computational Theory of Goal-Directed Style in Syntax</Title>
  <Section position="2" start_page="0" end_page="454" type="abstr">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"/>
    <Section position="1" start_page="0" end_page="453" type="sub_section">
      <SectionTitle>
1.1 An Advocacy of Style
</SectionTitle>
      <Paragraph position="0"> Understanding a text requires more than just understanding its propositional content.</Paragraph>
      <Paragraph position="1"> It requires a sensitivity to the interaction of semantic content, emotional expression, and interpersonal and situational attitudes. This interaction is reflected in the style of the text. Style in language is not just surface appearance, a decorative veneer. Rather, it is an essential part of meaning, part of the author's communication to the reader. So to fully understand the nuances of a text, one must determine not only the propositional content, but also how its communicative effect is colored by the form, which reflects affective content. While propositional content provides the basic tone, the expressive  * Department of Computer Science, University of Waterloo, Waterloo, Ontario, Canada N2L 3G1. t Department of Computer Science, University of Toronto, Toronto, Ontario, Canada M5S 1A4. (c) 1993 Association for Computational Linguistics Computational Linguistics Volume 19, Number 3 form provides the tonal quality. Together, form and content create style, that which distinguishes both an individual text and a collective body of writing.</Paragraph>
      <Paragraph position="2"> Style is created through subtle variation, seemingly minor modulations of exactly what is said, the words used to say it, and the syntactic constructions employed, but the resulting effect on communication can be striking. Consider the following versions of the same text, Matthew 7:27: 1. And descended the storm and came the floods and blew the winds and beat against that house and it fell and the fall of it was great. (Literal translation of the Hellenistic Greek.) 2. And the rain descended, and the floods came, and the winds blew, and beat upon that house; and it fell: and great was the fall of it. (Revised Standard Version, 1952.) 3. The rain came down, the floods rose, the wind blew, and beat upon that house; down it fell with a great crash. (The New English Bible, 1970.) 4. Rain came down, floods rose, gales blew and struck that house, and it fell; and what a fall it had! (The New Jerusalem Bible, 1985.) 5. The rains fell, the torrents came, the winds blew and lashed against his  house. It collapsed under all this and was completely ruined. (The New American Bible, 1973.) The first variation, a word-by-word translation from the Hellenistic Greek, is actually a good deal less striking than subsequent, widely accepted versions. The placement of the verbs before their subjects is quite normal and the closing is also quite usual. The sense of discord and resulting poetic effect is not as evident as it is in the second variation, from the Revised Standard Version, which is resoundingly poetic in the imitative form of the first five clauses, followed by the inverted form of the final clause. The text begins in strong concord and dissolves into discord, but not unpleasingly so. Discord, as we use the term in this paper, refers to a deviation from the norm, but such deviations can be used to good effect. As we will see, in language as in music, it is often through the construction of patterns of concord and discord, particular combinations of order and disorder that create an overall harmonious arrangement, that certain stylistic effects are achieved.</Paragraph>
      <Paragraph position="3"> In versions 3 and 4, the dramatic effect of the closing has been retained from the RSV translation but through different choices of words and structure. In 3, the last clause is inverted, but the translator has chosen to place more emphasis upon the fall itself (down it fell) than its magnitude (great). Example 4 ends on an intense note, achieved through the use of an exclamation (what a fall it had!), rather than an inversion of syntactic structure. And no single word expresses the degree of magnitude of the fall. In the final example, 5, the drama and intensity of the RSV has been lost with the removal of the dissolution from initial concord to final discord. In this case, the translator has opted for plainness and clarity, even at the expense of beauty.</Paragraph>
      <Paragraph position="4"> In English, example 1 sounds odd and disjointed; and examples 2, 3, and 4 are dramatic in a way that example 5 is not. But all have the same essential content. What then causes the differences in effect? What is being varied? There are at least four parameters that play a role in these stylistic variations: lexical, syntactic, thematic, and semantic aspects.</Paragraph>
      <Paragraph position="5">  Chrysanne DiMarco and Graeme Hirst Goal-Directed Style in Syntax Lexical aspects: Compare example 4 to the following constructed versions: 1 6. Rain descended, floods rose, gales raged and beat upon that dwelling, and it collapsed; and what a fall it had! 7. Rain fell, the water level rose, winds blew and hit that house, and it fell; and what a fall it had! The differences between 4, 6, and 7 are primarily lexical. Example 6 uses rather elegant words, while 7 opts for a more commonplace vocabulary, and 4 lies somewhere between the two in lexical formality.</Paragraph>
      <Paragraph position="6"> Syntax: Comparing example 2 to example 5, we see that the former uses syntactic structures that create more dramatic effects: a short climactic sentence, and it fell, and a striking inversion, great was the fall of it, close the text on a powerful note. In contrast, 5 ends with a very ordinary, unremarkable, construction.</Paragraph>
      <Paragraph position="7"> Thematic aspects: Comparing example 2 to example 3, we observe that the two sentence structures bring different elements into focus: great was the fall of it, in contrast to down it fell with a great crash. In general, variations in thematic structure can create different stylistic effects.</Paragraph>
      <Paragraph position="8"> Semantics: The exact choice of what is said, or not said, also has stylistic consequences. Compare example 4 to the following two constructed versions: .</Paragraph>
      <Paragraph position="9"> .</Paragraph>
      <Paragraph position="10"> A serious storm, with rain and gales and floods, struck that house, which collapsed.</Paragraph>
      <Paragraph position="11"> Rain fell and fell and eventually caused a flood, which rose up to that house; also the winds kept blowing until eventually the combined forces of rain, flood, and wind were too great and caused so much structural damage that the house collapsed.</Paragraph>
      <Paragraph position="12"> These variations differ in their semantic content: example 4 merely reports the events, but 8 evaluates, a serious storm, while 9 emphasizes technical details, structural damage. These semantic differences are reflected in the texts; and stylistic differences result. Given that these four parameters--lexical choice, syntax, theme, and semantics-control stylistic variations, two questions arise: * How do we characterize each type of variation? * How does each variation contribute to an overall stylistic effect? In considering these questions in this paper, we will concentrate on variations of the syntactic parameter.</Paragraph>
      <Paragraph position="13"> We emphasize that our intent here is not any form of literary analysis or literary theory. Our main concern will be ordinary, everyday text. It is true that, in the Biblical texts above, we saw how different syntactic forms carry different stylistic import to the point that one form may be poetry while another is just dull, plodding prose. But style  Computational Linguistics Volume 19, Number 3 isn't just a matter of achieving poetry or not. Every text, large or small, interesting or dull, effective or not, has its own style. Nor are we talking about style in any of its normative senses: the common tenets dictating standard forms (e.g., Chicago 1982), or the textbook prescriptions for 'good' style: be clear, be simple, be precise (e.g., Strunk and White 1979). For people aim to convince, to persuade, to impress, and, even, sometimes to obscure, and standard textbooks tell us very little about such varied and subtle stylistic and pragmatic goals.</Paragraph>
      <Paragraph position="14"> Rather than a study of literary or normative style, our intent is to determine what gives any ordinary piece of text its stylistic 'feel.' The following examples will demonstrate what we have in mind. These texts are all from newspaper feature articles: 10. Silvia, a commanding woman in her 50% a shrew falsely mellowed by religion, promptly organized prayer sessions on the lines of Tupperware meetings. 2 11. The artist provides a dreamy background done in yellow and bistre brushstrokes to a blue gown with woodenly rigid folds or the profile of a brown angel painted so mineral hard and modeled so carefully that the incoherence of virtue does it injury. 3 12. Crazed with fear, he tried to purify her by dunking her in the ocean and holding her under the water; then in desperation he threw her on the still-smoking pyreo 4 In a newspaper, we might have expected the writer and translator to have simply aimed for clarity. In fact, we find a variety of effects. The first text, 10, emphasizes a sense of harmony by repeating the same kind of structure, a nominal group, in the postmodification of Silvia. The second, 11, is more complex, and achieves a certain balance in the judicious use of conjunctions; but the result is so difficult to understand that it doesn't really make sense. Text 12 has a stark initial participle clause, crazed with fear, that emphasizes the intensity of the subject's emotional state.</Paragraph>
      <Paragraph position="15"> To account for the kinds of complex stylistic effects that occur even in everyday writing, we propose a goal-directed understanding of style. That is, an author's intent can vary with respect to a number of stylistic goals, such as clarity or obscurity, abstraction or concreteness, staticness or dynamism. Particular choices of words, syntactic structure, and semantic structure make a text more--or less--stylistically varied and effective. We believe that these choices, goals such as abstraction or concreteness, and the stylistic elements that are used to realize them can be recognized and represented in a formal notation. They are, in a word, codifiable. It is this codification that is at the heart of computational stylistics. S</Paragraph>
    </Section>
    <Section position="2" start_page="453" end_page="454" type="sub_section">
      <SectionTitle>
1.2 The Function of Style
</SectionTitle>
      <Paragraph position="0"> Propositional content alone is insufficient to determine the nature and form of a sentence (Halliday 1985; McDonald and Pustejovsky 1985; Jameson 1987; Hovy 1988; Scott  computer-aided stylistics, in which computer-generated data and pattern-matching aid human analysis and judgment of style in literary studies (see Section 2.2). In contrast, our use of the term entails fully automatic computer analysis of the style of any kind of text.</Paragraph>
      <Paragraph position="1">  Chrysanne DiMarco and Graeme Hirst Goal-Directed Style in Syntax and de Souza 1990). Even after the propositional content has been decided upon, there are still many linguistic, syntactic, and even semantic decisions that the language producer must make before a sentence can be formed. These decisions are assumed by the audience not to have been made randomly, but rather in specific, deliberate ways that encode additional information, such as opinion, emotional affect, and interpersonal relationships. 6 To the extent that a piece of text exhibits a particular, recognizable style, it also reflects the author's presumed intent to convey the effect associated with that style. Therefore, full understanding of a text must represent not only propositional content, but also stylistic effects.</Paragraph>
      <Paragraph position="2"> This becomes most apparent in the case of machine translation. If a translation is to be faithful, the stylistic effects of the source language text must be transferred to the target language text, making appropriate use of the stylistic conventions of the target language. But a dilemma arises: * One wants to preserve the original author's stylistic intent, the information being conveyed through the manner of presentation.</Paragraph>
      <Paragraph position="3"> However, different languages might realize this effect in different ways.</Paragraph>
      <Paragraph position="4"> So the source and target texts should both aim for the author's stylistic goal, but might have to achieve it through different linguistic means. 7 * Yet one wants to produce a text whose style is appropriate and natural to the particular target language. Languages differ as to the most 'natural' way to express an idea. For example, French tends to prefer more abstraction, English more concreteness (Vinay and Darbelnet 1958).</Paragraph>
      <Paragraph position="5"> The best translation, therefore, might modify the original author's stylistic intent and express a different effect.</Paragraph>
      <Paragraph position="6"> Sometimes, there is no way to resolve this dilemma, and one is left with an unsatisfactory translation. But, with a knowledge of the comparative stylistics of a language pair, and of the stylistic resources of each language and the possible range of effects they can create, one can substantially improve the quality of a translation.</Paragraph>
    </Section>
    <Section position="3" start_page="454" end_page="454" type="sub_section">
      <SectionTitle>
1.3 The Structure of the Paper
</SectionTitle>
      <Paragraph position="0"> Our goal is to create a formal representation of stylistics for use in natural language systems, and, moreover, to do so in a manner applicable to different languages. The solution we will propose is a codification of syntactic stylistic knowledge in the form of a stylistic grammar.</Paragraph>
      <Paragraph position="1"> In the next section, we will review the current status of the codification of style. In Section 3, we will construct a vocabulary for stylistics. The definition of concepts and the attempt to organize them into a recognizable structure is a necessary first step toward understanding the problem. Then we will develop in Section 4 a methodology for converting stylistic knowledge into a formal representation. The methodology will</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>