<?xml version="1.0" standalone="yes"?>
<Paper uid="W94-0313">
  <Title>Towards an Account of Causation in a Multilingual Text Generation System</Title>
  <Section position="3" start_page="0" end_page="109" type="intro">
    <SectionTitle>
1. Introduction
</SectionTitle>
    <Paragraph position="0"> Causation 1 has been a topic of linguistic investigation for many years. I believe two main reasons underly this interest of linguists in the phenomenon: (i) causative constructions build a good proving ground for many questions of theoretical interest (Herschensohn 1981), and (ii) causation can be expressed in natural language by numerous linguistic alternatives. Altenberg (1984) has in fact identified nearly one hundred possible explicit links for encoding a causal relation between two propositions. Any language user is thus confronted with the problem of choosing one alternative over another. But, how is this choice constrained? Why does a language user choose one link over another? This same question arises in natural language  linguistic term &amp;quot;causation&amp;quot; from the scientific notion of &amp;quot;causality&amp;quot; in the physical world. &amp;quot;For the latter, the totality of phenomena constitutes a causal continuum of which any conceptually delimited portion, an &amp;quot;event&amp;quot;, is understood as relating causally outside itself and containing causal relations within&amp;quot; (Talmy 1976:47). In contrast, causation designates a relation that is interpreted by the speaker as being a relation between a cause and an effect, cf. definition below.</Paragraph>
    <Paragraph position="1">  generation, where it is crucial that the rules to select a specific linguistic structure be made explicit. Text generation is a process in which meaning represented as non-linguistic knowledge at a higher level of abstraction than wordings - is organised and re-expressed over a number of steps so that it can be presented as worded units. So, in 9rder to use and generate the different possible causative constructions properly, we have to know what their underlying meaning is.</Paragraph>
    <Paragraph position="2"> In accordance with the 'one meaning, one form' principle advocated by linguists working in a semiotic approach (e.g. Bolinger 1968; Wierzbicka 1988; Halliday 1978), I assume that all these different causative constructions have their specific meanings and functions - different semantics being encoded in different lexicogrammatical structures. This view on language pervades in the Systemic Functional approach where it is posited that the relation between semantics and lexicogranunar is a 'natural', non-arbitrary one (cf. Halliday 1985:xiii-xxxv), in other words, semantics and lexicogrammar are dependent on one another. Language is interpreted as a resource for making meanings, which is organised functionally, textually, and communicatively. This makes it also suitable as a theoretical foundation for (monolingual and multilingual) text generation (see also, Matthiessen and Bateman 1991; Batemam Matthiessen and Zeng in prep.).</Paragraph>
    <Paragraph position="3"> In this paper, I will discuss some aspects of the generation of Dutch analytic causative constructions in a systemic multilingual grammar for generation (Bateman, Matthiessen, Zeng in preperation), currently under development at Sydney University and in the project KOMET at GMD/IPSI, Darmstadt. It is based on Nigel (Mann 1985), a systemic computational grammar of English for generation developed in the Penman Project at ISI/USC. Languages worked on currently include English (Matthiessen 1990), German (Teich 1992), Dutch (Degand 1993a), Chinese, Japanese, and French. The multilingual grammar follows the assumption of Systemic Functional Linguistics (SFL) that commonality across languages is functional in the first instance, not structural or realizational: functionality has to be preserved across languages, but structural realizations may very well differ (Bateman, Matthiessen, Nanri, Zeng 1991). This means that within a functional framework different languages can be treated in a similar way. The common functionality is realized stratum by stratum in the linguistic system (composed of the three strata: semantics, lexicogrammar, and phonology). Within a particular stratum, it is expressed by the paradigmatic, systemic organization of that strattun in the first instance, and only secondarily by the syntagmatic organization (structural realization). With this kind of organization of (multilingual) linguistic resources, it is ensured that the multilingual system is more than a loosely coordinated set of separate generation</Paragraph>
    <Section position="1" start_page="108" end_page="109" type="sub_section">
      <SectionTitle>
7th Intemationai Generation Workshop
</SectionTitle>
      <Paragraph position="0"> systems and that new language components can be added to the system without revising the overall organization (see also, Degand 1993b).</Paragraph>
      <Paragraph position="1"> After a brief sketch of the lexicogrammafical potential for expressing causation in Dutch, I will concentrate on the so-called analytic (or periphrastic) causative construction and show how this particular type of causative construction can be realized in a systemic mulfilingual grammar for generation, taking into account the theoretical elaborations developed in Degand (to appear). The importance of this theoretical basis for the computational account will be amply illustrated.</Paragraph>
      <Paragraph position="2"> 2. The lexiC/ogrammatical potential for expressing causation The expression of causation is ubiquitous in all forms of communication and in natural language in particular. The examples given below illustrate this multitude of different types of causal expressions in Dutch. Note that these examples have been grouped according to their structural realization, since it was my aim to illustrate the wide lexicogrammafical variation in this area. This does not mean that the subexamples can .always be used as alternatives for each other, neither does it mean that the English translations could be used in a same context (see Section 3. I.).</Paragraph>
      <Paragraph position="3"> . i. Hij blijft thuis, want hij is ziek.</Paragraph>
      <Paragraph position="4"> He stays at home, for he is ill.</Paragraph>
      <Paragraph position="5"> ii. Ik ben te laat, omdat ik mijn trein gemist heb.</Paragraph>
      <Paragraph position="6"> I am too late, because I missed my train.</Paragraph>
      <Paragraph position="7"> . i. Ze heefl een ongeluk veroorzaakt.</Paragraph>
      <Paragraph position="8"> She has caused an accident.</Paragraph>
      <Paragraph position="9"> ii. Hij bewoog zijn hand open neer.</Paragraph>
      <Paragraph position="10"> He moved his hand up and down.</Paragraph>
      <Paragraph position="11"> iii.Piet liet Marie een auto kopen.</Paragraph>
      <Paragraph position="12"> Piet had Marie buy a car.</Paragraph>
      <Paragraph position="13"> . Jij bent de oorzaak van al onze problemen.</Paragraph>
      <Paragraph position="14"> You are the cause of all our problems.</Paragraph>
      <Paragraph position="15"> 4. i. Doorde gladde wegen is bijna niemand op tijd kunnen zijn.</Paragraph>
      <Paragraph position="16"> Because of the slippery roads hardly anybody could make it on time.</Paragraph>
      <Paragraph position="17"> ii. Ten gevolge van bet ongeluk ontstond een lange file.</Paragraph>
      <Paragraph position="18"> As a result of the accident there was a traffic j am.</Paragraph>
      <Paragraph position="19"> In traditional descriptive grammars, these different constructions are classified according to the grammatical category of the causal element, i.e. the linguistic element which endows the clause with a</Paragraph>
      <Paragraph position="21"> causal meaning. Thus, in example (1) the causal element is realized by a coordinating or subordinating conjunction; example (2) displays different types of verbs with a causal meaning: causative verbs, intransitive verbs that are used transitively, causal auxiliaries. In example (3), the causal element is expressed in a nominal phrase and in example (4) in a prepositional phrase. All of these structures have a common underlying meaning (at least partially) which needs to be captured, namely causation, but which is not accounted for in these types of grammars, since semantics is described independently of the grammar.</Paragraph>
    </Section>
  </Section>
class="xml-element"></Paper>