On the Dimensions of Discourse Salience

This paper describes results of two corpus studies of information packaging of discourse referents in German dedicated to the following questions: • Do sentence-initial position, pronominalization and subject role assignment reflect a single underlying dimension of discourse salience or multiple dimensions ? • If there are multiple dimensions of salience, is it possible to associate them with a forwardlooking and a backward-looking perspective on discourse, as proposed, e.g., in the context of Centering (Grosz et al., 1995) ? This paper presents empirical findings from TüBa-D/Z, a corpus of German newspaper articles, that provide evidence against a unidimensional model of discourse salience, and support the claim that (at least) two dimensions of salience are to be distinguished and that these dimensions are associated with different temporal orientations on discourse. 1 Salience and Information Packaging In the last 30 years, the notion of “salience” has been employed in many accounts for the information packaging of discourse referents, especially with respect to the choice of referring expressions (e.g., realization as definite NP or as a pronoun), and with respect to the pragmatic function of grammatical roles and word order preferences: Personal pronouns are assumed to represent more salient referents than nominals (Sgall et al., 1986; Ariel, 1990; Grosz et al., 1995), the left periphery of sentences (and in particular, the sentence-initial position) are associated with a high degree of salience (Sgall et al., 1986; Sridhar, 1988; Rambow, 1993), and the grammatical subject is assumed to serve a similar function (Fillmore, 1977; Tomlin, 1995; Grosz et al., 1995). Consider the German example sentence in (1). The expression sie ‘they’ is a subject pronoun in preverbal (vorfeld ) position; according to the aforementioned theories, its referent is to be regarded as highly salient, whereas the nominals auf einem Tandem ‘on a tandem’ and ins Stadion ‘to the stadion’ are postverbal non-subjects and thus non-salient. (1) Sie they wollen want.to auf on einem a Tandem tandem ins into.the Stadion stadion radeln go.by.bike ‘They want to go to the stadion by tandem.’ (TüBa-D/Z, sentence 113) Despite the apparent agreement on the relevance of salience to different information packaging phenomena, researchers disagree on determinants and the actual nature of salience. Already Sridhar (1988, p.38) noted that ‘a number of different factors have been claimed to contribute to salience’, that ‘[r]esearchers are (...) divided on the effects of salience to sentences’, and further that ‘salience is obviously (...) characterized by a number of superficially dissimilar properties’. Since then, three major views on the nature of salience have been established: unidimensional: In traditional unidimensional models as advocated by Sgall et al. (1986, word order and referring expressions), Gundel et al. (1993) and Ariel (1990, referring expressions), and Tomlin (1995, word order and grammatical roles), salience is seen as a single dimension of cognitive states: Every referent is assigned a particular degree of salience and this degree of salience determines the packaging preferences for this referent. multifactorial: The radical antithesis to the unidimensional view is to abandon the idea of a generalized notion of salience, and to focus on the study of individual factors. This has been the premise of the psycholinguistic research of Osgood and Bock (1977) and Sridhar (1988), and has been recently revived by Kaiser and Trueswell (2004, to appear 2011) and Brown-Schmidt et al. (2005). multidimensional: Multidimensional models of salience postulate the existence of multiple dimensions of salience as independent generalizations over a certain range of different factors. Typically, two dimensions are distinguished (Givón, 1983; Pattabhiraman and Cercone, 1990; Clamons et al., 1993; Mulkern, 2007): One dimension that is primarily defined with respect to the preceding discourse and/or the common ground, and that is thus primarily backward-looking. The other dimension is more concerned with the intentions and goals of the speaker and takes into consideration how these are manifested in subsequent discourse, and that is thus (at least partially) forward-looking.1 Psycholinguistic experiments and corpus studies on personal pronouns and demonstrative pronouns conducted by Kaiser and Trueswell (2004, to appear 2011), BrownSchmidt et al. (2005) and Ellert and Hopp (2010) indicate that personal pronouns and demonstrative pronouns deviate in their antecedent selection preferences (in Finnish, Estonian, Dutch, English and German). In other words, certain salience factors contribute independently to the choice and interpretation of referring expressions in these languages. This observation can be seen as direct counterevidence for a unidimensional model that would postulate that demonstrative pronouns and personal pronouns reflect cognitive states organized in one uniform dimension of salience. However, it is not necessary to conclude that multiple cognitive dimensions are involved: A functionalist with a unidimensional salience model might argue that the specific preference of personal pronouns to take subject antecedents can be attributed 1A representative multidimensional model of salience is the proposal of Clamons et al. (1993) and Mulkern (2007). They postulate that every referent is characterized by the degree of salience arising from the preceding context (‘givenness’, ‘inherent salience’) on the one hand, and on the other hand by the degree of salience imposed on this referent by the speaker (‘importance’, ‘emphasis’, ‘imposed salience’) in order to increase its accessibility in subsequent discourse. Similar approaches (with different terminologies) have been described by Givón (1983, 2001); Pattabhiraman and Cercone (1990), and are also suggested by Levelt (1989) and Chafe (1994).