Discourse Coherence, Information Structure and Additive Relations

Discourse coherence concerns the level at which the speaker, putting together her discourse, needs to enable the hearer to build an ongoing representation where each upcoming ‘idea’ - theme or proposition - finds its place. Information structure refers here to thematic progression, in the sense of structuring given and new information, as well as informational salience: means used by the speaker to foreground or background ideas, creating an information contour for the discourse.

Both coherence relations and information structure may be encoded in some linguistic device (such as prosodic pattern, lexical expression/construction or syntactic structure /construction), or may be left implicit for the hearer/reader to pragmatically infer. Some particular linguistic device may mark simultaneously a coherence relation and an information structural relation. In fact, some approaches to discourse tie the two together so that each coherence relation has an inherent information contour or grounding relation. This is the case, for instance, of Rhetorical Structure Theory (RST) (Mann and Thompson 1986). Others, such as Relational Discourse Analysis (RDA) (Oberlander and Moore 2001), distinguish ‘semantic’ coherence relations from ‘functional’ information structure.

Coherence relations (also known as discourse relations or rhetorical relations) include such notions as ‘contrast’, ‘concession’, ‘result’, ‘elaboration’, ‘exemplification’, ‘addition’, ‘justification’ and so on. They refer to the various ways in which the segments (or groups of segments) of a text or discourse fit into the rest of the text or discourse; that is, how each part relates to the parts that precede and follow it, and thus contributes to the overall meaning of the text.

These types of meaning can themselves be thought of as propositional. (In fact, they are referred to by Mann and Thompson (1986) as ‘relational propositions’, an area of meaning that is relatively grammaticalized into particles and adverbs, but

Table 1 A partial simple model of discourse coherence relations

Consonant relations

Dissonant relations














on the contrary, ..



so that..

Even then..

which can also be ‘propositionalized’.) Attempts to draw up empirically satisfactory taxonomies of coherence relations, using labels such as the ones above (contrast, concession, etc.), have foundered on three main difficulties: the issue of constraining the number of relations, the degree to which the taxonomy is hierarchical and the relationship between coherence and information structure. Moreover, each language will have its own network of relations depending on the way relations are typically drawn in the language in question. We do not adopt a taxonomic approach here; descriptions of relations in Sects. 4 and 5 are not to be interpreted as labels belonging to a particular taxonomy of predefined coherence relations, but simply as indications of the types of meaning expressed in the corpus data.

For practical purposes, nevertheless, a working model is needed to delimit an area for investigation. The approach adopted here is to view relations as a consonant- dissonant cline from total or high consonance to low or zero consonance. High consonance occurs where the ideas or sets of ideas expressed in consecutive discourse segments co-exist happily, being wholly compatible with one another (e.g., reformulation, exemplification). High dissonance occurs where adjacent discourse segments express ideas that are wholly incompatible (e.g., polar opposites). (This model is comparable to Murray’s (1997) model of continuous vs discontinuous relations; we prefer different terms to avoid confusion with Continuative relations, which Murry subsumes along with causal relations under ‘continuous’). Table 1 illustrates such a simple partial model.

Relations may be explicitly marked or left implicit (v. Taboada 2009). Marking takes many forms, more or less grammaticalized: syntactic pattern, subordinating conjunction, non-subordinating conjunction, adverb, adverbial phrase, clause, modal particle, and so on. Dedicated discourse markers are adverbial lexemes and phrases such as however, even so, besides, for instance, moreover, and similar expressions in other languages. A further function of many, if not all, discourse markers is to signal the relative informational salience of the discourse segment they attach to. They thereby help the hearer to appreciate the speaker’s evaluation of the relative importance of the states of affairs related in the discourse. The expression of discourse coherence is thus both subjective, indicating the speaker’s vision of how the ideas expressed inter-relate, and intersubjective insofar as the speaker must anticipate the expectations of the hearer.

The focus here is on the discourse marking of additive relations. An additive relation will be said to exist where a new idea in the upcoming discourse develops the topic of the preceding discourse and is compatible with the preceding idea(s); sim?ply put, it is ‘more in the same vein’. (This use of ‘additive’ differs from that of other authors such as Halliday (1994), for example.) The relation may be between two states of affairs (‘content’ use) or between two speaker arguments (‘presentational’ use); often both types of relation obtain between two ideas (cf Hasselgard 2014: 72). A single occurrence of a discourse marker might therefore be interpreted as encoding a state-of-affairs relation, an argumentational relation and an information structural relation. In (1), for instance, What’s more can be interpreted as introducing an additional event and an additional speaker argument, as well as signalling that the upcoming event/argument is more salient (rhetorically stronger for the speaker) than the previous idea that it links to.

(1) if they had been cheating I would have known. What’s more, I would have been the first to complain. [BNC CH7, newspaper]

The aim of the study is to compare the usages of additive coherence relation markers by speakers of the political speech genre in the two languages and to identify potential discourse constructions built around an additive coherence relation.

Consonant relations in general are expected to be less marked (for example, by a discourse marker) than dissonant relations. This is because ‘coherence’ in the lay sense excludes incompatibility: the bare assertion of two apparently incompatible ideas results in incoherence. Where a proposition may appear to the hearer to be either at odds with what went before or irrelevant to it, some marker is called for to at least acknowledge the counterexpectation. But where an idea follows on naturally and unsurprisingly, it will usually be enough to use discourse continuity intonation, a discourse continuity marker such as English ‘and’, or simple juxtaposition, for the coherence to be understood. This can be seen from example (1), where the removal of What’s more does not render the sequence incoherent. As Patterson and Kehler point out, “the more difficult recovering the correct relation would be without a connective, the more necessary it is to include one” (2013: 915). Additive markers are therefore more optional than markers of other relation types.

This notion of uneven marking of relations is compatible too with the uniformity of information density (UID) hypothesis, according to which predictability largely explains variability in reduction. That is, the more predictable an upcoming item is, the more likely it is to be reduced (phonetically, syntactically, discoursally) (Levy and Jaeger 2007). Asr and Demberg (2012: 84) apply this hypothesis to discourse marking and observe that easily inferable relations are on average marked more ambiguously than relations which are less expected, in a fashion that arguably reflects discourse-level information density smoothing.

< Prev   CONTENTS   Source   Next >