Hierarchical levels Hierarchical organization

a conceptual unit cannot be assumed: “the material included in even a well-formed sentence does not necessarily represent a unit of perception, storage, or remembering, but results from an on-line, one-time decision that something has been completed” Chafe 1994:143. So as addressees process discourse by constructing mental representations, at what level do they stop opening new substructures and simply fill out the current discourse space? At what level is it true that “each successive utterance updates the current discourse space” Langacker 2001a:171 without subdividing it? 61 If one could discover a particular level of discourse which is qualitatively different from other levels, which by its nature could have no embedded discourse units, then that would be a minimal discourse unit, regardless of the number of sentences it might have. Although minimal discourse units may not exist in this strict sense, it does appears that language processing commonly utilizes a discourse level that can be described as a MINIMAL COMPLETE DISCOURSE UNIT , in the sense that if a such unit embeds any other discourse units, they will predictably lack some property or properties that are characteristic of complete discourse units. The minimal complete discourse unit is what I call the paragraph. 62 Reasons for this characterization will shortly be presented, along with the related hypothesis that paragraphs are the lowest-level discourse units that have subspaces in mental representation.

2.6.1 Hierarchical levels

This treatment uses the following hierarchical levels in discourse form: • The GLOBAL LEVEL is that of the text as a whole as, for example, the Gospel of Matthew. A text may never be completely closed off to the possibility of being embedded in one of a higher level as through “just-in-time coherence,” §3.5.4. The global level is simply that of whole texts as we receive them. • E PISODIC LEVELS are major subdivisions of the text, smaller than the entire text and made up of paragraphs. So as not to multiply terms, we will consider that episodes can be embedded in episodes. One recent commentary on Matthew indicates the passion story 26:1–27:66 as a major episode, and embedded in that is the episode of the death of Jesus 27:27–66, and embedded within that, the burial 27:57–66 Morris 1992. • P ARAGRAPHS are minimal complete discourse units, and the basic-level conceptual unit of discourse. The paragraph is the lowest MACRO - LEVEL UNIT , other macro-level units being episodes and the text itself. The distinctiveness of the paragraph is discussed in §2.6.3. In the Gospel of Matthew, the episode of Jesus’s burial contains two paragraphs: the burial itself which was done by Joseph of Arimathea 27:57–61 and Pilate’s subsequent making the tomb secure 27:62–66. • M ICRO - LEVEL UNITS are “building blocks of the paragraph” Hinds 1979:146, that is, component units in a paragraph schema §2.2.4. They usually consist of one or two utterances, but can have more. Steps in the paragraph about Joseph of Arimathea Example text 6 are displayed in Figure 5. Although these are formal units, the perception that discourse is segmented into such levels uses, in general, both conceptual and formal evidence. Conceptual evidence has to do with various conceptual functions: initialization continuities and discontinuities along orientation dimensions; see §2.3.1 and Dooley and Levinsohn 2001, ch. 7, development such as schemas; see §2.2.4, and closure §2.3.3. • The global organization of the text largely reflects the development function, and knowledge management in particular. In the Book of Acts, for example, the development is basically geographical and referential, reflecting Luke’s intent as author §2.2.5. 61 In Gernsbacher’s Structure Building Framework, filling out the current discourse space is called “mapping” and opening a new substructure is called “shifting” Gernsbacher et al. 2004:146. These are described as “two general cognitive processes to build mental representations structures of texts such as narratives.” 62 The term “paragraph” referring to discourse units rather than simply to typographical paragraphs, has a long history: see, for example, Hinds 1977, 1979; Longacre 1979; Chafe 1994, ch. 23. • A text is often segmented into paragraphs on the basis of genre-related orientation dimensions. In Acts, “because the primary genre of the book is narrative, many sections naturally subdivide on the ground of changes of temporal setting” Levinsohn 2000, §17.1. • A paragraph’s internal schema generally depends heavily on that paragraph’s genre as it functions within the culture loc. cit. and §2.6.3. • Micro-level phenomena are generally internally structured by sequential aspects of knowledge management and attention management. In knowledge management, this largely refers to relations between propositions; in attention management, it means the kinds of pragmatic relations that are seen in information structure. 63 Why does the speaker begin a new paragraph at a particular place? Four possible reasons have been proposed, any or all of which might be operative in a given situation: • A new paragraph can be begun because a significant DISCONTINUITY is perceived in one or more of the orientation dimensions §2.3.1; Givón 1984:245, Chafe 1987:43. • There is an inherent limit as to how much semiactive information can be manipulated at once within a paragraph, so when that limit is exceeded a new paragraph is begun Chafe 1994:137–139. • In knowledge management, the speaker can begin a new paragraph “whenever one returns from a lower or less inclusive level to a higher or more inclusive level” in the schema Grimes 1975:106, citing Christensen 1965, or when a step cannot be taken as presupposed but needs to be argued. • In attention management, “Episodes are defined ultimately by the sustaining of attention on a particular paragraph level theme…. Episode boundaries represent major breaks, or attention shifts, in the flow of information” Tomlin 1987:458, 460. Possibly also, one can begin a new paragraph in order to return from a subtheme to some higher-level theme; see the discussion of the interview with a barber Appendix F in §2.6.5. In the discourse model of Grosz and Sidner 1986:177, “linguistic structure is not entirely DECOMPOSITIONAL . An individual segment [discourse unit; RAD] may include a combination of subsegments and utterances only in that segment and not members of any of its embedded subsegments.” Thus, for example, an episode could contain embedded episodes, paragraphs, and utterances which do not belong to any other discourse unit. The present treatment is agnostic on the question of structural decompositionality. One possible type of utterance-constituent in a discourse unit would be represented by introductory or closing statements, such as the macropredication labelled “Step 1,” lines 01–02 in the schema of the Guarani bow text Example text 9. This sentence pair corresponds to what is sometimes called a “thematic sentence.” The steps in Example text 9 indicate a flat structure, whereas if lines 01–02 were not considered a step, the structure would be hierarchical. There is another possibility for hierarchical structure, however, suggested by the three conceptual functions in §2.3: there could be a preliminary decomposition involving an initialization component, a development component, and a closure component, with commonly only the development component having embedded discourse units as internal components. On that analysis, Example text 9 would have an initialization component composed of lines 01–02 followed by a development component which would be decomposed of the other steps; in that example, there would be no closure component. Another possible type of utterance-constituent could be “transition” statements occuring between regular subunits of a discourse unit. There might be yet other possibilities, but the question of decompositionality does not seem to be crucial to discourse thematicity and topicality, hence is not pursued further here. 63 Although information structure can reflect the structure of paragraphs and larger units, its primary scope seems to be micro-levels. Thus, in Japanese oral narratives, Clancy and Downing 1987:46f. found that “the primary function of wa is to serve as a local cohesive device,” and that “the association between thematicity and wa-marking of a referent is one consequence of the use of wa as a locally motivated marker of discourse cohesion.”

2.6.2 Micro-level units as incomplete discourse units; centers of attention