|Theme in discourse: 'thematic progression' and 'method of development' re-evaluated
Fries (1981) hypothesises that the textual phenomena of 'thematic progression' (TP) (Daneš 1974) and 'method of development' (MOD) provide discourse evidence for the function proposed by Halliday (1967) for Theme, in particular that 'initial position in the sentence, or sentence-level Theme, means "point of departure of the sentence as message"'. This paper discusses the theoretical basis for this hypothesis, in particular the relation between TP and MOD, and reviews previous empirical research. Further research conducted by the author is described, into global proportions of TP, TP patterning, and the relation between TP and rhematic progression (RP) in a small corpus of 80 short argumentative texts. It was found that only small proportions of either argumentative text, or high-quality argumentative text could be considered as having a MOD. It was also found that texts had comparable levels of TP and RP. It is concluded that MOD is not a universal feature of discourse organisation, and therefore not conclusive evidence for Fries's original hypothesis.
Theme in discourse: 'thematic progression' and 'method of development' re-evaluated*
In this article I review the theoretical and empirical basis for two closely-related concepts which have been posited as being of use in discourse analysis, Thematic Progression (hereafter TP) and Method of Development (hereafter MOD). I also report some research of my own into the empirical basis for these concepts (part of a larger enquiry into Theme described in Crompton 2003).
TP and MOD are associated with claims made for the function of Hallidayan Theme, a concept which bridges the syntax and discourse levels of linguistic description. There are perhaps three groups of people to whom TP/MOD may be of interest: syntacticians interested in the claim that there is evidence from discourse to justify Theme as a unit of syntactic analysis; discourse analysts interested in the claim that syntactic Theme expounds discourse structure; composition theorists interested in the claim that certain kinds of Thematic behaviour are associated with rhetorical competence.
2Fries (1981) reconsidered
2.1The status of TP/MOD
TP/MOD first appeared in linguistic theory in Fries (1981). Fries's paper seeks to address a problem with Halliday's account of Theme: defining the meaning or function of Theme satisfactorily. Halliday (1967) and (1985) set out unambiguous accounts of the form of Theme but explained its function using metaphors: “the point of departure of the message…the starting point for the message… what the clause is going to be about” (1985:38-39). Fries (1981) can be seen as having anticipated the criticism that intuition-based metaphors such as these cannot be empirically validated. This criticism was in due course levelled:
Perhaps he [Halliday] is tuned into language in a way that the rest of us are incapable of, but those of us who can't easily pick out the parts of a clause which define "what it is going to be about", or its "point of departure" are simply unable to decide whether any of his claims about themes are right or wrong. (Hudson 1986:797)
A similar criticism of Hallidayan Theme appears in Givon (1995) the burden of which is that because the communicative function of Theme has not been defined independently of its form claims about Theme are empirically unfalsifiable and theoretically circular. Fries accepted the validity of such criticism: “no real argument has been brought forward to justify the statement that the Theme or beginning of a group, clause, or sentence means 'the point of departure of the message expressed by that unit'” (Fries 1981:4). Fries's remedy was to mount a defence based on "textual evidence" from "connected text". The detail of Fries's defence is complex and discussed in greater detail below but in essence he argues that the individual sentence Themes collectively may form noticeable patterns and that these patterns play a text-structuring role. In terms of discourse analysis, Fries’s (1981) implicit claim that a text's structure can be detected independently of the text's context is an unusually strong claim for a purely linguistic analysis of discourse.
Fries (1981) has been influential within the school of Systemic Functional Linguistics (SFL): Martin (2001) describes Fries's paper as "seminal" and "canonical" and Fries (1981) is often cited by SFL scholars, most notably perhaps Halliday (1985), as a key to explaining the function of Hallidayan Theme (e.g. Matthiessen 1992, 1995; Martin 1992; Thompson 1996). In discourse analysis literature outside SFL, by contrast, the MOD concept appears to have attracted little notice. The contributions of Hallidayan Theme to debate on information flow and structure within clauses and sentences and the Halliday and Hasan (1976) model of cohesion are well-known and commonly cited but the interaction of Theme and cohesion in MOD/TP theory appears not to be so well-known. A recent edited collection surveying discourse analysis (Schiffrin et al. 2001) refers to MOD in a contribution by a leading SFL scholar, entitled 'Cohesion and Texture' (Martin 2001). Other contributions entitled 'Discourse and Information Structure' (Ward and Birner 2001) and 'The Linguistic Structure of Discourse' (Polanyi 2001) make no reference to TP/MOD, however. In other recent introductions to discourse analysis, there are several references to Theme but none to TP/MOD (van Dijk 1997; Johnstone 2001).
Some scholars have questioned the text-structuring role claimed for TP/MOD. The authors of another survey of discourse analysis illustrate the three types of Danešian TP and argue that descriptive texts organised round time or location are well known for syntactic patterning of sentence-initial adverbials and clauses. They continue:
It is doubtful, however, whether we can generalise this technique to a topic development strategy for all non-narrative texts, as seems to be implied by Winter (1982) and Fries (1983). (Georgakopoulou and Goutsos 1997).
Models of discourse structure from scholars with a similar Firthian background to that of Hallidayan SFL also appear to discount TP/MOD as text-structuring. Hoey's (1991) strongly cohesion-influenced Patterns of Lexis in Text, contains no reference to TP/MOD. Sinclair (1993, 1994) proposes a dynamic model of text structure in which there are two kinds of cohesion: (i) that which accomplishes encapsulation of sentence-level constituents of previous text and (ii) that which does not. Into the latter category he assigns much cohesive patterning, including lexical cohesion, which is "not regarded as textual in nature […] not clearly structural" (Sinclair 1994:16).
From a composition theory perspective, Vande Kopple (1991: 339) drew attention to the potential applicability of TP/MOD theory and suggested several pertinent research questions, e.g. 'Do all texts have an identifiable method of development […]?' and 'What are the most prevalent methods of development in contemporary English prose?' To my knowledge, these questions have not been addressed.
In summary, Fries's (1981) claim that TP/MOD plays a role in expounding discourse structure, has been accepted within SFL but either disputed or largely overlooked outside SFL. What I would like to do in this paper is to re-evaluate TP/MOD, first considering their theoretical basis and then looking at empirical evidence relating to them.
2.2Problems in evaluating the core claims for TP/MOD
The core claims regarding TP and MOD in Fries (1981) are the following:
Step 1: Thematic progression correlates with the structure of a text.
Step 2: Thematic content correlates with the method of development of a text (and the nature of the text). (Fries 1981:4)
Evaluating these hypotheses is difficult for a number of reasons. One of these is that neither Fries nor subsequent Theme theorists thought it necessary to attempt definitions of the terms used: TP was a concept borrowed from Daneš (1974) and MOD is treated as a commonplace rhetorical term. I think it is helpful to recognise that following the adoption of Fries’s (1981) ideas, when these concepts have been referred to in SFL literature (e.g. Halliday 1985, Martin 1992, Matthiessen 1992, Thompson 1996) it is generally as components of a specifically Hallidayan model of Theme in discourse. For them to serve Fries’s original purpose, however, which was to provide independent evidence of the validity of Hallidayan Theme, they need to be viewed as concepts independent of Hallidayan Theme.
A second and related difficulty in evaluating claims about TP/MOD is that the relation between the two concepts is unclear. Some overlap would appear inevitable but the precise nature and the degree of the overlap is a matter for interpretation. TP and MOD are both semantic and structural text properties. In his abstract Fries states that thematic content correlates not only with TP but with "if the passage is outlinable, the outline structure of the passage" (1981:1). My own reading would lead me to interpret Fries (1981) as intending the following relation between TP and MOD: TP is the superordinate term and MOD refers to a text structure based on one of the three TP types described in TP theory (Constant). However, it seems clear from the SFL literature that a more prevalent interpretation (e.g. Halliday 1985) is the following: MOD is the superordinate term for a text structure based on any of the three TP types (Constant, Linear, Derived). The concepts of TP and MOD have coalesced within subsequent SFL literature: Ventola and Mauranen, for example, write of "all the major methods of thematic development" (1991:476) and the hybrid term "thematic development" appears in the title and introduction of an edited collection on Theme (Ghadessy 1995). To ensure completeness of coverage, I have chosen to designate the two concepts by the awkward but unified label 'TP/MOD'.
To these terminological difficulties in evaluating TP/MOD must be added two more concerning their epistemological status. First, there is some ambivalence in the literature as to what extent TP and MOD are everyday linguistic phenomena and to what extent they are ideal abstractions. Are they principles of linguistic description or rhetorical prescription? Do all texts have TPs and MODs – just as all sentences have Themes – or only rhetorically effective texts? Do all rhetorically effective texts have TPs and MODs or only some? The concept names themselves imply a degree of universality, i.e. that there is a semantic progression in the cumulative sentence Themes of any text and that this Theme-based progression is the method by which that text is developed. Fries (1981) could be interpreted narrowly as hypothesising that where texts have a clear text structure, that text structure correlates with TP and that where texts have MODs these correlate with thematic content. In some cases, however, Fries seems to have been interpreted as hypothesising more broadly that all texts have TPs which correlate with text structure and all texts have MODs which correlate with thematic content, as exemplified in the following statement:
The choice of clause Themes plays a fundamental part in the way discourse is organized; it is this, in fact, which constitutes what is often known as the "method of development" of the text. (Halliday 1985:62)
A second epistemological difficulty in evaluating TP/MOD is that the more universal statements italicised above seem in places (as in the last quotation) to have been interpreted not as hypotheses to test but as already proven principles. Matthiessen, to take another example, argues that: “thematic selections have been shown to key into [MOD] (see Fries 1981, for the original research)” (1995:26). MOD's discourse function has been further explained by means of new metaphors, for example “[a] lens, an orientation, a perspective, a point of view, a perch, a purchase” (Martin 1992:489). MOD has been positioned within larger theoretical frameworks, also often associated with new metaphors – periodicity (Halliday 1985; Matthiessen 1992, Martin 1995b), scaffolding (Martin 1995a), logogenetic ideational networks (Matthiessen 1995). Such assimilation into larger theory may suggest that TP/MOD are conceived of as universal rather than contingent properties of texts.
If TP/MOD were indeed textual universals, what would remain for research would simply be the details of variation across genres. In this paper, I would like to step backwards a little, treat Fries's Steps 1 and 2 as hypotheses to be tested, and examine the evidence for and against them.
2.3Evidence for claims in discourse
Regarding the general issue of evidence for research hypotheses in discourse analysis, Tomlin et al. (1997) suggest that there "are three principal methodological strategies employed in the analysis of text and discourse: (1) introspection-based analysis, (2) text counting methods, and (3) experimental and quasi-experimental methods”. In this typology, the methodology of Fries (1981) can be seen as a type (1) strategy – introspection-based:
Argumentation consists largely of documenting numerous examples congruent with one's [theoretical] definition and hypotheses (Tomlin et al. 1997:101).
Martin (1992; 1995a; 1995b) and Matthiessen (1992; 1995) have written defending and refining Theme using the same introspection-based methodology as Fries (1981), that is, by documenting examples of TP/MOD in selected texts. As well as introspection, insofar as Fries (1981) sought the evidence of third party informants as to the comprehensibility of texts with manipulated Themes, he ventured a little into type (3) strategy – quasi-experimental methods.
The main weakness of Fries’s (1981) own evidence for his hypotheses is its paucity. The claim about correlation of TP and text structure, for example, is supported by the evidence of two analysed paragraphs: how far these paragraphs can be considered representative of text perceived as well-structured is not clear. Significantly, Fries himself appears to regard Fries (1981) as in need of empirical support. Fries and Francis over ten years later argued there was a need "to collect and analyse far more data" (Fries and Francis 1992:52). In a 1995 review of work published on Theme since Fries (1981), Fries describes that article as having made two hypotheses (concerning TP/MOD) and adds two further hypotheses (concerning Theme and genre, and Theme and generic elements of structure). He argues that work published on Theme since 1981 "has generally supported the four hypotheses" but argues for "a considerable expansion of the data which are used to test them" (Fries 1995b:339), pointing out limitations in the size and spread of the data samples used in previous studies. In discussing the lack of "robustness"' of previous empirical studies Fries implies that, even in principle, introspection-based methods are insufficient to prove generalisability. It seems clear to me that analysis of further selected texts (e.g. Matthiessen 1995; Martin 1995a) per se cannot be relied on to support a general claim about TP/MOD in discourse, any more than, say, the continuing citation of selected texts written in blank verse would support a claim that all texts are written in blank verse. By the same token of course, the continuing citation of selected texts which do not fit a particular claim would not necessarily undermine it. What is required is to analyse representative samples of language collected independently of criteria suggested by the claim.
Rather than replicating Fries's (1981) introspection-based methodology, one way to confirm/disconfirm his hypotheses would be to employ a different methodology. Tomlin et al. argue that “the best overall strategy in studies of discourse semantics" is "to provide convincing evidence from an array of studies” (1997:102). The research reviewed below and my own research employed a text-counting (type 2) methodology, in which "critical theoretical notions are operationalised through a set of heuristic counting procedures" (Tomlin et al. 1997:101).
Because form and function are so tightly knit in TP/MOD theory, operationalising the "critical theoretical notions" is important. In discussing research on the relation between language form and communicative function Givon points out:
If two entities A and B are said to correlate, then neither can partake in the other's definition; otherwise stating that they "correlate" is stating a tautology. (Givon 1995:309).
To say anything worth saying about the relation between TP and text structure, then, a definition of text structure is required which is independent of Theme. Similarly, any claim about correlation of thematic content and MOD requires a Theme-independent definition of MOD. In order to carry out empirical research on TP/MOD, two sets of operationalisations are required: first, for the Theme-bound concepts (thematic progression and content); second, for the non-Theme-bound discourse constructs (text structure and textual method of development).
When the occurrence of these features in texts has been analysed and quantified, the claim that these features correlate (TP with text structure, thematic content with MOD) can be tested. To do this it is necessary to be specific about the details of the correlations hypothesised: as Givon (1995:306) argues, to carry out research into functional grammar it is necessary "to make hypotheses about form-function association explicit enough so that they generate explicit factual predictions". It is then possible "to subject such factual predictions to falsificatory testing" (Givon 1995:306).
In the next section I shall review the core concepts of TP/MOD from a theoretical perspective, and attempt to flesh out what form the correlations might take. I shall go on to formulate some specific predictions arising from the hypothesised correlations and review the findings from previous text-counting empirical work in the terms of these predictions. Finally, I shall describe further research of my own into the extent to which the predictions are fulfilled in a small corpus of argumentative texts.
2.4The core concepts of TP/MOD
2.4.1TP and text structure
If Fries's were the first use of the term, TP might be interpreted as simply the global set of Themes in a text considered from a dynamic perspective. However, Fries borrows from Daneš (1974) a typological modelling of how Themes are connected with previous textual content, reproduced here as Figure 1. What sometimes appears to have been overlooked is that because Fries was working from a different account of Theme from Daneš, his account of TP necessarily differs from Daneš's. Essentially, Daneš (1974) works within the Praguean tradition according to which Thematicity is determined contextually rather than syntactically. Theme, treated as non-Rheme, is discovered indirectly, employing "a procedure using wh-questions, prompted by the given context and situation, for eliciting the rheme" (Daneš 1974: 114). Practically this means that in Figure 1 "the formula TR the order of symbols does not necessarily correspond to the sequence of expressions in a particular sentential utterance based on this formula" (Daneš 1974:118). (It seems appropriate to assume, however, that the vertical arrows, indicating the "contextual connection" of utterances (Daneš 1974:112) transfer unchanged to Hallidayan TP.)
[INSERT FIGURE 1 ABOUT HERE]
Figure 1 Daneš's Thematic Progression Types (Daneš 1974:118-119) ( T= Theme, R= Rheme)
For Danešian Theme the TP concept is not a claim about word-order but about textual connectedness or connexity. For Hallidayan Theme, however, the TP concept is inevitably a claim about both textual connectedness and word-order. Fries (1981) effectively adds to Daneš’s claim – that all sentences in a discourse are connected linguistically to the preceding discourse – a new claim that the linguistic exponent of this connection invariably appears sentence-initially. The practical consequences of this difference between Danešian Theme and Hallidayan Theme can be illustrated by attempting rival analyses of the TP of the second of the following sequence of sentences from Fries's first sample text
 The process of learning is essential to our lives.
 All higher animals seek it deliberately.
(Bronowski (1959:111) cited in Fries (1981:8)).
For Daneš the (contextually-determined) Theme of  would be it (linked to the process of learning), and the TP type Constant . For Fries the (syntactically-determined) Theme of  would be All higher animals (linked to our lives) and the TP type Linear. This mis-match between the Danešian analytical apparatus of TP and the Hallidayan concept to which Fries has attempted to put the apparatus to work has practical consequences for empirical research which will be discussed later.
Fries glosses text structure as "what ideas are coordinate with or subordinate to what other ideas" (1981:9). Recognising and describing text structure is one of the core endeavours of discourse analysis and there are many rival models from a range of disciplines (cognitive science, sociology, natural language engineering) as well as linguistics. Fries chooses to employ a rhetorical analysis of texts, using two text extracts independently analysed by a scholar interested in rhetoric rather than linguistics (Christensen 1967). A feature of this analytical system is its simplicity. This could be a drawback in that the simple binary distinction subordination/co-ordination appears to conflate various notions pertaining to text structure such as dependency, hierarchy, weighting.
Fries's discussion of the two example texts seems to propose the following correlations:
TP type Structural feature
One feature of this correlation is that it leaves out mention of a structural correlate for Derived TP. Another feature is that there is no TP type to correlate with superordination, that is no TP to indicate that a sentence is structurally higher than its predecessor, unless co-ordinate with a higher previous Theme. Logically, this means that the opening sentence of a text must always be at the apex of the text’s structure: in terms of Fries's visual representation of structure as patterns of indentation it would be impossible for sentences to appear to the left of the opening sentence. This appears to be an inherited feature of Christensen’s original modelling of rhetorical structure, and accords with the privileging of linear priority inherent in TP. If the correlation holds good, the linearity of a text constrains the formation of its hierarchy. Such a linearity constraint is not a necessary feature of models of discourse organisation (e.g. the Problem-Solution-type patterning discussed in Hoey (1983), Mann and Thompson’s Rhetorical Structure Theory (1987), Polanyi's Linguistic Discourse Model (1988)). On the basis of the exemplar texts in Fries (1981) it would seem to be the case either (a) that Fries is implying that the first sentence is paramount structurally in well-formed texts, or (b) that there happens to be no TP type which correlates with non-text-initial sentences which are paramount structurally.
One of Fries's major arguments is that broadly narrative texts will be characterised by a structure of repeated co-ordination/Constant TP, while expository and argumentative texts by contrast will be characterised by a structure of repeated subordination/Linear TP. As well as citing Christensen's (1967) Theme-independent analyses of text structure, Fries also makes the following observation as to the model relation between TP and text structure:
ideally, in argumentative or expository prose, each sentence should follow logically from what has gone before. This implies in part that the point of departure of each sentence should relate in some way to what has preceded. If there are unexplained jumps in the sequence of starting points, that implies that there are breaks in the argument. (Fries 1981:6)
Such an account of the structure of argumentative texts is intuitively appealing. However, it offers no argument why points of departure or starting points of sentences should be found in their (Hallidayan) syntactic Theme. That this is actually the case would have to be demonstrated empirically.
2.4.2Thematic content and MOD
'Thematic content' seems a relatively straightforward concept, denoting the content of the global set of Themes in a text. Similarly to text structure, MOD is argued to be a perceived textual property. The origin of the term is not given by Fries. It sounds like a term from composition theory but I have been unable to trace an example of its use. Its function, to develop the text, is "obvious" or recognised by readers according to Fries (1981:15). MOD is distinct from other textual properties possibly perceived by readers, for example, topic and message, although Fries notes that topic and MOD may be identical.
Fries depicts the kind of content which correlates with MOD as collectively constituting (a) a general semantic concept, e.g. relative position, reference to component parts, contrast in time and/or (b) a lexical system, e.g. living-growing-changing, wisdom versus chance, concepts to do with government. Fries's hypothesis of correlation between thematic content and MOD suggests a simple two-term system:
…if the themes of most of the sentences of a paragraph refer to one semantic field (say, location, parts of some object, wisdom vs. chance, etc.), then that semantic field will be perceived as the method of development of the paragraph. If no common semantic element runs through the themes of the sentences of a paragraph, then no simple method of development will be perceived. (Fries 1981:20)
Mention of both a ‘simple’ and later a ‘single’ MOD suggests that there might be contrasting complex or multiple MODs but Fries does not pursue this suggestion. Presumably too great a complexity of MOD would militate against the functional requirement of MOD that it be obvious to the reader and too many semantic elements in an MOD would be the equivalent of zero-MOD. The textual meaning of not having a single MOD is not specified, but one implication of Fries’s article is that single MODs are associated with texts perceived as rhetorically effective.
In a later paper Fries clarifies that because MOD is a psychological phenomenon, the correlation hypothesis – that reader perceptions of MOD are associated with Theme content – should be "testable using standard psychological testing techniques" although this has not yet been done (1995a:9).