Header image  
How readable are your texts?  
line decor
line decor

Text Coherence and Cohesion

Text coherence
Most readers are aware that some texts, whatever their content, seem to "hang together" better than others and are therefore easier to read. In part this is a function of how they conform to expectations about text types (rhetorical organisation) but is mainly a function of how they "cohere".
"Understanding a discourse may be regarded as the construction of a mental representation of the discourse by the reader. An acceptable discourse representation has a property that distinguishes it from the representation a reader might make of an arbitrary set of utterances: The representations of the segments in the discourse are linked coherently." (Sanders et al 1997:1,2)

First of all we have to accept Carrell's (1982) assertion that cohesion is not coherence. Few would now dispute that:
"cohesion relates only to the interconnectedness of the 'components of the SURFACE TEXT' while coherence relates to 'how the configuration of CONCEPTS and RELATIONS which underlie the surface text, are mutually accessible and relevant' " (de Beaugrande and Dressler 1981:3-4).
Similar distinctions are made by Hoover, (1997:195) and Sanders et al (1997:2)
For the purposes of this discussion, a distinction can be made between cohesive devices operating on a surface, textual level, and discourse relations which may or may not be explicitly signalled.

Many researchers have looked at logical relations and conjunctions in investigating text comprehension. The Kintsch & van Dijk model of the reading process involves making inferences about how propositions are linked, (1978:365): the Just & Carpenter model provides for "interclause integrations" (1980:343). Meyer, (1977), Geva (1983), Winter, (1982), Meyer, Brandt & Bluth (1981) have all investigated the effects of signalling of relations on the perception of the organisational structure of texts.
Analysis of coherence relations would seem to offer insights into the difficulty of text because "coherence relations are ultimately cognitive relations" Sanders et al (1992:1). Winter and Hoey have made the same point:
 "A clause relation is the cognitive process whereby we interpret the meaning of a sentence or group of sentences in the light of its adjoining sentence or group of sentences." (Winter, 1971 and elsewhere) and its refinement by Hoey "A clause relation is also the cognitive process whereby the choices we make from grammar, lexis and intonation in the creation of a sentence or a group of sentences are made in the light of its adjoining sentence or group of sentences." (Hoey, 1983)

It seems unclear whether readers process text merely for local coherence or whether they try to integrate information into a rhetorical framework (global coherence) or with background knowledge (Kintsch, Mandel & Kozminsky (1977:552) "….the formation of macrostructures is an integral part of the comprehension process and occurs during reading…" and Albrecht & O'Brien (1993:1062) suggest they do; McKoon & Ratcliffe (1992) suggest they do not). Whether they do or not might also depend on the reader's ability, his motivation and his reading goal. It might also depend on the length and type of text. Texts which conventionally have a fairly fixed rhetorical structure might force more sensitivity to breaks in global coherence (in the way that it is easy enough to amuse a child by introducing a few inconsistencies into a well-known story) whereas other genres might be less constrained (e.g. academic texts -because they present a lot of new information). Kieras makes this point:
"…quite often readers are required to understand material whose content is basically novel, lacking in stereotypical organization, and about which they have few expectations." (1978:14)

Whatever the case, we need to be able to compare texts from a clausal relations point of view in order to be able to say that one text is more difficult than another. A simple way might be to count the ratio of conjunctions per number of sentences (marked relations presumably being easier to understand than unmarked ones). But there may be a problem here because relations are signalled by other devices than conjunctions such as lexis and repetition (see Hoey, 1983, Hoey & Winter 1986) and causal verbs (Singer 1995). A more sophisticated method would be to take into account;

  1. The ratio of explicitly marked relations to those that have to be inferred, and
  2. The type of relations used. (On the assumption that certain relations are easier to process than others)

Signalling of clause relations is generally discussed at a micro level (relations between propositions, clauses or sentences) but relationships at a super-ordinate or macro-level are also signalled by such markers. So these markers also help readers who know how to use them (Meyer, Brandt & Bluth, 1981) to establish the rhetorical framework of the text.   
A number of lists or classifications of coherence relations have been proposed (e.g. Martin's conjunctive relations (1992), Mann & Thompson's Rhetorical Structure Theory, (1987), Winter (1977, 1982), Lascarides & Asher (1991).
Bateman & Rhonduis (1997) have tried to draw some of these together in a synthesis.
It is clear that understanding a text involves understanding the explicit and implicit relations that bind the text together. But how readers actually do this on line as they read is far from clear. Actually analysing a text to make explicit these relations takes an extremely long time and not all analysts will come up with exactly the same analysis. Any thorough comparison of texts would require such an analysis using one of the systems mentioned above. O'Brien (1995) used RST to investigate a college essay and such an analysis could be used to investigate any text. Those which were less coherent could be presumed to be more difficult and the analysis would point out directions for improvement. Britton & Gülgöz (1991) used a propositional analysis to investigate and improve the readability of text. Both these methods require considerable time and no little expertise.
What we really need to identify is whether explicitly stated relations are easier to process than ones which have to be inferred, whether certain relations are more difficult to process than others, and if so which relations they are. 
Irwin and Pulver (1984) suggest some answers to these questions:
"results … indicate that students ….are more likely to comprehend a causal relationship when it is stated explicitly"
"this seems to indicate a way in which the language of … textbooks might be made more comprehensible."
"irreversible causal relationships are more likely to be understood than are reversible ones." (Irwin and Pulver 1984:405)

A list of relations which are more difficult to process is one thing but it would be even better to know why they are more difficult. Sanders et al (1992) attempt at drawing up a taxonomy of coherence relations seems promising in this regard.
They propose that relations can be defined in terms of four factors; a) basic operation (causal or additive) b) source of coherence (semantic or pragmatic) c) order of the segments and d) polarity. It may be that the last two provide clues as to difficulty, non basic order and negative polarity being more difficult to process. This would be an interesting research question.  
Text difficulty might then be assessed in two ways:
1) If we accept that some texts are better written than others and that texts can be rewritten or edited to present the same information in a more accessible way, then we can assess readability by looking at how easy or difficult it is to assign coherence relations to segments of text. Texts which are difficult to analyse might be classified as more difficult.
2) Different types of texts are likely to exhibit different coherence relations. It may be the case that certain relations are more difficult to process than others. For example, Black & Bern (1981:267):
 "Causally related events in narratives were remembered better than events that were not causally related"

Sanders et al (1992:31) state
 "…..in general it takes longer to verify denials than affirmatives (see Wason & Johnson-Laird, 1972), and it takes longer to judge the truth or falsity of unless sentences than that of equivalent if sentences (Clark & Lucy, 1975; Noordman, 1979)…." 

 If we are concerned with particular readers it may also be the case that as in L1 acquisition (Bloom, et al 1980: Wing & Scholnick, 1981, Kail & Weissenborn, 1985, cited in Sanders et al 1992:30) acquisition of coherence relations follows a fixed order. Geva (1992:732) states "There is evidence in the L1 literature to suggest that understanding conjunctions as marking the focus of topical relations between sentences is a gradual process that is mastered by literate adults (Johnson & Pearson, 1982; McClure & Geva, 1983; Zinar, 1990)."
She also states that "skilled and less skilled readers have been shown to differ in the degree to which they infer logical relations in text (Bridge & Winograd, 1982; Evans & Balance, 1980; Geva, 1986, Geva & Ryan, 1985; Irwin, 1980)." (Geva, 1992:732).
If this is the case it might be possible to identify texts too rich in relations the readers have not acquired for them to be readable (a process teachers already do on an intuitive basis).   

We have seen that coherence relations may be inferred or explicitly signalled by conjunctions or other devices. Other forms of signalling give the text cohesion and indicate that it is coherent without giving actual clues to the precise relations that hold between propositions. These cohesive ties may also affect readability as we shall now see.
Whereas coherence is assigned to a text by a reader, cohesion is a property of the text itself.
Of the cohesive devices (reference, substitution, ellipsis, conjunction and lexical) identified by Halliday and Hasan (1976), the most common by far are reference and lexical (see Binkley,1988:112). 

Not surprisingly studies have found the closer referents were to their pronouns, the easier they were to process (e.g. Clark & Sengul, 1979: Cirilo, 1981). Since then, others factors influencing the ease of resolution of anaphors have been found:
".. whether the antecedent is in focus (e.g. Garrod, Freudenthal, & Boyle, 1994) ……syntactic constraints (e.g., Nicol & Swinney, 1989; Swinney & Oesterhout, 1990), syntactic prominence (e.g., Gernsbacher & Shroyer, 1989), as well as pragmatic constraints (e.g., Garnham & Oakhil, 1985; Hirst & Brill, 1980; Tyler & Marslen-Wilson, 1982)"  (O'Brien, et al 1997:2)

Hoover (1997) has also reviewed the literature on this, and concludes that factors which facilitate reading are whether there is continuity of reference, whether pronouns are "in focus" and whether there was a parallel syntactical function between the pronoun and the referent (pronouns and referent which were grammatical subjects were easier to process). Also "pronouns that referred to the agent rather than the patient of the preceding sentence were easier to process, regardless or their syntactic position." (1997:201)  
But it is unclear whether cohesive ties in general have a serious effect on readability in their own right, or whether the difficulty that might arise is caused by vocabulary effects or the inability to establish the semantic relations that obtain in the text.
Freebody and Anderson found "weak and inconsistent effects of cohesion" (1983:291) in their experiments on reading comprehension. They suggest that "cohesion, in the specific sense of linguistic ties, simply is not very important in reading" (1983:291)
This may not be a very popular or accepted conclusion given all the effort and emphasis put on this topic in teaching reading of late, but they may be right. Interesting though analysis of text is from this point of view, we have to remember that we are analysing a product (the text itself) and this might not throw much light on what a reader actually does (with these features) as he processes a text. Brown and Yule make this point:
"It is important, however, that the discourse analyst should be clear just what it is that Halliday & Hasan are doing and should not assume that the account of textual relations produced as a post hoc analysis of the structure of a completed text should necessarily be revealing about how a processor working 'on-line' as a the discourse unfolds experiences that discourse." (Brown & Yule 1983:204)

The most important function of cohesive ties may not be the links themselves but simply the reduction they afford the reader: repeating antecedents in full (some of which may be long noun phrases or even a concept outlined in a whole paragraph or more in the case of anaphoric nouns) would render the discourse so unwieldy as to be uninterpretable, placing demand on memory which might become intolerable.
Tempting as it might be to use cohesion as an indicator of text difficulty (since it is analysable in an objective manner and not difficult to do) cohesion does not have much bearing on text difficulty.
Binckley (1988) has made a case for using cohesion in this way and although he rightly relates the analysis to the readers for whom the text is intended, it probably falls down on two counts; a) that cohesion is not a serious indicator of text difficulty, b) the effects measured in the cloze tests he uses are vocabulary effects not cohesion.   
It is very tempting to agree with Morgan & Sellner that cohesion is "an epiphenomenon of coherence content" (1980:181)
Although cohesive devices are clues to the coherence of a text, a text is only coherent to the reader if the textual relations are clear to him. This can only be investigated with the co-operation of the reader not by just examining a text.
However, It may be that we can identify some features of cohesion which are more difficult for some readers to interpret. Cooper (1984) in a comparison of practised and unpractised readers found:
"The features which discriminated most clearly between practised and unpractised readers were lexical cohesion (in particular hyponymy,{..}) and cataphoric reference; but both groups were unsure with synonyms. Grammatical cohesion achieved by anaphoric reference (e.g. locative reference..), substitution and ellipsis did not appear to present much difficulty to either group. We concluded again that practised readers are distinguished from unpractised readers by their relatively superior lexical competence. Practised readers not only have larger vocabularies, but have greater knowledge of lexical relationships. In particular, they have a better grasp of the ways in which writers use words to create and maintain textual relationships by exploring features like hyponymy and synonymy." (Cooper, 1984:131)

And Berman identified and number of factors she thought might cause problems:
"We suggest, next, that the FL reader needs maximal 'transparency' in marking the relations between one part of the text and another. That is certain kinds of cohesive devices [ ] may render a text opaque to the FL reader. These may take the form of deletion - for instance, by means of gapping, lack of relative pronouns in English relative clauses, wh+be deletion in post-nominal modifiers, etc. - or of substitution of, say, nominal one or verbal do as grammatical substitutes for repeated lexical material, as well as of lexical substitution." (Berman, 1984:42)

Whilst these may cause difficulty, they do not seem to be promising candidates on which to base any indicator of readability. So we can probably dismiss reference, substitution and ellipsis as major factors contributing to text difficulty. What remains (conjunction having been dealt with under clause relations) is lexical cohesion and is a vocabulary effect.

      © 2007 Essem Educational Limited