Coherence without cohesion?

[Simon Corston-Oliver]

I grow increasingly uncomfortable with the Hallidayan distinction between
cohesion and coherence. It is my belief that naturally occuring texts (i.e.
not constructed two sentence microtexts) are dripping in cohesive devices,
and that our sense of coherence is created by observing how those devices
cluster. The reader is not aribtrarily assessing whether a text hangs
together, but is being led by the nose by the writer to construct the
intended internal mental representation of the relations between
propositions. We perhaps resolve ambiguous indicators by reference to world
knowledge etc, but those indicators are at least massively constraining the
search space for a valid interpretation of a text.

To make a strong formulation, I doubt that coherence without cohesion occurs
in nature (e.g. in natural texts of more than clauses), or that cohesion
without coherence occurs either. Of course, we can _imagine_ texts in either
category, but do we observe them?

The claim is occasionally made (e.g. even in Mann and Thompson 1987) that
RST relations can be observed even without overt indicators. Certainly, they
can occur without overt cue phrases, but the writer's rhetorical goals are
going to have implications that ripple through the rest of the text --
influencing the selection of tense, aspect, the form of referring
expressions, even derivational morphology (e.g. selecting verbs for use in
Contrast relations that have derivational prefixes reflecting a difference
in polarity).

Recent work on the automated analysis of RST structure (see for example
Daniel Marcu's work at http://www.isi.edu/~marcu/ -- hey what happened to
the "portrait" of the monkey?, my own work at
http://research.microsoft.com/~simonco, or a host of other recent
computational studies) suggests that constructing models of text structure
is feasible on the basis of aspects of form, so I prefer to not view
"coherence" as a mysterious, intangible, unobservable property of a text.