Mutiple levels/analyses

[Simon Corston-Oliver]

As follow up to Ali's message of 18 Jan.

Is there one level of analysis / always one compelling RST relation that
wins out in each case when we are combining text spans? I I think not, but I
believe that most written text that we examine will tend to make us think
so. I like to think of an analogue to prosody: in general intonation, pauses
etc correlate well with syntactic constituency. Declaratives tend to have
falling intonation in English, pause-defined units often look like XPs etc.
However, prosody has functions other than to signal discourse constituency:
a rising final intonation on a declarative might signal a desire to maintain
the floor, for example.

Perhaps something similar is going on in discourse: in general the
organization of textual material is subordinate to interpersonal goals that
the lg producer is trying to achieve. In written monologues, the textual and
ideational materials predominate or are aligned with reasonably
straightforward interpersonal goals. As Maier and Hovy (1991) note, some
texts are subordinate to a single speech act like DESCRIBE -- as is
certainly the case in the encylopedia prose that I have analysed: DESCRIBE
or EXPLAIN are the overarching speech act goals.

As we move into texts where lg producers are making more complex
interpersonal moves, we might expect to see divergences of the sort that
Ford (1986) or Moore and Pollack (1992) and noted -- even to the extent that
the headedness of the relationship between two text spans is problematic. In
dialogue, where RST is at its descriptive weakest, the mismatches between
levels become most pronounced. The communicative goals that we as discourse
linguists imagine the speaker has in mind are more fluid and are being
subverted by interlocutors.

Given that different forces are shaping the text, and that these are not
always in synch, I don't think we should insist that RST has the one true
analysis for each text. So this weakens our claims about the theory? Oh
well.

RST hums along in describing ideational and textual relations when the
interpersonal stuff is nicely in synch, which accounts for a lot of text
(most of the text that I have been interested in to date).