Mixed and pure schemas in RST

[William Mann]

There is a feature of RST that seems to be often missed, which greatly
simplifies the analysis of text.  It was given in the definitive paper (Text
8(3), 1988) but not particularly prominent.  In this message I want to make
it clearer, because I am sure that for some people working with RST it has
not been noticed.

When a relation links a satellite to a nucleus,  only that relation can be
used to link to that particular nucleus.  It can be used indefinitely many
times, both left and right, but the relations may not be mixed.

The most basic reason for this restriction is to avoid having two different
sets of constraints, possibly incompatible,  on one nucleus.  So, for
example, having both Motivation and Evidence point to the same nucleus would
require that it express both an action and a true claim.

Another reason for the restriction is that in the relation definitions, the
constraints on satellites and nuclei form a kind of matched set; they often
refer to each other.  Underlying that is the idea that the N+S pair is doing
one kind of job for the author, one expressive function, satisfying one
intent.  Usually there are scope considerations as well, so that for example
a Summary normally scopes over both the most important nuclei but their
satellites as well.  Thus the Summary relation should not point to a nucleus
that has other satellites.

The only exception to this is the Motivation and Enablement pair, most often
seen in advertising.  Both the motivating statements and the enablement
support, in different ways, the same action, e.g. buying a particular
product.  Here the matching has already been done in the definitions.

I am grateful to Hanny den Ouden for pointing out that this had been left
obscure.

Happy analysis.

Bill Mann