conclusion relation

[Manfred Stede]

Hi Ed, and all,

thx for the reply, and sorry if my writings were obscure at times... let's
see
if I can make my case a litte clearer. Let me respond to your first point
(themes & purposes) now and get back to the other in a day or two.

> At 8:47 PM +0200 8/24/08, Manfred Stede wrote:
>>As everybody in the discussion seems to agree, there are other things
>>going on in text besides constructing the argument / realizing the
>>discourse purpose. One important thing is the thematic development,
>>and I'd argue (and maybe not everybody agrees...) that once we decide
>>to use RST trees to capture writer's purposes, we can't simultaneously
>>capture thematic development with those same trees very well. cf. the
>>discussion of ELABORATION by Knott et al. 2001, and others.
>
> I can sort-of imagine 'constructing the argument / realizing the
> discourse purpose'.  The speaker wants to convey some information
> (semantic) or affect some attitude (interpersonal) and builds the
> discourse to do so, step by step.
>
> I am confused about what 'thematic development' might be, as a thing
> on its own.  Isn't it the case that when one follows a purpose, one
> has to develop one's argument, and therefore one necessarily has  to
> introduce themes and rhemes and move to new themes, etc.?  That is,
> isn't thematic development an automatic effect of realizing one's
> purpose?
>
> If the answer is yes, then it should be possible to show strong
> predictive correlation from discourse purpose structure to thematic
> development structure (insofar as the latter is not isomorphic to the
> former).  That is, given the former, one should by simple rules be
> able to predict the latter.  (Their differences might arise from
> syntactic or prosodic considerations, but not intentional ones.)
>
> If the answer is no, then may I ask what is the difference between
> the kind of theme/rheme development introduced as an automatic
> side-effect of following one's purpose, and this other 'real'
> thematic development?  How do the two interact in order to decide
> which clause element becomes the theme in each new sentence?  On what
> basis does the speaker make thematic development decisions?  And so
> on.
>
> (To me this latter case, in which thematic development is a truly
> independent thing, looks very problematic.  I'd love to learn more
> about it.)
>
> Please note, in either case one would be able to annotate two
> separate layers, purpose-oriented and thematic.  In the former case,
> however, the thematic would be 'derivable' from the purpose-oriented,
> and not be a truly independent notation; its structure would be a
> homeomorphism of the purpose-oriented one.

I share your intuition that the answer to the question (5 paragraphs up)
should be "yes": thematic structure is to be expected to (strongly, maybe)
correlate with a structure based on intentions/purposes. This said, there
are two aspects to be settled: (i) the handling of thematic structure within
an RST-style single-tree framework, and (ii) consequences for annotation
practice.

As for (i), the only _explicit_ way of representing topic continuity in RST
is presumably the Elaboration relation. It seems, however, that
staying-on-topic
can quite easily co-occur with a "deeper" relation such as causal one. Thus,
while

(a) Barack Obama gave his acceptance speech the other night. (b) He
spoke in front of a Roman-style column structure. (c) The crowds were
highly excited throughout the speech.

can very plausibly be represented with Elaboration(a,b) and Elaboration(a,c),
we would want to capture the causality in

(a) Barack Obama gave his acceptance speech last night. (b) He spoke in front
of a Roman-style column structure, (c) because it made him look truly
presidential.

So let's link (c) to (b) with "Explanation" or some other causal relation.
Here,
(c) also stays on topic, but we can imagine the text to give a different
reason
and thereby moving on to a different topic:

(c) because the Democrats were eager to make him look truly presidential.
(d) This party has always been obsessed with (...)

Thus when we again link (c) to (b) using Explanation, resulting in the same
configuration as above, we just don't have the information whether the topic
"continues" or "breaks". Hence my point that I'd prefer to do
topic-continuity
on a level of representation that is separate from a level where causality
and
other "deeper" semantic and pragmatic things are being captured.

(Granted, when the texts get larger und you come up with an RST tree, you can
very often guess the thematic blocks, because their breaks tend to co-occur
with discernible "branches" in the RST. I just don't see how to make this
vague
notion ("guess", "tend to") more precise. As you said in your reply, "one
should
be able to predict the former from the latter by simple rules" -- and for a
single-tree representation, I don't see these rules, due to the inherent
conflation
of topic-related analysis and rhetorics-based analysis.)

(ii) Then what does this mean for annotation practice?  My primary concern is
to end up with corpora whose annotations are as useful as possible, and I
think
to that end, conflation of information should be avoided where possible.
Hence
a preference for doing topic-based annotation and intention-based
annotation on
separate layers. Your point whether the former can be derived from the
latter is
then a very important one, and as I said, my intuition is "yes"; but from the
methodological viewpoint, I'd be reluctant to code this assumption a
priori into
the annotation framework. Instead, seeing it fall out from two independently
crafted annotations would be much more satisfactory. Or, who knows, maybe
even
discovering interesting and unexpected points of divergence between the
two...

best,
Manfred