[Rstlist] RST versus issue trees?

[David Wojick]

My interest in the RST list is that I have developed a method that does 
something like RST, but is different, so I want to discuss it with the RST 
group. It is called the issue tree.
See 
https://scholarlykitchen.sspnet.org/2013/07/10/the-issue-tree-structure-of-expressed-thought/

My impression is that RST (about which I know little) is based on a 
relatively small constructed taxonomy of relations between "spans" of text. 
See http://www.sfu.ca/rst/01intro/intro.html for a listing of these relations.

Issue tree theory has no such taxonomy. It is based on the following 
fundamental observation:

With certain important exceptions, every sentence in a text (except the 
first) is answering a specific question posed to a specific prior sentence.

Thus the set of relations between sentences is the set of all possible 
questions. The tree structure occurs because more than one question can be 
asked of a given sentence and this frequently occurs. The questions are 
often quite simple, such as how?, why?, such as?, what evidence?, etc.

For example consider this string of sentences: We have to go. The cops are 
coming. Use the back door.

The second sentence is answering the question why? of the first, while the 
third sentence is answering the question how? of the first. This is a 
simple issue tree.

Note that these are reasoning relations, not rhetorical relations.

When there are many sentences, as in a journal article, the issue tree can 
be difficult to grasp just by reading the string of sentences. Here the 
issue tree diagram becomes useful. One can see the reasoning. One can also 
measure it in various useful ways.

Also the RST analysis looks to be applicable only to individual documents, 
while any set of documents on a given topic will have a unique combined 
issue tree structure. Moreover, the issue tree can be scaled to show just 
the reasoning relations between documents rather than sentences. Let's say 
we have 400 recent journal articles on a given topic, which is a fairly 
typical number. An issue tree diagram of a few thousand nodes could show 
the collective reasoning that ties this corpus together. The state of the 
reasoning, as it were. The technology is pretty powerful.

I welcome your thoughts.

David

David Wojick, Ph.D.
https://scholarlykitchen.sspnet.org/author/dwojick/