Rhetorical Structure Theory

RST ANALYSES

+ Published Analyses
+ Unpublished Analyses

The published and unpublished analyses pages contain detailed analysis of a variety of texts. Click on the links on the left for more detail on the analyses.

You can also download all the analyses in a pdf file (50 pages, 2.3 Kb), or in an exe file (Windows only for the executable), to be used with the RSTTool. After you have installed RSTTool, move the downloaded file of analyses (named rst-analyses-all.exe) to the Analyses directory (part of the installed tool) and execute it there. It will put the diagram files, which all have extension .rs2, into that directory. You can then start the tool and use the Load RST File button to examine particular analyses.

Signaled and Unsignaled Relations

People sometimes assume that whenever a text has some particular kind of discourse structure, there will be a signal indicating that structure. A typical case would be a conjunction such as but. What structure is seen depends vitally on the words and sentences of the text are, but the relationship between words and text structure is extremely complex. Phrases and syntactic patterns can also be used to signal discourse structure.

People expect discourse structure to be conveyed by signals. So the idea that discourse structure can be conveyed without signals is unexpected. Even more unexpected is the fact that for the discourse structure that RST represents, more than half is conveyed without explicit signals. To see a simple example of this, look at the not laziness text from the published analyses list on this site. It uses these signals (in text order:)

when (in unit 2, helping to signal a Circumstance) and
but (in unit 6, helping to signal an Antithesis).

The relations Volitional-Result, Background, Evidence and Concession are unsignalled.

To highlight and demonstrate this unexpected situation, the lists of published and unpublished RST analyses also show the proportions of signaled relations for each text. For the current set of text analyses, only about 30% of the relations are signalled.

On a relation-by-relation basis, it appears that every relation can be signaled in some contexts, and also that every relation can be conveyed without an explicit signal.

There are tables showing the number of signaled and unsignaled relations for each of the analyses on this website. Navigate to these two tables using published analyses or unpublished analyses.

On the RSTlist, there is a discussion of relations that are seldom or never signalled, starting on January 18, 2000. To find that discussion, click here: discussion archives.