21.2.1 Rhetorical Structure Theory

Another theory of coherence relations that has received broad usage is Rhetorical Structure Theory (RST), a model of text organization that was originally proposed for the study of text generation (Mann and Thompson, 1987).

RST is based on a set of 23 rhetorical relations that can hold between spans of text within a discourse. Most relations hold between two text spans (often clauses or sentences), a nucleus and a satellite. The nucleus is the unit that is more central to the writer's purpose, and that is interpretable independently; the satellite is less central, and generally is only interpretable with respect to the nucleus.

Consider the Evidence relation, in which a satellite presents evidence for the proposition or situation expressed in the nucleus:

(21.22) Kevin must be here. His car is parked outside.

RST relations are traditionally represented graphically; the asymmetric Nucleus-Satellite relation is represented with an arrow from the satellite to the nucleus:

Kevin must be here. His car is parked outside

In the original (Mann and Thompson, 1987) formulation, an RST relation is formally defined by a set of constraints on the nucleus and satellite, having to do with the goals and beliefs of the writer (W) and reader (R), and by the effect on the reader (R). The Evidence relation, for example, is defined as follows:

Relation Name: Evidence

Constraints on N: R might not believe N to a degree satisfactory to W

Constraints on S: R believes S or will find it credible

Constraints on N+S: R's comprehending S increases R's belief of N

Effects: R's belief of N is increased

There are many different sets of rhetorical relations in RST and related theories and implementations. The RST TreeBank (Carlson et al., 2001), for example, defines 78 distinct relations, grouped into 16 classes. Here are some common RST relations, with definitions adapted from Carlson and Marcu (2001).

Elaboration: There are various kinds of elaboration relations; in each one, the satellite gives further information about the content of the nucleus:

[N The company wouldn't elaborate,] [S citing competitive reasons]

Attribution: The satellite gives the source of attribution for an instance of reported speech in the nucleus.

[S Analysts estimated,] [N that sales at U.S. stores declined in the quarter, too]

Contrast: This is a multinuclear relation, in which two or more nuclei contrast along some important dimension:

[N The priest was in a very bad temper.] [N but the lama was quite happy.]

List: In this multinuclear relation, a series of nuclei is given, without contrast or explicit comparison:

[N Billy Bones was the mate; ] [N Long John, he was quartermaster]

Background: The satellite gives context for interpreting the nucleus:

[S T is the pointer to the root of a binary tree.] [N Initialize T.]

Just as we saw for the Hobbs coherence relations, RST relations can be hierarchically organized into an entire discourse tree. Fig. 21.4 shows one from Marcu (2000a) for the text in (21.23) from the Scientific American magazine.

With its distant orbit–50 percent farther from the sun than Earth–and slim atmospheric blanket, Mars experiences frigid weather conditions. Surface temperatures typically average about -60 degrees Celsius (-76 degrees Fahrenheit) at the equator and can dip to -123 degrees C near the poles. Only the midday sun at tropical latitudes is warm enough to thaw ice on occasion, but any liquid water formed in this way would evaporate almost instantly because of the low atmospheric pressure.

See the end of the chapter for pointers to other theories of coherence relations and related corpora, and Ch. 23 for the application of RST and similar coherence relations to summarization.