Going beyond sentence boundaries, this paper explores recent research about “processing discourse--treating texts or dialogs as whole units composed of interrelated parts, not merely sequences of isolated sentences.…the comprehension and production of language” (p. 102).
This is one of the best survey papers I have read in some time. While I occasionally wished for more depth, the available space has been allocated properly. All the relevant subtopics are treated; none is shortchanged. The writing is clear, concise, and consistent in style. The examples are appropriate and understandable. My only complaint is that the sentences composing an example should be numbered in the same manner as their in-text references (1.1, 1.2, and so on, not 1, 2, and so on).
The references will lead the researcher to all the depth necessary. Also, the majority of the other papers included in this special section of this issue of CACM are deep rather than broad, dealing with building real rather than toy systems.
Often, papers such as this are only understood by the specialist. That does not apply here. A single linguistics course, or a willingness to use the dictionary, is sufficient background to read and appreciate the material. The automated systems that such research may lead to are of interest (or at least concern) to all: automated medical history takers; conversing automatic teller machines; real, functioning automated language translators; and so forth.
I recommend this paper to the researcher and to the merely curious.