The Interpretation of Nominal Compounds in Discourse

In earlier work, we described a system that built semantic interpretations of nominal compounds - sequences of two or more nouns (or nominal adjectives) that function as a unit. Our initial system took the representations of the concepts making up the compound and synthesized a number of concepts that represented possible interpretations. Each candidate interpretation was then assigned an "appropriateness" measure based on a variety of factors. One source of knowledge that was not used was the discourse context. Nominal compounding is often used as a naming device to refer to a concept already introduced, either directly or indirectly, into the discourse. Our current work in this area centers around three questions: (1) How can we effectively use discourse context as a primary source of evidence for both hypothesizing candidate interpretations (roughly in order of plausibility) and for choosing a preferred interpretation from among them? (2) How much analysis is really required to do an adequate job of understanding nominal compounds used in discourse? We believe that many compound strings may require only a relatively shallow analysis. (3) When is it appropriate to use nominal compounds? We would like to formulate the conditions when a particular nominal compound is likely to be accurately and efficiently interpreted by a hearer. (An abbreviated version of this paper appears in Penn Linguistics Review, 1982)


University of Pennsylvania

Computer and Information Science


UMBC ebiquity