Technical reports

Computing presuppositions in an incremantal language processing system

Derek G. Bridge

212 pages

This technical report is based on a dissertation submitted April 1991 by the author for the degree of Doctor of Philosophy to the University of Cambridge, Wolfson College.

DOI: 10.48456/tr-237

Abstract

This thesis describes the design and implementation of a natural language analysis system for the computation of presuppositions. The system is one in which syntactic, semantic and pragmatic processing are interleaved with feedback to syntactic analysis from semantic and pragmatic processing. The thesis begins by illustrating how the system processes definite noun phrases. The mechanisms used for this are then shown to be easily extensible to processing other parts of speech such as indefinite noun phrases and verb phrases.

Definite noun phrases have been said to be presupposition triggers. This means that traditionally they have been seen as licensing certain inferences — presuppositions. In the system described herein, presuppositions are treated as a special kind of inference: preconditions. This treatment for definite noun phrases can be extended to give a uniform account of all presupposition triggers (e.g. factive verbs). It is a view that makes it clear that presuppositions are not ‘optional extras’ that might or might not be derived once a semantic representation of an utterance has been produced. Rather, they play an essential role in driving the utterance analysis process: the failure of a presupposition, i.e. failure to satisfy a precondition, can direct the system to choose an alternative reading of an utterance of an ambiguous sentence.

As it processes an utterance, the system builds and regularly consults a representation of contextual knowledge referred to as a discourse model. Importantly, the system checks whether presuppositions are satisfied against the discourse model. Presupposition failure, i.e. a presupposition not being satisfied by the discourse model, is not necessarily the same as a presupposition being false in, e.g., the ‘real’ world. Checking presuppositions for satisfaction in a discourse model and not for truth in a possible world offers new ideas on the behaviour of presuppositions in utterances of negative and complex sentences.

In utterances of negative sentences, presuppositions must still be satisfied by the discourse model. Presuppositions cannot be cancelled as they can in other accounts. Rather, presupposition “cancellation” data is explained in terms of utterances that make metalinguistic statements about the model-theoretic interpretation of the discourse model. It is shown that computing presuppositions in an incremental system gives a simple account of most of the data relating to the behaviour of presuppositions in utterancesof compound sentences and longer stretches of text (the so-called “projection problem”). Presuppositions must again be satisfied by the discourse model, but they may be satisfied by virtue of changes made to the discourse model by earlier parts of the utterance or text.

Full text

PDF (25.0 MB)

BibTeX record

@TechReport{UCAM-CL-TR-237,
  author =	 {Bridge, Derek G.},
  title = 	 {{Computing presuppositions in an incremantal language
         	   processing system}},
  url = 	 {https://www.cl.cam.ac.uk/techreports/UCAM-CL-TR-237.pdf},
  institution =  {University of Cambridge, Computer Laboratory},
  doi = 	 {10.48456/tr-237},
  number = 	 {UCAM-CL-TR-237}
}