Topics in presupposition theory

Abstract

Presupposition theory within the dynamic semantics paradigm has been characterized by three developments: the systematization of the projection facts and the conceptual explanation in terms of context change, the development of representational means and algorithms in discourse representation theory, and the emphasis of the semantic foundations within dynamic semantics. The first development is constituted by Karttunen (1973), Karttunen (1974) and Stalnaker (1974) and Heim (1983b). Kamp and Roßdeutscher (1994b), Kamp and Roßdeutscher (1994a) have created DRT representations and coined the term presupposition justification, while van der Sandt (1992) has developed an algorithm for treating presuppositions. An extensive exposition of this second development is in Geurts (1999). The third development is the application of techniques from dynamic semantics to presupposition (Beaver 1995b), (Beaver 1997) and van Rooy (1997) and builds on dynamic predicate logic (Groenendijk and Stokhof 1991, Dekker 1993). The first development has led to detailed descriptive observations in Karttunen (1974) and to great conceptual clarity in Stalnaker (1974), but was not yet embedded in a semantic formalism which directly expresses the underlying ideas, which would be very important for the purposes of computational linguistics and for the integration with other areas of semantic theory. The second development has led to operational models of presupposition justification and to good empirical coverage. The model-theoretic interpretation of presuppositions seems to have been slightly neglected in this approach. The question whether the representational operations employed in the algorithms have a semantic or logical justification was not always pursued consequently. The third development can be understood as a reaction against this tendency. Because of the renewed emphasis on the model-theoretic interpretation of presuppositions, it has resulted in detailed expositions and refinements of the so-called satisfaction model of presupposition, in particular to an eliminative semantics for the presupposition operator within a generalized dynamic semantics using epistemic alternatives (Beaver 1995b). This revival of the semantic basis of presupposition theory has also led to new two-dimensional analyses of presuppositions (van Rooy 1997), and to an interest in the question how dynamic semantics analyses of presuppositions and analyses based on partial logic relate to each other (Krahmer 1998)1. With respect to empirical coverage and algorithmic realization, these new satisfaction models do not significantly improve on the discourse representation theory treatment. In addition to these developments within presupposition theory, the work on semantic formalisms in dynamic semantics has led to dynamic predicate logic (Groenendijk and Stokhof 1991) and a version of it which is - through the use of partial variable assignments - closer to the framework in which Heim (1983b) formulated her analysis of presuppositions and to original DRT (Kamp 1981). But this work on semantic formalisms concentrated on first-order fragments without a presupposition operator. The result were mathematically precise reconstructions of the semantic machinery underlying the dynamics of interpretation. The other main ingredient in the analysis of Heim (1983b), namely the semi-pragmatic accommodation mechanism, was not reconstructed in these formal models. In this situation, it was desirable to develop a theory of presupposition which integrates semantic foundations with an algorithmic treatment. A semantically well-founded algorithm is usually one which is based on logical inference. Attempts to analyze presuppositions as semantic entailment in first-order logic have met with difficulties, however. So another way of understanding presuppositional reasoning as based on logical inference was needed. Such an idea came up in the artificial intelligence literature, at least for the treatment of certain expressions that linguists classify as presuppositonal, in particular definite descriptions (Hobbs, Stickel, Appelt and Martin 1993). Hobbs proposed to use abductive reasoning quite generally to model text comprehension. The presuppositions of definite descriptions must be proved abductively from the context. The logical language was however not equipped with a dynamic semantics. David Beaver has argued for the relevance of this approach for semantics. Beaver emphasized the need for a possibility to let presuppositions interact with commonsense reasoning by giving examples in which two sentences with the same syntactic form, but different lexical entries can give rise to different perceived presuppositions because of the different plausibilities of the possible presuppositions.

Diese Arbeit betrifft die Präsuppositionstheorie und ihre computerlinguistische Umsetzung. Ein semantischer Formalismus aus dem Bereich der dynamischen Semantik wird so um Konstrukte zur Repräsentation von Präsuppositionen erweitert, dass die präsuppositionale Inferenz als abduktive Inferenz dargestellt werden kann. Die theoretischen Vorschläge sind implementierungsnah formuliert, um eine unmittelbare computerlinguistische Umsetzung zu ermöglichen. Die Arbeit hat zwei Hauptthemen aus dem Bereich der Präsuppositionstheorie. Das erste Thema ist die Anwendung der abduktiven Inferenz auf das Problem der Rechtfertigung von Präsuppositionen, das zweite Thema sind die Einzigkeitsbedingungen von definiten Kennzeichnungen. Das erste Thema wird in den Kapiteln 1, 2 und 3 behandelt, das zweite Thema in den Kapiteln 4 und 5. Diese Reihenfolge ergibt sich aus der Natur der Sache, da die vorgestellte Theorie zu Einzigkeitsbedingungen auf eine Theorie der Präsuppositionsrechtfertigung Bezug nehmen muß. Die Grundidee dieser Theorie könnte wohl auch mit Hilfe anderer Präsuppositionstheorien ausgedrückt werden, im Rahmen dieser Arbeit war es jedoch natürlich, die in den Kapiteln 1 bis 3 vorgestellte abduktionsbasierte Analyse zu verwenden, weil dort wichtige Grundbegriffe explizit definiert werden, die in anderen Analysen oft nicht bereitgestellt werden.