Received: from adler.ims.uni-stuttgart.de by moose.cs.indiana.edu (8.7.1/IUCS.1.39) id IAA13657; Sun, 31 Dec 1995 08:47:08 -0500 (EST) Received: by adler.ims.uni-stuttgart.de (8.6.10/IMS-1.0) id OAA08831; Sun, 31 Dec 1995 14:47:00 +0100 Date: Sun, 31 Dec 1995 14:47:00 +0100 From: Ede Zimmermann Message-Id: <199512311347.OAA08831@adler.ims.uni-stuttgart.de> To: ITALLC96@cs.indiana.edu Subject: ITALLC96 Submission Cc: ede@ims.uni-stuttgart.de Status: RO Dear organizers, what follows is my (rather short) submission for the London conference. It is based on my (as yet unpublished) comments on Dekker (1995). Happy New Year: Thomas Ede Zimmermann IMS Uni Stuttgart Azenbergstr. 12 D-70174 Stuttgart Remarks on the epistemic role of discourse referents (Abstract) According to so-called dynamic theories of meaning (e.g., Kamp [1981], Heim [1982]), the following two sentences differ in meaning: (1) A man is walking through the park. (2) It is not so that no man is walking through the park. The difference is usually described as one in the sentences' update potential: (1) and (2) may induce different information states in persons informed by these sentences. In particular, updating an information state s with (1) usually results in a state with more discourse referents (aka subjects) than the corresponding update of s by (2). However (see #1 in the appendix) it is not obvious in what sense someone who has been informed by (1) knows more than someone with the same background who has been informed by (2) instead: what, if any, is the informational contribution of discourse referents? Any attempt to explain meaning in terms of change of information states must given an answer to this question. In this paper I want to compare several possible answers to this question. Which of them is ultimately right will depend on a lot of factors outside the scope of the present investigation. The goal is merely to arrive at a better understanding of information states in dynamic semantics by comparing the following alternative accounts of discourse referents, some of which have been proposed in the literature: Closure account A discourse referent represents the information obtained by (statically) existentially quantifying it away. Dynamic account A discourse referent can distinguish between information states which themselves can only be individuated in terms of their update behaviour. Meta-discourse account A discourse referent represents the information that a certain kind of noun phrase (e.g. an indefinite) has been used (to introduce it). Causal chain account A discourse referent represents an individual that has played a (certain kind of) causal role in arriving at the information state in which it occurs. De origine account A discourse referent represents a source of the informational content of the information state in which it occurs. The comparison of the above accounts, which will be made in terms of the simple framework of Dekker [1995], leads to the following observations. According to the the closure account, updating by (1) and (2) may lead to (qualitatively) distinct information states containing the same information; it is therefore committed to representationalism, in the sense of a mismatch between informational content and information states. This is remarkable because certain dynamic reformulations of DRT have been given precisely to dodge representationalism; on the closure account, the don't. According to the the dynamic account (most recently defended by Spohn [1995]), some form of representationalism is needed in order to explain the dynamic aspects of epistemology. In particular, the information states induced by (1) and (2) have different update possibilities. I will show that any argument in favour of the dynamic account is likely to be circular: the dynamic account of updating involving (allegedly) essential discourse referents can be rephrased in terms of less fine- grained information states (as sketched in #2 in the appendix). According to the meta-discourse account (in its most straightforward form) the information expressed by a sentence like (1) concerns the use of certain English noun phrases, viz. indefinites, and is therefore language-specific. This makes the meta-discourse account susceptible to translation arguments. On the other hand, a language-independent reformulation would have to presuppose a cross-linguistic characterization of indefinites, which can only be given in terms of their update behaviour. According to the causal chain account, (1) induces a kind of de re knowledge about the object denoted by the subject. Although this interpretation may be correct for some discourse referents, it is not going to work in general. In particular, (1) does not have to be true in order to be believed. According to the de origine account, (1) induces information states that causally depend on events which the subject believes to be causally related to a man who fits the description given in (1). The main problem with this account is that it makes updates involving discourse anaphora complicated operations on information states. This is the price of resisting the temptations of representationalism. References Dekker, Paul (1995): On Context and Identity. In: J. Groenendijk (ed.): Ellipsis, Underspecification, Events and More in Dynamic Semantics. DYANA Report R2.2.C. Heim, Irene (1982): The semantics of definite and indefinite noun phrases. University of Massachusetts Doctoral Dissertation. Amherst. Kamp, Hans (1981): A theory of truth and semantic representation. In: J.A.G. Groenendijk et al. (eds.): Formal Methods in the Study of Language. Part 1. Amsterdam: 277-322. Spohn, Wolfgang (1995): Ueber die Gegenstaende des Glaubens. Ms., Universitaet Bielefeld. Appendix: two gedankenexperiments #1: The Dutch Twins This is the story of J and M, a pair of twins who prefer to stay anonymous. J and M are very fond of each other, and they are also very concerned about each other's knowledge: whenever one of them learns something, he immediately informs his brother, without omitting the least detail, let alone deceiving him. Since they have been together for longer than anyone can remember, they have ended up in (qualitatively) identical information states, which they certainly do not regret. Right now they are standing at the window of their common office overlooking the local green, which they are used to refer to as the park, when suddenly J makes a discovery and, as always, immediately informs M by uttering (1). At the same time M, who is in the same position as J, perceives the same disturbing scene and, consequently, immediately tries to update his brother's information state. But he prefers to achieve this goal by uttering (2). In spite of being occupied with producing a sentence, both J and M very well understand their respective brother's utterance and react with an immediate update. #2: Etypical Languages This is the story of an extraterrestrial species, the Etypes, who live in Etypeland. Etypes look a lot like human beings and they have also developed a rather human culture (cathedrals, video- games, etc.) P but the languages they speak are extremely different from and, in a way, inferior to ours: they are static. This is no accident: their genetic design prevents Etypes to ever be in information states involving subjects. So, e.g., an Etype never says anything as complex as (1), though they are quite able to say things like (2). Or rather, they could utter (1) but then what they would mean is the proposition consisting of all worlds in which there is a man walking in the park. Needless to say, the EtypesU only, extremely primitive mode of updating is by taking inter sections. Still, it turns out that communication among Etypes is pretty successful (see above) P and this in spite of the fact that they never continue a sentence like (1) by: (3) He is wearing blue suede shoes. Or when they do say such a thing, it is just short for: (4) One of the men walking through the park is wearing blue suede shoes. Finally, there are no donkeys in Etypeland.