ABSTRACTS

Presuppositions and Conventional Implicatures
Barbara Abbott
Michigan State University

    This paper investigates the differences between presuppositions (entailments of unembedded, unmodalized, positive, declarative sentences which are maintained throughout the "S family" (Chierchia & McConnell-Ginet 1990)) and conventional implicatures (which are not so entailed (Grice 1975)), especially with a view toward their suspension and/or cancellation behavior.

     Anyone who spent any time in the years 1955-1975 on the third floor of Princeton’s Firestone Library, where the Philosophy collection is housed, will be familiar with a figure that often lurked there, dressed in an overcoat, scarf, and hat, summer or winter.  Every book by or about Leibnitz has his autograph on its circulation card.  He was a Leibnitz buff.  No doubt about it.  He was also a great philosopher and logician. 

     This is a talk about his philosophical views.  It is purely expository – I will not try to assess them today.  The interest they hold for me stems somewhat from a sentimental attachment to the gigantic intellect that lurked inside that overcoat, under that hat.  But I am taken with them mostly, as I will argue, because they are much more complex and thoroughgoing than they are normally thought to be.  And in their complexity they illustrate the marvelous complexity of the philosophy of mathematics, as well as its connection to the philosophy of mind.

     Relevance Theory (RT) distinguishes two varieties of semantics: ‘linguistic’ vs. ‘real’. But ‘linguistics semantics’ seems both inscrutable and (in Cappelen & Lepore’s 2005 sense) unstable.  By ‘linguistic semantics’, I suggest, RT attributes either (A) not enough in the way of semantics to the expressions of natural languages or (B) too much.

     Given the inscrutability of ‘linguistic semantics’, I suggest we admit just one variety of semantics, namely real (truth-theoretic) semantics. On the standard assumption that the expressions of a language (e.g. English) do have a semantics, it is reasonable to resolve the instability by pursuing option (A) above, arguing that the expressions of English are possessed of real semantic properties. 

     However, the present paper seeks to resolve the instability by pursuing option (B), questioning the assumption that the expressions of English (for example) have a semantics in the first place. It is thoughts alone that have (any kind of) semantics. Defending this position, I note an ambiguity surrounding RT’s concept of ‘encoding’, especially apparent in Carston 2002, Ch 5.

‘Encoding’―sense II. An expression ‘encodes’ f only in virtue of pointing to (indicating, acting as a sign of, or representing) f. It does not follow that f is a property of E. Quite the contrary.  f is emphatically a (conceptual-intentional) property of―and only of―a constituent of thought that E merely (and conventionally) points to. Sense II distinguishes radically between the English word not and the logical functor ‘¬’.      
I call this ‘M-encoding’ (‘M’ for ‘Magritte’―to be explained).

I pursue option (B) by adopting sense II of ‘encoding’ (i.e. M-encoding, in line with the  ‘Representational Hypothesis’?\e.g. Burton-Roberts 2000).

     I show how this allows us to implement Jay Atlas’ (1989) proposal that there is no such thing as ‘scope of negation’ in the linguistic encoding. I also discuss its implications for numerical expressions and for ‘narrowing’ and ‘loosening’ of ‘lexically encoded concepts’ (Carston 2002).

    One of my long-term intellectual debts to Atlas arises from his role as my sparring partner on the semantics of only.  Since the 13th century, the symmetricalists (from Peter of Spain to Atlas 1991, 1993, 1996) have essentially analyzed (1) into the conjunction of (2) and (3), while the asymmetricalists (e.g. Geach 1980, McCawley 1981, Horn 1969, 1992, 1996) have argued that (1) entails (2) but at most presupposes or implicates (3).

(1)       Only God can make a tree.

(2)       Nothing distinct from God can make a tree.

(3)       God can make a tree.

     While the former view is persuasive in dealing with the intuition that (1) is false if no one can make a tree, it cannot explain why only NP licenses negative polarity items (Only God could ever have made that tree) and exhibits other grammatical behavior diagnostic for negativity.  The rival view, treating only NP as a downward-entailing operator, correctly predicts this negative behavior but misses the fact that, as Atlas has consistently stressed, (1) does entail (3) as well as (2). 

     This standoff motivated the construct of assertoric inertia (Horn 1996, 2002), which exploits the distinction between what is entailed and what is asserted.  Semantically entailed material that is outside the scope of the asserted, and hence potentially controversial, part of utterance meaning (Stalnaker 1978) is assertorically inert.  Such material will be effectively transparent to polarity licensing, as it is downward assertion and not downward entailment that licenses NPIs.  On this view, sentences with only+NP subjects are semantically symmetric (with both negative and positive conjuncts in their logical form) but function as if they were asymmetric. 

     Kai von Fintel (1999) has recently proposed a new model to account for the behavior of only, but his analysis incorporates the notion of semantic presupposition (Strawson 1952), reconstructed through the notions of Strawson-entailment and Strawson-downward entailment within dynamic semantics. Over the last two decades, however, Atlas has led a cogently argued opposition to the notion of semantic presupposition, and his objections hold sway against von Fintel’s position.  Futhermore, as I shall argue, the behavior of only is paralleled by that of barely and other particles (Atlas 1997, Horn 2002), where the Strawson-entailment approach is less viable.

     Study of dialogue has been proposed by Pickering and Garrod (2004) as the major new challenge facing both linguistic and psycho-linguistic theory.  Two of the phenomena which they highlight as common in dialogue, but posing a significant challenge to theoretical linguists, are alignment between conversational participants, and shared utterances, with participants   mirroring each other's choices at many levels (lexical, syntactic and semantic).  Shared utterances as in (1) are those in which participants exchange parser and producer roles mid-sentence:

(1)       A:  Where have you got to

           B:   in reading your book?  Chapter 3.

     This talk will introduce Dynamic Syntax (DS: Kempson et al 2001, Cann et al forthcoming), a grammar formalism which directly reflects the way interpretation of a sentences is built up by hearers in real time.  I will show how: (i) the  framework  allows a psycholinguistically plausible model of incremental context-dependent generation, (ii) the resulting tightly coordinated systems of generation and parsing directly reflect dialogue alignment patterns and shared-utterance phenomena. These accounts depend on the progressive building up of partial representations of content throughout the interpretation process, and I use these to argue that though the account is broadly relevance-theoretic,  there is no intervening level of logical form representing sentence meanings.

     What is central to DS are the concepts of under-specification and growth of semantic representation reflecting the time-linearity of parsing.   The system is essentially procedural.  Parsing is defined in terms of actions updating partial semantic tree structures, driven by the goal of establishing some proposition conveyed by the speaker, using procedures dictated by the serial order of words in a string implemented in conjunction with context-dependent choices.  Generation reflecting word-by-word incrementality is defined to use the very same actions as the parser. With both parsing and generation defined in terms of actions updating partial trees, the minimal context is defined in the same terms, as made up of trees  that have been constructed prior to the current point of the parse process, together with the associated actions used in creating such trees.  Alignment results directly from the use of such context as a generation strategy for minimizing general-lexicon search. Switch of speaker-hearer roles in shared utterances can be straightforwardly modelled in virtue of both speaker and hearer constructing tree-structures representing content and then adopting them as the minimal context for the next update.

References
Cann, R., Kempson, R., & Marten, L. (fcmg).The Dynamics of Language.Elsevier.
Kempson, R., Meyer-Viol, W. and Gabbay, D. (2001).Dynamic Syntax. Blackwell.
Pickering, M  & Garrod. (2004) Towards a mechanistic psychology of dialogue BBS 27.
Sperber, D. & Wilson, D. (1995) Relevance: Communication and Cognition. Blackwell.

     In the golden age of radical pragmatics (GARP) various scholars, including our honoree, attempted to remove as much language description as possible from the arbitrary realm of grammar and lexicon to the principled kingdom of pragmatics. This began with Horn’s proposal that the cardinal numbers, 1,2, 3, etc. were not semantically precise but were upper bounded by pragmatic implicature. Thus  three would not mean “no less and no more than 3”, but rather “three at least.” The upper bound would then be supplied by Gricean reasoning from the premise that one should say as much as is needed for the purposes of the talk exchange. The numerical scale, however, is situationally reversible, as Horn himself pointed out. If I told you that my golf score was 90, you are entitled to conclude that it was 90 at most, not 90 at least. In this case the cardinal number seems to mean “at most ninety”, the lower bound being supplied by pragmatic reasoning.

     If sometimes the cardinal numbers are upper bounded by implicature and sometimes lower bounded, we are lead to conclude that they have no semantic content at all, everything about them being deducible from pragmatic principles. As absurd as such a theory seems at first, I will travel back in time to GARP and suggest that it is possibly correct and that there is a way of constructing a coherent system of mathematics based on the use of the cardinals, rather than their meaning.

     According to the traditional picture, semantics tells us what assertions sentences are used to make. Although it remains controversial what sorts of things are asserted, it is common to refer to them as propositions, and to ascribe three important properties to them:  being asserted, bearing of truth conditions, and being the semantic contents of sentences.  Thus, when a sentence S is uttered, the proposition asserted is standardly taken to be the semantic content of S.  After the assertion has been semantically determined, pragmatics kicks in, as conversational maxims and other pragmatic principles generate further implicatures, presuppositions, or suggestions carried by the utterance.  So much for the traditional picture.

     I will argue that this picture needs to be revised to allow for systematic pragmatic enrichment to play a major role in determining the propositions asserted by utterances.  On the new view, the semantic content of S is sometimes, but not always, a complete proposition – i.e. a proper candidate for being asserted and believed.  In other cases it is only a skeleton, or partial specification, of such a proposition.  Typically, the semantic content of S interacts with pragmatic factors to generate the proposition(s) that the speaker primarily intends to assert.  Other propositions count as asserted only when they are relevant, unmistakable, necessary and apriori consequences of the speaker’s primary assertion(s) (together with relevant background).  On this view, the meaning M of a nonindexical sentence S is what is common to that which is asserted by normal, literal utterances of S -- in the sense that such utterances result in assertions of propositions that are proper pragmatic enrichments of M.  When M is a complete proposition, it counts as asserted only if M is an obvious and relevant consequence of the enriched proposition(s) asserted in uttering S.  On this view, the semantic content of the sentence uttered is often not itself asserted by the speaker’s utterance – even when one is speaking literally and unmetaphorically.

     The advantages of this view will be illustrated by discussions of possessives, linguistically simple names, partially descriptive names, and incomplete definite descriptions.  The view will also be used to throw light on referential uses of names and descriptions.