1Gennaro ChierchiaU. of Milan - BicoccaMay 2001Scalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interface*Revised version (May 2002).Forthcoming in A. Belletti (ed.) Structures and Beyond, Oxford University Press2Gennaro ChierchiaU. of Milan - BicoccaScalar Implicatures, Polarity Phenomena, and the Syntax/Pragmatics Interface*1. Introduction.Negative Polarity phenomena, as exemplified in the behavior of English any, and scalarimplicatures, as exemplified by, say, the interpretations of some (i.e. “al least one and possiblyall” vs. “at least one but not all”) have often been felt to be closely related. 1 However, the exactnature of such relationship remains as of now not fully understood, in so far as I can tell. And infact, some important empirical generalizations pertaining to it, if not altogether missed, haveperhaps not been properly appreciated. In the present paper, I address the issue of what are therelevant factual connections between scalar implicature and negative polarity and what we canlearn from this concerning the grammatical mechanisms at the basis of these phenomena. One ofthe features that makes the analysis of negative polarity items (NPIs) and scalar implicatures(SIs) particularly interesting is that they raise a number of key questions concerning how syntax,semantics and pragmatics interact with each other. We will mostly focus on the interface ofpragmatics with syntax and semantics. More specifically, here is a widespread view of the latter.Grammar (which includes syntax and semantics) is a computational system that delivers, say,pairs of phonetic representations and interpreted logical forms. The output of the computationalsystem is passed onto the conceptual/pragmatic system that employs it for concretecommunication. The computational system of grammar and the conceptual/pragmatic system areseparate units and work in a modular way: each unit is blind to the inner workings of the other.Things like agreement or c-command belong to grammar, things like relevance or conversationalmaxims belong to the conceptual/pragmatic system. This view is very plausible, and has beenquite successful in explaining things. Yet, I would like to make a case that, in certain importantrespects, it is actually wrong.Let me try to give, in broad outline, the structure of the main claims to be developed. Aninfluential account of scalar implicatures stems from Grice (1989) and work more or less directlyinspired by it (e. g. Horn (1972, 1989), Atlas and Levinson (1981)). I will take the (neo)Griceanview as our starting point. I will then try to establish a factual generalization relating scalarimplicatures to polarity phenomena. The generalization, in rough terms, is the following:ordinary scalar implicatures are systematically suspended in the very contexts that licenseelements like any. This seems to entail that the two phenomena in question must be based on adevice governed by uniform principles. At the same time, scalar implicature computation andNPI licensing are are influenced by structural (i.e. locality) considerations in very different ways.So much so that it is hard not to draw the conclusion that they are driven by very differentmechanisms, after all. This is, then, the problem: SIs and NPIs are so similar in certain respectsand so different in others. Why? How come? After discussing the limits within which currentapproaches manage to provide an account of this puzzle, we will make a specific proposal onhow SIs are computed, and on how NPIs are licensed. We will try to make a case that theinteraction of these two proposals actually takes us some way to a better understanding of the *The first version of this work was presented in the Fall of 1999 at the Center for CognitiveScience in Lyon and at the Pontignano workshop (which concluded Chomsky’s visit to Italy).Subsequent versions were presented at NELS, UC Irvine, UCLA, U. of Tuebingen (Fall 2000); atthe DGfS meeting in Leipzig and at MIT (Spring 2001). All those audiences contributedsignificantly to give shape to the original ideas. I also would like to thank I. Caponigro, C.Cecchetto, S. Crain, J. Gajewski, A. Gualmini, T. Guasti, E. Guerzoni, L.Meroni, F. Panzeri, O.Percus and U. Sauerland.1 See, e.g. Fauconnier (1975) Horn (1989, pp. 230 ff.), Krifka (1995), Landman (1998), Israel(1998) among many others.3relevant phenomena. As we will see, our proposal has interesting (though not uncontroversial)consequences for the overall architecture of grammar at the above mentioned interfaces. Inparticular, we will argue that pragmatic computations and grammar driven ones are“interspersed”. Implicatures are not computed after truth-conditions of (root) sentences have beenfigured out; they are computed phrase by phrase in tandem with truth-conditions (or whatevercompositional semantics computes).The structure of the paper is the following. In section 2 we discuss the main empiricalproperties of SIs. Then in section 3, we will put forth a theory of SIs that hopefully sheds somelight on such properties. We will argue that, contrary to the dominant view, SIs are introducedlocally and projected upwards in a way that mirrors the standard semantic recursion. In section 4we turn to NPIs. In this area, recent proposals have been developed, which make us understandwhy NPIs are licensed in negative context (Kadmond and Landman 1993, Krifka 1995, Lahiri1998). All such proposals are based on the idea that the semantics NPIs involves a comparisonamong relevant alternatives (much like SI implicature computation involve a comparison withscalar alternatives). I will propose a modification of such views in order to overcome somedifficulties they run into. The two proposals (on SIs and NPIs) will have an unexpected kind ofinteraction having to do with the so called intervention effect which has been observed inconnection with NPIs. In the rest of this introduction, in order to set our baseline, I summarize (a version of) theNeoGricean stand on implicatures. The phenomenon is well known. The truth conditional contentof a sentence like (1a) is taken to be (1b). Yet, such a sentence typically conversationallyimplicates (1c). Similarly for the sentences in (2)(1) a. John is singing or screamingb. singing(j) ⁄ screaming(j)c. ¬ (singing(j)