t M p L r c A T U R Epso,l A R t r y P H E N o M E N AA,N D T H Es y N T A x / r n n c u n r t c sT N T E R F A C E 7 5

SCALAR
74 S T R U C T U R EASN D B E Y O N D

constitutesa seriousproblem for approachesthat try to reduceNPIs and SIs to iden- not disturbedby the interventionof the embeddedevery. Example(111) forms a
tical phenomena.zz minimal pair with (109c and d). So there are no interventioneffectsbetween scalar
Th. ,..ond problemis relatedto the first but adds,possibly,an extra twist. Since terms and what determines the presenceor absenceof the scalar implicature. Nor
would we expect it in the theory I have outlined. Implicature calculation is a purely
the work of Linebarger(1981), it has beenknown that NPI licensingis subjectto an
,.interventiono'or"minimality" effect. The problem is illustrated by paradigms of the semanticprocessthat involves the comparison of two potential values.For example,
following sort: the upper every in (11lb) comparesthe two available values of its argument and
choosesthe strongest,period.
(109) a. I doubtthatSuehaspotatoes. Summing up, NPIs, unlike SIs, display both roofing and minimality effects (terms
b. I doubtthat Suehasany potatoes. that I am using here and throughout in a purely descriptive manner).Now, it is con-
c. I doubtthat everyhousemateof Suehaspotatoes' ceivable,even a priori desirable,that phenomenaof this sort, so similar to others
d. ??I doubtthat everyhousemateof Suehasany potatoes' that arise elsewherein grammar, are to be accountedfor in terms of a uniforrn theory
of locality (but cf. section4.3). Be that as it may, we seethat the optimistic picture
In (109a),we have a plain indefinite. In (109b), we have the samesentencewith an we had at the beginning of this section must be substantially qualified. We hoped we
NpI replacing the plain indefinite (which, in the hypothesis we are entertaining, in- had a unified way to understandNPIs and SIs in terms of relative inforrnational
ducesdomain widening). In (109c), a universalquantifier intervenesbetweennega- strength.An NPI broadensthe quantificational domain of a plain indefinite; but this
tion and the (plain) indefinite, and the result is perfectly grammatical. In (109d) we is informative (and hence felicitous) only in negative contexts.A SIs strengthensan
do the samewith the NPI, and the sentencebecomesnoticeablydegradeil.Every,tn assertion(taking into account the relevant scale); but this is informative (and hence
that position, seemsto interfere in the relationtionship between the NPI and the li- felicitous)only in positive contexts.This makesus seewhy NPIs and SIs are sensi-
..nring context. This is surprising. Effects of this sort are quite familiar from syn- tive to the samecontexts.But we also seethat the locality conditions in the phenom-
tax. Often enough we see syntactic relations that are affected by interveners. Often enaat hand are very different from each other, and it is hard to understandwhy. In
particular,if the key to NPIs is a pragmatic condition and if suchconditions kick in
enough a relation between X and Y is disturbed by an interveningZ, which might in
its own right establish the relevant relation with Y, as schematically illustrated in theconceptuaVintentionalsystemafter thecomputationalsystemof grammarhas done
(110a) (taken from Rizzi 2001): its job, locality effects are unexpectedto begin with. If, instead,pragmaticswork as
in our proposal, the locality effects would be different from what they in fact are.
(110)a. ...X ...2.'.Y.'. This is the dilemma.
b. I wonderwho couldsolvetheproblemin this way'
c. *How do you wonderwho couldsolvethe problemt ? 4.2. Theory outline

For example,wh-island effects [suchas the contrastbetween(110b) and (110c)] can The main insight of recent semanticproposalson NPIs is that they make explicit the
be explained in this way. The relation between how and its trace in (110c) is dis- intuition that such items involve a comparisonof some sort. In K&L, NPIs involve a
turbed by the intervention of anotherwh-element. The effect in (109d) appearsto be comparisonof two domains;in Krifka's (1995) proposal,a comparisonamonga whole
of a similar nature.But if the relation in questionis semanticin nature (i.e., check scaleof domains; in Lahiri's (1998) approach,a comparisonof likelihood (analo-
gousto that involved in the analysis of focal particles hke even).I will stick to this
whether domain widening induced by A leads to strengtheningunder C), there is no
generalview and propose a variant of K&L's approach, which blends with aspects
obvious reason why we should expect effects of this sort. This is particularly so in
view of the fact that though everyis a potentialNPI licensor,any in (109c) occurson of the other two. The result, coupled with my approachto Sls, should lead to a better
its wrong side, so to speak,that is, on the side in which every is upward entailing and understandingof the relevant phenomena.
could not licensean NPI anYhow. Here is the basic idea. NPIs are parasitic upon indefinites. They differ from
A point worth underscoring in this connection is that SIs aren't subject to any- the meaning of basic indefinites (say, the meaning of some)in that they invite us
thing like intervention, as seen in the following example: to consider possible domain expansions-not necessarily a specific one: any
reasonabledomain expansion will do. Let me elaborate on this idea to make it
(111) a. Every girl invited every boy three times. (exactly) clearer.The meaning of NPIs comes, as it were, in two parts. First, as in the ver-
b. Every boy whom every girl has invited three times pleasestand up. (at least) sion of K&L's idea discussedin the previous section, they introduce into the
pl domain variable associatedwith indefinites a variable ranging over increasing
functions:
Sentence(11la) clearly has a highly prominent "exactly" reading.Sentence(11lb)
(lI2) 4ry'= l,PxQsomeg@;'(PxQ) (where,for any D, g(D) r (D)
does not. The upper every removes the embeddedimplicature, and in doing so it is