42 s T R U C T U RA

ENs D BEYoND S C A L A RI M P L I C A T U R E SP/o L A R I T YP H E N O M E N A /A N D

rHr svNrnx/pRACMATtcS TNTERFACE 43

Supposea hearergets(6b) asan answerto question(6a); sftrewill then typically come we chooseto assertour sentence.That is, we add it to a contextc,
which will include
to concludethat the answerin (6b) implies that (6c) doesnot hold (i.e., that John and a sharedbody of information (the conversationalbackground). It is
at this point that
Bill are not both in the room) in the following (idealized) way: SIs are factored in becauseto choose a proposition from a given set
of alternatives
will, reasonably,calry along the weaker ones (i.e., the entailed
alternatives) and
(7) i. The speakersaid(6b) andnot (6c), which would havebeenalsorelevant excludethe stronger ones (i.e., the entailing alternatives)-by
somethinglike the
ii. (6c) entails(6b) t or andand arepart of a scalel Griceanreasoning sketchedabove in (7). Krifka formalizes this
by defining a notion
iii.If the speakerhad the info that (6c), she/hewould havesaidso [quantity] of scalarcontext incrementation whereby adding (8a) to a context
c amountsto add-
iv. The speaker hasno evidencethat (6c)holds ing to c the following:
v. The speakeris well informed
Therefore, (10) [earn(, $200)n -' earn(i, $n)] ($200< n)
vi. It is unlikely/notthe casethat (6c) holds
Thus,there are two parts to this process:the recursivecomputation
of meaning(truth-
Whether one goes for the stronger or the weaker version of the conclusion in (8vi) conditional content plus alternative set) and context incrementation
(where SIs are
will depend on various pragmatic factors. It is important to notice that in the view addedin). The second part necessarily follows the first in
time. I refer to Krifka,s
just sketched,SIs are computed "globally," that is, after graflrmar has done its job. (1995)article for details.
One first computes the (plain) meaning of the sentences;then, taking into account I have sketcheda neo-Griceanmodel of how implicatures
are computed that to
the relevant alternatives, one strengthensthat meaning by adding the implicature. thebestof my knowledge pretty much representsthe ievel of
our currentunderstand-
It might be useful to have a more explicit model of how SIs are computed. An ing of the phenomenon.(The existing variants of it, to the extent
they are or can be
interestingproposalin this sensemay be found in Krifka (1995). Following recent madeequally explicit, sharethe basic architectureof Krifka's
1995pioposal).To be
discussionsof the semanticsof focus, Krifka iugues that a sentenceS is generally sure,I have offered no argumentsin favor of the neo-Gricean view.
And, in fact, to
consideredagainstthe backgroundof a relevant setof alternatives,that is, other state- do so would take us too far afield. However, I believe that there
area number of things
ments that might have been made in place of S. When scalar items are involved, the that that approach explains reasonablywell.
relevant set of alternativesis constituted by propositions built up by using the other
items on the scale.Consider, for example, a sentencelike (8a). Its truth conditional (11) Whatthe (neo-)Gricean approach explains:
content is given in (8b). a. Defeasabilityof scalarimplicature
b. Sistematicityandcross-linguistic
stabilityof the phenomenon
(8) a. Johnearns$200an hour c. Lack of lexicalambiguityof scalaritems
b. earn(j, $200) (in the 'oatleast"sense) d. Metalinguistic/echoic
usesof negation(andotheroperators)

The relevant set of alternativesare the following: Let us briefly review how a neo-Griceanaccount of the properties
of SIs listed in
(11) would go. Scalarimplicaturesare defeasablefor
a number of reasons.perhaps
(9) Relevantalternatives: the most straightforward one is that something in the context may
make the stronger
{ . . . earn(j, $100),_, earn(j,$300),earn(i, $400) . . } alternativesirrelevant, thereby undermining thr canonical reasons
for assumingthat
entailment:e they do not hold. [Imagine, e.g., uttering (8a) in a situation
in which our explicit
concernis to find out who earns at least $200 an hour]. Moreover,
the neo_Gricean
The members of the relevant alternatives in (9) are presentedin their natural order, reasoningapplies in an equal manner to every item that may be
construedas belong-
going from the weakestto the strongest.That is, every item in (9) entails the items to ing to a scale of the type illustrated in (6); hence, whenever we
find a set of items in
its left. For example, if it is the casethat John earns$400 an hour, it must also be the anylanguagethat naturally forms a Horn scale,we will expect them
to display simi-
case that he earns $300 an hour, and so on. The affow beneath (9) indicates the di- lar behavior. And the reason that one does not find distinct lexical
entries for the
rection of entailment. The slashindicates where the assertionwould fit (for simplic- alternativeinterpretations of scalar items is also clear: the two interpretations
of, for
ity I am assuming only multiples of $100 to be relevant). According to Krifka, the example,some can be derived by means of a fully general mechanism.
Finally, the
recursive part of the semanticsis set up in such a way as to compute,next to the truth- neo-Griceanview mesheswell with the independeniobservationthat
negation (and,
conditionalcontentof a sentencellSll,alsoits relevantsetof alternativesllsllAlr (along possibly,other connectives)may be usedin a "echoic" or "metalinguistic'iway
(again
lines similar to thoseproposedby Rooth 1985for focus).Thus, we keep track simul- seeHorn 1989). I think it is desirableto hold on to theseresulis.
However, some
taneously of truth conditions and alternative sets,which is tantamount to saying that further empirical generalizationsrelevant to SIs will lead us to change
rather radi-
what we call "meaning" is in fact a multidimensional phenomenon.At some point, cally certain aspectsof the picture I have just sketched.To these I
now turn.