In the rich literature on Negative Polarity Items (NPIs) the focus particle either has received little attention. This paper demonstrates that either is a fairly standard NPI, albeit one with a restricted set of licensers. In particular, it shows that either must be in the scope of its licenser, rather than have scope over it as assumed by earlier analyses (Green 1973, McCawley 1988).

Either can be licensed by many of the standard NPI triggers: negation (not), certain downward entailing determiners (no, few), adverbs (scarcely, hardly, seldom, rarely) (Klima 1964), verbs and adjectives (doubt, cease, refuse, unlikely) (Green 1973). It has even been attested in yes/no questions. However, either is not licensed in some standard NPI contexts such as conditionals and comparatives, a fact for which I offer no explanation.

Like its positive counterpart too, either is a focus sensitive particle which introduces a presupposition. For instance, while (1) presupposes that I saw someone other than John, (2) presupposes that there was someone other than John whom I didn’t see:

(1) I saw John_F too (the subscripted F indicates intonational focus)
(2) I didn’t see John_F either

Assuming Rooth’s (1992) semantics for focus, an expression of the form [X too] (where X expresses a proposition) carries the presupposition that [X]^f (the focus-semantic value of X) contains a (contextually salient) true proposition other than [X]^o (the ordinary semantic value of X).Given this, there are two competing hypotheses for either:

1. either always has scope over its licenser and carries the same presupposition as too;

2. either is always in the scope of its licenser but has a different semantics than too: [X either] carries the presupposition that [X]^f contains a false proposition other than [X]^o.

The paper spells out and defends hypothesis 2 which is supported by the following arguments:

I. If either could be licensed by having scope over a negative element, it is unclear why it wouldn’t be licensed by the prefix un- in (3):

(3) *Tom is not happy and Mary is unhappy either

II. Hypothesis 1 wrongly predicts that the attested example (4) presupposes "they refuse to let their children participate in the oral society", whereas hypothesis 2 predicts the correct presupposition "their children do not participate in the oral society":

(4) [Preceding context: Their children can’t participate in the oral society.]
They refuse to let their children participate in the deaf_F society either.

III. Hypothesis 1 wrongly predicts that (5a) presupposes "none of the male students liked the course"; as (5b) shows the right presupposition is the much weaker "someone didn’t like the course", a fact which can be explained under hypothesis 2:

(5) a. None of the female_F students liked the course
b. John didn’t like the course and none of the female_F students liked it either

IV. Hypothesis 2 explains McCawley’s (1988) observation that (6b) is not ambiguous in the way (6a) is, because if either is in the scope of the negation the because-clause must be too:

(6) a. Tom doesn’t beat his wife because he loves her (ambiguous)
b. Tom doesn’t beat his wife because he loves her, and Dick doesn’t beat his wife because he loves her, either (not ambiguous)

The analysis has the conceptual advantage that it assimilates the properties of either to those of other NPIs (although certain peculiarities remain). In particular there is a strong parallel to behavior of NPI counterparts of even (e.g. German auch nur and Dutch ook maar), which can be shown to be in the scope of their NPI licenser as well (Rooth 1985; contra Wilkinson 1996).