Cumulative readings of /every, /as in (1a) below, do not arise in every
configuration (e.g. 1b).
(1)
a. The three copy-editors caught every mistake
b. Every copy-editor caught the twenty mistakes
The nature of this restriction is debated. Distribution aside, the
cumulative reading is unexpected given a standard denotation for every.
I make two empirical contributions: first, looking at ditransitives, I
provide an additional argument that the restrictions on this reading can
be explained in terms of c-command of thematic roles (following
Champollion 2010, pace Kratzer 2001). Second, we show two lines of
evidence that underlyingly, the reading of the sentence (1a) does not
require all copy-editors to have caught a mistake (Ivlieva, 2013), i.e.
the "leakage" reading in Bayer (2013). My analysis explains these facts
while maintaining a standard entry for /every/. The leakage reading
obtains underlyingly and the inference that all copy-editors caught a
mistake arise through a standard strengthening process.
