(39a) is a question, so the indefinitea carlacks existential force. However, since (39b) is declarative, it has an assertion
arrow, which in turn leads to an implication of existential force for the pronounit. This turns out to be unacceptable.
The basic principle seems to be thatan anaphoric expression on the referential tier cannot have stronger existential force than its
antecedent. This is of course just a descriptive generalization. One would hope to base it on more fundamentalfirst
principles, but I will not do so here.
To see that this is the proper formulation of the principle, consider the other continuations in (39). In (39c) the
pronounitis acceptable,unlike(39b). Thereason is that here the pronoun also fallsunder thequestion, so it also lacks
existentialforce. Hencethepronoun canacceptablycorefer. In(39d), thepronounonehas itsown existentialforce.But
it is acceptable because it is not coreferential with its antecedent; its anaphora is on the descriptive tier.
Negation, thefuturetense, and themayofpossibility havean effecton thereferentialtiersimilar tothat ofquestioning.
In (40), there is no clai mthat an event of Joan's buying a car takes place, nor that any specific car is referred to.^212
(40) a.Joan didn't buy a car.
b. Joan will/may buy a car.The tests with referential and descriptive anaphora work exactly as they do in questions.
(41) Joan didn't buy a car...
a. *She parked it on the street.
b. and park it on the street.
c. Fred bought one.PHRASAL SEMANTICS 401
(^212) The car becomes specific ifwereplacea car bya certain car, that car, or variousotherexpressionsthatservetoground thisentityindependentlyoftheeventdescribedin the
sentence.The modalswill andmay, like questions, cause indefinite subjects as well as objects to be nonspecific:A woman will/may buy a car. With negation, however, an
indefinite subject simply sounds odd: ??woman didn't buy a car. We leave this difference aside here.