The Semantics of Determiners

care a citit cartea scrisă de Lessing.
who has read book.Def written by L.
‘Every professor talked to every student who read the book by Lessing.’

This is unsurprising given that definites have to have determined reference. If the interpretation of the definite has to be fixed in context, it cannot co-vary. But as expected, if the context is such that determined reference is assured relative to every entity in the Restrictor of the universal, co-variation (and therefore narrow scope) is possible:

(46) Fiecare student a citit o carte scrisă de Lessing şi o poezie scrisă de Eminescu.

every student has read a book written by L. and a poem written by E.
‘Every student read a book written by Lessing and a poem written by Eminescu.’
Fiecare student a vorbit despre cartea scrisă de Lessing.
every student has spoken about book.Def wriiten by L.
‘Every student spoke about the book written by Lessing.’

In the context of the first sentence in (46) interpreted under a ‘direct’ scope reading, the second sentence may well be interpreted as saying that every student spoke about that book written by Lessing that they read. We have seen similar examples with demonstratives. We therefore conclude that although definite and demonstrative descriptions are naturally reluctant to being given co-varying interpretations precisely because of their determined reference requirement, such interpretations are not in fact ruled out, once the context satisfies the determined reference requirement for each input function relative to which the definite is interpreted. Unmarked indefinite and definites then are alike in their special inverse scope abilities.

How to account for this freedom of scope is a matter that is still hotly debated. Reviewing and discussing all the major proposals would take us much too long. For some recent and not so recent proposals see Fodor and Sag (1982), Abusch (1993), Reinhart (1997), Winter (1997), Farkas (1997), Kratzer (1998), Chierchia (2001), Schwarzschild (2002), Farkas and Brasoveanu (2007).

Another property unmarked indefinites are well-known for, and which is often elevated to the rank of an ‘indefinite’ test is their ‘quantificational variability’, i.e., their ability to have their quantificational force influenced by their linguistic environment. An indefinite in the antecedent of a conditional, for instance, interpreted as having narrow scope relative to the conditional, appears to acquire a universal force:

(47) Dacă o persoană e în Atena, (ea) nu poate fi la Rhodos.

if a person is in A. (she) not can be at R.
‘If a person is in Athens, she cannot be in Rhodes.’

The indefinite is said to have narrow scope relative to the antecedent under the quite natural interpretation in which (47) is understood as claiming that under all cases in which some person or other is in Athens, that person cannot be in Rhodes. Here then the person co-varies with the cases and in fact, there are as many cases as persons.

This example serves to illustrate a second property of indefinites, namely the ability to serve as antecedents to so-called ‘donkey’ pronouns.²⁰ The indefinite a person serves as antecedent to the ‘donkey’ pronoun she. The anaphoric relation is problematic (and the problem has been recognized since the Middle Ages) because for such a link to be possible, the indefinite should c-command the pronoun, but in this example it does not.²¹ My contention here is that this property does not differentiate unmarked indefinites from definites, as is often assumed in the literature. As long as the context is such that the presupposition of definites is met, a narrow scope definite is possible in the antecedent of a conditional sentence and such an indefinite may well serve as an antecedent to a pronoun in the consequent:

(48) În familiile de azi din China, dacă copilul e băiat (el) e foarte alintat.

in family.Pl.Def of today in C. if child.Def is boy (he) is very spoiled
‘In today’s families in China, if the kid is a boy he is very spoiled.’

In order to have a definite within the scope of the conditional, we have to make sure that the domain of the conditional (which acts like a universal quantifier over ‘cases’ or worlds) contains only cases made up of families that have a unique child. This is what the restrictive adverbial phrase accomplishes. We can treat the definite the kid here as accommodated (via bridging to the introduced families) in each of the cases within the domain of quantification. We need such a special context because otherwise the presupposition of the definite article would not be met. But once we have it, there is no difference between definite and indefinite DPs with respect to donkey pronouns.

Moreover, note that not all types of indefinites may have narrow scope relative to a conditional. The special indefinite a certain in English may not, and therefore it must keep its ‘pure’ existential force in conditionals:

(49) If a certain professor is angry, we are all upset.

This sentence can only be interpreted as claiming that there is a particular professor such that if that professor is angry, we are all upset and thus a certain, unlike unmarked indefinites, does not allow a universal-like interpretation. The same is true for the closest Romanian equivalent, un anume profesor. Treating this kind of quantificational variability as a test for ‘indefiniteness’ would force us to class a certain/un anume DPs as indefinite, and unmarked definites as indefinite, an unwelcome result.

A final observation concerns the possibility of unmarked indefinites to occur in generic generalizations, exemplified in (50):

(50) Un cîine nu-şi atacă stăpînul.

a dog not-Refl attack owner.Def
‘A dog does not attack its owner.’
In the most natural interpretation of both the Romanian example and its English counterpart, this sentence is understood as stating a general fact about dogs rather than a fact about a particular dog. The issue of the proper analysis of generic sentences in general and this kind of generic statements involving a singular indefinite in particular has spawned an entire subfield of linguistics, originating in the seminal work of Carlson (1977), which we cannot review even the highlights of for reasons of space. Sticking to our narrow concern, a popular (though by no means universally accepted) treatment of singular indefinites in examples such as (50) is to take them as ordinary indefinites, introducing a discourse referent which ends up being indirectly bound by a silent generic operator Gen.²² Under this approach then, (50) is a special case of ‘quantificational variability’ we find in antecedents of conditionals, were the indefinite is also indirectly bound by an implicit or explicit quantifier. The interpretation of (50) says that typical cases involving a typical dog in them are not extendable to cases where that dog attacks its owner. Under this analysis, it is not surprising that a certain/un anume indefinites cannot be used as generics in such statements since we found them resisting the parallel interpretation in antecedents of conditionals. The fact that definites resist it too, as shown in (51), should not be surprising out of context since what we need for a definite in a generic generalization is a context that both satisfies the definite requirement and allows the discourse referent to be indirectly bound by the generic operator.²³
(51) Cîinele nu-şi atacă stăpînul.
dog.Def not-Refl attacks owner.Def
‘The dog does not attack its owner.’

Changing the context, however, yields plausible candidates for definites in the role of indefinites in (50):

(52) În case cu un cîine şi o pisică, cîinele nu-şi atacă stăpînul.

in house.Pl with a dog and a cat dog.Def not-Refl attacks owner.Def
‘In households with a dog and a cat, the dog does not attack its owner.’

We conclude that generic generalizations exemplified in (50) should be treated as special cases of indirect binding by an adverb of quantification. The source of the ‘genericity’ of these examples is the implicit adverb rather than the article. We see below that unmarked indefinites contrast with marked ones in that the latter do not allow such generic binding.