Discourse and Donkey Anaphora
Structural Condition on Pronoun Binding
Natural language pronouns are translated as variables. They must receive interpretations either anchored to some individual made available in the discourse context or by being bound by operator phrases. Just as quantificational operators bind only the occurrences of their target variable in the formula which is their scope, quantificational NPs can only bind pronouns which they c-command.
The examples in (1) - (3) allow for a bound reading of the pronoun. However the examples in (4) - (6) do not. Based upon examples like these, we assume that:
In order for a pronoun to receive a bound variable interpretation, it must be c-commanded by its binder.
This is illustrated also by examples from coordination and between sentences in discourse:
All of these examples consider the relation between pronouns and quantificational noun phrases which are universal (every applicant, each new employee) or negative (no applicant ). Referential terms like proper names and definite and demonstrative descriptions of course do allow co-reference in these cases.
Indefinites allow intersentential anaphoric reference
Indefinite noun phrases (some employee(s), an applicant, two students , exactly one student ) have so far been given quantificational translations, but they do not show the same restriction on apparent pronoun binding. The sentences above with coordinations and two sentence discourses allow the pronoun to be construed as coreferential with the indefinite NP:
Based upon these and other considerations, we maintain the assumption that there is a structural condition on true pronoun binding. However we view the examples with indefinites above as showing that indefinites alter the discourse context to introduce entities to which the pronouns can be anchored. These entities are called discourse referents.
Donkey Pronouns
Within sentences, indefinites show this difference from true quantificational noun phrases by allowing for apparent pronoun binding.
In (1) and (3), the pronouns must be interpreted as anchored in some way in the context and cannot be treated as bound by the phrase suggested by the coindexation. This is exactly what we expect given the condition on bound pronoun interpretation.
However in (2) and (4), the pronouns can be interpreted as related to the coindexed indefinite phrase and not in a way which gives it a single anchor in the context. This is surprising given that there is no c-command relation here.
Pronouns like these, in which a pronoun is apparently bound to an indefinite which does not c-command it are known as donkey pronouns, after the sentences below:
Crucial to the occurrence of the donkey pronoun type of binding is the fact that the indefinite which is apparently binding the pronoun itself occurs in the restriction of a quantificational NP which has the pronoun in its scope. This is evident in (7). The same account can be given to the conditional sentence in (8) once we view it as an instance of universal quantification.