What leads speakers to choose, for example, a pronoun over a name, and how does this choice affect language comprehension? Although many approaches to this problem have been taken, the one thing all approaches all have in common is the claim that pronouns are preferred when the referent is salient. But what determines this salience? Many diverse factors have been proposed, but they seem to have very little to do with each other except that they are purported to affect the thing called "salience".

In this paper, I will propose that comprehenders prefer certain forms of reference depending on how easy it is to establish reference. The ease of establishing reference, in turn, is influenced by multiple constraints, one of which is the frequency of reference patterns. This frequency information is available because speakers and writers produce language in a highly systemmatic way.

I will support the claim about frequency information by looking at four sources of information with both experimentation and corpus analysis. I will demonstrate that pronouns are preferred when the referent is the grammatical Subject (a topic position), or when it is in the focus of a cleft. NPs in these positions are prominent with respect to the following discourse, even though topics and foci have traditionally been opposed. I will also demonstrate that thematic role information affects reference form preferences, where pronouns are used more often to refer to goals than sources. In addition, I will present detailed evidence that pronouns are preferred more for things which have been mentioned recently.

For all four sources of information (Subjects, the focus of clefts, goals, and recent NPs), I will also show that subsequent reference to these items is more frequent than subsequent reference to other elements of the utterance. In this way, I will show that four very different types of information are similar in two ways: frequent types of reference correspond to increased preference for pronouns. Based on this evidence, I will propose an activation framework, in which frequency information affects the activation of the mental representations of discourse entities. High activation means that establishing reference is easier, and less specified forms of reference are preferred.