Abstract Since Prince (1981) and Givón (1983), studies on discourse reference have explained the grammatical realization of referents in terms of general concepts such as “assumed familiarity” or “discourse coherence.” In this paper, we develop a complementary approach based on a detailed statistical tracking of subjects in Emirati Arabic, from which two major categories of subject expression emerge. On the one hand, null subjects are opposed to overt ones; on the other, subject-verb (SV) is opposed to verb-subject (VS). Although null subjects strongly correlate with coreferentiality with the subject of the previous clause, they can also index more distant referents within a single episode. With respect to SV vs. VS, morpholexical classes are found to be biased toward one or the other: nouns are typically VS, pronouns SV. We conclude that the null subject variant is the norm in Emirati Arabic, and when an overt subject is appropriate, lexical identity biases the subject into SV or VS order, generating word order as a discourse-relevant parameter. Overall, our approach attempts to understand Arabic discourse from a microlevel perspective.