Centering, accessibility and the next mention

Producing and interpreting natural language, reference included, relies on linguistic codes and pragmatic inferences. Can't have one without the other. Moreover, the classical Gricean view, whereby inferences only come in late to "clean up" after grammar has done its duty, has given way to a view where addressees are constantly engaged in predicting what's ahead. The role of this pragmatic work in reference is the focus of Kehler and Rohde's research. Kehler and Rohde propose a complementary division of labor between coherence-based prior predictions regarding next-mention subjects and a linguistic account for the referring expressions chosen for these subjects. They argue that neither a coherencebased theory (Hobbs, 1979 and onwards), nor a linguistic theory (they discuss Centering theory, Grosz et al., 1995) can do it alone. I couldn't agree more. Still, accepting Kehler and Rohde's division of labor, my goal here is to propose that unlike Accessibility theory (Ariel, 1990, 2001 inter alia), Centering theory is not complex enough to account for referential forms (1), and similarly, that the Hobbsian coherence predictions are not enough to account for next-subject mention biases (2). A study of natural discourse is needed before we can establish next-mention predictions (3).