cross-linguistically valid claims about which discourses are easier to process, abstracting away from specific algorithms for anaphora resolution or anaphora generation (although many such algorithms are based on the theory). The result is a very different theory from those one usually finds in Computational Linguistics. In central papers such as (Grosz, Joshi, and Weinstein, 1995) no algorithms are provided to compute notions such as `utterance', `previous utterance', `ranking,' and `realization' that play a crucial role in the theory. The researchers working on Centering argue that while these c 2000 Association for Computational Linguistics