Whenever speakers refer to entities or events, they are tasked with making a series of decisions: which entity to refer to, how to embed it into syntactic structures, and which type of expression to use for the reference. This study concerns itself with the last of these decisions, the selection of the linguistic exponents of reference, or referential choice. While it is well known that languages differ greatly in their preference for either pronominal (*she*, *this*) or zero anaphora (ellipsis) to refer back to previously mentioned discourse referents, the overall rates of occurrence of lexically-headed anaphora (*the woman*, *Jane*) in monological discourse turn out to be remarkably stable across languages. This study examines the circumstances in which speakers opt for the more informative but less econominal choice of lexical references over reduced alternatives. It does so from a typological and comparative angle, charting the cross-linguistic stability of certain classes of explanatory factors and the parametricization of others across languages. Rather than adopting a specific theoretical framework in its approach to the question, it instead explores a number of empirical bottom-up approaches, deriving complex analytical categories from multiple levels of relatively basic annotations. Where earlier research on referential choice has been focused predominantly based on written data, or else on small data sets of spoken language from English and other overrepresented languages, this study addresses issues of typological representativity by employing spoken corpora from a diverse set of ten languages, many of which are understudied and endangered.