The extent to which children's cohesive oral text varied as a function of listener status and the extent to which children could be trained to generate decontextualized oral text was examined in a study of children in kindergarten through second grade. Independent variables were children's age; exposure to a play, discussion, or drawing condition; and the naive or informed status of the audience to which a story was retold. The dependent measure consisted of individual children's oral texts, generated on a recall task. Recall was scored for cohesion according to endophoric or exophoric reference and ellipsis. Endophorics are defined as linguistically encoded ties between the presupposed element and the presupposing element (e.g., "John's here. He'll leave later."). Exophorics rely on context or unencoded assumptions to convey meaning (e.g., the statement "It's nice," when "it" is defined by the speaker's pointing to an object). Reference is composed of three subcategories: pronominals, demonstratives, and comparatives. Ellipsis is also composed of three subcategories: nominal, verbal, and causal ellipsis. Children assigned to same-age groups of four were read children's books on three occasions. Following the readings, subjects were exposed to their respective treatments, then were asked to retell the story to a naive or informed listener. Results supported the proposition that children's register is affected by participants in discourse. Additionally, dramatic play appeared to be a context within which children learn to put meaning into words. (RH)