This article reports an experimental study of the differential effects of premodified input, interactionally modified input, and modified output on the comprehension of directions in a listen-and-do task and the acquisition of new words embedded in the directions. The modified output group achieved higher comprehension and vocabulary acquisition scores than either of the input groups. There was no difference between the premodified and interactionally modified input groups. The advantage of the modified output group is explained in terms of the qualitatively superior dialogic interaction that occurred in this condition rather than in terms of actual language production.