Five teenagers with severe intellectual impairments and no discernible communication skills were enrolled in training to teach a conditional request for assistance using a speech-generating device (SGD). All were either blind or severely visually impaired since birth. All learned to operate an adaptive switch to control sensory outcomes, next showed preferences among sensory outcomes, and then demonstrated the ability to use their switch to signal for assistance with an SGD when the sensory outcome was remotely disabled. During the signaling phase, or subsequent attempts to generalize its use outside the laboratory, three participants began vocalizing. Most notably, they began imitation of the word "song" or the word "help" emitted by the SGD. The potential role of cause-and-effect training with adaptive switches is discussed. (Contains 4 figures.)