We are all familiar with robots and other computers producing linguistic expressions, starting with the Turing machine, and including more or less human-shaped machines producing more or less human-like speech sounds. I wish to address the question in what sense these speech-like phenomena can be regarded as an outcome of what might be called learning to talk. The question might also be rephrased as follows: in what sense can a talking robot be considered a speaker. What concerns me in this question is not so much the ongoing debate about what robots can and cannot do. Rather, it is the light it may throw on what it means for a human being to learn to talk.