The literature on phonetic learning for speech has shown that listeners are quite flexible in their perception of speech sounds. Specifically, research has shown that the boundaries separating similar speech sounds can be altered for a particular listener through repeated exposure to manipulated speech tokens. This particular form of learning is often cited as supporting listeners’ abilities to understand unfamiliar accents and to learn new phonetic inventories in the process of learning a second language. In this talk, I will discuss work examining whether the information supporting phonetic learning for speech is actually available in naturally produced speech and whether this information may support learning more effectively than manipulated tokens.