Understanding a spoken utterance requires identifying the spoken words that compose the utterance. A critical characteristic of speech is that it extends over time; spoken words are temporal sequences that become fully available only after a few hundred milliseconds. The recognition of spoken words must proceed incrementally, as phonetic information becomes available. Two sources of information can be distinguished in this process: the phonetic input available at each moment in time, and a set of expectations that the listener has formed, based on previous phonetic input and on the context in which this input occurred. In this talk, I will present results from studies that address how listeners monitor and evaluate these sources of information, both independently and in concert. Taken together, they support a view of speech perception in which all sources of information are continuously updated and integrated to support the best matching lexical hypotheses.