The purpose of the proposed research is to investigate 3 basic assumptions that are commonly used in knowledge-based automatic speech recognition (ASR):
(1) the linguistic units to be recognized are distinctive features; (2) the relevant acoustic information for these features is located close to instants of consonantal opening or closure;
(3) the features must be measured in a hierarchical fashion. These assumptions will be investigated by means of two perception experiments in which listeners are presented with gated natural utterances, while they have to guess the upcoming consonant. A first experiment will lead to a description of the distribution of featural information along the time dimension. This description will enable us to reevaluate the first two assumptions. In the second experiment it will be investigated whether providing information on manner-of-articulation features changes the perception of place-of -articulation features. This is achieved through manipulation of the response set that is available to the listener.