A method for analyzing and adjusting the performance of a speech-enabled application includes selecting a number of user utterances that were previously received by the speech-enabled application. The speech-enabled application receives such user utterances and associates each user utterance with an action-object based on one or more salient terms in the user utterance that are associated with the action-object. The method further includes associating one of a number of action-objects with each of the selected user utterances. Furthermore, for each action-object, the percentage of the utterances associated with the action-object that include at least one of the salient terms associated with the action-object is determined. If the percentage does not exceed a selected threshold, the method also includes adjusting the one or more salient terms associated with the action-object.