Generating referring expressions is a task that has received a great deal of attention in the natural-language generation community, with an increasing amount of recent effort targeted at the generation of multimodal referring expressions. However, most implemented systems tend to assume very little shared knowledge between the speaker and the hearer, and therefore must generate fully-elaborated linguistic references. Some systems do include a representation of the physical context or the dialogue context; however, other sources of contextual information are not normally used. Also, the generated references normally consist only of language and, possibly, deictic pointing gestures.

Speakers engaging in dialogue with another conversationalist must create and execute plans with respect to the content of the utterance. An analysis of disfluencies from Map Task monologues shows that a speaker is influenced by the pressure to communicate with a distant listener. Speakers were also subject to time-pressure, thereby increasing the cognitive burden of the overall task at hand. The duress of the speaker, as determined by disfluency rate, was examined across four conditions of variable feedback and timing. A surprising result was found that does not adhere to the predictions of the traditional views concerning collaboration in dialogue.

Disfluencies repetitions and reformulations midsentence in normal spontaneous speech are problematic for both psychological and computational models of speech understanding. Much effort is being applied to finding ways of adapting computational systems to detect and delete disfluencies. The input to such systems is usually an accurate transcription. We present results of an experiment in which human listeners are asked to give verbatim transcriptions of disfluent and fluent utterances. These suggest that listeners are seldom able to identify all the words “deleted” in disfluencies. While all types suffer, identification rates for repetitions are even worse than for other types. We attribute the results to difficulties in recall or coding for recall items which can not be identified with certainty. This inability seems to make human speech recognition more robust than current computational models.

This paper tests the hypothesis that listeners miss disuencies or fail to transcribe them accurately because disuencies interfere with the normal relationship between speech sound and linguistic context in human spoken word recognition.In a word-level gating experiment 16 listeners heard a total of 56 disuent utterances selected from a corpus of spontaneous speech, 56 length-matched uent controls, and 56 uent foils.The proportion of words never recognized was greater in disuent utterances than in controls.The failures clustered around the point where the disuency interrupted the utterance, ocurring particularly within the reparanda, but were not found at corresponding locations in uninterrupted controls.Repetition disuencies, where preand post-interruption portions might easily be construed together, allowed more successful word recognitions than recast disuencies, where reconstruction of a single intended utterance would be dicult, if not impossible.The results have implications both for understanding human speech recognition and for improving the robustness of ASR systems.