Gayle Ayers Elam, Ph.D.
|Home||Research||Projects||Industry Experience||Publications & Patent||About Me||Contact Info|
Very high quality prompts can be made for computer systems by using prerecorded units. The trick is in carefully determining the recording units that are the basic building blocks and what prosody they must be produced with to sound fluent and natural when the units are concatenated. See Elam & Wayland (1999), Prosody and prompt design in a computer dialog system, for our solution to numbers from 0.00 to 999,999,999,999.99, which worked for both English and German dialog systems.
Also, to have an effective computer dialog system, the system prompts and the userís speech inputs must be a conversation between the two parties. In the situation where the system requires a syntactically constrained response from the user, the system prompt must guide the user what to say, by proper use of vocabulary and prosody.
This was the topic of my dissertation research. Read the abstract of my dissertation.
Discourse structure is signalled by a hierarchical organization of pitch range as well as by speech rate and pause structure. In principle this is true of spontaneous speech as well as read speech. However, in spontaneous dialogue these are used for other purposes (turn-taking, etc.) and are moreover disrupted by false starts, interruptions, etc. Therefore, if we compare spontaneous and reenacted dialogue, we expect the discourse structure to be signalled more clearly in the reenacted versions, and this hypothesis is confirmed. See OSU Working Papers in Linguistics, 44.
Work done with Paul Touati of Lund University, Sweden. In French two levels of accentual prominence seem to be relevant: a default accent and a focal accent. The default accent is assigned to the final syllable of a prosodic phrase; it can be identified with final phrase juncture and has a demarcative function. The optional focal accent (accent d'insistance) is assigned to the initial syllable of a polysyllabic prosodic word or prosodic phrase; it can be related to a semantic or pragmatic prominence. Reiterant speech data were recorded for a corpus varying focus location. Main results include:
Guest researcher on Gosta Bruce's project "Prosodic Segmentation and Structuring of Dialogue" at Lund University, Sweden.
|Last Modified: May 2011|