Automatic Acquisition of Names Using Speak and Spell Mode in Spoken Dialogue Systems

Automatic Acquisition of Names Using Speak and Spell Mode in Spoken Dialogue Systems Grace Chung � , Stephanie Seneff � and Chao Wang �� � Corporation for National Research Initiatives 1895 Preston White Drive, Suite 100, Reston, VA 22209 gchung@cnri.reston.va.us � Spoken Language Systems Group MIT Laboratory for Computer Science 200 Technology Square, Cambridge, MA 02139 � seneff,wangc � @sls.lcs.mit.edu Abstract This paper describes a novel multi-stage recog- nition procedure for deducing the spelling and pronunciation of an open set of names. The overall goal is the automatic acquisition of un- known words in a human computer conver- sational system. The names are spoken and spelled in a single utterance, achieving a con- cise and natural dialogue flow. The first recog- nition pass extracts letter hypotheses from the spelled part of the waveform and maps them to phonemic hypotheses via a hierarchical sub- lexical model capable of generating grapheme- phoneme mappings. A second recognition pass determines the name by combining information from the spoken and spelled part of the wave- form, augmented with language model con- straints. The procedure is integrated into a spo- ken dialogue system where users are asked to enroll their names for the first time. The acqui- sition process is implemented in multiple paral- lel threads for real-time operation. Subsequent to inducing the spelling and pronunciation of a new name, a series of operations automatically updates the recognition and natural language systems to immediately accommodate the new word. Experiments show promising results for letter and phoneme accuracies on a preliminary dataset. �The research at CNRI was supported by DARPA un- der contract number N66001-00-2-8922, monitored through SPAWAR Systems Center, San Diego. The research at MIT was supported by DARPA under contract number NBCH1020002 monitored through the Dept. of the Interior, National Business Center, Acquisition Services Div., Fort Huachuca, AZ. 1 Introduction Spoken dialogue systems are emerging as an effective means for humans to access information spaces through natural spoken interaction with computers. These sys- tems are usually implemented in such a way that their knowledge space is static, or is only augmented through human intervention from the system developers. A sig- nificant enhancement to the usability of such systems would be the ability to automatically acquire new knowl- edge through spoken interaction with its end users. Such knowledge would include both the spelling and pronunci- ation of a new word, as well as an understanding of its us- age in the language (e.g., a semantic category). However, this is a difficult task to carry out effectively, challeng- ing both with regard to the automatic acquisition of the sound-to-letter mapping from typically telephone-quality speech, and the system level aspect of integrating the usually off-line activities of system upgrade while seam- lessly continuing the conversation with the user. The research reported here is concerned with the ac- quisition of the user’s name, which is entered via a “speak and spell” mode, spoken sequentially for the first and last names respectively. It is our belief that this would be …

Content truncated to 500 words