A robust system for human-machine dialogue in telephony-based applications

• Published in: International journal of speech technology Vol. 2; no. 2; pp. 101 - 111
• Main Authors: Albesano, D., Baggia, P., Danieli, M., Gemello, R., Gerbino, E., Rullent, C.
• Format: Journal Article
• Language: English
• Published: 01.12.1997
• Subjects: Computer Applications; Human Computer Communication; Italian; Natural Language Processing; Parsing; Speech Synthesis; Telephone; Voice Recognition
• ISSN: 1381-2416; 1572-8110
• DOI: 10.1007/BF02208822

A real-time system is presented for human-machine spoken dialogue on the telephone in task-oriented domains. Dialogos understands spoken Italian in the domain of railway inquiry. The system involves modules that perform acoustic processing (feature extraction, acoustic modeling, & recognition modeling), language modeling (class-based bigram models & dialogue prediction), a linguistic processor that performs local grammatical analysis iteratively & semantic analysis in two steps, & a dialogue module that is pragmatically based & tolerates spontaneous speech phenomena. Integration of the different modules allows the system to deal with partial or total breakdowns at other levels of analysis. The system was tested with inexperienced users (N = 245 male & 248 female Italian speakers, aged 18-65) each of whom made three calls for a total of 1,404 dialogues. Results are reported with respect to speech recognition metrics of word accuracy & sentence understanding rate, time to completion, time to acquisition of crucial parameters, & degree of success of the interactions in providing the speakers with the information they required. Evaluation data show that most subjects were able to interact fruitfully with the system. It is suggested that usable spoken language systems can result from design choices that enhance robustness. 3 Tables, 6 Figures, 14 References. Adapted from the source document