Spoken-style explanation generator for Japanese kanji using a text-to-speech system

In this paper we describe a spoken explanation generator, PLANET, for Japanese Kanji (ideograms), especially Kanji used in people's names. A number of text-to-speech systems for Kanji texts have been proposed but this is the rst one that can explain Kanji characters so as to disambiguate characters from many homophone Kanji candidates. To accomplish this the generator explains the Kanji by using both internal composition and use in other words. The system has a database of over 6,000 Kanji characters that breaks down any given Kanji into its components and a text corpus of explanations for explaining two or more Kanji characters. It is capable of generating both other words that include the Kanji characters in question, and identifying information. Using these other words and the information the system makes phrases and sentences. Furthermore, this system generates natural prosodic information by classifying the pattern of semantic connections between words and phrases. The explanations are output through a natural-sounding voice synthesizer. Hearing examinations con rmed that this system achieves high accuracy in disambiguating Kanji characters from among many candidates. This system will make it possible to provide advanced and user-friendly human-computer interfaces.