| 摘要 |
PROBLEM TO BE SOLVED: To provide a word string recognition device that can prevent inconvenience due to erroneous recognition such that a user is at a loss. SOLUTION: Multistage processing is carried out. Here, spoken words and word attributes are classified in such a way that "I want to eat (request keyword) shao-mai (request object) at China Town (facility name) in Yokohama (place)". In the 1st stage, the request keyword "want to eat" is spotted to grasp what is requested and a topic is determined. In the 2nd stage, a vocabulary is limited on the basis of the topic determined in the 1st stage and a dictionary is switched to the one wherein the majority of 100 words are composed of the food name related to the names of restaurants based upon the destination place. Namely, it is known from the request keyword "want to eat" that the user wants to dine out, so it can be predicted that other words constituting the word string are a restaurant name and a dish name. COPYRIGHT: (C)2005,JPO&NCIPI
|
