Abstract:
다수의발화의도예측성능을향상시킬수 있는사용자발화의도추출방법및 사용자발화의도추출방법을수행하는장치가개시된다. 훈련말뭉치로부터문맥자질을추출하는단계와, 추출된문맥자질에대해순차적으로문맥자질을분류하는분류기를학습하여발화의도추출을위한적어도하나의모델을생성하는단계와, 생성된모델을저장하는단계와, 입력된적어도하나의사용자발화로부터적어도하나의문맥자질을추출하는단계와, 추출된적어도하나의문맥자질에대해생성된적어도하나의모델을이용하여사용자발화의도를예측하는단계및 예측된사용자발화의도에기초하여사용자발화의도를추출하는단계를포함한다. 따라서, 발화의도지시자를이용해하나의발화에포함된다수의발화의도를용이하게찾을수 있으며, 발화의도를예측성능을향상시킬수 있다.
Abstract:
Disclosed is a voice dialogue system and method using humor utterance. The voice dialogue system comprises an utterance analyzer for receiving an input of user utterance to be converted to a character format and analyzing the intention of a user; a humor utterance generator for using abbreviations or keywords included in the utterance of the user, based on the intention of the user, to produce humor utterance; a chatting utterance generator for generating a chatting utterance as a response corresponding to the intention of the user; and a final utterance selector for selecting a final utterance from the humor utterance and the chatting utterance. Therefore, the present invention is able to reduce boredom and allow the user to have fun while using the chatting system, by providing humor utterance to the user.
Abstract:
PURPOSE: An entity name tagging method and a device thereof are provided to improve the performance of a conversation system by tagging an accurate entity name to a word included in a corpus. CONSTITUTION: An acquisition unit(21) acquires an entity name candidate group from a word included in a corpus based on a dictionary for a predetermined domain. A tagging unit(22) tags an entity name to the entity name candidate group by applying an unsupervised learning method including a restriction condition to the entity name candidate group. The acquisition unit acquires the entity name candidate group according to a characteristic in which words included in a corpus are repeated. The restriction condition is a number which indicates the entity name in a sentence in which the words belong to the corpus. [Reference numerals] (21) Acquisition unit; (22) Tagging unit;
Abstract:
PURPOSE: An information search method by using the web and a voice conversation method using the method are provided to supply a better search result for a user query by extending knowledge information and information about the user query based on the web. CONSTITUTION: A basic word vector for a user query and a language analysis result is generated to search a vector space database for a vector space corresponding to the basic word vector(S420). Similarity between the basic word vector and the searched vector space are determined(S430). If the similarity is less than standards, an extended word vector is generated from a web search result performed by using the user query and the language analysis result(S440). The vector space database is searched for the vector space corresponding to the extended word vector by using the extended word vector. Knowledge information is searched based on the vector space. [Reference numerals] (1000) Knowledge information DB; (2000) Vector space; (2100) Vector space basic DB; (2200) Vector space extension DB; (AA) No; (BB) Yes; (CC) Search result; (S410) User query and language analysis result; (S420) Basic search; (S430) Determination?; (S440) Extended search; (S450) Generating vector space
Abstract:
PURPOSE: A confirmation enabled probabilistic and example-based spoken dialog system is provided to enable a conversation manager to determine whether information is unclear when a voice error occurs in a voice conversation interface, thereby providing an information confirmation conversation to a user. CONSTITUTION: A conversation state managing unit(112) of a confirmation conversation managing unit(110) calculates reliability of current conversation states using reliability in recognizing a user speech, reliability of understanding voice language, and reliability of a previous conversation state. A confirmation conversation request unit(114) of the confirmation conversation managing unit determines whether information is unclear by a confirmation conversation strategy about the reliability of the current conversation states. [Reference numerals] (10) Voice recognizer; (100) Conversation managing unit; (110) Confirmation Conversation managing unit(probability-based); (112) Conversation state managing unit; (114) Confirmation Conversation request unit; (120) Work related conversation managing unit(example-based); (20) Voice language comprehension unit; (200) Confirmation conversation strategy DB; (300) Conversation example DB
Abstract:
본 발명은 대화 로그를 이용한 학습 기반 대화 시스템 성능 향상 방법 및 그 장치에 관한 것이다. 본 발명에 따른 방법은 사용자와 대화 시스템 사이의 대화 로그를 수집하는 단계, 수집된 대화 로그에서 오류 발화를 추출하여 대화 시스템의 음성 인식 모델을 향상하기 위한 음성 인식 후보군을 생성하는 단계, 수집된 대화 로그에서 오류 발화를 추출하여 대화 시스템의 언어 이해 모델을 향상하기 위한 언어 이해 후보군을 생성하는 단계, 수집된 대화 로그에서 대화 시스템의 대화 패턴에 존재하지 않는 새로운 대화 패턴을 추출하여 대화 시스템의 대화 모델을 향상하기 위한 대화 패턴 후보군을 생성하는 단계, 음성 인식 후보군, 언어 이해 후보군 및 대화 패턴 후보군을 검증하는 단계, 그리고 검증된 후보군을 음성 인식 모델, 언어 이해 모델 및 대화 모델의 향상에 적용하는 단계를 포함한다.