http://chineseinput.net/에서 pinyin(병음)방식으로 중국어를 변환할 수 있습니다.
변환된 중국어를 복사하여 사용하시면 됩니다.
허성환(Seonghwan Heo),이원기(Wonkee Lee),이종혁(Jonghyeok Lee) 한국정보과학회 2021 한국정보과학회 학술발표논문집 Vol.2021 No.6
목적 지향 대화 시스템(task-oriented dialogue system)의 한 가지 중요한 구성 요소인 발화 이해(Spoken Language Understanding, SLU)는 사용자 발화의 의도(intent)와 의미적 슬롯(semantic slot)을 찾는 것을 목표로 한다. 본 연구에서는 의미적 슬롯을 찾아내는 슬롯 채우기(slot filling)문제를 해결하는 데 있어, 학습 데이터의 부족 문제를 지적하고, 이러한 문제를 해결하기 위해 크로스도메인 슬롯 채우기(cross-domain slot filling) 방법을 통해 접근한다. 본 연구에서는 이전까지 적용되지 않았던 사전학습 언어 모델인 BERT를 크로스도메인 슬롯 채우기에 적용하였고, 기존의 연구와 비교하여 그 성능이 향상됨을 확인하였다.
엔그램 사용량 조절을 통한 딥러닝 기반 Chit-chat 대화시스템의 상투적 응답 생성 제어
오재영(JaeYoung Oh),이원기(WonKee Lee),방지수(Jeesoo Bang),신재훈(Jaehun Shin),이종혁(Jong-Hyeok Lee) Korean Institute of Information Scientists and Eng 2022 정보과학회논문지 Vol.49 No.1
Chit-chat dialogue systems, the systems for unstructured conversations between humans and computer, aim to generate meaningful and diverse responses. However, training methods based on the maximum likelihood estimation have been reported to generate too many generic responses by the model; thus, reducing the interest in these systems. Recently, a new training method using unlikelihood training was proposed to generate diverse responses by penalizing the overuse of each vocab. However, it has a limitation that it only considers the usage of a token when penalizing each word, and does not consider in what context each token is used. Therefore, we propose a method by extending this work, which is penalizing the overuse of each n-gram. This method has the advantage of using information about the surrounding context in n-gram to penalize each token.
심층학습 기반의 Predictor-Estimator 모델을 이용한 영어-한국어 기계번역 품질 예측
김현(Hyun Kim),신재훈(Jaehun Shin),이원기(Wonkee Lee),조승우(Seungwoo Cho),이종혁(Jong-Hyeok Lee) Korean Institute of Information Scientists and Eng 2018 정보과학회논문지 Vol.45 No.6
Quality Estimation (QE) for machine translation is an automatic method for estimating the quality of machine translation output without the need to use reference translations. QE has recently grown in importance in the field of machine translation (MT). Recent studies on QE have mainly focused on European languages, whereas fewer studies have been carried out on QE for Korean. In this paper, we create a new QE dataset for English to Korean translations and apply a neural network based Predictor-Estimator model to a QE task of English-Korean. Creating a QE dataset requires manual post-edited translations for MT outputs. Because Korean is a free word order language and allows various writing styles for translation, we provide guidance for creating manual post-edited Korean translations for English-Korean QE data. Also, we alleviate the imbalanced data problem of QE data. Finally, this paper reports on our experimental results of the QE task of English-Korean by using the Predictor-Estimator model trained from the created English-Korean QE data.