• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.47-52
• /
• 2021

최근 지문을 바탕으로 답을 추론하는 연구들이 많이 이루어지고 있으며, 대표적으로 기계 독해 연구가 존재하고 관련 데이터 셋 또한 여러 가지가 공개되어 있다. 그러나 한국의 대학수학능력시험 국어 영역과 같은 복잡한 구조의 문제에 대한 고차원적인 문제 해결 능력을 요구하는 데이터 셋은 거의 존재하지 않는다. 이로 인해 고차원적인 독해 문제를 해결하기 위한 연구가 활발히 이루어지고 있지 않으며, 인공지능 모델의 독해 능력에 대한 성능 향상이 제한적이다. 기존의 입력 구조가 단조로운 독해 문제에 대한 모델로는 복잡한 구조의 독해 문제에 적용하기가 쉽지 않으며, 이를 해결하기 위해서는 새로운 모델 훈련 방법이 필요하다. 이에 복잡한 구조의 고차원적인 독해 문제에도 대응이 가능하도록 하는 모델 훈련 방법을 제안하고자 한다. 더불어 3가지의 데이터 증강 기법을 제안함으로써 고차원 독해 문제 데이터 셋의 부족 문제 또한 해소하고자 한다.

• The Journal of the Korea Contents Association
• /
• v.11 no.4
• /
• pp.477-484
• /
• 2011

The purpose of this study is to research middle school student's reading problem and to suggest how to improve their reading competence through the use of the NIE. For this research, we proposed three research questions of study as below. First, what difference is there in English reading competence between experimental and control group? Second, what difference is there in the English reading competence according to English proficiency? Third, what are the effects of interest and satisfaction through NIE? The subject of a sample was 34 experimental group and 36 control group, total 70 eighth graders in Kwangju. The experimental group had been taught through NIE for 10 weeks. Grammar Translation Method was applied while teaching English to the control group. The data analyzing method was t-test through the statistics program SPSS12.0. Based on the result of this study, NIE approach was effective in improving the students' reading competence. Most of the students who were taught reading by NIE responded that they had more interest and satisfaction in the reading lesson.

• Journal of Digital Convergence
• /
• v.17 no.6
• /
• pp.1-9
• /
• 2019

In an effort to explore the effects of Youtube video-watching activities on Korean college students' English reading comprehension skills, 148 undergraduate students who enrolled in a General English class at a university in Korea participated in the present study. Participants were randomly classified into four groups - three experimental groups and one control group - according to when they watch videos: before class (n = 33), during class (n = 42), after class (n = 36), and none (n = 37). Over 16 weeks, the three experimental groups engaged in Youtube video-watching activities for about 10 minutes before, during, and after the class, while the control group did not. Pre- and post-tests were administered to confirm the effects of the use of Youtube videos on improving English reading comprehension skills. To compare the improvement between groups, a one-way ANOVA was also run. Major findings are as follows: First, participants in all the three experimental groups significantly improved their English reading comprehension skills, indicating the beneficial effects of Youtube video-watching activities. However, there was no statistically significant difference in the mean improvement between the groups. Based on this, limitations and suggestions for the future research are discussed at the end.

• Journal of the Korean BIBLIA Society for library and Information Science
• /
• v.21 no.3
• /
• pp.17-27
• /
• 2010

This study aimed to identify the problems of college students in reading the literature and on the basis of the identified problems, to suggest the approaches to solve the problems. To this end, time required for reading passages, reading patterns, understanding, memory and reading habits and attitudes were analyzed with the freshmen in A university. In accordance with the analysis results, 58% of subjects was good and 42% was not sufficient on the basis of the averages in Scholastic Aptitude Test. Second, 77% of subjects had the good patterns but 23% showed certain problems in reading patterns. Third, 69% and 67% of subjects illustrated good results in the analysis on understanding and memory, respectively. However, 31% and 33% were evaluated as being on the general level or requiring efforts in the analysis on understanding and memory, respectively. Next, according to the analysis on reading habits and attitudes, 77% had no problems but 23% required improvement. For solving the problems identified through the analysis, it is recommended to develop the scientific and standardized evaluation tools for evaluating the reading ability of college students. Second, it is necessary to evaluate the reading ability, habit and attitude during the screening process for admission or after admission. Finally, it is required to operate the Fundamental Academic Ability Learning Center(tentative name) to improve the ability of students who show the insufficient results in evaluation.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.273-277
• /
• 2018

기계가 주어진 텍스트를 이해하고 추론하는 능력을 기계독해 능력이라 한다. 기계독해는 질의응답 태스크에 적용될 수 있는데 이것을 기계독해 질의응답이라 한다. 기계독해 질의응답은 주어진 질문과 문서를 이해하고 이를 기반으로 질문에 적합한 답을 출력하는 태스크이다. 본 논문에서는 구조화된 표 형식 데이터로부터 질문에 대한 답을 추론하는 TableQA 태스크를 소개하고, $S^3-NET$을 이용하여 TableQA 문제를 해결할 것을 제안한다. 실험 결과, 본 논문에서 제안한 방법이 EM 96.36%, F1 97.04%로 우수한 성능을 보였다.

• Journal of the Korea Society of Computer and Information
• /
• v.5 no.3
• /
• pp.58-63
• /
• 2000

Since some methodologies of using Internet on English reading have been appeared, most of them have just led students to find some articles on the Internet and translate them into their first language. However, these methodologies have been criticized in that they can not provide naturalistic environment for practical English reading. There are some problems in using Internet for practical English reading. First, the level of vocabularies and grammar of articles from the Internet has not been proved to be appropriate for students. Usually, their level is too high for most students. Second, it needs computer using ability as well as English proficiency if a student successfully finds an article which he or she wants to on the Internet in a limited time. Finally, a teacher should be trained to lead students to participate in a classroom discussion to get, appropriate gists of articles. With all these problems, it is difficult only to use articles from the Internet for successful English reading. Therefore, this study tries to find out some critical problems and solve them, and construct English reading courseware system on the Internet.

• The Journal of the Korea Contents Association
• /
• v.9 no.7
• /
• pp.411-418
• /
• 2009

The purpose of this study is to research the effects of English reading strategies on English reading comprehension by Korean college students. Reading strategy use was assessed through Oxford's self-report questionnaire in reading strategies. This study has three research questions. The first question was to investigate some reading strategies used by college students. The second question was to investigate the differences in reading strategies between two groups in gender. The third question was to investigate the differences in reading strategies of three college student groups according to their English proficiency estimated by reading scores. Some major findings of this study are as follows. First, college English learners use memory strategies most frequently of the six strategies, while using metacognitive strategies least frequently. Second, there exists a significant difference in reading strategies between the gender group. Third, there also exists a significant difference in reading strategies among the three groups divided according to English proficiency. This study shows that students' reading ability can be strengthened and motivated by some reading strategies in reading practice. It also means that it is necessary for English teachers to take into consideration the reading strategies suitable for the students in their reading classes.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.213-218
• /
• 2019

기계독해는 자연어로 표현된 질문과 단락이 주어졌을 때, 해당 단락 내에 표현된 정답을 찾는 태스크이다. 최근 기계독해 태스크도 다른 자연어처리 태스크와 유사하게 BERT, XLNet, RoBERTa와 같이 사전에 학습한 언어모델을 이용하고 질문과 단락이 입력되었을 경우 정답의 경계를 추가 학습(fine-tuning)하는 방법이 우수한 성능을 보이고 있으며, 특히 KorQuAD v1.0 데이터셋에서 학습 및 평가하였을 경우 94% F1 이상의 높은 성능을 보이고 있다. 본 논문에서는 현재 최고 수준의 기계독해 기술이 학습셋과 유사한 평가셋이 아닌 일반적인 질문과 단락 쌍에 대해서 가지는 일반화 능력을 평가하고자 한다. 이를 위하여 첫번째로 한국어에 대해서 공개된 KorQuAD v1.0 데이터셋과 NIA v2017 데이터셋, 그리고 엑소브레인 과제에서 구축한 엑소브레인 v2018 데이터셋을 이용하여 데이터셋 간의 교차 평가를 수행하였다. 교차 평가결과, 각 데이터셋의 정답의 길이, 질문과 단락 사이의 오버랩 비율과 같은 데이터셋 통계와 일반화 성능이 서로 관련이 있음을 확인하였다. 다음으로 KorBERT 사전 학습 언어모델과 학습 가능한 기계독해 데이터 셋 21만 건 전체를 이용하여 학습한 기계독해 모델에 대해 블라인드 평가셋 평가를 수행하였다. 블라인드 평가로 일반분야에서 학습한 기계독해 모델의 법률분야 평가셋에서의 일반화 성능을 평가하고, 정답 단락을 읽고 질문을 생성하지 않고 질문을 먼저 생성한 후 정답 단락을 검색한 평가셋에서의 기계독해 성능을 평가하였다. 블라인드 평가 결과, 사전 학습 언어 모델을 사용하지 않은 기계독해 모델 대비 사전 학습 언어 모델을 사용하는 모델이 큰 폭의 일반화 성능을 보였으나, 정답의 길이가 길고 질문과 단락 사이 어휘 오버랩 비율이 낮은 평가셋에서는 아직 80%이하의 성능을 보임을 확인하였다. 본 논문의 실험 결과 기계 독해 태스크는 특성 상 질문과 정답 사이의 어휘 오버랩 및 정답의 길이에 따라 난이도 및 일반화 성능 차이가 발생함을 확인하였고, 일반적인 질문과 단락을 대상으로 하는 기계독해 모델 개발을 위해서는 다양한 유형의 평가셋에서 일반화 평가가 필요함을 확인하였다.

• Journal of Digital Convergence
• /
• v.19 no.2
• /
• pp.21-32
• /
• 2021

The current study investigates how Malaysian ESL learners' L2 (English) speaking fluency is reflected in advanced L2 reading and what difficulties they encounter in reading comprehension. Nine Malaysian students attending a Korean university participated in qualitative research using in-depth and semi-structured interviews. The data revealed that L2 was a very familiar language, and their speaking fluency in L2 reduced the anxiety of L2 reading in general. However, it did not play a significant role in reading at an advanced level. Their difficulties in reading were mainly due to a lack of vocabulary knowledge. However, insufficient background knowledge and interest also frustrated their reading tasks. These factors lowered their reading comprehension, causing inaccurate interpretations or discouraging their endeavors to find messages from the given text. Thus, these findings should be carefully addressed in reading classes for Korean L2 learners as well as international students.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2019.10a
• /
• pp.922-925
• /
• 2019

기계 독해 기술은 기계가 주어진 비정형 문서 내에서 사용자의 질문을 이해하여 답변을 하는 기술로써, 챗봇이나 스마트 스피커 등, 사용자 질의응답 분야에서 핵심이 되는 기술 중 하나이다. 최근 딥러닝을 이용한 기학습 언어모델과 전이학습을 통해 사람의 기계 독해 능력을 뛰어넘는 방법론들이 제시되었다. 하지만 이러한 방식은 사람이 인식하는 질의응답 방법과 달리, 개체가 가지는 의미론(Semantic) 관점보다는 토큰 단위로 분리된 개체의 형태(Syntactic)와 등장하는 문맥(Context)에 의존해 기계 독해를 수행하였다. 본 논문에서는 기존의 높은 성능을 나타내던 기학습 언어모델에 대규모 지식그래프에 등장하는 개체 정보를 함께 학습함으로써, 의미학적 정보를 반영하는 방법을 제시한다. 본 논문이 제시하는 방법을 통해 기존 방법보다 기계 독해 분야에서 높은 성능향상 결과를 얻을 수 있었다.