• Korean Journal of Cognitive Science
• /
• v.11 no.3_4
• /
• pp.23-32
• /
• 2000

The distinction between main structure and side structure in discourse which was central to narrative studies has lacked an adequate. formal definition. This study supports the contention that there exists a hierarchical structure between discourse units constituting main structures, substructures, and side structures. The aim of this study is twofold: (j) to present an adequate. formal definition that provides a general identification criterion for distinguishing main structure from substructure and side structure proposed by Kuppevelt, and (jj) to propose conceptual relations representing hierarchical structures in discourse based on Sowa's Conceptual Structure Theory. The proposed conceptual relations which represent hierarchy and pragmatic relations of discourse segments are: DIGR (digression). T-SHFT (topic shift), and FRAM (frame). This s study shows pragmatic functions can be incorporated within CST in a systematic way.

• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.59-64
• /
• 2015

최근 온라인 텍스트 자료를 이용하여 대중의 의견을 분석하는 작업이 활발히 이루어지고 있다. 이러한 작업에는 주관적 방향성을 갖는 텍스트의 논증 구조와 중요 내용을 파악하는 과정이 필요하며, 자료의 양과 다양성이 급격히 증가하면서 그 과정의 자동화가 불가피해지고 있다. 본 연구에서는 정책에 대한 찬반 의견으로 구성된 한국어 텍스트 자료를 직접 구축하고, 글을 구성하는 기본 단위들 사이의 담화 관계를 정의하였다. 각 단위들 사이의 관계는 기계학습과 규칙 기반 방식을 이용하여 예측되고, 그 결과는 합성되어 하나의 글에 대응되는 트리 구조를 이룬다. 또한 텍스트의 구조상에서 주제문을 직접적으로 뒷받침하는 문장 혹은 절을 추출하여 글의 중요 내용을 얻고자 하였다.

• 한국어정보학회:학술대회논문집
• /
• 2016.10a
• /
• pp.33-38
• /
• 2016

수사구조는 텍스트의 각 구성 성분이 맺고 있는 관계를 의미하며, 필자의 의도는 논리적인 구조를 통해서 독자에게 더 잘 전달될 수 있다. 따라서 독자의 인지적 효과를 극대화할 수 있도록 수사구조를 고려하여 단락과 문장 구조를 구성하는 것이 필요하다. 그럼에도 불구하고 지금까지 수사구조에 기초한 한국어 분류체계를 만들거나 주석 코퍼스를 설계하려는 시도가 없었다. 본 연구에서는 기존 수사구조 이론을 기반으로, 한국어 보도문 형식에 적합한 30개 유형의 분류체계를 정제하고 최소 담화 단위별로 태깅한 코퍼스를 구축하였다. 또한 구축한 코퍼스를 토대로 중심문장을 비롯한 문장 구조의 특징과 분포 비율, 신문기사의 장르적 특성 등을 살펴봄으로써 텍스트에서 응집성의 실현 양상과 구문상의 특징을 확인하였다. 본 연구는 한국어 담화 구문에 적합한 수사구조 분류체계를 설계하고 이를 이용한 주석 코퍼스를 최초로 구축하였다는 점에서 의의를 갖는다.

• Journal of the Korean Geographical Society
• /
• v.45 no.6
• /
• pp.820-844
• /
• 2010

This study investigates the effect of GIS-integrated lessons on spatial thinking abilities and geographical skills using discourse analysis along with a comparative three-case case study method. A series of curricula were designed and implemented in an 8th grade classroom for a semester. The data collected consist of the dialogue transcripts of six consecutive GIS-integrated lessons. The transcripts were analyzed to identify the moves (speech acts) used by each student and to classify discourse content of spatial thinking and geographical skills. Based on three individual case studies, a cross-case study was performed to uncover any relationship between the phenomenon and the contexts. The empirical evidence from discourse analysis demonstrated that students were able to generate appropriate terms representative of spatial thinking and geographical skills although students appeared to possess primarily lower-order spatial abilities, followed by a moderate-level of spatial abilities. Considering that the unit was implemented in a biology class rather than a geography class the result reflected the fact that the student's spatial thinking and geographical skills were attributable to the GIS-integrated lessons. Thus, the results have a great implication for GIS-integrated lessons and geography education as an innovative tool for improving student's spatial thinking and geographical skills.

• KIISE Transactions on Computing Practices
• /
• v.23 no.12
• /
• pp.698-702
• /
• 2017

Since the introduction of Deep Neural Networks to the Natural Language Processing field, two major approaches have been considered for modeling text. One method involved learning embeddings, i.e. the distributed representations containing abstract semantics of words or sentences, with the textual context. The other strategy consisted of composing the embeddings trained by the above to get embeddings of longer texts. However, most studies of the composition methods just adopt word embeddings without consideration of the optimal embedding unit and the optimal method of composition. In this paper, we conducted experiments to analyze the optimal embedding unit and the optimal composition method for modeling longer texts, such as documents. In addition, we suggest a new discourse-based composition to overcome the limitation of the sequential composition method on composing sentence embeddings.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.33-38
• /
• 2016

수사구조는 텍스트의 각 구성 성분이 맺고 있는 관계를 의미하며, 필자의 의도는 논리적인 구조를 통해서 독자에게 더 잘 전달될 수 있다. 따라서 독자의 인지적 효과를 극대화할 수 있도록 수사구조를 고려하여 단락과 문장 구조를 구성하는 것이 필요하다. 그럼에도 불구하고 지금까지 수사구조에 기초한 한국어 분류체계를 만들거나 주석 코퍼스를 설계하려는 시도가 없었다. 본 연구에서는 기존 수사구조 이론을 기반으로, 한국어 보도문 형식에 적합한 30개 유형의 분류체계를 정제하고 최소 담화 단위별로 태깅한 코퍼스를 구축하였다. 또한 구축한 코퍼스를 토대로 중심문장을 비롯한 문장 구조의 특징과 분포 비율, 신문기사의 장르적 특성 등을 살펴봄으로써 텍스트에서 응집성의 실현 양상과 구문상의 특징을 확인하였다. 본 연구는 한국어 담화 구문에 적합한 수사구조 분류체계를 설계하고 이를 이용한 주석 코퍼스를 최초로 구축하였다는 점에서 의의를 갖는다.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.229-235
• /
• 2019

동시통역에서는 번역이 즉각적으로 빠르게 이루어지면서 원천텍스트의 의미가 정확히 전달되는 것이 핵심이다. 따라서 실시간 동시통역 시스템의 개발을 위해서는 번역정확도와 번역속도가 균형적으로 최적을 이루는 지점에서 분절하는 방법론이 필요하다. 이를 위해 본 연구에서는 운율 정보, 문법·통사 규칙, 의미 단위, 담화구조 표지, 분절단위의 길이 등 다양한 언어학적 자질을 제시하였다. 또한 본 논문에서 제안한 방법론을 검증하는 실험을 진행하였으며 그 결과 영한 데이터는 82%, 한영 데이터는 90%의 정확도를 보였다.

• Journal of The Korean Association For Science Education
• /
• v.20 no.4
• /
• pp.505-509
• /
• 2000

The kinetic structure of the content of instructions at the unit of 'heart and circulation' in "the common science", which was respectively given by two teachers majored in biology and two teachers majored in earth science was investigated. Concerning of the mean of the fundamental coefficient (B1), the mean of weighted coefficients (B2) and the progression of instruction (P), T4 (B1, 0.14; B2, 0.82; P; 0.86) teacher showed relatively lower kinetic structure and faster progression than those of the others. The other teachers showed a different level of the kinetic structure and progression: T1(B1, 0.21; B2, 0.86; P; 0.79), T2 (B1, 0.02; B2, 0.85; P; 0.80), and T3 (B1, 0.22; B2, 0.83; P; 0.78). T1 teacher who showed the largest total frequency of verbal elements, the great number of sorts of verbal elements and secondary span, had a higher level of the kinetic structure, but T4 teacher showed the lowest level of total frequency of verbal elements, the great number of sorts of verbal elements and secondary span. The main reason why T4 teacher's instruction achieved a lower level of kinetic structure was estimated that he neglected to reorganize teaching materials and he showed a small number of discourse units. In addition, the unit of 'heart and circulation' was less related to T4 teacher's major and T4 teacher had a shorter teaching experience than those of the others. T4 teacher also took charge of only one class, and this did not allow to instruct repeatedly.

• Journal of The Korean Association For Science Education
• /
• v.44 no.4
• /
• pp.301-311
• /
• 2024

We developed a small-group training program for in-service teachers focused on scientific modeling. We collected the discourses of the teachers who participated in the activity and analyzed them by type. The training program employed a collaborative approach in which a small group completed tasks and produced outputs based on the theme of 'galaxies and the Universe' to enable practical application in classes. Three in-service science teachers participated in the training program. Their discourses were recorded, transcribed, and classified into types based on individual turns and interaction units. The language expressions of the teachers reflected the unique characteristics of the teaching profession, with each participant having preferred language expression types, albeit with a generally low prevalence of specific language expression types across the participants. Differences in discourse characteristics related to the modeling theme, task presentation method, and model types, revealed that variations in the proportion of interaction unit types during the modeling design, build, and evaluation stages were primarily influenced by the teachers' familiarity with the modeling theme. While the task presentation method also influenced interaction types, model types had little impact on the distribution of interaction types. Considering these findings, training programs on modeling for in-service teachers should include a checklist to encourage sufficient interaction between participants as well as propose proper questions that can be effectively addressed through collaboration.

• Journal of The Korean Association For Science Education
• /
• v.41 no.3
• /
• pp.203-220
• /
• 2021

We explored the performance improvement method of automated scoring for scientific argumentation. We analyzed the pattern of argumentation using automated scoring models. For this purpose, we assessed the level of argumentation for student's scientific discourses in classrooms. The dataset consists of four units of argumentation features and argumentation levels for episodes. We utilized argumentation clusters and n-gram to enhance automated scoring accuracy. We used the three supervised learning algorithms resulting in 33 automatic scoring models. As a result of automated scoring, we got a good scoring accuracy of 77.59% on average and up to 85.37%. In this process, we found that argumentation cluster patterns could enhance automated scoring performance accuracy. Then, we analyzed argumentation patterns using the model of decision tree and random forest. Our results were consistent with the previous research in which justification in coordination with claim and evidence determines scientific argumentation quality. Our research method suggests a novel approach for analyzing the quality of scientific argumentation in classrooms.