A Study on Post-editing Guidelines for Korean-Japanese and Japanese-Korean Machine Translated Informative Texts - Focused on the suitability of implementing post-editing guidelines in translation courses -

JuRiAe Lee 한국외국어대학교 통번역연구소 2019 한국외국어대학교 통번역연구소 학술대회 Vol.2019 No.07

Artificial neural network based machine translations have become common since 2016 and the advancements in quality of said translations call for a change in the translation and interpretation education sector. It is true that some machine translations are comprehensible to a certain degree, but they also tend to translate texts into something that is very different from the original. Therefore, they cannot be said to have reached a reliable level. However, considering the speed at which the quality of machine translation is being improved, there is a need to learn about it in translation classes. Machine translations require human post-editing. Therefore, it is necessary to train students to become familiar with the task of post-editing and to offer them some guidelines with which to manage said task with speed. This study attempts to draw such guidelines using Korean press articles as data. Press articles cover various current topics related to politics, economics, social issues, cultural issues and others. They also contain reports, columns and editorials on different subjects and genres in various different forms of texts, such as headline texts and main texts, which are very useful in practicing translation in class. The practice of translation using press articles can help students set a strong foundation for developing the capacity to manage diverse forms of text. This study uses Google Translation, and Naver search engine's PAPAGO. The study uses as reference the results of a preceding research(Park, 2018) on post-editing class training and student reaction. According to said case study, students felt heavily burdened when they had to post-edit machine translations. There were also students that wondered about the range and extent in which they could modify the translations. In this study, time was limited, as real life post-editing requests from clients tend to be time-pressing. Also, limited to news article texts, a corpus was created according to article content and errors were analyzed applying the machine translation editing code(Lee, 2018) to develop post-editing guidelines. News article translations were divided into three groups for the purpose of comparing results and quality: human translations, post-edited machine translations that were done without any guideline, and post-edited machine translations that followed the developed guidelines. Afterwards, translators of each group were interviewed in-depth to assess the suitability of implementing said guidelines. There are no post-editing guidelines for machine translations as of yet as machine translation is still in its early, developing stages. However, this study is significant in that it proposes a set of guidelines for post-editing Korean-Japanese and Japanese- Korean machine translations under the assumption that the role of human translators will extend to co-working with machine translations as the use of these translations becomes more and more common. The guidelines should be implemented in translation teaching courses. These will need to be modified along with advancements and changes in machine translation. They need to be continuously evaluated and adjusted for the purpose of training effective, fast working post-editing professionals.

经验主义时期国际机器翻译演化路径及前沿探析

金蒙 窦,宝霞 王 한국중국언어학회 2019 중국언어연구 Vol.0 No.81

With the advent of the age of big data, machine translation has been receiving more and more attention as an important branch of artificial intelligence. The bibliometric study of international machine translation can help domestic scholars to better grasp the development of machine translation and find out the inadequacies of current machine translation research. This paper uses CiteSpace V as a tool to systematically analyze the status, hotspots and frontier trends of research in international machine translation from 1993 to 2017 by using the map of dual-map overlays of journal, the cooperation maps of authors-institutions-country, and the clustering map of the co-citation literatures. In the field of international machine translation, chronological order of the hotspots is machine translation based on statistical principle only, statistical machine translation using linguistic information, domain adaptation, neural machine translation, research frontier includes translation of low-resource language, post-editing, quality estimation of machine translation. This paper holds that the field of machine translation at the moment needs further strengthening in the development of the application of the existing machine translation technology, the fusion of cognitive linguistics and machine translation technology. 随着大数据时代的到来,机器翻译作为人工智能领域的一个重要分支正在受到越来越多的关注。对国际机器翻译进行文献计量研究有助于学者更好地把握机器翻译的发展动态,发现当前机器翻译研究的不足之处。本研究以CiteSpace V为工具,运用期刊双图叠加图谱、作者-机构-国家合作图谱、文献共被引聚类图谱系统地分析了1993-2017年经验主义时期间国际机器翻译领域的研究状况、热点主题以及前沿趋势等问题。经研究发现国际机器翻译领域的热点依次为仅依靠统计学原理的机器翻译、综合运用语言学信息的统计机器翻译、领域自适应、神经机器翻译,研究前沿为低资源语言的翻译、译后编辑、质量评估标准。最后,当前国际机器翻译领域在既有机器翻译技术应用模式的开发、认知语言学等先验知识与机器翻译技术的融合两个方面有待进一步加强。

기계번역 프로그램 품질에 대한 사용자 평가와 사용자의 L2 수준 간 상관관계 고찰 -한중 언어 쌍을 중심으로-

공수 한국통역번역학회 2019 통역과 번역 Vol.21 No.3

While the quality of machine translation is getting better and better, it is still not perfect. In this case, how the user treats the imperfect machine translation results determines whether the user achieves the purpose for using machine translation. In this process, user evaluations are a key factor. User evaluations are not objective nor accurate. User evaluations are a subjective evaluation of the user and is related to the user. Therefore, this paper attempts to analyze the relationship between the level of the user's L2 and user evaluations. This paper surveyed 69 Chinese users to understand the current status of their use of machine translation, including frequency of use, purpose of use, favorite machine translation, reasons for preferences, and satisfaction with the quality of the machine translation used. At the same time, for specific translation articles, it lets users evaluate the accuracy and fluency of machine translation articles. The survey results show that users with low L2 levels used machine translation at a higher frequency and were more inclined to evaluate machine translation from the perspective of ease of use. In addition, statistical analysis of the evaluation results found that users’ evaluation of machine translation was related to the user's own level L2. The lower the L2 level, the higher the evaluation of the adequacy and fluency of machine translation, and the higher the assessment of the overall quality of the machine translation.

기계번역, 저작권법에서 자유로운가?

김윤명 한국법제연구원 2023 법제연구 Vol.- No.64

Translation is also for social contributions or public interests in that it allows people to enjoy information or culture in various fields. Translators increase the utilization of language and play a role as cultural propagators. Machine translation is much lower wavelength. The right to the translation, which is the result of translation, should be considered to belong to the user who used the machine translation service as a tool, not to the business operator that provides the machine translation service. However, machine translation is only a simple reproduction, not the creation of a secondary work called the right to translate. This is because the machine translation process has no room for the user's creative contribution to the translation, independent of the original work's creative performance. Therefore, since machine translation is only a conversion of language, not a secondary work creation act, machine translation in which permission is granted is a reproduction under the Copyright Act and violates the copyright holder's right to reproduce. However, users may be directly liable for infringement, but there is a high possibility that they will be exempted in accordance with the private replication regulations. Therefore, it is questionable whether it is reasonable to hold the online service provider (OSP) legally responsible even though the user is exempted as a regular offender. Therefore, I think it is reasonable to indemnify the translation service provider in consideration of the public interest of the translation service as to whether the copyright holder violates the right to reproduce. The problem is that the current OSP exemption regulations are difficult to apply to online service providers that provide machine translation. This is because it is questionable whether the translation service meets the type of OSP exemption regulations. Therefore, online service providers that provide machine translation are bound to be responsible for infringement, so it is necessary to review whether the translation service can be exempted in accordance with the fair use regulations. In conclusion, the conclusion made after reviewing the utility that users can obtain through translation compared to the transaction cost of obtaining permission from the right holder for the public interest provided by the translation service is highly likely to be exempted. However, for the social contribution of translation service providers and the legal stability of online service providers accordingly, I would like to propose a legislative proposal that includes a new type of OSP exemption in the form of general provisions. Separately, it is necessary to clarify the rights relationship of important machine translation such as universal services in the intelligent information society through in-depth research or policy review on machine translation such as translation quality. In the future, when a generative AI model, such as ChatGPT, directly translates as a policy task that may appear in the machine translation process, it is necessary to specifically discuss matters such as attribution of the rights of translation data(or translation memory). 번역은 다양한 분야의 정보나 문화를 향유할 수 있도록 한다는 점에서 사회적 기여나 공공의 이익을 위한 것이기도 하다. 번역가는 언어의 활용도를 높이고, 문화적 전파자로서 역할을 한다. 기계번역(machine translation)은 훨씬 더 전파력이 크다. 번역에 따른 결과물인 번역물에 대한 권리는 기계번역 서비스를 제공하는 사업자가 아니라 기계번역 서비스를 도구로써 활용한 이용자에게 귀속된다고 보아야 한다. 다만, 기계번역은 번역권이라는 2차적저작물의 작성이 아닌 단순한 복제에 불과하다. 기계번역 과정은 원저작물의 창작성과는 별개로 번역물에 대한 이용자의 창작적 기여가 발생할 여지가 없기 때문이다. 따라서, 기계번역은 2차적저작물 작성행위가 아닌 언어의 변환에 불과하므로 허락이 이루어지는 기계번역은 저작권법상 복제로서 저작권자의 복제권을 침해하는 구조이다. 다만, 이용자는 직접적인 침해책임을 질 수 있으나 사적복제 규정에 따라 면책될 가능성이 높다. 따라서, 정범으로서 이용자가 면책됨에도 온라인서비스제공자(OSP)에게 법적 책임을 지우는 것이 합리적인지 의문이다. 그렇기 때문에 저작권자의 복제권 침해 여부에 대해서는 번역서비스가 갖는 공익성 등을 고려하여 번역서비스 제공자를 면책하는 방안이 합리적이라고 생각한다. 문제는 현행 OSP 면책규정은 기계번역을 제공하는 온라인서비스제공자에게 적용하기 어렵다는 점이다. 번역서비스가 OSP 면책규정의 유형에 부합한 것인지 의문이기 때문이다. 따라서, 기계번역을 제공하는 온라인서비스제공자는 침해책임을 질 수밖에 없어, 차선으로써 번역서비스가 공정이용 규정에 따른 면책될 수 있는지 검토가 필요하다. 번역서비스가 제공하는 공익이 권리자로부터 이용허락 등을 얻는 거래비용에 비해 번역으로 이용자가 얻을 수 있는 효용 등을 검토한 후에 내린 결론은 면책가능성이 높다고 본다. 다만, 번역서비스 제공자의 사회적 기여 및 이에 따른 온라인서비스제공자의 법적안정성을 위해서라도 OSP 면책규정이 일반조항 형태로 새로운 유형을 포함하는 입법안을 제안하고자 한다. 이와 별개로, 번역품질 등 기계번역에 대해서는 심도 있는 연구 내지 정책적 검토를 통해 지능정보사회에서 보편적 서비스와 같이 중요한 기계번역에 따른 권리관계를 명확히 하거나 정보격차 등의 해소방안으로써 접근하여 예상되는 법적인 문제를 해결하는 방안을 고려하는 것도 필요하다. 향후, 기계번역 과정에서 나타날 수 있는 정책적 과제로서 ChatGPT와 같이 생성형 AI모델이 직접 번역하는 경우, 번역데이터(또는 번역메모리)의 권리 귀속, 번역결과물에 대한 편향·오류 등에 대한 사항을 구체적으로 논의할 필요가 있다.

古典文言文 기계번역의 현황과 과제

김우정 한국중국어문학회 2021 中國文學 Vol.109 No.-

This article examines the current status of machine translation in Classical Chinese and suggests future tasks. In order to understand machine translation, this article introduced the development process of machine translation technology from rule-based machine translation(RBMT) to neural machine translation(NMT), and also introduced techniques applied to improve the principles and performance of neural machine translation. There are two types of Classical Chinese machine translation: a translator provided by Baidu and a translator developed by Institute for the Translation of Korea Classics(ITKC). Various errors are found in Baidu translators, but overall translation performance is somewhat higher than that of ITKC translators, and ITKC translators perform better than Baidu translators in certain fields, but lags behind Baidu translators in various types of Classical Chinese translations. In order to improve the performance of Classical Chinese machine translation, it is necessary to build a cloud-based translation platform, expand high-quality parallel corpus, develop additional systems to improve machine translation performance, and develop reliable translation evaluation methods. 이 글은 한문(고전문언문) 기계번역의 현황을 살펴보고, 향후 과제를 제언한 것이다. 기계번역에 대한 이해를 돕기 위해 규칙 기반 기계번역(RBMT: Rule-Based Machine Translation)부터 신경망 기반 기계번역(NMT : Neural Machine Translation)에 이르기까지 기계번역 기술의 발달 과정과 신경망 기계번역의 원리와 성능 향상을 위해 적용되고 있는 기법들을 소개하였다. 고전문언문 기계번역은 바이두에서 제공하는 기계번역기와 한국고전번역원에서 개발한 번역기 2종이 있다. 바이두 번역기는 다양한 오류가 발견되나 전반적인 번역 성능은 한국고전번역원 번역기에 비해 다소 높은 편이며, 한국고전번역원 번역기는 특정 분야에서는 바이두 번역기에 비해 나은 성능을 발휘하지만 다양한 유형의 문언문 번역에서는 바이두 번역기에 비해 뒤지는 성능을 보여준다. 문언문 기계번역의 성능 향상을 위해서는 클라우드 기반 번역 플랫폼 구축, 양질의 병렬코퍼스 확충, 기계번역 성능 향상을 위한 부가 시스템 개발, 신뢰도 높은 번역평가 방식 개발 등이 이루어져야 한다.

외국어 학습 도구로서의 기계 번역: 프랑스어 학습자의 인식 및 태도에 대하여

임순정,임현수 학습자중심교과교육학회 2023 학습자중심교과교육연구 Vol.23 No.22

Objectives The objective of this study was to examine the perceptions and attitudes of undergraduate French language majors about neural machine translation through a survey. Methods The study involved students enrolled in courses such as “French Literature and Translation” and “Advanced French Composition 2” at a university in Seoul. Over the course of a semester, students actively participated in translation and composition exercises aided by machine translation technology and draw up Integrated Problem and Decision Reporting (IPDR) detailing the use of machine translation for problem-solving in the translation process. Subsequently, a survey was conducted to assess the participants' perceptions and attitudes toward machine translation. Results The participants exhibited highly positive responses towards the effectiveness and convenience of using machine translation for learning, highlighting benefits such as time-saving and enhanced learning outcomes. However, the participants displayed a flexible attitude toward the machine-generated results and modified them instead of accepting unconditionally. While statistical analysis did not reveal significant differences in perception and attitude based on enrolled courses or grades, open-ended responses indicated that high-achieving students emphasized the learning benefits and effectiveness of machine translation, whereas lower-achieving students focused on its convenience, particularly in terms of time-saving. Conclusions With the increasing integration of machine translation into foreign language education, understanding how French language majors perceive machine translation can aid educators in devising teaching methods and learning strategies that optimize efficient and effective learning outcomes. This study contributes to the exploration of using machine translation as a tool to enhance language learning.

언어학과 기계 번역-한문학 텍스트의 기계 번역과 관련하여

정성훈 근역한문학회 2019 한문학논집(漢文學論集) Vol.53 No.-

This study examines the history of machine translation in relation to linguistics and briefly introduces algorithms for rule based machine translation and data based machine translation. And the purpose of this study is to propose and prospect a machine translation of Korean texts in classical Chinese. Machine translation is a technology that automatically transforms a language into another language. Recently, it has been actively researched in artificial intelligence(AI) and computational linguistics. Machine translation started with Weaver(1949), and until the early 1980’s, rule based machine translation developed which is a system that applied many rules for vocabulary, grammar, and meaning. Since the 1980’s, the attempts using corpus for machine translation come up with computer development and large corpus construction, then data base machine translation has developed. In recent years, the machine translations with AI has become very popular. One of them is ‘Neural Machine Translation’. However, it is difficult to use rule based machine translation because it is not the agreed-on situation about vocabulary and grammar on Korean literature text in classical Chinese. Also, since there are not enough parallel corpora, it is difficult to use statistical machine translation or neural machine translation. Therefore, the most reasonable method of machine translation on Korean literature text in classical Chinese is to use ‘translation memory’. It is possible to minimize the time and cost for translating the current Korean literature text in classical Chinese, and to generate a large amount of parallel corpus which is required in neural network machine translation in the future. 본 연구는 언어학과 관련하여 기계 번역의 역사를 살펴보고, 규칙 기반 기계 번역과 자료 기반 기계 번역의 알고리듬을 간략히 소개한다. 또한 이를 통해 한문학 텍스트의 기계 번역에 대한 제안과 전망을 하는 것이 본 연구의 목적이다. 기계 번역은 컴퓨터를 이용하여 하나의 언어를 다른 언어로 자동으로 변환하는 기술인데, 최근 인공 지능(AI)과 전산언어학 분야에서 활발히 연구되고 있다. 기계 번역은 Weaver(1949)에서 출발하였으며, 1980년대 초까지 언어학의 영향으로 어휘, 문법, 의미 생성에 필요한 많은 규칙을 적용한 시스템인 규칙 기반 기계 번역이 발전하였다. 1980년대 이후에는 컴퓨터의 발달과 대규모 코퍼스(corpus)의 구축이 가능해지면서 코퍼스를 기계 번역에 이용하려는 시도들이 나타났는데, 코퍼스를 기반으로 하는 자료 기반 기계 번역이 발전하였다. 최근에는 딥러닝(deep learning)을 통한 기계 번역의 인기가 매우 높아지고 있다. 그 중 주목받고 있는 기술은 ‘신경망 기계 번역(Neural Machine Translation)’이다. 그런데 한문학 텍스트는 어휘와 문법에 대한 정확한 정의와 분류가 합의된 상황도 아닐 뿐만 아니라 이를 규칙화한 시스템도 구축하지 못한 상황이기 때문에 규칙 기반 기계 번역을 활용하기 어렵다. 한편 충분한 병렬 코퍼스도 부족하기 때문에 통계적 기계 번역이나 신경망 기계 번역을 활용하기도 어렵다. 따라서 현재 한문학 텍스트의 기계 번역에서 가장 합리적인 방법은 번역 메모리를 활용하는 방법이다. 이를 통해 현재 한문학 텍스트의 번역에 대한 시간과 비용을 최소화 할 수 있고, 향후 신경망 기계 번역에서 필요로 하는 대용량의 병렬 코퍼스를 생성해 낼 수 있을 것이다.

한영 기계 번역에서 ST의 유형적 특징에 따른 번역 오류 분석

박옥수 동아인문학회 2017 동아인문학 Vol.41 No.-

Over the past few years, the technology of machine translation has made remarkable progress from Statistical Machine Translation (SMT) to Neural Machine Translation (NMT). With the introduction of deep learning, the study of machine translation has been performed mainly in the field of computer engineering, which develops systems and algorithms for artificial intelligence. However, the quality and quantity of parallel corpus composed of human translation has a decisive influence on the quality of NMT. In the process of NMT, the model translation that neural network learns is input by human. Also, input of source text is an important factor. The translator must be able to input the correct original text and modify the translation to get excellent quality results in machine translation. Especially, in the case of Korean-English translation, the accuracy of the original text is the main variable for the translation quality. To reduce errors, the translator must enter the correct source text. Inputting source text that contains incorrect spelling, imprecise phrases, and non-grammatical sentences will, of course, increase the probability of error. In the natural language processing, the characteristics of the source language are also the main variables. For example, if the source language is Korean, translation errors may occur depending on the use of the postposition or contextual situation. Given this situation, the accuracy of the source text becomes a major factor in the quality of translation. While the review and revision of the original text is a preprocessing process, evaluating the results of the machine translation and suggesting ways to improve the quality is a postprocessing process. Considering this situation, it is necessary to study the relation between the source text and the error type of machine translation, and it is desirable to perform it in the field of humanities. In this study, we examine human preprocessing and postprocessing issues in Korean-English machine translation. The purpose of this study is to present the error problem of machine translation by analyzing the error in line with the accuracy of the source text, and discussing ways to reduce translation errors. 지난 몇 년 동안 기계번역의 기술은 통계 기계번역(SMT: Statistical Machine Translation)에서 신경망 기계번역(NMT: Neural Machine Translation)으로 괄목할만한 발전을 보여주었다. 딥러닝(deep learning)이 도입되면서 기계번역의 연구는 주로 인공지능의 시스템과 알고리즘을 개발하는 컴퓨터 공학 분야로 제한되었다. 그러나 NMT의 품질은 인간 번역으로 구성된 병렬 코퍼스의 질과 양이 결정적인 영향을 미친다. NMT의 작동 과정에서 신경망이 학습하는 모범 번역은 인간이 투입한다. 그리고 정확한 원문의 입력도 중요한 요소가 된다. 기계번역을 통해 우수한 품질의 결과물을 생산하기 위해서는 번역자가 정확한 원문을 입력하고, 번역물을 수정할 수 있어야 한다. 특히 한영 번역의 경우 원문의 정확도는 번역물 품질에 주요 변수가 된다. 틀린 맞춤법이나 구문, 비문 등이 포함된 원천 텍스트를 입력하면 당연히 오류의 확률이 더 높아진다. 자연언어처리에서 원천 언어의 특징도 주요 변수가 된다. 예를 들면 원천 언어가 한국어인 경우, 조사의 사용이나 언어사용 환경에 따라 번역 오류가 발생할 수 있다. 이런 상황을 고려해 볼 때 기계번역의 오류 유형을 제시하고, 원천 텍스트와의 관련성을 고찰하는 연구가 필요하고, 이는 인문학 분야에서 수행되는 것이 바람직하다. 원문의 검토가 사전 처리 과정이라면 기계번역의 결과물을 평가하고, 품질을 개선하는 방안을 제시하는 것은 후처리 과정이 된다. 이 연구에서는 한영 기계번역에서 인간의 사전 처리와 후처리 문제를 살펴본다. 연구의 목적은 원천 텍스트의 정확도와 연계해서 오류를 항목별로 분석함으로써 기계번역의 오류 문제에 접근하는 방식을 제시하고 오류를 줄이는 방안을 논의하는 데 있다.

한국어 번역을 위한 문화소의 기계번역 연구-중국 외교연설문 번역의 정확도 평가를 중심으로-

임형재,왕첨 대구대학교 다문화사회정책연구소 2018 현대사회와 다문화 Vol.8 No.2

The development of translation machines represented by machine translation and AI translation has recently become a hot topic in the society. Particularly, in recent years, the accuracy and fluency of translation machines have been greatly improved thanks to the use of neural network technology (NMT,Neural Machine Translation), and translation machines have become the main focus of public attention. In addition, the translation machine was previously considered to be dedicated to professional translators because of technical problems, but now the translation machine has already become an accessible tool that people can easily use in daily life. This study will analyze the implementation of machine translation between Chinese and Korean by targeting at Papago, which is a translation machine developed by Korean famous online company Naver that Chinese-speaking learners can easily have access to, and Baidu, which is a well-known translation machine in China. Besides, in order to evaluate the translation accuracy between Chinese and Korean of the famous Google Translate which is currently used worldwide, Google Translate is also included in the study. Ten translators were invited to evaluate the translation results of the machine and in this way the accuracy of the machine translation has been obtained. The analysis objects selected in this study are representative cultural vocabulary arising from language differences: idioms, poems, proverbs and proper nouns, all of which are cultural vocabulary that are the most dependent on the culture in translation. Although the object of evaluation is linguistic expression, considering the contextual information of translation and interpretation, the entire sentence containing cultural vocabulary is presented to the evaluator so that they can realize the environment in which the expression is used. In this study, the evaluation scores, evaluation items, and evaluation criteria of the translation are listed, and the results are obtained from a combination of all these factors. This study reveals the development direction of machine translation and seeks the common development between translators and machines, through the comparison between machine translation and human translation. Therefore, if the translation pattern such as the error pattern of the translation machine can be understood before using, it is expected to further improve the efficiency.

기계번역 활용 한국어 쓰기 수업에서 나타난 학습자 인식과 번역문의 특징

박수진 한국교양교육학회 2023 교양교육연구 Vol.17 No.3

This study examined how Korean learners perceive machine translation in Korean writing class, what the characteristics of machine-translated texts are, and what patterns appear depending on the level of Korean proficiency. Based on these results, this study aimed to suggest how machine translation in Korean writing class would help both of instructors and students. According to a survey of 77 Korean learners, 96% use machine translation and about 90% find it convenient. For beginners, most used machine translation when translating their native language into Korean, while intermediate and advanced learners used machine translation when translating Korean into their native language. Machine translation was mainly used for learning written language. In the second survey of same population, more than 98% of learners recognized that machine translation was convenient but inaccurate, and 97% required that there would be activities to use machine translation which could also provide feedback during class time. In sum, advanced level learners reviewed and modified machine-translated results more carefully than beginners and intermediate level learners, while beginners reviewed and modified less carefully than intermediate and advanced level learners. Thus based on this study, the teaching and learning methods for using machine translation in the writing class were presented as ‘1) finding problems and correcting one’s own language knowledge through self-correction after using machine translation, 2) discovering the differences between one’s mother tongue and Korean through back-translation activities, 3) discovering and using ways to reduce machine translation errors, where 4) the instructor should guide learners to discover cultural elements and provide explicit feedback., discovering various translations according to translation purpose and intention through cooperative activities.’ 이 연구에서는 한국어 쓰기 교양 수업에서 한국어 학습자들이 기계번역에 대해 어떻게 인식하는지, 기계번역을 활용하여 생성한 번역문의 특징은 어떠한지, 한국어 수준에 따라 어떤 양상이 나타나는지에 대해 살펴보았다. 그리고 이를 바탕으로 기계번역 활용 한국어 쓰기 수업의 교수⋅학습 방안을 제시하였다. 한국어 학습자 77명의 설문 조사 결과, 96%가 번역기를 사용하고 있고 90% 정도가 번역기 사용이편리하다고 하였다. 초급 학습자의 경우, 대다수가 모국어를 한국어로 번역할 때 기계번역을 사용하는 반면에 중급, 고급 수준의 학습자들은 한국어를 모국어로 번역할 때 기계번역을 사용하였다. 주로 문어 학습에 기계번역을 많이 사용하였다. 2차 설문 조사 결과, 98% 이상의 학습자들이 기계번역사용은 편리하나 부정확한 것으로 인식하였고 97%의 학습자들이 수업 시간에 기계번역을 사용하고 피드백하는 활동이 있기를 희망하였다. 초급 수준의 학습자보다 중급 학습자가, 그리고 중급보다 고급학습자가 기계번역 결과물을 더 세밀하게 검토하고 수정하는 것을 알 수 있었다. 초급 학습자들은 자기 수정 빈도가 가장 적게 나타났으며 기계번역 결과를 그대로 수용하는 경향이 강했으며 수정 시 문체에 집중하는 것을 알 수 있었다. 중급 학습자는 자기 수정의 성공률과 실패율이 고급 수준의 학습자와 큰 차이가 나타나지 않았으나 고급 수준의 학습자들은 기계번역 산출물의 대부분을 수정하였고 문장을 재구성하여 의미를 상세히 나타내려고 하였다. 이를 바탕으로 기계번역 활용 수업의 교수⋅학습 방안을 ‘1) 기계번역 활용 후 자기 수정을 통해 문제점을 찾고 자신의 언어지식 수정하기, 2) 역번역활동을 통해 학습자 스스로 모국어와 한국어의 차이 발견하기, 3) 기계번역 오류를 줄이는 방법을 발견하고 사용하기, 이때 4) 교수자는 특히 문화적 요소를 학습자들이 발견하도록 유도하고 명시적 피드백하기, 5) 협력 활동을 통해 번역 목적, 의도에 따른 다양한 번역문 발견하기’로 제시하였다.