• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.80-85
• /
• 2023

기계 번역에서 품질 검증은 정답 문장 없이 기계 번역 시스템에서 생성된 번역의 품질을 자동으로 추정하는 것을 목표로 한다. 일반적으로 이 작업은 상용화된 기계 번역 시스템에서 후처리 모듈 역할을 하여 사용자에게 잠재적인 번역 오류를 경고한다. 품질 검증의 하위 작업인 치명적인 오류 탐지는 번역의 오류 중에서도 정치, 경제, 사회적으로 문제를 일으킬 수 있을 만큼 심각한 오류를 찾는 것을 목표로 한다. 본 논문은 치명적인 오류의 유무를 분류하는 것을 넘어 문장에서 치명적인 오류가 존재하는 부분을 제시하기 위한 새로운 데이터셋과 모델을 제안한다. 이 데이터셋은 거대 언어 모델을 활용하는 구축 방식을 채택하여 오류의 구체적인 범위를 표시한다. 또한, 우리는 우리의 데이터를 효과적으로 활용할 수 있는 다중 작업 학습 모델을 제시하여 오류 범위 탐지에서 뛰어난 성능을 입증한다. 추가적으로 언어 모델을 활용하여 번역 오류를 삽입하는 데이터 증강 방법을 통해 보다 향상된 성능을 제시한다. 우리의 연구는 기계 번역의 품질을 향상시키고 치명적인 오류를 줄이는 실질적인 해결책을 제공할 것이다.

• The KIPS Transactions:PartA
• /
• v.15A no.2
• /
• pp.125-134
• /
• 2008

We implemented a system being able to suggest example sentences for multiple choice tests, considering the level of students. To build the system, we designed an automatic method for sentence generation, which made it possible to control the difficulty degree of questions. For the proper evaluation in the multiple choice tests, proper size of question pools is required. To satisfy this requirement, a system which can generate various and numerous questions and their example sentences in a fast way should be used. In this paper, we designed an automatic generation method using a linguistic resource called WordNet. For the automatic generation, firstly, we extracted keywords from the existing sentences with the morphological analysis and candidate terms with similar meaning to the keywords in Korean WordNet space are suggested. When suggesting candidate terms, we transformed the existing Korean WordNet scheme into a new scheme to construct the concept similarity matrix. The similarity degree between concepts can be ranged from 0, representing synonyms relationships, to 9, representing non-connected relationships. By using the degree, we can control the difficulty degree of newly generated questions. We used two methods for evaluating semantic similarity between two concepts. The first one is considering only the distance between two concepts and the second one additionally considers positions of two concepts in the Korean Wordnet space. With these methods, we can build a system which can help the instructors generate new questions and their example sentences with various contents and difficulty degree from existing sentences more easily.

• Annual Conference on Human and Language Technology
• /
• 2022.10a
• /
• pp.10-15
• /
• 2022

온라인상에서 발생하는 혐오 표현은 사회가 직면한 주요 문제 중 하나이다. 이러한 필요성에 입각해, 최근 인공지능을 활용하여 발화에 대한 교화 목적을 가진 대응 발화 쌍을 통해 혐오 표현에 대한 실질적인 완화를 진행하는 연구들이 생겨나고 있다. 그러나 각 혐오 표현에 적합한 대응 발화의 구축은 다수의 전문 인력이 요구되므로 데이터를 구축함에 있어 시간과 비용이 많이 소요되며 대응 발화 생성 또한 어려운 문제로 여겨진다. 해당 문제를 완화하기위해, 본 논문은 사전에 기 구축되어 있는 혐오 표현 데이터를 기반으로 의미 기반 검색을 적용하여 자동으로 데이터를 증강할 수 있는 쉽고 빠른 데이터 증강 방법론을 제안한다. 제안하는 프로세스의 타당성과 증강된 문장의 효과를 검증하기 위해 사전학습 모델을 기반으로 비교 실험을 진행하였다. 실험 결과, 제안하는 프로세스를 적용하였을 시, 그렇지 않은 모델 대비 높은 폭의 성능 향상을 보였다.

• Design Convergence Study
• /
• v.15 no.6
• /
• pp.73-92
• /
• 2016

As one of empirical research and developments on BIM applications to improve design quality and productivity of building, efforts have been devoted to automated compliance checking of building design for building permit. KBimLogic is a mechanism that translate Korean Buidling Act to the computational language called KBimCode. KBimCode aims to standardized and neutral language that can be applied to various design rules. This paper focuses on testing expandability of KBimCode by appling it on Chinese Building Act. We analyzed Chinese national regulation on fire protection and evacuation, based on 1) Object·property, 2) function for predicate processing, 3) relationship of sentences. As a result, Chinese Building Act were successfully translated to KBimCode with some important implications for further application. Based on the finding of the paper, KBimCode is expected to be applicable to kinds of design rules.

• Information Systems Review
• /
• v.25 no.2
• /
• pp.125-144
• /
• 2023

Generative Artificial Intelligence (AI) technology is gaining global attention as it can automatically generate sentences, images, and voices that humans previously generated. In particular, ChatGPT, a representative generative AI service, shows proactivity and accuracy differentiated from existing chatbot services, and the number of users is rapidly increasing in a short period of time. Despite this growing interest in generative AI services, most preceding studies are still in their infancy. Therefore, this study utilized LDA topic modeling and keyword network diagrams to derive success factors for generative AI services and to propose successful business strategies based on them. In addition, using ChatGPT, a new research methodology that complements the existing text-mining method, was presented. This study overcomes the limitations of previous research that relied on qualitative methods and makes academic and practical contributions to the future development of generative AI services.

• Journal of the Korean Society for information Management
• /
• v.29 no.4
• /
• pp.191-208
• /
• 2012

Two fundamental aspects of speech summary generation are the extraction of key speech content and the style of presentation of the extracted speech synopses. We first investigated whether acoustic features (speaking rate, pitch pattern, and intensity) are equally important and, if not, which one can be effectively modeled to compute the significance of segments for lecture summarization. As a result, we found that the intensity (that is, difference between max DB and min DB) is the most efficient factor for speech summarization. We evaluated the intensity-based method of using the difference between max-DB and min-DB by comparing it to the keyword-based method in terms of which method produces better speech summaries and of how similar weight values assigned to segments by two methods are. Then, we investigated the way to present speech summaries to the viewers. As such, for speech summarization, we suggested how to extract key segments from a speech video efficiently using acoustic features and then present the extracted segments to the viewers.

• Journal of Korean Library and Information Science Society
• /
• v.47 no.4
• /
• pp.289-307
• /
• 2016

This paper introduces a software system and process model for constructing domain-specific relation extraction datasets semi-automatically. The system uses a set of terms such as genes, proteins diseases and so forth as inputs and then by exploiting massive biological interaction database, generates a set of term pairs which are utilized as queries for retrieving sentences containing the pairs from scientific databases. To assess the usefulness of the proposed system, this paper applies it into constructing a genic interaction dataset related to Alzheimer's disease domain, which extracts 3,510 interaction-related sentences by using 140 gene names in the area. In conclusion, the resulting outputs of the case study performed in this paper indicate the fact that the system and process could highly boost the efficiency of the dataset construction in various subfields of biomedical research.

• Journal of Digital Convergence
• /
• v.19 no.7
• /
• pp.199-208
• /
• 2021

Automatic Post Editing(APE) is the study that automatically correcting errors included in the machine translated sentences. The goal of APE task is to generate error correcting models that improve translation quality, regardless of the translation system. For training these models, source sentence, machine translation, and post edit, which is manually edited by human translator, are utilized. Especially in the recent APE research, multilingual pretrained language models are being adopted, prior to the training by APE data. This study deals with multilingual pretrained language models adopted to the latest APE researches, and the specific application method for each APE study. Furthermore, based on the current research trend, we propose future research directions utilizing translation model or mBART model.

• The Journal of the Acoustical Society of Korea
• /
• v.20 no.3
• /
• pp.24-33
• /
• 2001

this paper describes a baseline for an implementation of a corpus-based Korean TTS system. The conventional TTS systems using small-sized speech still generate machine-like synthetic speech. To overcome this problem we introduce the corpus-based TTS system which enables to generate natural synthetic speech without prosodic modifications. The corpus should be composed of a natural prosody of source speech and multiple instances of synthesis units. To make a phone level synthesis unit, we train a speech recognizer with the target speech, and then perform an automatic phoneme segmentation. We also detect the fine pitch period using Laryngo graph signals, which is used for prosodic feature extraction. For break strength allocation, 4 levels of break indices are decided as pause length and also attached to phones to reflect prosodic variations in phrase boundaries. To predict the break strength on texts, we utilize the statistical information of POS (Part-of-Speech) sequences. The best triphone sequences are selected by Viterbi search considering the minimization of accumulative Euclidean distance of concatenating distortion. To get high quality synthesis speech applicable to commercial purpose, we introduce a domain specific database. By adding domain specific database to general domain database, we can greatly improve the quality of synthetic speech on specific domain. From the subjective evaluation, the new Korean corpus-based TTS system shows better naturalness than the conventional demisyllable-based one.

• Annual Conference on Human and Language Technology
• /
• 2017.10a
• /
• pp.317-320
• /
• 2017

개체명 인식은 자연어 문장에서 장소, 제작물, 사람 등 분류를 통한 의미 부여가 가능한 단어를 파악하는 기술로서 의미 분석을 위한 핵심 기술이다. 현재 많은 개체명 분석 관련 연구들은 형태소 분석 결과에 의존적인 형태를 갖고 있어서, 형태소 분석 결과의 정확성이 개체명 분석 결과의 성능에 영향을 미치고 있다. 본 연구에서는 형태소 분석 과정을 거치지 않는 음절 기반의 개체명 분석 기술을 제안하여 형태소 분석의 정확도가 낮은 통신어, 신조어 분석 성능을 향상하였다. 또한, 자동화된 방법으로 음절 단위 개체명 말뭉치 및 개체명 사전을 구축하는 프로세스를 정의하여 개체명 분석의 정확도 향상 및 인지 범주의 확대를 도모하였다. 본 연구에서 제안한 개체명 인식 기술은 한국어 개체명 표준에 기반한 129가지의 개체명 분류가 가능하며, 이는 자연어 처리 기술이 필요한 산업계에서 상용화하는데 큰 기여를 할 것으로 판단된다.