• Journal of KIISE
• /
• v.42 no.4
• /
• pp.475-482
• /
• 2015

Developing a high-performance Semantic Role Labeling (SRL) system for a domain requires manually annotated training data of large size in the same domain. However, such SRL training data of sufficient size is available only for a few domains. Performances of Korean SRL are degraded by almost 15% or more, when it is directly applied to another domain with relatively small training data. This paper proposes two techniques to minimize performance degradation in the domain transfer. First, a domain adaptation algorithm for Korean SRL is proposed which is based on the prior model that is one of domain adaptation paradigms. Secondly, we proposed to use simplified features related to morphological and syntactic tags, when using small-sized target domain data to suppress the problem of data sparseness. Other domain adaptation techniques were experimentally compared to our techniques in this paper, where news and Wikipedia were used as the sources and target domains, respectively. It was observed that the highest performance is achieved when our two techniques were applied together. In our system's performance, F1 score of 64.3% was considered to be 2.4~3.1% higher than the methods from other research.

• Smart Media Journal
• /
• v.12 no.2
• /
• pp.56-65
• /
• 2023

In broadcasts such as news and coverage programs, voice is modulated to protect the identity of the informant. Adjusting the pitch is commonly used voice modulation method, which allows easy voice restoration to the original voice by adjusting the pitch. Therefore, since broadcast voice modulation methods cannot properly protect the identity of the speaker and are vulnerable to security, a new voice modulation method is needed to replace them. In this paper, using the Lightweight speech de-identification model as the evaluation target model, we compare speech de-identification performance with broadcast voice modulation method using pitch modulation. Among the six modulation methods in the Lightweight speech de-identification model, we experimented on the de-identification performance of Korean speech as a human test and EER(Equal Error Rate) test compared with broadcast voice modulation using three modulation methods: McAdams, Resampling, and Vocal Tract Length Normalization(VTLN). Experimental results show VTLN modulation methods performed higher de-identification performance in both human tests and EER tests. As a result, the modulation methods of the Lightweight model for Korean speech has sufficient de-identification performance and will be able to replace the security-weak broadcast voice modulation.

• The Journal of the Korea Contents Association
• /
• v.12 no.12
• /
• pp.479-488
• /
• 2012

The study examined acculturation patterns of Korean-Americans living in Portland, USA and their relationship to the consumption of media contents. Under the academic guidance of Berry's acculturation model, the four strategies of acculturation model were extensively inquired that immigrants and sojourner would adopt in host culture: Integration, Assimilation, Separation and Marginalization. Furthermore, how differently the four acculturation modes are related to media contents was examined. For this study, the survey questionnaires were administered among Korean Americans in Portland, USA. The total 371 data were collected and analyzed. The result indicated that 1) they preferred integration(M=4.57) and separation modes(M=4.49) over assimilation (M=2.92); 2) as separation tendency increased, the preference of Korean contents over American contents increased while as assimilation tendency increased, the preference of American contents over Korean contents; 3) high assimilation persons consumed more news than entertainments in American media contents. Compared to it, high separation persons used more entertainments than news in American media contents and more news than entertainments in Korean media contents.

• Annual Conference on Human and Language Technology
• /
• 2009.10a
• /
• pp.145-150
• /
• 2009

이 논문은 한글 뉴스 기사의 댓글에 대한 감정 분류 방법을 제안한다. 제안된 방법은 기계학습을 이용하는데 본 논문에서는 자질의 가중치를 재조정하는 좀 색다른 방법을 제안한다. 일반적으로 댓글은 독자들이 특정 기사에 대해서 어떠한 감정을 가지고 있는지를 파악하는 중요한 단서가 된다. 그런데 독자들의 감정은 가사에 어떤 분야에 속하느냐에 영향을 받는다. 예를 들면 정치 기사는 부정적인 댓글은 많이 포함하고 있으며 인물 기사는 긍정적인 기사를 많이 포함한다. 이 논문은 이와 같은 댓글의 속성을 이용해서 기사의 원문과 기사의 분야 정보를 이용하여 가중치를 조정한다. 제안된 시스템의 성능을 평가하기 위해 신문 기사와 댓글을 수집하여 감정 말뭉치를 구축하였으며 감정자질을 추출하기 위해 감정 사전을 구축하였다. 제안된 시스템의 $F_1$ 척도는 92.2%였으며 원문의 감정 단어와 분야 정보가 댓글의 감정을 분류하는데 중요한 자질임을 알 수 있었다.

• Annual Conference on Human and Language Technology
• /
• 2016.10a
• /
• pp.243-245
• /
• 2016

광학 문자 인식(OCR)을 통해 문서의 글자를 인식할 때 띄어쓰기 오류가 발생한다. 본 논문에서는 이를 해결하기 위해 OCR의 후처리 과정으로 동적 프로그래밍을 이용한 분절(Segmentation) 방식의 띄어쓰기 오류 교정 시스템을 제안한다. 제안하는 시스템의 띄어쓰기 오류 교정 과정은 다음과 같다. 첫째, 띄어쓰기 오류가 있다고 분류된 어절 내의 공백을 모두 제거한다. 둘째, 공백이 제거된 문자열을 동적 프로그래밍을 이용한 분절로 입력 문자열에 대하여 가능한 모든 띄어쓰기 후보들을 찾는다. 셋째, 뉴스 기사 말뭉치와 그 말뭉치에 기반을 둔 띄어쓰기 확률 모델을 참조하여 각 후보의 띄어쓰기 확률을 계산한다. 마지막으로 띄어쓰기 후보들 중 확률이 가장 높은 후보를 교정 결과로 제시한다. 본 논문에서 제안하는 시스템을 이용하여 OCR의 띄어쓰기 오류를 해결할 수 있었다. 향후 띄어쓰기 오류 교정에 필요한 언어 규칙 등을 시스템에 추가한 띄어쓰기 교정시스템을 통하여 OCR의 최종적인 인식률을 향상에 대해 연구할 예정이다.

• Korean Linguistics
• /
• v.72
• /
• pp.199-225
• /
• 2016

This study has two main purposes: to establish a list of source domains in the metaphors for Middle East respiratory syndrome (MERS) and to uncover ideological meanings embedded in them in Korean news reports from KBS and JTBC. The first part of this study presents metaphors such as [MERS IS WAR], [MERS IS WAVE], [MERS IS A LIVING THING], and [MERS IS A THING], which were found in the data. The latter part of this study deals with how the two broadcasting companies use these metaphors differently according to their ideologies. In the metaphor of [MERS IS WAR], KBS tends to show less of the agents who controls the war since the war against MERS has failed which casts responsibility to the controlling agents, the government and big hospitals. In this, KBS tries to present less of the information of the responsible agents that presented in JTBC. Through the metaphor of [MERS IS WAVE], KBS presents the aftermath of MERS as something not serious. Compared to JTBC, KBS tends to suggest that the aftermath of MERS is predominantly an economic effects by metaphorically suggesting that predominantly the economic sector got hit by MERS.

• KIISE Transactions on Computing Practices
• /
• v.23 no.3
• /
• pp.183-187
• /
• 2017

Smart phone users prefer fast reading and texting. Hence, users frequently use abbreviated sequences of words and phrases. Nowadays, abbreviations are widely used from chat terms to technical terms. Therefore, gathering abbreviations would be helpful to many services, including information retrieval, recommendation system, and so on. However, manually gathering abbreviations needs to much effort and cost. This is because new abbreviations are continuously generated whenever a new material such as a TV program or a phenomenon is made. Thus it is required to generate of abbreviations automatically. To generate Korean abbreviations, the existing methods use the rule-based approach. The rule-based approach has limitations, in that it is unable to generate irregular abbreviations. Another problem is to decide the correct abbreviation among candidate abbreviations generated rules. To address the limitations, we propose a method of generating Korean abbreviations automatically using sequence-to-sequence learning in this paper. The sequence-to-sequence learning can generate irregular abbreviation and does not lead to the problem of deciding correct abbreviation among candidate abbreviations. Accordingly, it is suitable for generating Korean abbreviations. To evaluate the proposed method, we use dataset of two type. As experimental results, we prove that our method is effective for irregular abbreviations.

• The Journal of the Acoustical Society of Korea
• /
• v.24 no.2
• /
• pp.117-126
• /
• 2005

This study investigates the prosodic characteristics of Korean through the analysis of a large database. One female and one male speakers each read 650 sentences and they were segmentally and prosodically labeled. Statistical analyses were done on these utterances regarding the tonal pattern and the size of prosodic units, correlation between the size of higher level prosodic units and the number of lower level prosodic units. and the slope and F0 of the falling and rising contours of an accentual phrase. The results showed that the duration and the number of words and syllables of a prosodic unit were significantly different not only between speakers but also between its positions within a higher level prosodic nit. The munber of a prosodic unit showed a high correlation with the duration and the number of syllables of its higher level units. The slope of the falling contour within an accentual phrase was inversely Proportional to the number of its syllables. The slope was different depending on the first tone type of an accentual phrase, which could be explained with the F0 rising and the different amount of rising between tones when an accentual phrase starts with an H tone. The slope of the falling contour across an accentual phrase boundary showed a constant and larger value compared to one within an accentual phrase. The rising contours in the beginning and end of an accentual Phrase were similar in their slopes but they differ in the amount of F0 change : the former showed a larger amount of change. The slope of the rising contour which forms an accentual Phrase on its own was inversely Proportional to the number of its syllables.

• Journal of Korea Entertainment Industry Association
• /
• v.13 no.8
• /
• pp.131-142
• /
• 2019

This paper is intended to examine and analyze the new Chinese words as the result of culture content. The development of the Korean entertainment industry has created a Korean wave around the world. Through this, many Korean words, Internet vocabulary, and cultural concepts have begun to enter China. Among them, there are many new words that have appeared on the Chinese Internet due to the culture content. As the number of Korean fans and Korean learners increases, new words on the Internet are widely used. The new Chinese words, which are influenced by Korean cultural content, are considered an important part of new Chinese vocabulary. To accurately recognize and understand this, first of all six categories of the new Chinese words were analyzed, which were figurative meaning, substitution, loan of foreign words, abbreviation, compound word, derivation. This formulation also works on the Chinese words with the influence of cultural content. There are three types of the Internet new words form Korean cultural. Which were new words in Chinese characters, new words in alphabets, extended meanings. And had analyzed new words through the acquisition of new meanings. Also took specific news titles and songs according to each category. Through new Chinese words, The influence of cultural content had been confirmed. It is expected that these new Chinese words enrich Chinese vocabulary, also help to facilitate communication. And these new Chinese words are often used in public media or in everyday life. We should recognize the existence of these new Chinese words, and have an accurate perception of them.

• Korean Journal of Cognitive Science
• /
• v.26 no.4
• /
• pp.435-452
• /
• 2015

Recent researches on Entity Linking(EL) have attempted to disambiguate entities by using a knowledge base to handle the semantic relatedness and up-to-date information. However, EL for tweets using a knowledge base is still unsatisfactory, mainly because the tweet data are mostly composed of short and noisy contexts and real-time issues. The EL system the present work builds up links ambiguous entities to the corresponding entries in a given knowledge base via exploring the news articles and the user history. Using news articles, the system can overcome the problem of Wikipedia coverage (i.e., not handling real-time issues). In addition, given that users usually post tweets related to their particular interests, the current system referring to the user history robustly and effectively works with a small size of tweet data. In this paper, we propose an approach to building an EL system that links ambiguous entities to the corresponding entries in a given knowledge base through the news articles and the user history. We created a dataset of Korean tweets including ambiguous entities randomly selected from the extracted tweets over a seven-day period and evaluated the system using this dataset. We use accuracy index(number of correct answer given by system/number of data set) The experimental results show that our system achieves a accuracy of 67.7% and outperforms the EL methods that exclusively use a knowledge base.