전자 술어사전 구축을 위한 ‘만족’과 ‘불만’의 형태ㆍ통사적 기술

남지순(Nam, Jeesun) 한국사전학회 2013 한국사전학 Vol.- No.21

This study aims at describing morpho-syntactic properties of mancok (satisfaction) and pwulman (dissatisfaction) for building an electronic dictionary of Korean predicates, notably predicates of sentiment expressions. The nouns mancok/pwulman are not only significant in the network construction of general sentiment nouns, but also essential in the classification of subjective sentences such as positive and negative ones in opinion mining research area. To begin with, representative Korean dictionaries have been consulted and large-scale corpora have been analyzed to provide linguistic properties of these nouns such as derivational particularities, contextual collocations and syntactic structures. Although the derivational aspects of these nouns are relatively poor, they occur in a complex set of sentence types: the predicates made of the verbs or adjectives derived from these nouns may constitute 8 syntactic structures according to the argument types and noun distributions; the predicates composed of these nouns accompanied by other verbs or adjectives can be classified into 7 syntactic classes. On the basis of this morpho-syntactic description, a formal method named Local-Grammar Graph (LGG) model has been introduced to transform the linguistic description into an electronic database directly applicable to the automatic text processing.

오피니언 극성 결정에 관여하지 않는 대립관계 감성 관용구 연구

남지순 ( Jeesun Nam ) 한국외국어대학교 언어연구소 2016 언어와 언어학 Vol.0 No.70

This paper aims at describing the Opinion-Free Frozen expressions (OFFs) composed of two contrastive sentiment keywords. Although sentiment words occur in OFFs, they do not determine the polarity of the given sentences, which is a problematic issue for the current opinion classification systems that basically work on the detection of sentiment keywords in the sentence. As a matter of fact, OFFs consist of two predicative sequences ending with the repeated inflectional suffixes. In this study, 93 inflectional suffixes are obtained from the list of 360 elements in the Korean electronic lexicon DECO and classified into 4 sub-categories.

영화평 문서의 오피니언 자동 분류를 위한 언어학적 패턴 연구

남지순(Jeesun Nam) 사단법인 한국언어학회 2010 언어학 Vol.0 No.58

In this study, we describe the linguistic characteristics of online review texts on Movie in order to develop an efficient opinion evaluation system of online documents: we analyze especially lexico-semantic features such as positive keywords or negative keywords, syntactic properties such as negative constructions, distribution of collocated nouns, conditional constructions, comparative constructions, and some frozen expressions conveying a positive or negative polarity. We underline the current approach to opinion mining using mostly statistical methods should be reinforced with a more serious linguistic description of review texts. We introduce a finite grammar named LGG(Local-Grammar Graph) which allows to formalize linguistic description into finite-state graph equivalent to FST(finite-state transducer): this task can be performed in an efficient way by the support of the UNITEX developed for this purpose. We present a result of the application of the LGG-MOVIE which represents a part of the linguistic patterns we discussed in this paper on a new sub-set of review texts on Movie.

한국어 활용형 전자사전 구축을 위한 조사 복합형 구성단위에 대한 연구

남지순(Nam, Jeesun) 한국사전학회 2014 한국사전학 Vol.- No.24

Postpositions that are attached to nouns in Korean sentences have been named by various terms and classified into several sub-types according to scholars’ theoretical viewpoints. Nonetheless, relatively little work has been devoted to providing an exhaustive nomenclature of postpositions and an empirical classification of these units. In order to construct an electronic dictionary of all surface forms in Korean, a complete list of simple units that constitute complex sequences of postpositions is required as well as a complete list of restrictions in their combinatorial properties. This study has been conducted to build an accurate list of simple postpositions that would be indispensable to the implementation of a dictionary of complex forms of postpositions. In this study, 142 simple postpositions have been discerned and classified into 4 major classes. These 4 classes have been sub-classified into 22 types, which will be indicated as POS tags in an electronic dictionary, and therefore, marked as the output of morphological analysis.

DECO 한국어전자사전에 기반한 외국인 학습자용 한국어 활용형사전 구축 방법론 연구

윤소은(Yun, Soeun),남지순(Nam, Jeesun) 한국사전학회 2020 한국사전학 Vol.- No.36

This study aims to introduce a methodology for building a Korean learners’ inflectional dictionary that overcomes the limitations of the existing approaches based on the frequent inflectional patterns in corpus. In this study, two main tasks were conducted: ‘simplifying’ and ‘restructuring’ the inflected entries represented in the Korean Electronic dictionary DECO. In this way, a BASIC learners’ dictionary and an EXPANDED one were proposed. For the former, three criteria were considered: the number of morphemes in inflectional endings, the combination of different types of morphemes and the complex endings that appear at low frequency. For the latter, there were also three criteria: colloquial expressions, endings that are particular to social media texts and the contracted forms of morphemes. The result was represented as a finite-state transducer like DECO inflectional endings, but in a human-friendly form. By consulting this dictionary, learners may understand and generate Korean inflectional forms in a more systematic way.

이커머스 후기글 평가분석 트리플에 기반한 평가주석 데이터셋 EVAD 구축 방법론

최수원(Choi, Suwon),남지순(Jeesun Nam) 언어과학회 2021 언어과학연구 Vol.- No.99

This study aims to introduce a methodology of building evaluation-annotated datasets based on the evaluation-triple schema that we extracted from the review texts of fashion e-commerce apps. The evaluation triple defined in this study consists of ‘TARGET’, ‘ASPECT’, and ‘VALUE’. We classified the pair {ASPECT-VALUE} into 35 categories, which were characterized as three categories: ‘Information-, Judgment- and Suggestion- related" types. The triple elements were represented under a set of Local Grammar Graphs based on the domain-specific dictionary DECO-DOM. By applying these linguistic resources to a target review text, we generated a large-scale annotated dataset. In this study, the Semi-automatic Symbolic Propagation (SSP) methodology proposed by Nam (2021) was adopted to generate the EVAD dataset of about 200,000 review texts. The evaluation of the SSP approach shows the performance of the F1- Score 0.91, which confirms the reliability of the process that we propose in this study.

오피니언 마이닝 연구를 위한 정치기사 댓글의 ‘아니다’ 부정 의문문 분석

신동혁(Shin, Donghyok),남지순(Jeesun Nam) 언어과학회 2016 언어과학연구 Vol.0 No.78

This study analyzes the sentences where a negator anita ‘not be’ is not used to convert the opinion polarity of the main predicative elements when it appears in certain types of interrogatives in Korean political news reply texts. This study mainly provides the following findings: First, from morphological viewpoints, a typical interrogative suffix is used: -neunka takes about 54.4%. Second, from syntactic viewpoints, certain abstract nouns or bound nous such as kes without any postpositions occur most frequently with anita. Finally, from semantic viewpoints, the interrogatives based on anita express mostly a negative opinion, which suggests that this negator occurs with negative nouns and emphasizes that negative opinion. We suggested that anita occurring in these contexts is not used as a PSD (Polarity-Shifting Devices), but a rhetorical marker, which should be properly recognized in opinion mining.

SNS텍스트의 오피니언 마이닝을 위한 언어자원 기반 감성 분석 플랫폼 연구

유광훈 ( Yoo Gwanghoon ),남지순 ( Nam Jeesun ) 서강대학교 언어정보연구소 2017 언어와 정보 사회 Vol.32 No.-

This study aims to introduce DecoTex, a language resource-based sentiment analysis platform, implemented for opinion mining of social media texts. DecoTex supports several functions such as the Twitter crawler, Preprocessing module, and Sentiment Analysis modules. The Sentiment Analysis modules consist of two parts: Supervised Machine Learning options requiring Sentiment- Annotated Corpora as training data and Lexicon-based algorithmic options based on Sentiment Lexica and Local Grammars. By illustrating a process of classifying positive/negative opinions on ‘China’ and ‘Japan’ in Twitter texts through DecoTex platform, this study emphasizes the importance of a computational platform for humanities researchers. We believe that it is crucial to free them from making efforts to learn programming skills for obtaining experimental results or evaluating their studies since they may focus on constructing linguistic resources that require an enormous amount of time, energy, and knowledge.

한국어 감성 사전 DecoSelex 구축을 위한 영어 SentiWordNet 활용 및 보완 논의

신동혁(Shin, Donghyok),조동희(Cho, Donghee),남지순(Nam, Jeesun) 한국사전학회 2016 한국사전학 Vol.- No.28

In this study, we present the Korean Sentiment Lexicon DecoSelex that we constructed for the sentiment analysis of userᐨgenerated texts. This study starts with an examination of the results of Google translation of English SWN into Korean sentiment word lists. Among these Korean words, nonᐨtranslated words or irrelevant ones are observed, and therefore, after the manual elimination of these inappropriate terms, we obtained 3,665 candidates for Korean sentiment words. However, the nouns, being important in number in this list, are mostly medical disease names, and the adjectives, the most significant part of speech in general in the sentiment lexicon, appear extremely small in number. This situation led us to examine the DECO Korean electronic dictionary, a more sizable and reliable electronic resource for Korean. We obtained 35,452 sentimentᐨrelated words based on the DECO entries and classified them into several subᐨclasses according to their polarity properties and psychological meanings. For evaluating our sentiment lexicon DecoSelex, a dataset of 1,200 review texts of IT products was used. The current version of DecoSelex showed 83.6% precision & 83.2% recall for the positive words, and 72.9% precision & 69.1% recall for the negative words.

Building Chinese Request Pattern Graphs for Chinese-Korean Translation

Jiabin Li(이가빈),Changhoe Hwang(황창회),Jeesun Nam(남지순) 언어과학회 2020 언어과학연구 Vol.0 No.95

This paper aims to describe diverse request patterns in Chinese for the automatic classification and the Chinese-Korean translation of music-request sentences in AI assistant platforms. These patterns are formalized by Local-Grammar Graphs (LGGs), a finite-state grammar, and represented under directed acyclic graphs based on the main function categories. The UNITEX graph editor has been used to construct Chinese LGGs, and the request pattern tags that are proposed for Korean sentences based on DECO Korean Electronic dictionaries was applied. The result of this study has been examined and evaluated by comparison with Google Translate results at the end of the paper. Experimental results prove that the LGG grammar constructed in this paper can effectively cover and annotate various request patterns. This will provide the possibility to generate a large-scale bilingual annotation corpus in the field of music listening.