한국어 의미역 기술 모형 개선을 위한 연구 - Korean PropBank 프레임파일의 구조 개선을 중심으로 -

신서인 한국어의미학회 2020 한국어 의미학 Vol.70 No.-

The purpose of this paper is to analyze the structure of the Korean PropBank framefiles and suggest several ways to improve it. In the paper, we discuss the features and limitations of the Korean PropBank framefiles and propose an alternative model for semantic role description. In terms of macro structure, the Korean PropBank framefiles have a peculiar structure. The lemma representing several words becomes the entry, and the related vocabulary is described in one file. Therefore, the lemma setting criteria and related vocabulary processing criteria should be clearly established, and homonyms should be processed in a consistent way. In terms of micro structure, there is no place in the Korean PropBank framefiles to describe postposition(josa) information, which is essential for Korean semantic role labeling. Therefore, this lack has to be supplemented. In describing the sematic role information, it is more useful to describe it as a label-type sematic role rather than as a number-type sematic role or a narrative-type sematic role. In the Korean PropBank framefiles, arguments and adjuncts are analyzed in different ways in the analysis of example sentence, but they need to be analyzed parallelly. In order to take care of the suggestions for improvement, we propose an alternative model for Korean semantic role description. This new model will be able to describe information that is useful for a more effective analysis of Korean semantic role.

네트워크 분석을 이용한 복지 담화 연구

신서인 한림대학교 한림과학원 2017 개념과 소통 Vol.0 No.20

본고는 네트워크 분석방법을 이용하여 복지 담화의 변화양상을 분석하는 것을목적으로 한다. 1992년부터 2012년에 이르기까지 5년 단위로 대통령 선거가 치러지기 직전 한 달 동안의 신문기사 중 ‘복지’를 포함한 기사를 수집하고 네트워크 분석을 이용하여 복지 담화에 등장한 단어들의 관계를 파악하였다. 복지 담화는 20여 년에 걸쳐 일정한 변화양상을 보였다. 첫째, 노인·장애인 복지에서 여성·아동 복지로 주로 논의되는 복지대상의 변화가 감지되었다. 둘째, 복지가 경제성장 및 일자리와 관련하여 논의되게 되었다. 본고에서는 네트워크 분석방법의 의의와 한계를 밝히는 데에도 힘을 기울였다. 네트워크 분석은 단어 간 관계를 시각화하여 보여 주며 검색어와 공기어의 관계뿐 아니라 공기어 사이의 관계를 보여 준다는 점에서 유용한 면이 있었다. 그러나 코퍼스언어학에서 발달해 있는 다른 연구방법들, 즉 키워드 분석·연어 분석·클러스터 분석·용례 분석 등과 비교할 때 맥락에 대한 고려가 상대적으로 적고, 해당 단어의 전체적인 분포에 대한 고려가 적다는 것이 한계로 드러났다. 코퍼스언어학의 연구방법과 네트워크 분석방법을 결합한다면 대규모 코퍼스를 대상으로 한 담화 분석을 효율적으로 수행할 수 있을 것이다. This paper examines changes in the welfare discourse, using network analysis. Newspaper articles containing the word bokji (welfare) were collected which had been published during the presidential election held every five years between 1992 and 2012. Analysis of related words shows two ways in which the welfare discourse was changing over this 20 year period. Firstly, there is a shift in the main beneficiaries of welfare from the disabled and the elderly to women and children. Secondly, there is a increasing tendency for the issue of welfare to be discussed in relation to economic growth and jobs. The paper also demonstrates the strengths and weaknesses of network analysis when applied to discourse analysis. Thus,network analysis is useful in revealing connections between the search word and its collocates, but also shows linkages between the collocates, through visualization. In comparison with other well-established methods of analysis in corpus linguistics, however, such as keyword, collocate, cluster, and concordance analyses, the major limitation of network analysis is that it provides relatively little information about the context in which a given word is used nor about its overall distribution. Nevertheless, it is likely that in the future, combining traditional corpus analysis methods with network analysis will boost the efficiency of discourse analysis, especially when using large corpuses.

담화 구성 요소를 고려한 문장부사 하위분류

신서인 한국어의미학회 2014 한국어 의미학 Vol.44 No.-

This study aims to clarify the characteristics of sentence adverbs in the Korean language by re-subcategorizing them. Sentence adverbs are divided into four sub-categories: propositional adverbs, domain adverbs, speech act adverbs, and connecting adverbs. Propositional adverbs are sub-divided into modal adverbs, event-oriented evaluatives, and participants-evaluatives, whereas speech act adverbs are sub-divided into speech act-oriented adverbs and speaker/ hearer-oriented adverbs. Previous studies focused on the relationship between sentence adverbs and other constituents at the sentence level. In contrast, the current study attempts to define the characteristics of sentence adverbs by investigating at the discourse level the relationship between sentence adverbs and other factors such as propositions, speech acts, the speaker, and the hearer.

『한국어 의미학』 72집에 대한 총평

신서인 한국어의미학회 2021 한국어 의미학 Vol.73 No.-

문장부사의 위치에 대한 고찰

신서인 국어학회 2011 국어학 Vol.61 No.-

This paper examines the unmarked word order of sentences with sentence adverbs, and analyses what the sentences with sentence adverbs mean when they appear in marked word order. The unmarked position of a sentence adverb that modifies a whole sentence is at the sentence initial position. If a sentence with a sentence adverb has marked word order, it is not because the sentence adverb is positioned in different places, but because other constituents are placed before the sentence adverb. The constituents that come before sentence adverbs are those which are topicalized.

코퍼스를 이용한 담화 분석 방법론

신서인 한국어의미학회 2017 한국어 의미학 Vol.55 No.-

This paper aims to examine the methodology of corpus-assisted discourse analysis. Methodologically, corpus linguistics is frequently criticised for its tendency to disregard context, while the major criticism of CDA(critical dicourse analysis) is its arbitrariness in selecting texts to be analysed. Combining the two approaches can offer a way to overcome the weaknesses of each method and enhance analytical rigour. In this paper, collocational analysis, cluster analysis, and concordance analysis are applied to the corpus of newspaper articles on distribution. The collocational and cluster analyses are effective for processing a large amount of data and extracting recurring patterns in which specific words are used. They further provide the starting point of the concordance analysis, which allows a qualitative look into the context that is difficult to observe in the collocational analysis. The combined methodology has proven to be useful in revealing the intention of the usage of specific words or phrases, and shown how corpus analysis can contribute to discourse analysis.

한국어의 어순 변이 경향과 그 요인에 대한 연구

신서인 국어학회 2007 국어학 Vol.0 No.50

This paper presents the results of an analysis on word order variations in Korean using an annotated Korean corpus. It is found that basic word orders are dominantly found in the sentences. One of the factors which decide the extent of the word order variations is obligatoriness. It is found that ‘NP-ro’ occurs in front of a predicate when it is an argument in an argument structure of the predicate, while its position is flexible when ‘NP-ro’ is an adjunct. Another factor is a thematic role. According to the thematic role, one word order is preferred: a ‘Theme’ precedes a ‘Final State’, while a ‘Instrument’ does not affect the position of the preceding ‘Theme’ or the following ‘Theme’. The factors which cause word order variation is found in information structures: left-movement represent topicalization and right-movement represent focalization.

어순 변이와 문장 의미 해석

신서인 한국어의미학회 2009 한국어 의미학 Vol.28 No.-

Shin Seo-in. 2009. Word Order Variation and Analyzing Sentence Meaning. Korean Semantics, 28. This paper has looked over various approaches to word order variation in Korean. Language typological approach shows generalized insight on basic word order; but it is not enough for word order variation in individual sentences. In syntactic approach, word order information of lexicon interact with other kinds of information such as case-marker or selectional restriction. In pragmatic approach, topic and focus, main concepts of a theory of information structure offer useful explanation for marked order. In some cases, scrambling is occurred for successful communication.

한국어 의미역 말뭉치 구축을 위한 의미역 표지 연구

신서인 한국어의미학회 2019 한국어 의미학 Vol.66 No.-

The purpose of this paper is to discuss what should be considered in the process of establishment of the semantic role label set and semantic role annotation for a semantic role annotated Korean corpus. This paper first compares the semantic role label lists of dictionaries or corpora that are built on a large scale, and examines the status of the semantic role annotation, and examines the issues related to the semantic role annotation of arguments. Accordingly, semantic role labels for adjunct are proposed. For the practical purposes, such as the establishment of an automatic response system, it would be advisable to use the same semantic role set to identify the agreement between the two sentences, but the current paper proposes the semantic role annotation suitable for expressing the relationship between the predicate and the arguments in a sentence. Decisions on how to solve the semantic annotation problems of Korean sentences should be made before a balanced list of semantic role label sets and clear work instructions can be drawn up.

개체명 분석 및 개체 연결 말뭉치 구축의 현황과 문제점

신서인,이수현,박진현,김나현 한림대학교 한림과학원 2022 개념과 소통 Vol.- No.30

Named-entity analysis searches for a named-entity in texts, classifying the occurrences into specified subclasses, and attaching appropriate tags. Named-entity corpora derived from this process can be used for various purposes, such as information retrieval, document summarization, automatic question-answering system construction, and knowledge base construction. Entity linking connects a named-entity to a knowledge base document such as Wikipedia. Entity linking corpora can be used to differentiate the meanings of named-entities. This paper discusses and seeks solutions to some problems encountered in the process of constructing named-entity corpora and entity linking corpora. First, for named-entity corpora, the choice of analysis target, analysis unit, and named-entity tag is discussed, and alternatives approaches considered. Next, there is a discussion of entity linking corpus construction, problems with the entity linking unit, with the knowledge base itself, and with a named-entity corresponding to multiple Wikipedia documents: again alternative approaches are suggested. Lastly, problems which occur in the construction of a named-entity dictionary are examined, and ways to overcome them are presented. The problems identified in this paper and the solutions proposed contribute towards building high-quality data by securing clarity and consistency in name-entity analysis. Ultimately, by improving the consistency of the entity linking and making it more systematic, it should be possible to provide information suitable for easier and more widespread utilization. 개체명 분석은 텍스트에서 개체명을 탐색해 일정한 하위 부류로 분류하고 적절한 태그를 부착하는 것으로, 자연언어처리를 이용한 정보 검색, 문서 요약, 자동 질의응답 시스템 구축, 지식베이스 구축 등의 다양한 목적으로 활용될 수 있다. 개체 연결은 분석된 개체명에 위키피디아와 같은 지식베이스 문서를 연결하는 것으로, 사용자에게 사전적 의미를 넘어서는 다양한 정보를 제공할 수 있다. 언어학뿐만 아니라 산업 전반에 걸친 이러한 활용도를 고려하면, 보다 양질의 데이터를 구축할 필요가 있음이 분명하다. 이를 위해 이 논문은 개체명 분석 말뭉치 및 개체 연결 말뭉치 구축 과정에서 발생한 문제들을 논의하고 해결 방안을 모색하고자 했다. 먼저 개체명 분석 말뭉치 구축 과정에서는 개체명 분석 대상 결정의 문제, 개체명 분석 단위 결정의 문제, 개체명 분석 표지 부착의 문제 등이 있음을 논의하고 이 문제들에 대한 대안을 제시했 다. 다음으로 개체 연결 말뭉치 구축 과정에서는 개체 연결 단위의 문제, 지식베이스 자체가 갖는 문제, 동일한 언어 표현의 개체명이 여러 위키피디아 문서에 대응하는 문제 등이 있음을 논의하고 이 문제들에 대한 대안을 제시했다. 마지막으로 개체명 사전 구축 과정에서 발생하는 문제를 살펴보고 이를 극복하는 방안을 제시했다. 이논문에서 제시한 문제점을 극복한다면 개체명 분석 단계에서 명확성과 일관성을 확보해 양질의 데이터를 구축할 수 있을 것이다. 또한 개체 연결 단계에서 체계성 및 일관성을 제고해 활용 목적에 알맞은 정보를 제공할 수 있을 것이다.