Simple and effective neural coreference resolution for Korean language

박천음,임준호,류지희,김현기,이창기 한국전자통신연구원 2021 ETRI Journal Vol.43 No.6

We propose an end‐to‐end neural coreference resolution for the Korean language that uses an attention mechanism to point to the same entity. Because Korean is a head‐final language, we focused on a method that uses a pointer network based on the head. The key idea is to consider all nouns in the document as candidates based on the head‐final characteristics of the Korean language and learn distributions over the referenced entity positions for each noun. Given the recent success of applications using bidirectional encoder representation from transformer (BERT) in natural language‐processing tasks, we employed BERT in the proposed model to create word representations based on contextual information. The experimental results indicated that the proposed model achieved state‐of‐the‐art performance in Korean language coreference resolution.

Processing Scrambled Wh-Constructions in Head-Final Languages: Dependency Resolution and Feature Checking

( Hye Ryeong Hahn ),( Seung Jin Hong ) 한국언어정보학회 2014 언어와 정보 Vol.18 No.2

This paper aims at exploring the processing mechanism of filler-gap dependency resolution and feature checking in Korean wh-constructions. Based on their findings on Japanese sentence processing, Aoshima et al. (2004) have argued that the parser posits a gap in the embedded clause in head-final languages, unlike in head-initial languages, where the parser posits a gap in the matrix clause. In order to verify their findings in the Korean context, and to further explore the mechanisms involved in processing Korean wh-constructions, the present study replicated the study done by Aoshima et al., with some modifications of problematic areas in their original design. Sixty-four Korean native speakers were presented Korean sentences containing a wh-phrase in four conditions, with word order and complementizer type as the two main factors. The participants read sentences segment-by-segment, and the reading times at each segment were measured. The reading time analysis showed that there was no such slowdown at the embedded verb in the scrambled conditions as observed in Aoshima et al. Instead, there was a clear indication of the wh-feature checking process in terms of a major slowdown at the relevant region. (Seowon University & University at Bu_alo, SUNY)

어순에 관한 언어 유형적 접근과 한국어의 기본 어순

임홍빈 서강대학교 인문과학연구소 2007 서강인문논총 Vol.22 No.-

"This paper aims to make it clear that the language typological approach to word orders as exemplified in Dryer(2005a-q) has intrinsic shortcomings for capturing the real nature of the word order of human languages, and instead to set up the proper principles by which to describe basic word orders embodied in Korean syntactic constructions. By fixing the parameters relative to word order varieties proposed by Dryer(2005a-q), this paper postulates the 14 individual word order theorems concerning the Korean word order, which applies separately to syntactic constructions. Dryer's(2005a-q) parameters relative to word order do not allow any implicational universals as shown in Greenberg's(1963) universals. However, most of the characteristic word order facts of Korean can be subsumed under the head-final nature of Korean syntactic constructions. As to the syntactic analysis, it is noted that it should be carried out on the basis of the lexical information given in the lexicon, in which argument(s) of a predicate or of a head is/are assumed to have the case markers and/or the relevant endings and the proper basic order(s) between the arguments and the head, etc. Dryer's(2005a-q) generalizations basically rely on the Chomskyan grammatical theory, which could be thought to have fatal flaws neglecting the lexical information. Applied Dryer's(2005a-q) principle, the obliques in Korean are posited before '(objects) + verb', which is schematized in X(O)V order. This means that the oblique comes before the object if the verb has a object, but otherwise the oblique comes before a verb. However, the dative comes before the object in dative verb constructions in Korean, due to the lexical information given in the lexicon, not due to Dryer's generalization. The same can be said to the object complement, in complex transitive constructions in Korean. As to the adverbials or adjuncts not given any lexical information from the lexicon, this paper takes much of the intrinsic nature of the adverbials or adjuncts to have the positional requirements. From this, the number of the obliques to be posited before (O)V in accordance to Dryer's(2005) generalization must be considered to be minimal. " 본고는 한국어의 기본 어순에 대하여 Dryer (2005a-q)적인 언어 유형적 접근이 가지는 한계를 지적하고, 그 대안으로 핵의 위치에 관한 매개 변수적 접근이 어느 정도 설득력을 가질 수 있음을 분명히 한 것이다. Dryer (2005a-q)에 의하면, 한국어의 기본 어순적인 사실은 적어도 14가지의 서로 구별되는 현상으로 나누어 기술된다. 각 현상은 다른 현상과 유기적인 관련을 이루지 못하고, 각각 독립된 현상의 성격을 가지며, Greenberg (1963)적인 어떠한 함축도 허용하지 않는다. 이에 대하여 핵에 관한 매개 변수적 접근에 의하면, 한국어는 핵말 언어로 성격지어진다. Dryer (2005a-q)적인 언어 유형론을 한국어에 적용한 어순적 사실의 대부분은 핵말 언어의 유형적 사실로 흡수될 수 있는 것이다. 그러나 그 유형론은 주어나 목적어와 같은 논항만을 대상으로 하는 체계에 입각한 것이다. 논항 외에도 부가어나 부사어의 순서가 문제된다. 이에 대해서는 부사적인 성분의 고유한 성격을 중시하였다. 이로써 사격어의 위치에 관한 언어 유형론적인 매개 변수의 적용은 최소화된다. 이러한 논의에는 ‘격 어휘론자 가설’과 ‘가변 중간 투사론’이 핵심적인 위치에 있다.

Multi-task learning with contextual hierarchical attention for Korean coreference resolution

Cheoneum Park Electronics and Telecommunications Research Instit 2023 ETRI Journal Vol.45 No.1

Coreference resolution is a task in discourse analysis that links several headwords used in any document object. We suggest pointer networks-based coreference resolution for Korean using multi-task learning (MTL) with an attention mechanism for a hierarchical structure. As Korean is a head-final language, the head can easily be found. Our model learns the distribution by referring to the same entity position and utilizes a pointer network to conduct coreference resolution depending on the input headword. As the input is a document, the input sequence is very long. Thus, the core idea is to learn the word- and sentence-level distributions in parallel with MTL, while using a shared representation to address the long sequence problem. The suggested technique is used to generate word representations for Korean based on contextual information using pre-trained language models for Korean. In the same experimental conditions, our model performed roughly 1.8% better on CoNLL F1 than previous research without hierarchical structure.

Processing Scrambled Wh-Constructions in Head-Final Languages: Dependency Resolution and Feature Checking

한혜령,홍승진 한국언어정보학회 2014 언어와 정보 Vol.18 No.2

This paperaims at exploring the processing mechanism of filler-gap dependency resolutionand feature checking in Korean wh-constructions. Based on their findings onJapanese sentence processing, Aoshima et al. (2004) have argued that theparser posits a gap in the embedded clause in head-final languages, unlike inhead-initial languages, where the parser posits a gap in the matrix clause. Inorder to verify their findings in the Korean context, and to further explorethe mechanisms involved in processing Korean wh-constructions, the presentstudy replicated the study done by Aoshima et al., with some modifications ofproblematic areas in their original design. Sixty-four Korean native speakerswere presented Korean sentences containing a wh-phrase in four conditions,with word order and complementizer type as the two main factors. Theparticipants read sentences segment-by-segment, and the reading times ateach segment were measured. The reading time analysis showed that therewas no such slowdown at the embedded verb in the scrambled conditionsas observed in Aoshima et al. Instead, there was a clear indication of thewh-feature checking process in terms of a major slowdown at the relevantregion.

Processing Scrambled Wh-Constructions in Head-Final Languages: Dependency Resolution and Feature Checking

Hahn, Hye-ryeong,Hong, Seungjin Korean Society for Language and Information 2014 언어와 정보 Vol.18 No.2

This paper aims at exploring the processing mechanism of filler-gap dependency resolution and feature checking in Korean wh-constructions. Based on their findings on Japanese sentence processing, Aoshima et al. (2004) have argued that the parser posits a gap in the embedded clause in head-final languages, unlike in head-initial languages, where the parser posits a gap in the matrix clause. In order to verify their findings in the Korean context, and to further explore the mechanisms involved in processing Korean wh-constructions, the present study replicated the study done by Aoshima et al., with some modifications of problematic areas in their original design. Sixty-four Korean native speakers were presented Korean sentences containing a wh-phrase in four conditions, with word order and complementizer type as the two main factors. The participants read sentences segment-by-segment, and the reading times at each segment were measured. The reading time analysis showed that there was no such slowdown at the embedded verb in the scrambled conditions as observed in Aoshima et al. Instead, there was a clear indication of the wh-feature checking process in terms of a major slowdown at the relevant region.

A Structural Account of Intrasentential Code-switching in Korean-English Bilingual Speech

이준규 대한언어학회 2010 언어학 Vol.18 No.3

Lee, Junkyu. 2010. A Structural Account of Intrasentential Code-switching in Korean-English Bilingual Speech. The Linguistic Association of Korea Journal. 18(3). 93-110. This study investigates intrasentential code-switching patterns by Korean-English bilinguals with special reference to their structural characterizations. An examination of Korean-English code-switching is theoretically significant since Korean and English have distinct parameters: Korean is head-final whereas English is head-initial. Data were collected from natural speech of three pairs of Korean-English bilingual speakers. Result showed that the bilinguals produced lexical items which have both a preposition and a postposition in a single word, which seems to stem from the interactions between two different head-parameters of the languages. The findings pose an empirical challenge to the proposed formulations in the literature, providing a new insight into intrasentential code-switching that cannot be found in the same language parameter.

'킥라니'형 신어 혼성어에 대하여

최형용 서강대학교 언어정보연구소 2022 언어와 정보 사회 Vol.46 No.-

This paper aims at revealing the properties of the khiklani-type, new blends. They have in common that the second half of the blend which have attribute meanings modify the first half with respect to the semantic structure. This property does not follow the general principles of Korean as a head-final language. However, the scale of such blends is not so small and they can be divided into five main types and ten subtypes according to the morphological structure. Moreover, the metaphor which is expressed by the latter elements of the khiklani-type blends is much more common in blends in general in contrast to compounds. As a result, these new khiklani-type blends have to be considered a new (productive?) kind of word formation rather than mere exceptions. 지금까지 후행 요소가 은유를 통한 속성 의미를 가지면서 선행 요소를 수식하는 ‘킥라니’형 신어 혼성어에 대해서는 그 특성이 제대로 밝혀졌다고 보기는 어렵다. 본고에서는 모두 64개 ‘킥라니’형 신어 혼성어 목록을 확인하였는데 형태 구조를 기반으로 할 때 다시 크게는 5개, 작게는 10개까지 구분이 가능하다. ‘킥라니’형 신어 혼성어는 한국어가 핵-끝머리 속성을 가진다는 사실에 대한 예외로서 특히 주목할 필요가 있는데 이때 후행 요소는 모두 은유를 통한 속성화로서 선행 요소를 수식하고 있다는 점에서 공통적이다. 그런데 ‘킥라니’형 신어 혼성어에서 나타나는 후행 요소 은유는 ‘킥라니’형에만 국한되는 것은 아닐 뿐만 아니라 혼성어에만 국한된 것도 아니다. 그러나 합성어에서 나타나는 은유의 양상은 선행 요소 은유가 더 일반적일 뿐만 아니라 후행 요소에서 은유가 나타나는 경우 선행 요소를 수식하는 것은 찾기가 어렵다. 이에 비하면 신어를 대상으로 할 때에도 후행 요소가 은유의 양상을 보이는 경우가 혼성어에서 두드러진다는 것은 ‘킥라니’형 신어 혼성어의 존재를 설명하는 실마리가 된다. 한편 ‘킥라니’형 신어 혼성어의 단어 형성적 특성에 대해 살피기 위해 김병건(2018)에서의 실험 결과에 주의를 집중하였다. 김병건(2018)의 내용을 보면 ‘킥라니’형 신어 혼성어의 존재가 우연한 결과가 아님을 보여 주기에 충분하기 때문이다. 이는 곧 ‘킥라니’형 신어 혼성어가 혼성어 형성에서 하나의 유형을 보여 주는 존재임을 의미하는 것으로 해석할 수 있다.

A Structural Account of Intrasentential Code-switching in Korean-English Bilingual Speech

( Jun Kyu Lee ) 대한언어학회 2010 언어학 Vol.18 No.3

유형론적으로 유사한 제2언어의 선행 학습경험이 제3언어로서의 한국어 습득에 미치는 효과 연구

조재현 ( Jaehyun Jo ) 연세대학교 언어정보연구원(구 연세대학교 언어정보개발원) 2011 언어사실과 관점 Vol.27 No.-

In this study the author conducted an experimental test of adult third language (L3) acquisition whose results provide concrete and clear evidence for the Cumulative Enhancement Model (CEM; Flynn, Foley and Vinnitskaya, 2004). The main research questions examined in the present study were as follows: (1) whether L1 or L2 has a privileged role in L3 acquisition as to syntactic transfer, (2) what particular categories of the syntactic knowledge in Korean were acquired most and least easily by those with previous experience; (3) if age factor matters with respect to syntactic transfer in L3 acquisition, and lastly, (4) whether syntactic transfer is limited only to the initial state of L3 acquisition. The current study reports data from three experimental groups as follows: (Group A) L1 English/L2 Korean, (Group B) L1 English/L2 Japanese/L3 Korean, (Group C) L1 Japanese/L2 English/L3 Korean. All subjects were tested on their knowledge of 6 different types of syntactic categories (10 subcategories) of Korean by the elicited imitation methodology (Lust, Chien and Flynn, 1987; Lust, Flynn and Foley, 1996) with Korean sentences of varying types.