딥러닝 기반 특허의 종속 청구항 인식 개선

박주연 ( Ju-yeon Park ),신예지 ( Yeji Shin ),김민수 ( Minsu Kim ),김동호 ( Dongho Kim ),김지희 ( Jihie Kim ) 한국정보처리학회 2020 한국정보처리학회 학술대회논문집 Vol.27 No.1

특허를 통해 기술의 권리를 정의하고 보호하는 일이 매우 중요해짐에 따라 특허 문서를 분석하는 연구 또한 중요해지고 있다. 특히 특허의 청구항을 종속항과 독립항을 구분하고, 관련된 인용을 찾아내는 일은 관련 특허들을 분석하는데 매우 중요하다. 본 연구는 최근 텍스트 분석 분야에 획기적 성능 개선을 이끈 BERT(Bidirectional Encoder Representations From Transformers) 언어 모델을 사용하고 Neural Network 의 파인 튜닝 과정을 통해 청구항의 독립과 종속을 구분하였고, 인용하는 항의 번호와 인용 문구로 이루어진 인용 패턴을 통해 종속항의 인용 항을 찾아내었다. 이 방법을 2003 년 이후의 xml 형식의 미국 특허 데이터에 사용한 결과, 정확도 99% 의 성능을 확보하였다.

인공지능의 문화번역에 대한 소고(小考): 속담번역을 중심으로

곽은주,탁진영,김동미 인문사회 21 2020 인문사회 21 Vol.11 No.1

오늘날 번역 분야에 있어서 인공지능의 출현은 번역작업의 새로운 패러다임의 출현을 야기한다. 그동안의 번역은 인간이 주축이 되어서 처음부터 끝까지 작업을 수행하는 형태였으나 인공지능의 출현으로 인간이라는 매개체는 번역작업에서 선택사항이 되었다. 본 논고는 인공지능 번역이 문화가 바탕이 되는 문화번역, 특히 속담을 어떻게 번역하는지 고찰하고 인공지능과 인간의 번역상의 차이점을 비교함으로써 앞으로 진행될 각 주체별 번역 패러다임을 예측해 보고자 한다. 이러한 연구를 위해 4가지 측면으로 인공지능 번역과 인간 번역을 비교해 보았다. 인공지능 번역인 파파고와 구글번역의 결과물을 살펴보면, 파파고가 3가지 비교 측면에서 구글번역보다 우수함을 보였으며, 고유문화 언어는 두 인공지능 모두 해독이 불가능한 것으로 나타났다. 이것은 문화적 배경을 바탕으로 하는 속담번역에서 인간이 인공지능보다 문화적 요소를 잘 살려서 번역하는 반면 인공지능 번역 엔진은 아직 문화번역이 가능할 만큼의 데이터를 축적하지 못하고 있다는 것을 보여준다. The emergence of artificial intelligence (henceforth AI) has induced a new paradigm in the field of translation. While translation has been done solely by human beings from beginning to end so for, the introduction of AI makes the medium of human beings optional in the process of translation. In the present study, it is analyzed how AI translation deals with cultural translation, especially in translating proverbs, where cultural background is indispensible. Furthermore, the direction of translation paradigm development is also speculated by comparing AI and human translation. For this purpose, four different cases are compared. The results show that Papago translates proverbs better than Google Translator in three cases and that both of them do not translate the language based on culture properly. Additionally, it is found that human translators accommodate cultural elements more appropriately than AI translators in the processes of proverb translation and that AI translation engines are not equipped with sufficient data for cultural translation yet.

순차 검증과 자료융합을 이용한 수중 표적 판별

곽은주,홍순목 경북대학교 전자기술연구소 1997 電子技術硏究誌 Vol.18 No.2

In this paper we discuss an algorithm to discriminate a target under track against multiple acoustic counter-measure (ACM) sources, based on sequential testings of multiple hypotheses. The ACM sources are separated from the target under track and generate, while drifting, measurements with false range and Doppler information. The purpose of the ACM is to mislead the target tracking and to help the true target evade a pursuer. The proposed algorithm uses as a test statistic a function of both the sequences of processed waveform signature and the innovation sequences from extended Kalman filters to estimate the target dynamics and the drifting positions of the ACM sources. Numerical experiments on various scenarios show that the proposed algorithm discriminates the target faster with a higher probability of success than the algorithm using only the innovation sequences from extended Kalman filters.

Interpretations of Plural Noun Phrases in Korean

곽은주 사단법인 한국언어학회 2003 언어학 Vol.0 No.35

Since plurality becomes a hot issue in semantics, the general consensus is that the domain of individuals includes both atomic and plural individuals as Link (1983) discusses. In his 1984 paper, however, Link argues that this domain is not rich enough to deal with a full range of interpretations of plural NPs and thus postulates group, a higher-ordered entity consisting of a sum of individuals. This new entity has raised an intense debate among semanticists; Link (1984) and Landman (1989) argue for the group readings of ordinary plural terms, and Schwarzschild (1992) and Barker (1992) are against the idea. Part of difficulties behind this discussion is that English does not provide conclusive evidence for or against groups, which becomes a stumbling block to further research. Interestingly, Korean has two forms of plurals: one with the plural marker tul and the other without the marker. For example, people is translated into either salam-tul person-tul or salam person in Korean. Both forms of NPs are widely used to deliver a plural reading. A general belief on these dual forms has been that they do not show any difference in their interpretations, which has served as evidence for non-sensitivity to number in Korean. (cf. Kang 1994 etc.) However, I will argue that Korean does distinguish singular and plural NPs in a way different from English-typed languages. A plural NP with tul is interpreted as a sum while a plural without tul is used as a group. I will argue for this distinction, considering distributivity, group properties, and contextual preferences. I will further discuss the fact that group readings are affected by the properties of predicates and thus propose in the framework of event semantics that groups are event-dependent entities.

The Distinction between Count and Mass Nouns

곽은주 현대영미어문학회 2009 현대영미어문학 Vol.27 No.2

In the classical theories, the distinction between count and mass nouns is mainly based on morpho-syntactic properties, which are based on the parallel between countability and plurality. However, morpho-syntactic criteria for countability may not work with languages that do not have prominent morphemes for count or mass nouns. In this study, I discuss problems with Chierchia (1998a,b)'s morpho-syntactic based criteria for countability with cross-linguistic data and argue for semantic based criteria such as cumulativity and homogeneity. Rothstein (2007) claims that both count and mass nouns are subject to cumulativity and that some mass nouns are not homogeneous in their denotations. However, since apparent homogeneous count nouns may be cumulative only under some contexts, context-free cumulativity may be used as a common property for mass nouns. It is also discussed that mass nouns denote homogeneous entities based on the distinctive notions of atomicity in semantics and the real world.

텍스트 유형 기반의 영한 게임 번역 특수성 고찰 - LQA 전후 텍스트 비교를 중심으로 -

곽은주,김정연,김동미 한국문화융합학회 2024 문화와 융합 Vol.46 No.-

본 연구는 일반 텍스트와는 다른 번역 특징을 지니는 게임 텍스트 번역만의 특징을 살펴보는 것으로 LQA 전후를 기준으로 TT1과 TT2를 비교하여 게임 전략을 설정하는 것을 목적으로 한다. 이러한 연구를 위해“Need for Speed: Unbound”라는 게임의 텍스트를 입수하여 O’Hagan과 Mangiron의 텍스트 유형(정보적, 표현적, 설득적)에 번역 애셋을 적용하여 분석하였다. 첫째, 정보적 텍스 유형에 속하는 퀘스트, 시스템 메시지, 튜토리얼의 애셋에서 나타나는 번역 전략의 분석에서는 일반 정보적 텍스트 번역과 차별화되는 부분으로써 게임이라는 특수성에 의해 게임의 어느 부분에 나오는지에 따라 명사구로 끝나거나 문장형으로 끝나야 하는 경우가 있었다. 둘째, 표현적 텍스트 번역 방법을 언급할 수 있는데, 다이얼로그나 내레이션에서는 번역자가 상황에 맞춰 첨가, 수정, 구체화, 어투 변경 전략을 사용한 것을 확인하였다. 마지막셋째, 설득적 텍스트 유형 번역 방법에서는 게임검색 시 게임 제목 전체가 아닌 일부를 검색하여 결과를확인하는 경우가 많은데, 쉽게 노출될 수 있는 어휘를 포함하는 전략을 사용하는 것을 알 수 있었다. This study examines the characteristics of game text translation, which has different translation characteristics from ordinary texts, and aims to establish a game strategy by comparing TT1 and TT2 before and after LQA. For this study, the text of the game "Need for Speed: Unbound" was obtained and analyzed by applying translation assets to the text types (informational, expressional, and persuasive) of O'Hagan and Mangiron. First, in the analysis of translation strategies that appear in quests, system messages, and tutorials assets belonging to the informative text type, there were cases in which it had to end in a noun phrase or a sentence type depending on which part of the game it appeared due to the specificity of the game as a part that was differentiated from general informative text translation. Second, in the expressive text translation method including dialog or narration, it was confirmed that the translator used the strategy of adding, modifying, concreting, and changing the tone according to the situation. Finally, in the persuasive text type translation method, it was found that when searching for a game, often only a part of the game title was searched to check the results, and a strategy that included vocabulary that could be easily exposed was used.

Optimal Forms of Noun Phrases in Korean

곽은주 대한언어학회 2012 언어학 Vol.20 No.1

Languages show a different strategy to mark the crosslinguistically prevalent grammatical features of plurality and definiteness. English makes use of a plural morpheme and articles to mark these features while Mandarin Chinese adopts unmarked forms for them. No matter which strategy a language takes, nominal forms make consistent patterns. Basically, Korean adopts a marking strategy for the features. However, marking itself is not mandatory for most categories of nouns, and thus alternate forms occur in Korean. To account for proper nominal forms in Korean, I resort to the Optimality Theory of de Swart & Zwarts (2009, 2010). I argue that a markedness constraint and faithfulness constraints are co-ranked in Korean unlike other languages and that some of the faithfulness constraints are subcategorized and situated in different positions in the hierarchical ordering of optimality. Hence, more than one nominal form turns out to be equally optimal and are used alternately.

불화 속 연화문양을 응용한 네일 디자인 연구

곽은주,박철호 코리아뷰티디자인학회 2011 코리아뷰티디자인학회지 Vol.7 No.1

One-sided Readings of Numbers in Modal Sentences

곽은주 한국영어영문학회 2011 영어 영문학 Vol.57 No.3

Numbers have been regarded as one-sided, and their exactly readings have been understood as the results of scalar implicature. This Neo-Gricean view on numbers becomes less persuasive due to theoretical and experimental counterarguments. In spite of growing evidence for theirtwo-sided readings, numbers are still one-sided in modal sentences. Moreover, the occurrence of a negative operator may worsen the acceptability of modal sentences with numbers. In the framework of Vector Space Semantics, I have derived two-sided readings of numbers with the simple notions of monotonicity of modals and scopal relations between modals and numbers. I have also argued that the awkwardness incurred by negation is the result of a split set of vectors for a number. The incoherent set of vectors is understood as the lack of an ideal behavior, which is against the deontic modality of the sentence.

Contrasted Meanings of Motun and On in Korean

곽은주 한국언어학회 2011 언어 Vol.36 No.1

Kwak, Eun-Joo. 2011. Contrasted Meanings of Motun and On in Korean. Korean Journal of Linguistics, 36-1, 25-47. Two universal quantifiers motun and on in Korean show similar distributivity in the quantification of non-collection terms, but they are distinct in the quantification of collection terms and proper terms. To capture the partial similarity, diverse patterns of distributivity are discussed for sentences with the quantifiers. Based on Kwak (2010), it is concluded that different distributivity between motun and on hinges on the additional restriction of integration on on and that apparent diverse distributivity with on is attributed to intricate mapping relations between groups and members in an interpretation domain. (Sejong University)