• 한국컴퓨터정보학회논문지
• /
• 제25권11호
• /
• pp.27-40
• /
• 2020

본 논문은 장단기기억신경망(LSTM)이 영어를 배우면서 학습한 암묵적 통사 관계인 필러-갭 의존 관계를 조사하여 영어 문장 학습량과 한국인 영어 학습자(L2ers)의 문장 처리 패턴 간의 상관관계를 규명한다. 이를 위해, 먼저 장단기기억신경망 언어모델(LSTM LM)을 구축하였다. 이 모델은 L2ers가 영어 학습 과정에서 잠재적으로 배울 수 있는 L2 코퍼스의 영어 문장들로 심층학습을 하였다. 다음으로, 이 언어 모델을 이용하여 필러-갭 의존 관계 구조를 위반한 영어 문장을 대상으로 의문사 상호작용 효과(wh-licensing interaction effect) 즉, 정보 이론의 정보량인 놀라움(surprisal)의 정도를 계산하여 문장 처리 양상을 조사하였다. 또한 L2ers 언어모델과 상응하는 원어민 언어모델을 비교 분석함으로써, 두 언어모델이 문장 처리에서 필러-갭 의존 관계에 내재된 추상적 구문 구조를 추적할 수 있음을 보여주었을 뿐만 아니라, 또한 선형 혼합효과 회귀모델을 사용하여 본 논문의 중심 연구 주제인 의존 관계 처리에 있어서 원어민 언어모델과 L2ers 언어모델간 통계적으로 유의미한 차이가 존재함을 규명하였다.

• 정보처리학회논문지B
• /
• 제15B권4호
• /
• pp.365-374
• /
• 2008

본 논문에서는 패턴의 다변화를 통하여 관계를 점진적으로 추출함으로써 온톨로지를 확장하는 모델을 제안한다. 패턴 다변화 과정에서 위키피디아로부터 추출한 관계 패턴 후보를 자질별로 다변화시킨다. 다변화된 패턴 후보로부터 말뭉치 빈도수에 따른 신뢰도를 이용하여 패턴을 선별한다. 선별된 패턴은 위키피디아로부터 관계를 추출하는 데 사용되며, 추출된 관계는 다시 관계 패턴 확장에 사용된다. 본 논문에서는 점진적 학습 과정에서의 패턴 다변화를 통하여 패턴 선택의 범위를 확장함으로써, 선택되는 패턴이 점진적으로 정제되는 모델을 제시한다. 이를 통하여, 관계의 확장성과 정확도를 향상시키고자 하였다. 단일 자질 패턴 모델에 대한 실험을 통하여, 어휘, 중심어, 상위어 정보는 신뢰도에, 품사, 구문 정보는 확장성에 유리하며, 구문 단위 유형별로 필요한 자질 유형이 다름을 관찰하였다. 이와 같은 특성에 기반하여 현재 연구 진행 중인복합 자질 패턴 모델을 제안한다.

• KSII Transactions on Internet and Information Systems (TIIS)
• /
• 제16권3호
• /
• pp.771-791
• /
• 2022

Protein-protein interaction (PPI) extraction from original text is important for revealing the molecular mechanism of biological processes. With the rapid growth of biomedical literature, manually extracting PPI has become more time-consuming and laborious. Therefore, the automatic PPI extraction from the raw literature through natural language processing technology has attracted the attention of the majority of researchers. We propose a PPI extraction model based on the large pre-trained language model and adversarial training. It enhances the learning of semantic and syntactic features using BioBERT pre-trained weights, which are built on large-scale domain corpora, and adversarial perturbations are applied to the embedding layer to improve the robustness of the model. Experimental results showed that the proposed model achieved the highest F1 scores (83.93% and 90.31%) on two corpora with large sample sizes, namely, AIMed and BioInfer, respectively, compared with the previous method. It also achieved comparable performance on three corpora with small sample sizes, namely, HPRD50, IEPA, and LLL.

• 한국인터넷방송통신학회논문지
• /
• 제22권4호
• /
• pp.111-117
• /
• 2022

수십년간 매우 많은 소프트웨어 결함 예측 모델에 관한 연구들이 수행되었으며, 그들 중 기계학습 기법을 사용한 모델들이 가장 좋은 성능을 보였다. 딥러닝 기법은 기계학습 분야에서 가장 각광받는 기술이 되었지만 결함 예측 모델의 분류기로 사용된 연구는 거의 없었다. 몇몇 연구들은 모델의 입력 소스나 구문 데이터로부터 시맨틱 정보를 얻어내는데 딥러닝을 사용하였다. 본 논문은 3개 이상의 은닉층을 갖는 MLP를 이용하여 모델 구조와 하이퍼 파라미터를 변경하여 여러 모델들을 제작하였다. 모델 평가 실험 결과 MLP 기반 딥러닝 모델들은 기존 결함 예측 모델들과 Accuracy는 비슷한 성능을 보였으나 AUC는 유의미하게 더 우수한 성능을 보였다. 또한 또다른 딥러닝 모델인 CNN 모델보다도 더 나은 성능을 보였다.

• 대한음성학회지:말소리
• /
• 제46호
• /
• pp.51-64
• /
• 2003

In order to produce high quality synthesized speech, it is very important to get an accurate grapheme-to-phoneme conversion and prosody model from texts using natural language processing. Robust preprocessing for non-Korean characters should also be required. In this paper, we analyzed Korean texts using a morphological analyzer, part-of-speech tagger and syntactic chunker. We present a new grapheme-to-phoneme conversion method for Korean using a hybrid method with a phonetic pattern dictionary and CCV (consonant vowel) LTS (letter to sound) rules, for unlimited vocabulary Korean TTS. We constructed a prosody model using a probabilistic method and decision tree-based method. The probabilistic method atone usually suffers from performance degradation due to inherent data sparseness problems. So we adopted tree-based error correction to overcome these training data limitations.

• 영어어문교육
• /
• 제4호
• /
• pp.179-206
• /
• 1998

Despite the fact that the phenomenal success of the classical model in phonology in terms of a small set of binary atomic primitives has encouraged the use of criterical features in the characterization of syntactic and semantic categories, cognitive linguists have recently found many evidences against it and proposed a valid alternative like prototype approach to classical model of linguistic categorization. In this paper the examples with prototype effects are shown in phonology, morphology, and syntax. In phonology there is no clear-cut borderline in initial sounds such as / r, m, n, v, ${\eth}$, z, w. j, b. d, g / and between vowels and consonants. In morphology family resemblance is shown between strong verbs and past tenses. In syntax there is no borderline among various kinds of sentences such as declarative, command, WH-question, yes-no question, etc.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2000년도 7월 학술대회지
• /
• pp.117-134
• /
• 2000

Kim, Key-Seop(2000). English Restructuring and A Use of Music in Teaching English Pronunciation. JSEP 2000 voU This study has two-fold aims: one is to clarify the restructuring of English in utterance, and the other is to relate it to teaching English pronunciation for listening and speaking with a use of music and song by suggesting a model of 10-15 minute pronunciation class syllabus for every period in class. Generally, English utterances are restructured by stress-timed rhythm, irrespective of syntactic boundaries. So the rhythmic units are arranged in isochronous groups, of which the making is to attach clitic(s) to a host or head often leftwards and sometimes rightwards, which results in linking, contraction, reduction, sound change and rhythm adjustment in utterance, just as in music and song. With English restructuring focused on, a model of English pronunciation class syllabus is proposed to be put forward in class for every period of a lesson or unit. It tries to relate the focused factor(s) in pronunciation to the integrated, with teaching techniques and music made use of.

• 음성과학
• /
• 제5권1호
• /
• pp.35-47
• /
• 1999

This paper describes procedures to model and synthesize emotional speech in a multimedia environment. At first, procedures to model the visual representation of emotional speech are proposed. To display the sequences of the images in synchronized form with speech, MSF(Multimedia Speech File) format is proposed and the display software is implemented. Then the emotional speech sinal is collected and analysed to obtain the prosodic characteristics of the emotional speech in limited domain. Multi-emotional sentences are spoken by actors. From the emotional speech signals, prosodic structures are compared in terms of the pseudo-syntactic structure. Based on the analyzed result, neutral speech is transformed into a specific emotinal state by modifying the prosodic structures.

• 대한음성학회:학술대회논문집
• /
• 대한음성학회 2003년도 5월 학술대회지
• /
• pp.87-90
• /
• 2003

In order to produce high quality synthesized speech, it is very important to get an accurate grapheme-to-phoneme conversion and prosody model from texts using natural language processing. Robust preprocessing for non-Korean characters should also be required. In this paper, we analyzed Korean texts using a morphological analyzer, part-of-speech tagger and syntactic chunker. We present a new grapheme-to-phoneme conversion method, i.e. a dictionary-based and rule-based hybrid method, for unlimited vocabulary Korean TTS. We constructed a prosody model using a probabilistic method and decision tree-based method.

• 한국언어정보학회지:언어와정보
• /
• 제18권2호
• /
• pp.1-29
• /
• 2014

Relativization is one of the common syntactic operations to merge two different clauses into a single information unit. This operation plays a pivotal role to structuralize multiple clauses cohesively as well as serves to specify the property an individual has within the context. That implies that relativization contributes to information structure of multiclausal sentences. In this context, this paper delves into information structure of relative clauses in English with an eye toward creation of a computational model from a standpoint of machine translation. The current work employs Head-driven Phrase Structure Grammar (HPSG, Pollard and Sag (1994)) as a theory of grammar and Minimal Recursion Semantics (MRS, Copestake et al. (2005) as a meaning representation system. Building upon these formalisms, this paper addresses how information structure of relative clauses can be represented and constrained. The current work makes use of Individual CONStraints (ICONS) for modeling relative clauses with respect to information structure. The current work also investigates which relative clause involves which information structure constraint. The present study argues that non-restrictive relative clauses impose a more specific constraint on information structure than restrictive relative clauses.