• 한국컴퓨터정보학회논문지
• /
• 제20권2호
• /
• pp.149-156
• /
• 2015

본 논문에서는 한국어 문장 기호의 처리를 통해 자연스러운 음성 합성 결과를 생성하는 방법에 대해서 논의한다. 이를 위해 한국어 위키피디아 문서를 분석하여 문장 기호의 사용을 8가지 형태로 분류하고, 11개의 정규표현식 규칙으로 문장 기호를 처리하는 방안을 제시한다. 그 결과 63,000 문장에 대해 56%의 정확도와 71.45%의 재현율을 달성하였으며, 문장 기호 처리 결과를 SSML 기반의 음성 합성 표현으로 변환하여 음성 합성 결과의 품질을 향상시키는 방법을 제안한다.

• 말소리와 음성과학
• /
• 제7권1호
• /
• pp.97-105
• /
• 2015

This research mainly focuses on intonational characteristics of synchronous English speech. Intonation contours were extracted from 1,848 utterances produced in two different speaking modes (solo vs. synchronous) by 28 (12 women and 16 men) native speakers of English. Synchronous speech is found to be slower than solo speech. Women are found to speak slower than men. The effect size of speech rate caused by different speaking modes is greater than gender differences. However, there is no interaction between the two factors (speaking modes vs. gender differences) in terms of speech rate. Analysis of pitch point features has it that synchronous speech has smaller Pt (pitch point movement time), Pr (pitch point pitch range), Ps (pitch point slope) and Pd (pitch point distance) than solo speech. There is no interaction between the two factors (speaking modes vs. gender differences) in terms of pitch point features. Analysis of sentence level features reveals that synchronous speech has smaller Sr (sentence level pitch range), Ss (sentence slope), MaxNr (normalized maximum pitch) and MinNr (normalized minimum pitch) but greater Min (minimum pitch) and Sd (sentence duration) than solo speech. It is also shown that the higher the Mid (median pitch), the MaxNr and the MinNr in solo speaking mode, the more they are reduced in synchronous speaking mode. Max, Min and Mid show greater speaker discriminability than other features.

• 품질경영학회지
• /
• 제48권3호
• /
• pp.421-431
• /
• 2020

Purpose: The purpose of this study is to compare and analyze the tokenizer in natural language processing for customer satisfaction in sentiment analysis. Methods: In this study, a supervised learning-based tokenizer Mecab-Ko and an unsupervised learning-based tokenizer SentencePiece were used for comparison. Three algorithms: Naïve Bayes, k-Nearest Neighbor, and Decision Tree were selected to compare the performance of each tokenizer. For performance comparison, three metrics: accuracy, precision, and recall were used in the study. Results: The results of this study are as follows; Through performance evaluation and verification, it was confirmed that SentencePiece shows better classification performance than Mecab-Ko. In order to confirm the robustness of the derived results, independent t-tests were conducted on the evaluation results for the two types of the tokenizer. As a result of the study, it was confirmed that the classification performance of the SentencePiece tokenizer was high in the k-Nearest Neighbor and Decision Tree algorithms. In addition, the Decision Tree showed slightly higher accuracy among the three classification algorithms. Conclusion: The SentencePiece tokenizer can be used to classify and interpret customer sentiment based on online reviews in Korean more accurately. In addition, it seems that it is possible to give a specific meaning to a short word or a jargon, which is often used by users when evaluating products but is not defined in advance.

• 컴퓨터교육학회논문지
• /
• 제21권6호
• /
• pp.105-115
• /
• 2018

서술형 문항은 수준 높은 사고능력 평가에 적합하나 채점이 쉽지 않다. 동일한 채점기준을 적용하더라도 채점자마다 서로 다른 채점을 할 수 있으므로 객관적인 채점 시스템이 필요하다. 그렇지만 채점 시스템을 구축하기 위해서는 표현 언어인 한국어 분석이 전제되어야 한다. 특히 서술형 문항의 답변은 단문형태가 아닌 복문이 내포된 복합 문장으로 표현되기 때문에 복합문의 한국어 분석이 이루어져야 한다. 본 연구는 이와 같은 서술형 문항의 자동 채점을 위해 한국어의 복합문을 구문의미 분석하는 시스템을 개발하고 시스템의 성능을 분석하였다. 이 시스템은 의미속성 사전을 활용하여 대상 구문의 의미적 제한조건을 검사하고 적합한 피수식 대상을 선별하여 구문의미분석을 실행한다. 실험 데이터에 대해 93%의 정확률을 얻어 제안한 시스템이 효과있음을 알 수 있었다. 본 연구의 시스템은 서술형 문항 채점에 활용할 수 있고 한국어 처리의 응용영역에 활용할 수 있을 것으로 기대된다.

• 한국언어정보학회지:언어와정보
• /
• 제4권1호
• /
• pp.97-109
• /
• 2000

Application systems of natural language processing such as machine translation system must deal with actual texts including the full range of linguistic phenomena. But it seems to be impossible that the existing grammar covers completely such actual texts because they include disruptive factors such as long sentences, unexpected sentence patterns and erroneous input to obstruct well-formed analysis of a sentence. In order to solve analysis failure due to the disruptive factors or incorrect selection of correct parse tree among forest parse trees, this paper proposes two-level computational grammar which consists of a constraint-based grammar and an error-tolerant grammar. The constraint-based computational grammar is the grammar that gives us the well-formed analysis of English texts. The error-tolerant computational grammar is the grammar that reconstructs a comprehensible whole sentence structure with partially successful parse trees within failed parsing results.

• KDI Journal of Economic Policy
• /
• 제39권3호
• /
• pp.1-18
• /
• 2017

The anchoring effect can be found when a decision shows cognitive prejudice towards the initial information given. Several studies have argued that such an effect is present even for judges in the courtroom. This paper seeks to find a relationship between judges' decisions on penalty sentences and the sentences recommended by prosecutors. In this study, 2,773 actual court cases are considered in the analysis, and quantile regression is used to show that the sentencing decisions judges make are anchored by the recommendations of prosecutors. However, this reliance on recommendations differs according to the seriousness of the crime committed. Specifically, at the lowest penalty levels, a one-month increase in the prosecutors' sentencing recommendation results in a 0.25-month increase in the judges' sentence, while at the highest sentence level, the judges' sentences increase by 0.78 months under an identical condition. The results of this research indicate the need to create more objective and clear sentencing guidelines in the future in an effort to mitigate the psychological pressure experienced by judges with regard to serious offences or heinous crimes.

• 음성과학
• /
• 제11권4호
• /
• pp.185-195
• /
• 2004

Spastic dysarthria is a motor speech disorder produced by a damage to the direct (pyramidal) and indirect (extrapyramidal) pathways of the central nervous system. This study attempted to determine the influences of the severity of spastic dysarthria on the stability of voice. A sustained /a/, /a/ produced at the beginning of a sentence and at the end of the sentence were analyzed. The Praat 4.0.51 was used for analysis. The results showed a significant difference between the severe spastic dysarthric group and normal group matched by age and sex at the beginning and the end of the sentence. In addition, the severe group showed a significant difference between the first /a/ and the second /a/ at the end of sentence.

• 한국정보통신학회논문지
• /
• 제21권2호
• /
• pp.461-467
• /
• 2017

인터넷의 발전과 스마트 디바이스의 확산으로 e-book에 대한 수요가 늘고 있다. 그러나 청각 손실로 정확한 언어를 배우기 어려운 청각장애인은 텍스트로만 이루어진 e-book 서비스를 사용하기 어렵다. 본 논문에서는 e-book의 문장을 읽어 수화 동영상으로 제공하는 안드로이드 기반 애플리케이션 프로그램을 설계하고 구현하였다. e-book의 한국어 문장을 수화 언어로 번역하기 위해 형태소 분석에 기반한 알고리즘을 사용하였다. 제안한 알고리즘은 3단계로 구성된다. 1단계는 수화 표현을 위한 문장 요소 제거, 2단계는 수화 표현의 변환 및 시제 표현, 3단계는 수화 높임말 용어 변경 및 위치 이동이다. 또한 수화 번역 품질에 대한 평가 방안을 제시하고 100개의 기준 문장에 대해 제안한 알고리즘을 통한 번역 결과의 우수성을 확인하였다.

• 전자공학회논문지B
• /
• 제31B권11호
• /
• pp.124-135
• /
• 1994

본 논문에서는 한국어 특수구문의 처리를 위한 국어 구문 분석 시스템을 제안한다. 문법의 기술은 동사와 의미를 통합적으로 처리하는 HPSG를 채택하고, 파싱 기법으로는 한국어에 유리한 단방향 활성 차트 파싱을 사용한다. 본 논문의 파서는 포괄적인 문장 구조(보어-중심어 구조, 수식어-중심어 구조, 중심어-중심어 구조)의 처리뿐만 아니라, 실용적인 문장에서 많이 나타나는 보조용언 구몬, 사동문, 피동문, 명사화 어미, 존칭, 화계와 같은 특수구문에 대해서도 파싱을 할 수 있도록 구현되었다.

• 인문언어
• /
• 제8집
• /
• pp.369-387
• /
• 2006

Which syntactic function should we assign to the 'ga-type' constituent which occurs in the morphological passive constructions in Korean, [N0-neun N1-i Vpass-ending]? This problem is very important in two respects. First, a small change of status of the particle 'i/ga' can exert an overall influence on the Korean grammar. Second, the particle '-i/ga' cannot guarantee that 'ga-type' constituents are subject of the sentence, so that the concept of syntactic category should be distinguished from that of syntactic function. This paper claims that the analysis of sentence has long been focused on the structure of proposition, namely the argument structure and that the direction of analysis should be turned to the 'person structure' which can be revealed on the pragmatic level. On the basis of this, this paper suggests that the specific type of the morphological passive constructions in Korean, [N0-neun N1-i Vpass-ending] should be analysed in line with the psych-verb constructions and that the modal meaning 'potential' of the passive constructions is correlated with sentence pattern and 'person structure'.