• The Journal of the Korea Contents Association
• /
• v.21 no.5
• /
• pp.285-294
• /
• 2021

Motivated by the mixed results in the previous studies on the relationship between speech production and perception, the current study aims to investigate the relationship between production and perception in depth through a case study on how Korean EFL university students produce and perceive American English vowels. To this end, 19 Korean students at a university located in the Seoul-metropolitan area participated in the production and perception tests on American English vowels to elucidate the precedence relationship and the correlation between production and perception. Results showed that precedence of neither perception nor production was found in the overall result. However, either precedence of perception or production was found for the vowels [ɛ], [α], [ɔ], [u], which implies that the precedence relationship between production and perception varies depending on individual vowels. As for the correlation between production and perception, no correlation was attested between production and perception, suggesting that production and perception skills are not closely linked for these participants. Given that mastering language requires to coordinate two distinct production and perception skills and that L2 learners' preception and production skills become more closely connected as the learners' L2 experience and proficiency increases, no correlation between production and perception attested by the current EFL students implies that the correlation between production and perception varies during the course of foreign language/L2 acquisition in such a way that production and perception skills become increasingly related. Implications of the findings were further discussed and pedagogical suggestions were provided.

• Annual Conference on Human and Language Technology
• /
• 2015.10a
• /
• pp.220-223
• /
• 2015

본 논문은 대화 시스템에서 사용되는 말뭉치의 구축을 위해 Object와 Action을 반자동으로 추출하는 도구에 대해 기술한다. 제안하는 추출 도구는 형태소 분석과 의존 구문 분석의 결과를 기반으로 적절한 Object와 Action을 추출하는 것에 목표를 두고 있다. 그러나 형태소 분석과 의존 구문 분석의 결과는 여러 가지 오류가 포함될 수 있다. 이러한 오류는 잘못된 Object와 Action의 추출로 이어질 수 있다. 그리고 Object의 추출에 있어 해당 명사의 격이 중요한 정보를 가진다. 하지만 한국어의 특성한 조사의 생략 등으로 인해 격 태깅의 모호성이 발생하게 된다. 따라서 본 논문에서 제안하는 반자동 추출기는 형태소 분석과 의존 구문 분석의 잘못된 결과를 사용자가 손쉽게 수정할 수 있도록 하고 모호성이 발생할 수 있는 Object를 사용자에게 알려주어 올바른 Object와 Action의 추출을 가능하게 한다. 추출기를 이용한 말뭉치의 구축은 1) 형태소 분석 2) 의존 구문 분석 3) Object-Action 추출의 단계로 진행된다. 실험에서 사용된 발화는 관광 회화용 대화 시스템의 숙박, 공항 영역의 500개의 발화이며, 이 중 259개의 발화가 태깅 시 모호성이 발생하는 발화이다. 반자동 추출기를 통해 모호성이 발생한 발화를 태깅한 결과 전체 발화 중 51.8%의 발화를 빠르고 정확하게 태깅할 수 있었다.

• Journal of KIISE:Software and Applications
• /
• v.30 no.3_4
• /
• pp.278-284
• /
• 2003

Speech act is the speaker's intentions indicated through utterances. It is important for understanding natural language dialogues and generating responses. This paper proposes the method of two stage that increases the performance of the korean speech act decision. The first stage is to select features from the part of speech results in sentence and from the context that uses previous speech acts. We use x$^2$ statistics(CHI) for selecting features that have showed high performance in text categorization. The second stage is to determine speech act with selected features and Neural Network. The proposed method shows the possibility of automatic speech act decision using only POS results, makes good performance by using the higher informative features and speed up by decreasing the number of features. We tested the system using our proposed method in Korean dialogue corpus transcribed from recording in real fields, and this corpus consists of 10,285 utterances and 17 speech acts. We trained it with 8,349 utterances and have test it with 1,936 utterances, obtained the correct speech act for 1,709 utterances(88.3%). This result is about 8% higher accuracy than without selecting features.

• The Journal of the Convergence on Culture Technology
• /
• v.8 no.1
• /
• pp.565-576
• /
• 2022

The scope and functions of a conversational agent are gradually expanding. In particular, research and technology development is being conducted on a conversational agent that can speak first without user calls. However, still in its early stages, there is a lack of research on how a preceded utterance conversational agent will affect users. Accordingly, this study conducted a 2×3 mixed design using the user's task performance condition and the agent's self-exposure as independent variables and measured Intimacy, Functional Satisfaction, Psychological Reactance, and Workload as dependent variables to identify the effects of preceded utterance conversational agent on user experience.

• Phonetics and Speech Sciences
• /
• v.12 no.3
• /
• pp.1-14
• /
• 2020

Previous studies have confirmed that /o/ and /u/ in Seoul Korean are undergoing a merger in the F1/F2 space, especially for female speakers. As a substitute parameter for formants, it is reported that female speakers use phonation (H1-H2) differences to distinguish /o/ from /u/. This study aimed to explore whether H1-H2 values are being used as perceptual cues for /o/-/u/. A perception test was conducted with 35 college students using /o/ and /u/ spoken by 41 females, which overlap considerably in the vowel space. An acoustic analysis of 182 stimuli was also conducted to see if there is any correspondence between production and perception. The identification rate was 89% on average, 86% for /o/, and 91% for /u/. The results confirmed that when /o/ and /u/ cannot be distinguished in the F1/F2 space because they are too close, H1-H2 differences contribute significantly to the separation of the two vowels. However, in perception, this was not the case. H1-H2 values were not significantly involved in the identification process, and the formants (especially F2) were still dominant cues. The study also showed that even though H1-H2 differences are apparent in females' production, males do not use H1-H2 in their production, and both females and males do not use H1-H2 in their perception. It is presumed that H1-H2 has not yet been developed as a perceptual cue for /o/ and /u/.

• Fire Science and Engineering
• /
• v.33 no.2
• /
• pp.1-8
• /
• 2019

This study was conducted to investigate the possibility of spontaneous ignition in Butadiene popcorn polymer, which is used as raw material and product in a chemical plant. A component analysis, thermogravimetric analysis, thermal stability analysis, spontaneous ignition point measurement and accelerated velocity calorimetric analysis were performed. As a result of analysis, various kinds of flammable components were measured and thermogravimetric analysis showed a weight loss of 95.6% in air and 89.2% in nitrogen. As a result of the thermal stability analysis, heat generation started at $88^{\circ}C$ in the air atmosphere, and the heat generation rate increased sharply in the vicinity of the natural ignition point ($220^{\circ}C$). The heat generation started at about $70^{\circ}C$ in nitrogen atmosphere, and the two exothermic peak values were observed up to $450^{\circ}C$. As a result of accelerated rate calorimetry, there was no exothermic phenomenon, and the lowest ignition temperature was $211.7^{\circ}C$ as a result of analysis of natural ignition point. Based on the results obtained from the thermal stability evaluation, it is considered that the possibility of inducing the thermal deformation of the column by the heat of reaction is sufficient.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.558-563
• /
• 2021

알츠하이머병 치매와 조현병 진단을 위한 2단계 분류 모델을 제안한다. 정상군과 환자군의 발화에 나타난 페어 언어 모델 간의 Perplexity 차이에 기반한 분류와 기존 단일 BERT 모델의 미세조정(fine-tuning)을 이용한 분류의 통합을 시도하였다. Perplexity 기반의 분류 성능이 알츠하이머병, 조현병 모두 우수한 결과를 보임을 확인 하였고, 조현병 분류 모델의 성능이 소폭 증가하였다. 향후 설명 가능한 인공지능 기법을 적용에 따른 성능 향상을 기대할 수 있었다.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.107-112
• /
• 2023

라벨 데이터 수집의 어려움에 따라 라벨이 없는 데이터로 학습하는 준지도학습, 비지도학습에 대한 연구가 활발하게 진행되고 있다. 본 논문에서는 그의 일환으로 Novel Intent Category Discovery(NICD) 문제를 제안하고 NICD 연구의 베이스라인이 될 모델을 소개한다. NICD 문제는 라벨이 있는 데이터와 라벨이 없는 데이터의 클래스 셋이 겹치지 않는다는 점에서 기존 준지도학습의 문제들과 차이가 있다. 제안 모델은 RoBERTa를 기반으로 두 개의 분류기를 추가하여 구성되며 라벨이 있는 데이터셋과 라벨이 없는 데이터셋에서 각각 다른 분류기를 사용하여 라벨을 예측한다. 학습방법은 2단계로 먼저 라벨이 있는 데이터셋으로 요인표현을 학습한다. 두 번째 단계에서는 교차 엔트로피, 이항교차 엔트로피, 평균제곱오차, 지도 대조 손실함수를 NICD 문제에 맞게 변형하여 학습에 사용한다. 논문에서 제안된 모델은 라벨이 없는 데이터셋에 대해 이미지 최고성능 모델보다 24.74 더 높은 정확도를 기록했다.

• Journal of the Society of Disaster Information
• /
• v.13 no.3
• /
• pp.296-304
• /
• 2017

In order to predict about forest fire behavior we constructed a database for combustion characteristic of forest fuels in Samcheok, Gangwon-do and prepared fire risk map and fire risk rating using GIS method in this study. For the mapping autoignition temperature, ignition time, flame duration time, total heat release and total smoke release are selected as the standardized parameters and the overall risk rating was made up of the ignition risk parameters(autoignition temperature, ignition time) and the spread risk parameters(flame duration time, total heat release, total smoke release). Forest fire risk was classified into 5 grades and lower grade of fire risk rating mean to correspond to more dangerous forest fire. As a result, the overall risk rating of Samcheok was classified into three grades from 1 to 3 and Nogok-myeon and Miro-myeon were turned out the most dangerous areas for forest fire. Because of the colony of pine and oak trees and the higher fire loads, the flame propagation will be carried out quickly in these areas.

• Fire Science and Engineering
• /
• v.19 no.4 s.60
• /
• pp.75-79
• /
• 2005

This study has developed Forest Fire Occurrence Danger Index (FFODI) using fuel and topographical characteristics for the practical purposes of forecasting forest fire occurrence danger rating. This was made on the basis of the 126 forest fire site according to field survey. The result of fire frequency analysis showed 87 sites on conifer $(69\%)$, 21 on mixed $(16.7\%)$ and 18 $(14.3\%)$ on non-conifer. The scale for Fuel Model Index(FMI) ranges from 1 to 10 and Topography Model Index(TMI) from 1 to 5. FMI is 10 on the conifer, 3 on the mixed and 2 on the non-conifer. In case of topographical analysis, it was estimated that 90 site $(71.4\%)$ of ignition point was bottom foot hill and 22 site $(17.5\%)$ was on the southwest. TMI in southwest direction is 5.0, 4.5 in the northwest and the northeast, 4.0 in the southeast and the south, 2.5 in the north and the west and 1.5 in the east. TMI in the bottom foot hill is 5 in the bottom foot hill, 1.5 in the upper foot hill, 1.0 in the bottom middle slope and 0.5 in the upper middle slope and bottom ridge.