• Journal of KIISE
• /
• v.42 no.3
• /
• pp.367-378
• /
• 2015

This study was directed at the design of a hybrid algorithm for competition relation extraction. Previous works on relation extraction have relied on various lexical and deep parsing indicators and mostly utilize only the machine learning method. We present a new algorithm integrating machine learning with various filtering methods. Some simple but useful features for competition relation extraction are also introduced, and an optimum feature set is proposed. The goal of this paper was to increase the precision of competition relation extraction by combining supervised learning with various filtering methods. Filtering methods were employed for classifying compete relation occurrence, using distance restriction for the filtering of feature pairs, and classifying whether or not the candidate entity pair is spam. For evaluation, a test set consisting of 2,565 sentences was examined. The proposed method was compared with the rule-based method and general relation extraction method. As a result, the rule-based method achieved positive precision of 0.812 and accuracy of 0.568, while the general relation extraction method achieved 0.612 and 0.563, respectively. The proposed system obtained positive precision of 0.922 and accuracy of 0.713. These results demonstrate that the developed method is effective for competition relation extraction.

• The Journal of the Korea Contents Association
• /
• v.20 no.8
• /
• pp.101-113
• /
• 2020

This study aimed to explore how effective explicit form-focused instruction (FFI) is in teaching the schwa vowel /ə/ to EFL students in a classroom setting. The participants were 25 female high school students, who were divided into the experimental group (n=13) and the control group (n=12). One female American also participated in the study for a speech sample as a reference. The treatment, which involves shadowing model pronunciation by the researcher and a free text-to-speech software and the researcher's feedback in a private session, was given to the control group over a month and a half. The speech samples, for which the participants read the 14 polysyllabic stimulus words followed by the sentences containing the words, were collected before and after the treatment. The paired-samples t test and non-parametric Wilcoxon signed-rank test were used for analysis. The results showed that the participants of the experimental group in the post-test reduced the duration of the schwa by around 40 percent compared to the pre-test. However, little effect was found in approximating the participants' distribution patterns of /ə/ measured by the F1/F2 formant frequencies to the reference point, which was 539 Hz (F1) by 1797 Hz (F2). The findings of this study suggest that explicit FFI with multiple repetitions and corrective feedback is partly effective in teaching pronunciation.

• Journal of the Korea Academia-Industrial cooperation Society
• /
• v.20 no.7
• /
• pp.158-165
• /
• 2019

Although evaluation of writing disabilities identification and timely intervention are required, clinicians adopt a manual scoring method and there is a possibility of error due to subjective evaluation. In this study, the size ratio and position of letters are digitized and quantified through image processing of offline handwritten characters. We tried to evaluate objectively and accurately the performance of writing through comparison with existing methods. From November 12th to 16th, 2018, 20 adults without neurological injury were selected. They used a pencil to follow the 10 words, 2 sentence stimuli after keeping the usual habit, and we collected the writing test data. The results showed that the height of the word was 1.2 times larger than the width and it tilted to the lower left. The spacing interval was 9mm on average. In the Paired T test, a high correlation was showed between our system and existing methods in the word and sentence 2. This demonstrated the possibility as a testing tool. This study evaluated objectively and precisely writing performance of offline handwritten characters through image processing and provided preliminary data for performance standards. In the future, it can be suggested as a basic data on writing diagnosis of various ages.

• Phonetics and Speech Sciences
• /
• v.14 no.2
• /
• pp.1-10
• /
• 2022

A wh-phrase positioned in an embedded clause can be interpreted as having a matrix scope if the sentence is produced with proper prosodic structures such as the wh-intonation. In a previous experiment, a sentence with a wh-phrase in an embedded clause was given to 40 speakers of Gyeongsang Korean. A script containing the sentence was provided to induce a matrix scope interpretation for the wh-phrase. These 40 utterances were prepared as stimuli for a perception test to verify whether the wh-phrases in the stimuli were perceived as having matrix scopes. Each utterance was played thrice to 24 subjects. The results showed that more than half of the 72 responses indicated a preference for an embedded scope rather than a matrix scope in 20 of the utterances. A multiple linear regression analysis showed that the matrix scope responses were best predicted by the magnitude of the pitch prominence in a prosodic word consisting of an embedded verb and a complementizer. The pitch prominence was calculated by subtracting the fundamental frequency (F0) at the right edge of the prosodic word from the peak F0 in the same prosodic word. The smaller the magnitude, the more matrix responses there were. These results suggest that the categorical perception of wh-scopes is based on the magnitude of pitch prominence.

• The Journal of the Acoustical Society of Korea
• /
• v.41 no.3
• /
• pp.359-366
• /
• 2022

In this paper, we present the Korean menu-ordering Sentence Text-to-Speech (TTS) system using conformer-based FastSpeech2. Conformer is the convolution-augmented transformer, which was originally proposed in Speech Recognition. Combining two different structures, the Conformer extracts better local and global features. It comprises two half Feed Forward module at the front and the end, sandwiching the Multi-Head Self-Attention module and Convolution module. We introduce the Conformer in Korean TTS, as we know it works well in Korean Speech Recognition. For comparison between transformer-based TTS model and Conformer-based one, we train FastSpeech2 and Conformer-based FastSpeech2. We collected a phoneme-balanced data set and used this for training our models. This corpus comprises not only general conversation, but also menu-ordering conversation consisting mainly of loanwords. This data set is the solution to the current Korean TTS model's degradation in loanwords. As a result of generating a synthesized sound using ParallelWave Gan, the Conformer-based FastSpeech2 achieved superior performance of MOS 4.04. We confirm that the model performance improved when the same structure was changed from transformer to Conformer in the Korean TTS.

• Proceedings of the Korea Institute of Convergence Signal Processing
• /
• 2005.11a
• /
• pp.233-237
• /
• 2005

수화의 인식 및 표현 기술에 대한 관련 연구는 수화 인식을 통한 건청인과의 의사 전달, 가상현실에서의 손동작 인식 등을 대상으로 여러 방면으로의 접근 및 연구 결과를 도출하고 있다. 그러나 이들 연구의 대부분 데스크탑 PC기반의 수신호(Hand signal) 제어 및 수화 - 손 동작 인식에 목적을 두었고 수화 신호의 획득을 위하여 영상장비를 이용하였으며 이를 바탕으로 단어 위주의 수화 인식 및 표현에 중점을 둔 수화 인식 시스템의 구현을 통해 비장애인과의 자유로운 의사소통을 추구하고 있다. 따라서 본 논문에서는 햅틱 장치로부터 사용자의 의미있는 수화 제스처를 획득하기 위한 접근 방식을 차세대 착용형 PC 플랫폼 기반의 유비쿼터스 환경으로 확대, 적용시켜 제스처 데이터 입력 모듈로부터 새로운 정보의 획득에 있어 한계성을 극복하고 사용자의 편의를 도모할 수 있는 효율적인 데이터 획득 방안을 제시한다. 또한 퍼지 알고리즘 및 RDBMS 모듈을 이용하여 언제, 어디에서나 사용자의 의미 있는 문장형 수화 제스처를 실시간으로 인식하고 표현하는 수화 제스처 인식기를 구현하였다. 본 논문에서는 수화 제스처 입력 모듈(5th Data Glove System과 $Fastrak{\circledR}$)과 차세대 착용형 PC 플랫폼(embedded I.MX21 board)간의 이격거리를 반경 10M의 타원 형태로 구성하고 규정된 위치로 수화 제스처 데이터 입력모듈을 이동시키면서 5인의 피실험자에 대하여 연속적으로 20회의 반복 실험을 수행하였으며 사용자의 동적 제스처 인식 실험결과 92.2% 평균 인식률을 도출하였다.

• Phonetics and Speech Sciences
• /
• v.3 no.4
• /
• pp.63-70
• /
• 2011

Recent studies on English $L^*$ (low pitch accent) have revealed the difference of changes in acoustic manifestation between utterances produced by Koreans and those produced by native speakers of English. However, not much effort has been made to compare $L^*$ focused constituents and non-focused constituents. At the same time, most previous works on focus realization are lacking in terms of normalization of acoustic measurement. Therefore, this research is dedicated to comparing the $L^*$ focused items and non-focused items realized by Koreans and Americans and to examining the realization of English $L^*$ produced by the two language groups with improved normalization of the acoustic features (F0, intensity and duration). Within-group analysis comparing focused words and non-focused words showed both Americans and Koreans prolonged the $L^*$ focused syllables but the effect size of syllable lengthening made by Koreans was far less than that made by Americans. Furthermore, significant F0 lowering was found in Americans but not in Koreans. However, the effect of intensity change caused by $L^*$ focus was not significant within each group. The effect of focused words was tested between the two groups revealing that Koreans implemented English $L^*$ focus with higher F0, lower intensity and shorter duration than Americans. In the instances in which a significant Group x Focus Location (initial, middle and final of a sentence) interaction was found, further analysis testing the effect of Group on each Focus Location was conducted. The testing showed that the Koreans produced shorter syllables at initial and middle of a sentence and higher F0 at initial of a sentence than Americans. Implications for the intonation training were also discussed.

• Journal of KIISE
• /
• v.44 no.8
• /
• pp.774-781
• /
• 2017

Mention detection systems use nouns or noun phrases as a head and construct a chunk of text that defines any meaning, including a modifier. The term "mention detection" relates to the extraction of mentions in a document. In the mentions, a coreference resolution pertains to finding out if various mentions have the same meaning to each other. A pointer network is a model based on a recurrent neural network (RNN) encoder-decoder, and outputs a list of elements that correspond to input sequence. In this paper, we propose the use of mention detection using pointer networks. Our proposed model can solve the problem of overlapped mention detection, an issue that could not be solved by sequence labeling when applying the pointer network to the mention detection. As a result of this experiment, performance of the proposed mention detection model showed an F1 of 80.07%, a 7.65%p higher than rule-based mention detection; a co-reference resolution performance using this mention detection model showed a CoNLL F1 of 52.67% (mention boundary), and a CoNLL F1 of 60.11% (head boundary) that is high, 7.68%p, or 1.5%p more than coreference resolution using rule-based mention detection.

• The KIPS Transactions:PartB
• /
• v.19B no.1
• /
• pp.43-52
• /
• 2012

Semantic role labeling is the research area analyzing the semantic relationship between elements in a sentence and it is considered as one of the most important semantic analysis research areas in natural language processing, such as word sense disambiguation. However, due to the lack of the relative linguistic resources, Korean semantic role labeling research has not been sufficiently developed. We, in this paper, propose an automatic predicate-argument analyzer to begin constructing the Korean PropBank which has been widely utilized in the semantic role labeling. The analyzer has mainly two components: the semantic lexical dictionary and the automatic predicate-argument extractor. The dictionary has the case frame information of verbs and the extractor is a module to decide the semantic class of the argument for a specific predicate existing in the syntactically annotated corpus. The analyzer developed in this research will help the construction of Korean PropBank and will finally play a big role in Korean semantic role labeling.

• The Journal of the Acoustical Society of Korea
• /
• v.16 no.5
• /
• pp.50-59
• /
• 1997

In this paper, we implement an automatic speech segmentation and labeling system which marks phone boundaries automatically for constructing the Korean speech database. We specify and implement the system based on conventional speech segmentation and labeling techniques, and also develop the graphic user interface(GUI) on Hangul $Motif^{TM}$ environment for the users to examine the automatic alignment boundaries and to refine them easily. The developed system is applied to 16kHz sampled speech, and the labeling unit is composed of 46 phoneme-like units(PLUs) and silence. The system uses both of the phonetic and orthographic transcription as input methods of linguistic information. For pattern-matching method, hidden Markov models(HMM) is employed. Each phoneme model is trained using the manually segmented 445 phonetically balanced word (PBW) database. In order to evaluate the performance of the system, we test it using another database consisting of sentence-type speech. According to our experiment, 74.7% of phoneme boundaries are within 20ms of the true boundary and 92.8% are within 40ms.