• Proceedings of the KSPS conference
• /
• 2007.05a
• /
• pp.173-175
• /
• 2007

The lexical decision task (LDT) commonly postulates the activation of semantic level. However, there are few studies for the feedback effect from semantic level. The purpose of the present study is to investigate whether the feedback effect from semantic level is facilitatory or inhibitory in Korean LDT. In Experiment 1, we manipulated the number of phonological syllable neighbors (PSN) and the number of semantic neighbors (SEN) orthogonally while orthographic syllable neighbor (OSN) is dense. In the results, the significant facilitatory effect was shown in words with many SEN. In Experiment 2, we examined same conditions as Experiment 1 but OSN was sparse. Although the similar lexical decision latency pattern was shown, there was no statistical significance. These results can be explained by the feedback activation from semantic level. If a target has many SENs and many PSNs, it receives more feedback activation from semantic level than a target with few SENs and PSNs.

• Journal of the Korean Institute of Telematics and Electronics B
• /
• v.31B no.9
• /
• pp.133-141
• /
• 1994

Conventional competitive learning algorithms compute the Euclidien distance to determine the winner neuron out of all predetermined output neurons. In such cases, there is a drawback that the performence of the learning algorithm depends on the initial reference(=weight) vectors. In this paper, we propose a new competitive learning algorithm that dynamically generates output neurons. The proposed method generates output neurons by dynamically changing the class thresholds for all output neurons. We compute the similarity between the input vector and the reference vector of each output neuron generated. If the two are similar, the reference vector is adjusted to make it still more like the input vector. Otherwise, the input vector is designated as the reference vector of a new outputneuron. Since the reference vectors of output neurons are dynamically assigned according to input pattern distribution, the proposed method gets around the phenomenon that learning is early determined due to redundant output neurons. Experiments using speech data have shown the proposed method to be superior to existint methods.

• Proceedings of the KSPS conference
• /
• 2006.05a
• /
• pp.67-70
• /
• 2006

Prosody can be used to resolve syntactic ambiguity of a sentence. English relative clause construction with complex NP(the N1, N2, and RC sequence) has syntactic ambiguity and the clause can be interpreted as modyfying N1(high attachment) or N2(low attachment), Speakers and listeners can disambiguate those sentences based on the prosody. In this paper, we investigate the Korean English learners production on the prosodic structure of English relative clause construction. The production experiment shows that the beginner learners use the phrasing frequently and the advanced learners depend on both the phrasing and the accent. One of the characteristic of the Korean English learners' intonation is that the Korean accentual phrase tone pattern LHa is transferred to their production.

• MALSORI
• /
• no.42
• /
• pp.71-88
• /
• 2001

This study is on the constraints of English Poetic Meter. In English poems, the metrical pattern doesn't always match the linguistic stress on the lines. These mismatches are found differently among the poets. The peaks mismatched with the weak metrical position are divided into the two ways according as they are adjacent to the boundary of a phonological domain or not. PAF and $^*UV$] are suggested for the mismatched peak which are not adjacent to the boundary of a phonological domain ; $^*Peak$] and BT for the mismatched peak which are adjacent to the boundary of a phonological domain. For the lexical stress mismatched with the weak metrical position, $^*W{\;}{\Rightarrow}{\;}Strength$ is set up by the concept of the strong syllable. $MPS{\;}{\Rightarrow}{\;}\Phi_{max}$ for the metrical position size can replace the resolution which is used to control the number of syllables in English poems. These constraints show the different hierarchies among the poets.

• The Journal of the Acoustical Society of Korea
• /
• v.13 no.1E
• /
• pp.64-70
• /
• 1994

Predictive neural network models are powerful speech recognition models based on a nonlinear pattern prediction. But those models suffer from poor discrimination between acoustically similar words. In this paper we propose an discriminative training algorithm for predictive neural network models. This algorithm is derived from GPD (Generalized Probabilistic Descent) algorithm coupled with MCEF(Minimum Classification Error Formulation). It allows direct minimization of a recognition error rate. Evaluation of our training algoritym on ten Korean digits shows its effectiveness by 30% reduction of recognition error.

• Proceedings of the Korean Information Science Society Conference
• /
• 2008.06c
• /
• pp.456-461
• /
• 2008

클래스별 원형상(prototype)의 분포가 선형분리 불가능하고 동시에 분산이 서로 다르고 희박한 분포의 원형상에 있어서 입력패턴에 대한 고정밀도의 식별을 행하기 위해 클래스별 최근방 원형상과 그 k 근방 원형상에 있어서 노름(norm) 평균에 기초한 최근방 결정법에 의한 패턴식별방법을 제안한다. 제안하는 방법의 유효성을 평가하기위해 인공적인 패턴과 실제 패턴에 대해 일반적인 k-NN법, 매해라노비스 거리(maharanobis distance), CAP, kCAP, SVM의 각각에 기초한 방법과 제안하는 방법을 적용하여 식별률에 의한 평가를 행하였다. 그 결과 특히, 원형상의 분포가 희박한 경우 제안하는 방법이 다른 방법들에 비해 높은 식별률을 나타냈다.

• Proceedings of the Korean Information Science Society Conference
• /
• 2000.04b
• /
• pp.441-443
• /
• 2000

연속 음성 인식을 하는 경우에 많은 에러가 발생한다. 특히 기능어의 경우나 서술어의 경우에는 동시 조음 현상에 의한 음운 변화에 의해 빈번한 에러가 발생한다. 이러한 빈번한 에러를 수정하기 위한 방법에는 언어 모델의 개선과 음향 모델의 개선등을 통한 인식률 향상과 여러 단계의 인식과정을 두어 서로 다른 언어 모델을 적용하는 등의 방법이 있지만 모두 시간과 비용이 많이 들고 각각의 상황에 의존적인 단점이 있다. 따라서 본 논문에서 제안하는 방법은 이것을 수정하기 위해 음성 인식기로부터 인식되어 나온 결과 문장을 정답과 비교, 학습함으로써 빈번하게 에러 패턴을 통계적 방법에 의해 학습하고 후처리 모듈을 이용하여 인식시에 발생하는 에러를 적은 비용과 시간으로 수정할 수 있도록 하는 것이다. 실험은 3000 단어급의 한국어 낭독체 연속 음성을 대상으로 하여 형태소와 의사형태소를 각각 인식단위로 하고, 언어모델로 World bigram과 Tagged word bigram을 각각 적용 실험을 하였다. 형태소, 의사 형태소일 경우 모두 언어 모델을 tagged word bigram을 사용하였을 경우 N best 후보 문장 중 적당한 단어 후보의 분포로 각각 1 best 문장에 비해 12%, 18%정도의 에러 수정하여 문장 인식률 향상에 상당한 기여를 하였다.

• Proceedings of the KIEE Conference
• /
• 1988.07a
• /
• pp.624-627
• /
• 1988

This paper describes the implementation of recognition of speaker - dependent Korean spoken continuous digits. The recognition system can be divided into two parts, acoustic - phonetic processor and lexical decoder. Acoustic - phonetic processor calculates the feature vectors from input speech signal and the performs frame labelling and phone labelling. Frame labelling is performed by Bayesian classification method and phone labelling is performed using labelled frame and posteriori probability. The lexical decoder accepts segments (phones) from acoustic - phonetic processor and decodes its lexical structure through phone network which is constructed from phonetic representation of ten digits. The experiment carried out with two sets of 4continuous digits, each set is composed of 35 patterns. An evaluation of the system yielded a pattern accuracy of about 80 percent resulting from a word accuracy of about 95 percent.

• Proceedings of the Korea Society for Industrial Systems Conference
• /
• 2003.11a
• /
• pp.721-727
• /
• 2003

This study designed and implemented a system to extract of vowel from a word. The system is comprised of a voice feature extraction module and a neutral network module. The voice feature extraction module use a LPC(Linear Prediction Coefficient) model to extract a voice feature from a word. The neutral network module is comprised of a learning module and voice recognition module. The learning module sets up a learning pattern and builds up a neutral network to learn. Using the information of a learned neutral network, a voice recognition module extracts a vowel from a word. A neutral network was made to learn selected vowels(a, eo, o, e, i) to test the performance of a implemented vowel extraction recognition machine. Through this experiment, could confirm that speech recognition module extract of vowel from 4 words.

• Proceedings of the Korean Institute of Intelligent Systems Conference
• /
• 1995.10b
• /
• pp.85-91
• /
• 1995

This paper presents a system which recognizes the Korean Sign Language(KSL) and translates into normal Korean speech. A sign language is a method of communication for the deaf-mute who uses gestures, especially both hands and fingers. Since the human hands and fingers are not the same in physical dimension, the same form of a gesture produced by two signers with their hands may not produce the same numerical values when obtained through electronic sensors. In this paper, we propose a dynamic gesture recognition method based on feature analysis for efficient classification of hand motions, and on a fuzzy min-max neural network for on-line pattern recognition.