• Proceedings of the Acoustical Society of Korea Conference
• /
• spring
• /
• pp.7-10
• /
• 2000

Studies of prosody and sentence processing have demonstrated that prosodic phrasing can exhibit strong effects on processing decisions in English. In this paper, we tested Korean sentence fragments containing syntactically ambiguous Adj-N1-N2 strings in a cross-modal naming task. Four accentual phrasing patterns were tested: (a) the default phrasing pattern, in which each word forms an accentual phrase; (b) a phrasing biased toward N1 modification; (c) a phrasing biased toward complex-NP modification; and (d) a phrasing used with adjective focus. Patterns (b) and (c) are disambiguating phrasings; the other two are commonly found with both interpretations and are thus ambiguous. The results showed that the naming time of items produced in the prosody contradicting the semantic grouping is significantly longer than that produced in either default or supporting prosody, We claim that, as in English, prosodic information in Korean is parsed into a well-formed prosodic representation during the early stages of processing. The partially constructed prosodic representation produces incremental effects on syntactic and semantic processing decisions and is retained in memory to influence reanalysis decisions.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.537-540
• /
• 2018

우리는 단어 임베딩에 외부지식을 내재할 수 있는 Ontofitting 방법을 제안한다. 이 방법은 retrofitting의한 방법으로 유의어, 반의어, 상위어, 하위어 정보를 단어 임베딩에 내재할 수 있다. 유의어와 반의어 정보를 내재하기 위해서 벡터의 각 유사도를 사용하였고 상하위어 정보를 내재하기 위해서 벡터의 길이 정보를 사용하였다. 유의어 사이에는 작은 각도를 가지고 반의어 사이에는 큰 각도를 가지게 된다. 하위어는 상위어보다 상대적으로 작은 길이를 가지게 된다. SimLex와 HyperLex로 실험하여 효과와 안정성을 검증하였다. 의미정보를 내재한 임베딩을 사용할 수 있다면 QA, 대화 등 응용에서 보다 좋은 성능을 보일 수 있을 것이다.

• Speech Sciences
• /
• v.13 no.2
• /
• pp.107-120
• /
• 2006

The main purpose of this study is to investigate whether Korean learners can use the vowel duration cue to distinguish voicing contrasts in word-final consonants in English. Given that the Korean group's performance on the auditory task was much better than their performance on the identification task or on the production task, we conclude that the AX discrimination task makes contact with a different layer of perception. In particular, the AX discrimination task can be done at the auditory or phonetic level, where differences in vowel length are still encoded in the representation. In contrast, the identification and production tasks are probing the mental representation of vowel length and voicing. It was also founded that Korean speakers stored neither vowel length nor voicing in memorized representations and did not internalize the lengthening of the preceding vowel as a rule to differentiate the voicing contrasts of final consonants, even though they were able to detect the acoustic differences in vowel duration provided that they were tested in an appropriate task.

• Speech Sciences
• /
• v.2
• /
• pp.159-172
• /
• 1997

This paper proposes a prosodic representation system of the Kyungnam dialect of Korean, based on the ToBI system. In this system, diverse intonation patterns are transcribed on the four parallel tiers: a tone tier, a break index tier, an orthographic tier, and a miscellaneous tier. The tone tier employs pitch accents, phrase accents, and boundary tones marked with diacritics in order to represent various pitch events. The break index tier uses five break indices, numbered from 0 to 4, in order to represent degrees of connectiveness in speech by associating each inter-word position with a break index. In this, each break index represents a boundary of some kind of constituent. This system can contribute not only to a more detailed theory connecting prosody, syntax, and intonation, but also to current text-to-speech synthesis approaches, speech recognition, and other quantitative computational modellings.

• Journal of Information Processing Systems
• /
• v.13 no.4
• /
• pp.863-875
• /
• 2017

The traditional text similarity measurement methods based on word frequency vector ignore the semantic relationships between words, which has become the obstacle to text similarity calculation, together with the high-dimensionality and sparsity of document vector. To address the problems, the improved singular value decomposition is used to reduce dimensionality and remove noises of the text representation model. The optimal number of singular values is analyzed and the semantic relevance between words can be calculated in constructed semantic space. An inverted index construction algorithm and the similarity definitions between vectors are proposed to calculate the similarity between two documents on the semantic level. The experimental results on benchmark corpus demonstrate that the proposed method promotes the evaluation metrics of F-measure.

• Proceedings of the Korea Information Processing Society Conference
• /
• 2018.10a
• /
• pp.613-616
• /
• 2018

Sequence tagging is an important task in Natural Language Processing (NLP), in which the Named Entity Recognition (NER) is the key issue. So far the most widely adopted model for NER in NLP is that of combining the neural network of bidirectional long short-term memory (BiLSTM) and the statistical sequence prediction method of Conditional Random Field (CRF). In this work, we improve the prediction accuracy of the BiLSTM by supporting an aligned word representation mechanism. We have performed experiments on multilingual (English, Spanish and Dutch) datasets and confirmed that our proposed model outperformed the existing state-of-the-art models.

• Annual Conference on Human and Language Technology
• /
• 2019.10a
• /
• pp.161-164
• /
• 2019

본 논문은 BERT 모델을 이용하여 동형이의어의 단어 표현(Word Representation) 차이에 대한 실험을 한다. BERT 모델은 Transformer 모델의 인코더 부분을 사용하여 양방향을 고려한 단어 예측과 문장 수준의 이해를 얻을 수 있는 모델이다. 실험은 동형이의어에 해당되는 단어의 임베딩으로 군집화를 수행하고 이를 Purity와 NMI 점수로 계산하였다. 또한 각 단어 임베딩 사이를 코사인거리(Cosine Distance)로 계산하고 t-SNE를 통해 계층에 따른 변화를 시각화하였다. 군집된 결과는 모델의 중간 계층에서 점수가 가장 높았으며, 코사인거리는 8계층까지는 증가하고 11계층에서 급격히 값이 변하는 것을 확인할 수 있었다.

• Phonetics and Speech Sciences
• /
• v.10 no.4
• /
• pp.77-89
• /
• 2018

Speech articulators are coordinated for the purpose of segmental constriction in terms of a task. In particular, vertical jaw movements repeatedly contribute to consonantal as well as vocalic constriction. The current study explores vertical jaw movements in conjunction with bilabial constriction in bilabial stop /p/ in the context /a/-to-/a/. Revisiting kinematic data of /p/ collected using the electromagenetic midsagittal articulometer (EMMA) method from seven (four female and three male) speakers of Seoul Korean, we examined maximum vertical jaw position, its relative timing with respect to the upper and lower lips, and lip aperture minima. The results of those dependent variables are recapitulated in terms of linguistic (different word boundaries) and paralinguistic (different speech rates) factors as follows. Firstly, maximum jaw height was lower in the across-word boundary condition (across-word < within-word), but it did not differ as a function of different speech rates (comfortable = fast). Secondly, more reduction in the lip aperture (LA) gesture occurred in fast rate, while word-boundary effects were absent. Thirdly, jaw raising was still in progress after the lips' positional extrema were achieved in the within-word condition, while the former was completed before the latter in the across-word condition. Lastly, relative temporal lags between the jaw and the lips (UL and LL) were more synchronous in fast rate, compared to comfortable rate. When these results are considered together, it is possible to posit that speakers are not tolerant of lenition to the extent that it is potentially realized as a labial approximant in either word-boundary condition while jaw height still manifested lower jaw position in the across-word boundary condition. Early termination of vertical jaw maxima before vertical lower lip maxima across-word condition may be partly responsible for the spatial reduction of jaw raising movements. This may come about as a consequence of an excessive number of factors (e.g., upper lip height (UH), lower lip height (LH), jaw angle (JA)) for the representation of a vector with two degrees of freedom (x, y) engaged in a gesture-based task (e.g., lip aperture (LA)). In the task-dynamic application toolkit, the jaw angle parameter can be assigned numerical values for greater weight in the across-word boundary condition, which in turn gives rise to lower jaw position. Speech rate-dependent spatial reduction in lip aperture may be able to be resolved by means of manipulating activation time of an active tract variable in the gestural score level.

• ETRI Journal
• /
• v.45 no.1
• /
• pp.93-104
• /
• 2023

Coreference resolution is a task in discourse analysis that links several headwords used in any document object. We suggest pointer networks-based coreference resolution for Korean using multi-task learning (MTL) with an attention mechanism for a hierarchical structure. As Korean is a head-final language, the head can easily be found. Our model learns the distribution by referring to the same entity position and utilizes a pointer network to conduct coreference resolution depending on the input headword. As the input is a document, the input sequence is very long. Thus, the core idea is to learn the word- and sentence-level distributions in parallel with MTL, while using a shared representation to address the long sequence problem. The suggested technique is used to generate word representations for Korean based on contextual information using pre-trained language models for Korean. In the same experimental conditions, our model performed roughly 1.8% better on CoNLL F1 than previous research without hierarchical structure.

• Annual Conference on Human and Language Technology
• /
• 1995.10a
• /
• pp.211-214
• /
• 1995

At what level of orthographic representation is phonology linked in thelexicon? Is it at the whole word level, the syllable level, letter level, etc? This question can be addressed by comparing the two scripts used in Korean, logographic Hanmoon and alphabetic/syllabic Hangul, on a task where judgements must be made about the phonology of a visually presented word. Four experiments are reported using a "homophone decision task" and manipulating the sub-lexical relationship between orthography and phonology in Hanmoon and Hangul, and the lexical status of the stimuli. Hangul words showed a much higher error rate in judging whether there was another word identically pronounced than both Hangul nonwords and Hanmoon words. It is concluded that the relationship between orthography and phonology in the lexicon differs according tn the type of script owing to the availability of sub-lexical information: the process of making a homophone derision is based on a spread of activation exclusively among lexical entries, from orthography to phonology and vice versa (called "Orthography-Phonology-Orthography Rebound" or "OPO Rebound"). The results are explained within the mulitilevel interactive activation model with orthographic units linked to phonological units at each level.