• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.544-547
• /
• 2021

한국어의 품사 태깅 문제는 입력 어절의 형태소 분석 후보들로부터 통계적으로 적절한 품사 태그를 가지는 후보들을 찾는 방식으로 해결하여 왔다. 어절을 형태소 단위로 분리하고 품사를 부착하는 기존의 방식은 품사태그 정보를 딥러닝 feature로 사용할 때 문장의 의미를 이해하는데 복잡도를 증가시키는 요인이 된다. 본 연구에서는 품사 태깅 문제를 단순화 하여 한 어절을 Head와 Tail이라는 두 가지 유형의 형태소 토큰으로 분리하여 Head와 Tail에 대해 품사를 부착한다. Head-Tail 품사 태깅 방법을 Sequence-to-Sequence 문제로 정의하여 Transformer를 이용한 Head-Tail 품사 태거를 설계하고 구현하였다. 학습데이터로는 KCC150 말뭉치의 품사 태깅 말뭉치 중에서 788만 문장을 사용하고, 실험 데이터로는 10만 문장을 사용하였다. 실험 결과로 토큰 정확도는 99.75%, 태그 정확도는 99.39%, 토큰-태그 정확도는 99.31%로 나타났다.

• Smart Media Journal
• /
• v.11 no.5
• /
• pp.17-25
• /
• 2022

Korean part-of-speech taggers decompose a compound morpheme into unit morphemes and attach part-of-speech tags. So, here is a disadvantage that part-of-speech for morphemes are over-classified in detail and complex word types are generated depending on the purpose of the taggers. When using the part-of-speech tagger for keyword extraction in deep learning based language processing, it is not required to decompose compound particles and verb-endings. In this study, the part-of-speech tagging problem is simplified by using a Head-Tail tokenization technique that divides only two types of tokens, a lexical morpheme part and a grammatical morpheme part that the problem of excessively decomposed morpheme was solved. Part-of-speech tagging was attempted with a statistical technique and a deep learning model on the Head-Tail tokenized corpus, and the accuracy of each model was evaluated. Part-of-speech tagging was implemented by TnT tagger, a statistical-based part-of-speech tagger, and Bi-LSTM tagger, a deep learning-based part-of-speech tagger. TnT tagger and Bi-LSTM tagger were trained on the Head-Tail tokenized corpus to measure the part-of-speech tagging accuracy. As a result, it showed that the Bi-LSTM tagger performs part-of-speech tagging with a high accuracy of 99.52% compared to 97.00% for the TnT tagger.

• IEMEK Journal of Embedded Systems and Applications
• /
• v.17 no.4
• /
• pp.199-208
• /
• 2022

Korean is an agglutinative language, and one or more morphemes are combined to form a single word. Part-of-speech tagging method separates each morpheme from a word and attaches a part-of-speech tag. In this study, we propose a new Korean part-of-speech tagging method based on the Head-Tail tokenization technique that divides a word into a lexical morpheme part and a grammatical morpheme part without decomposing compound words. In this method, the Head-Tail is divided by the syllable boundary without restoring irregular deformation or abbreviated syllables. Korean part-of-speech tagger was implemented using the Head-Tail tokenization and deep learning technique. In order to solve the problem that a large number of complex tags are generated due to the segmented tags and the tagging accuracy is low, we reduced the number of tags to a complex tag composed of large classification tags, and as a result, we improved the tagging accuracy. The performance of the Head-Tail part-of-speech tagger was experimented by using BERT, syllable bigram, and subword bigram embedding, and both syllable bigram and subword bigram embedding showed improvement in performance compared to general BERT. Part-of-speech tagging was performed by integrating the Head-Tail tokenization model and the simplified part-of-speech tagging model, achieving 98.99% word unit accuracy and 99.08% token unit accuracy. As a result of the experiment, it was found that the performance of part-of-speech tagging improved when the maximum token length was limited to twice the number of words.

• Proceedings of the KIEE Conference
• /
• 2007.10a
• /
• pp.155-156
• /
• 2007

This paper presents the vision-based camber and optimal cutting line detection algorithm for hot-rolling process. It is important to measure the camber of head and tail part of strips because many problems are caused by the camber in the hot-rolling process. The hot-rolling process has time constraints. The camber detection algorithm of head and tail parts requires fast and less complex for satisfying time constraints. The proposed algorithm consists of two parts: measurement of the camber in the head and tail part of strips and decision part of the optimal cutting line of hot-rolled strip. First, we obtain the camber value of the strip from the difference between the real center line and the center line of head, tail part. Second, the head and tail part of strips isn't suitable for strips connections. Therefore, the cutting process is needed in the hot-rolling process. The optimal cutting line is determined by the head and tail images obtained from cameras. The algorithm is applied into the vision system with two area cameras, Matrox image processing board and host PC for verification.

• Journal of The Korean Astronomical Society
• /
• v.21 no.2
• /
• pp.117-132
• /
• 1988

Multi-frequency polarization observations of Abell 695 are reported here. The brightest radio source 0838+325, which was previously classified as a wide angled head-tail radio source, is, in the present observations, resolved into two separable sources, a head-tail source 0838+325 AB and a diffuse one 0838+325C. The radio-tail shows high degree of linear polarization(${\sim}50$-25%) along the radio-tail, suggesting that the interaction with its surrounding intracluster medium (ICM) is not highly turbulent. With the present data, thermal particle densities at the locations of these sources are estimated to be n > $10^{-5}\;cm^{-3}$.

• Applied Microscopy
• /
• v.12 no.1
• /
• pp.33-40
• /
• 1982

Several phages of Bacillus thuringiensis distributed in Korea were isolated. The distribution and morphological characteristics of phages were studied. The results are as follows; 1. The isolated phages were highly specific for Bacillus thuringiensis var. thuringiensis. They were classified as YM series phages and designated as phage YM-1, YM-2 and YM-3 according to their morphological characteristics. 2. Most of these YM series phages were isolated from compost including domestic animal dung and soil under sewage. 3. The YM-1 phage was similar to Bacillus subtilis ${\phi}25$ in morphology. It has 94nm x 86nm head, contractile tail sheath and base plate with four cornered structure. 4. The YM-2 phage was similar to Bacillus subtilis GA-1 phage in morphology. It had 70nm x 56nm head and tail without contractile tail sheath. 5. The YM-3 phage was similar to Bacillus subtilis ${\phi}29$ phage. It had 56nm x 43nm head and tail with distal enlargement.

• Transactions of the Korean Society of Mechanical Engineers A
• /
• v.23 no.7 s.166
• /
• pp.1196-1204
• /
• 1999

adaption of the model predictions is highly desirable. In general, the width deviation at the head and tail ends of strip may be different from that of the steady state region. Therefore, the dynamic edger corrections can be used to compensate the width deviations which would otherwise occur. For the precise width control, the effect of edger roll gap and rolling conditions on the width deviation of head and tail ends of strip has been investigated and the effective method to decrease width deviation has been proposed. On-line application of dynamic edger control method in this study shows about 50% width compensation at the head end of the strip, and near perfect compensation at the tail end of strip.

• Journal of Animal Environmental Science
• /
• v.5 no.2
• /
• pp.93-100
• /
• 1999

Pig's original image data was transformed to a binary image, an image excluding head and tail portion from the whole binary image, and a projected image associated with pig's height. Then the length of body, width of shoulder, and area of pig were calculated and the relationships among the above characteristics and pig's weight were analyzed. The results obtained from this study were as follows: 1. Whole binary image data was considered to be improper to determine the pig's weight because the movement of pig's head and tail portion affected the image data. 2. Binary image data excluding head and tail portion from the whole binary image showed a better estimation of the pig's weight than the whole binary image. 3. Pig's should width was analyzed to be improper factor to determine the pig's weight. 4. The projected image associated with pig's height showed the highest correlation between the pig's area of the image and pig's weight(R2=0.9965). From this research the projected image associated with pig's height, which is excluding head and tail portion from the whole body of pig's image, was considered to be the prime factor to measure the pig's weight by the noncontact measurement.

• Applied Microscopy
• /
• v.26 no.3
• /
• pp.277-291
• /
• 1996

The structures of the egg envelope in fertilized eggs of three species of characidae, head and tail light fish (Hemigrammus ocellifer), black tetra (Gymnocorymbus ternetzi), and buenos aires tetra (Hemigrammus caudovittatus) were studied using light and electron microscopes. The fertilized eggs in all species were colorless, transparent, spherical and non-floted type. The egg envelopes have a single micropyle resembling the pathway of sperm in the area of the animal pole. The micropyle was surrounded by protruded lines of the egg envelope in a radiated form. Egg envelopes of fertilized eggs in both head and tail light fish and buenos aires tetra consisted of three distinct layers; an outer layer, a middle layer and an inner layer. And that of blacktetra consisted of two layers; an outer layer and an inner layer. Also, an outer layers of both head and tail light fish and black tetra were adhesive types but, in that of buenous aires tetra was non-adhesive type. An outer surface of egg envelope in black tetra was arranged by pores regularly. In that of head and tail light fish and buenos aires tetra have a rough side. An inner layer of egg envelope in fertilized eggs consisted of lamellae alternating with interlamellae of lower electron density; an inner layer of fertilized eggs in head and tail light fish consisted of three layers, that of black tetra was four layers, and that of buenos aires tetra was five layers.

• Annual Conference on Human and Language Technology
• /
• 2023.10a
• /
• pp.726-729
• /
• 2023

본 논문에서는 대화 중 발화에서 페르소나 트리플을 추출하는 방법을 연구한다. 발화 문장과 그에 해당하는 트리플 쌍을 활용하여 발화 문장 혹은 페르소나 문장이 주어졌을 때 그로부터 페르소나 트리플을 추출하도록 모델을 멀티 태스크 러닝 방식으로 학습시킨다. 모델은 인코더-디코더 구조를 갖는 사전학습 언어모델 BART [1]와 T5 [2]를 활용하며 relation 추출과 tail 추출의 두 가지 태스크를 각각 인코더, 디코더 위에 head를 추가하여 학습한다. Relation 추출은 분류로, tail 추출은 생성 문제로 접근하도록 하여 최종적으로 head, relation, tail의 구조를 갖는 페르소나 트리플을 추출하도록 한다. 실험에서는 BART와 T5를 활용하여 각 태스크에 대해 다른 학습 가중치를 두어 훈련시켰고, 두 모델 모두 relation과 tail을 추출하는 태스크 정확도에 있어서 90% 이상의 높은 점수를 보임을 확인했다.