• Korean Journal of Cognitive Science
• /
• v.20 no.4
• /
• pp.507-533
• /
• 2009

This paper examines the time course and processing patterns of filler-gap dependencies in Korean relative clauses, using an eyetracking method. Participants listened to a short story while viewing four pictures of entities mentioned in the story. Each story is followed by an auditorily presented question involving a relative clause (subject relative or dative relative). Participants' eye movements in response to the question were recorded. Results showed that the proportion of looks to the picture corresponding to a filler noun significantly increased at the relative verb affixed with a relativizer, and was largest at the filler where the fixation duration on the filler picture significantly increased. These results suggest that online resolution of the filler-gap dependency only starts at the relative verb marked with a relativiser and is finally completed at the filler position. Accordingly, they partly support the filler-driven parsing strategy for Korean, as for head-initial languages. In addition, the different patterns of eye movements between subject relatives and dative relatives indicate the role of case markers in parsing Korean sentences.

• Journal of Broadcast Engineering
• /
• v.23 no.6
• /
• pp.914-924
• /
• 2018

In this paper, we implement an Android application for infant language education using marker-based augmented reality. Combining animal word markers (noun), size/color word markers (adjective), and action word markers (verb) in puzzle form to make a simple sentence, the application shows virtual contents related to the content of the sentence. For example, when an animal marker is showed up on a camera, the corresponding animal appears. Additionally, when the motion markers are combined, the animal's appearance changes into an animation in which it acts. When a user touched a marker, user can hear the sound of the word, which gives an auditory effect, and by adding the rotation function, user can see the animation in any direction. Our goal is to increase infants' interest in learning language and also increase the effectiveness of education on the meaning of words and the structure of simple sentences, by encouraging them to actively participate in language learning through visual and auditory stimuli.

• Journal of Aquaculture
• /
• v.11 no.4
• /
• pp.475-485
• /
• 1998

To clarify annual reproductive cycle of Koran dark sleeper, odontobutis platycephala, we examined the seasonal changes of gonadosomatic index(GSI), testicular development stages and sex steroid hormones in blood from December 1995 to November 1997. Testis was podlike shape from July to October, and tadpole-like shape from November because of its expanded posterior part. GSI was 0.14~0.18 from July to September and increased to $0.43{\pm}0.04$ in October and then was not changed significantly until February. GSI was reincreased to $0.52{\pm}0.09$ from March and then was kept at similer levels until May, but fell down to $0.28{\pm}0.05$ in June. As results of histological observation, testis was divided into 3 parts(anterior, boundary, posterior) in the development progress of germ cells. In July, the testis was composed of only spermatogonia without seminiferous tubules in most fishes. In the anterior part of testis, the ferquency of spermatogenesis stage seminiferous tubules appearing in August was more than 80% from September to December. decreased gradually from January to March and drastically in April, and then disappeared in June. The frequency of spermiogenesis stage seminiferous tubules appearing in December, increased gradually from January to March and drastically to 80% in April, and reached to 90% the highest levels of the year in June. Post-spawning stage seminiferous tubules did not appear throughout the year. The frequency of spermatogonia was 100% and 65% in July and August, and less than 20% in the rest period of the year. In the boundary part, the frequency of spermatogenesis stage seminiferous tubules appearing in August increased from September and reached to 82% in November, decreased from December, adn disappeared in March. The frequency of spermiogenesis stage seminiferous tubules appearing in November was less than 18% until February, and increased to 29%~57% from March to June. The frequency of post-spawning stage seminiferous tubules appeared 12%~25% only from March to June. The frequency of spermatogonia was 100% in July, decreased to 85% in August and 10% in November, and increased gradually from December to 50% in April, and decreased again from May to June. In the posterior part, seminiferous tubules with some seminiferous tubules increased drastically 80%~85% in August and September, decreased drastically from October to November and remained below 10% until February, and disappeared after March. The frequency of spermiogenesis stage seminiferous tubules appearing in August increased sharply from October and reached to 75% in November. decreased to 15% in December and no significant changes until March, and disappeared after April. The frequency of post-spawning stage seminiferous tubules appearing very early in November increased to 82% in December and 85%~95% until June. The frequency of spermatogonia was 100% in July, decreased drastically to 15% in August, disappeared from October to Mrch, but reappeared from April and kept at less than 10% until June. The blood level of testosterone (T) increrased gradually from August was $0.61{\pm}0.09 ng/m\ell$ in November, increrased drastically to $3.99{\pm}1.22 ng/m\ell$ in December and maintained at in similar level until March, and decreased to $0.25{\pm}0.14 ng/m{\ell} ~ 0.17{\pm}0.13ng/m{\ell}$ in April and May and no significant changes until July (P<0.05). The blood level of 17, 20 -dihydroxy-4-pregnen-3-one $ng/m{\ell}$in the rest of year without significant changes(P<0.05). Taken together these results, the germ cell development of testis progressed in the order of posterior, boundary, anterior part during annual reproductive cycle in Korean dark sleeper. The testicular cycle of Korean dark sleeper was as follows. The anterior part of testis : i.e. spermatogonial proliferation period (July), early maturation period (from August to November), mid maturation period (from December to March), late maturation period (from April to May) and functional maturation period (June) were elucidated. The boundary of testis, i.e. spermatogonial proliferation period (July), early maturation period (from August to October), mid maturation period (from November to February) and the coexistence period of late maturation, functional maturation and post-spawn (from March to June) were elucidated. The posterior of testis, i.e. spermatogonial proliferation period (July), mid maturation period (from August ot September), late maturation period (October), functional maturation period (November) and post-spawn period (from December to June) were elucidated. It was showed that the changes of sex steroid hormone in blood played a important roles in the annual reproductive cycle of Korean dark sleeper.

• Journal of Intelligence and Information Systems
• /
• v.23 no.2
• /
• pp.71-88
• /
• 2017

Language models were originally developed for speech recognition and language processing. Using a set of example sentences, a language model predicts the next word or character based on sequential input data. N-gram models have been widely used but this model cannot model the correlation between the input units efficiently since it is a probabilistic model which are based on the frequency of each unit in the training set. Recently, as the deep learning algorithm has been developed, a recurrent neural network (RNN) model and a long short-term memory (LSTM) model have been widely used for the neural language model (Ahn, 2016; Kim et al., 2016; Lee et al., 2016). These models can reflect dependency between the objects that are entered sequentially into the model (Gers and Schmidhuber, 2001; Mikolov et al., 2010; Sundermeyer et al., 2012). In order to learning the neural language model, texts need to be decomposed into words or morphemes. Since, however, a training set of sentences includes a huge number of words or morphemes in general, the size of dictionary is very large and so it increases model complexity. In addition, word-level or morpheme-level models are able to generate vocabularies only which are contained in the training set. Furthermore, with highly morphological languages such as Turkish, Hungarian, Russian, Finnish or Korean, morpheme analyzers have more chance to cause errors in decomposition process (Lankinen et al., 2016). Therefore, this paper proposes a phoneme-level language model for Korean language based on LSTM models. A phoneme such as a vowel or a consonant is the smallest unit that comprises Korean texts. We construct the language model using three or four LSTM layers. Each model was trained using Stochastic Gradient Algorithm and more advanced optimization algorithms such as Adagrad, RMSprop, Adadelta, Adam, Adamax, and Nadam. Simulation study was done with Old Testament texts using a deep learning package Keras based the Theano. After pre-processing the texts, the dataset included 74 of unique characters including vowels, consonants, and punctuation marks. Then we constructed an input vector with 20 consecutive characters and an output with a following 21st character. Finally, total 1,023,411 sets of input-output vectors were included in the dataset and we divided them into training, validation, testsets with proportion 70:15:15. All the simulation were conducted on a system equipped with an Intel Xeon CPU (16 cores) and a NVIDIA GeForce GTX 1080 GPU. We compared the loss function evaluated for the validation set, the perplexity evaluated for the test set, and the time to be taken for training each model. As a result, all the optimization algorithms but the stochastic gradient algorithm showed similar validation loss and perplexity, which are clearly superior to those of the stochastic gradient algorithm. The stochastic gradient algorithm took the longest time to be trained for both 3- and 4-LSTM models. On average, the 4-LSTM layer model took 69% longer training time than the 3-LSTM layer model. However, the validation loss and perplexity were not improved significantly or became even worse for specific conditions. On the other hand, when comparing the automatically generated sentences, the 4-LSTM layer model tended to generate the sentences which are closer to the natural language than the 3-LSTM model. Although there were slight differences in the completeness of the generated sentences between the models, the sentence generation performance was quite satisfactory in any simulation conditions: they generated only legitimate Korean letters and the use of postposition and the conjugation of verbs were almost perfect in the sense of grammar. The results of this study are expected to be widely used for the processing of Korean language in the field of language processing and speech recognition, which are the basis of artificial intelligence systems.