• 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.

• Electronics and Telecommunications Trends
• /
• v.39 no.1
• /
• pp.95-105
• /
• 2024

In natural language processing, large language models such as GPT-4 have recently been in the spotlight. The performance of natural language processing has advanced dramatically driven by an increase in the number of model parameters related to the number of acceptable input tokens and model size. Research on multimodal models that can simultaneously process natural language and image data is being actively conducted. Moreover, natural-language and image-based reasoning capabilities of large language models is being explored in robot artificial intelligence technology. We discuss research and related patent trends in robot task planning and code generation for robot control using large language models.

• Journal of The Korean Radiological Technologist Association
• /
• v.28 no.1
• /
• pp.161-166
• /
• 2002

Purpose: To make a comparison evaluated of the auditory instrument and visual instrument language generation task in the fMRI, on the adult volunteers. Materials and Methods: Total of 6 normal adult volunteers(men;4, women;2, mean age;24) performed in 1.5

• Investigative Magnetic Resonance Imaging
• /
• v.5 no.1
• /
• pp.24-32
• /
• 2001

Purpose : We investigated the reproducibility of language lateralization by 4 different word generation paradigms or the rest contents in each paradigm using functional magnetic resonance imaging in normal volunteers Materials and Methods Nine normal volunteers with left-handedness (mean age: 25 yrs) were examined on a 1.57 MR unit using a single-shot gradient echo epibold sequence. Four different word generation paradigms of noun, verb, adjective and adverb were used in each normal volunteer for investigating language system. In each paradigm, two different rest contents consisted of only seeing the " +" symbol or reading the meaningless letters. Each task consisted of 96 phases including 3 activations and 6 rests of 2 different contents. Two activation maps in one task were obtained under two different rest contents using the correlation method. We evaluated the detection rates of Broca and Wernicke areas and the differences of language lateralization among four different word generation paradigms, or between the rest contents. Results : The detection rates of Broca and Wernicke areas were over 67 % in 4 different language paradigms and there was no significant difference of them among language paradigms, or between two different rest contents. Language dominances, in all 4 different language paradigms, were shown to be consistent in 66 %, but were contrary with language paradigms in some subjects. The rest contents made no significant effect on dominant language dominance determination, but the success rates of the dominant language dominances determined from 4 language paradigms were higher in reading the meaningless letter (100%, n=9) than in only seeing "+" on screen at the rest task (78%, n=7).

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.596-599
• /
• 2018

본 논문은 일정 등록을 위한 대화 시스템 개발에 대한 연구를 수행하였다. 기계는 사용자가 요구하는 일정 등록, 일정 수정 및 일정 삭제 등 다양한 목적에 따라 이에 맞는 API를 호출하게 된다. DSCT 6가 제안한 방법을 활용하여 호출되는 API의 종류에 따라 사람과 기계와의 대화를 task 라 불리는 여러 종류의 소규모 목적 대화로 분류하였다. 그 후 분류된 목적 task를 위해 Memory Network 개발에 대한 연구를 수행하였다. 첫 번째로 분류된 task에 대한 실행 결과 75%, 두 번째 task 88%, 세 번째 task 89%, 마지막 모든 task를 합쳤을 때 90%의 성능을 확인할 수 있었다.

• Annual Conference on Human and Language Technology
• /
• 2018.10a
• /
• pp.305-309
• /
• 2018

딥러닝에서 층을 공유하여 작업에 따라 변하지 않는 정보를 사용하는 multi-task learning이 다양한 자연어 처리 문제에 훌륭하게 사용되었다. 그렇지만 우리가 아는 한 공유 공간의 상태와 성능과의 관계를 조사한 연구는 없었다. 본 연구에서는 공유 공간과 task 의존 공간의 자질의 수와 오염 정도가 성능에 미치는 영향도 조사하여 공유 공간과 성능 관계에 대해서 탐구한다. 이 결과는 multi-task를 진행하는 실험에서 공유 공간의 역할과 성능의 관계를 밝혀서 시스템의 성능 향상에 도움이 될 것이다.

• The Journal of Korean Association of Computer Education
• /
• v.9 no.2
• /
• pp.89-99
• /
• 2006

This paper proposes a computational model which can simulate peculiar language phenomena observed in human lexical decision task. The model is designed to mimic major language phenomena such as frequency effect, lexical status effect, word similarity, and semantic priming effect. The experimental results show that the propose model replicated the major language phenomena and performed similar performance with that of human in LDT.

• Phonetics and Speech Sciences
• /
• v.12 no.1
• /
• pp.51-63
• /
• 2020

Phonological awareness is a metalinguistic awareness ability of phonology and is known to predict language skills, such as reading and vocabulary skills. The purpose of this study was to investigate the relationship between phonological awareness, phonological processing, and language skills in 4- to 6-years-old typically developing (TD) children and children with language delay (LD). A total of 32 children (TD=18, LD=15) participated in this study. They performed a phonological awareness task consisting of counting, deletion, and discrimination at syllable level. Nonword Repetition, Digit Backward, Receptive & Expressive Vocabulary Test, and Grammaticality Judgment Task were performed to analyze the correlation between phonological awareness, phonological processing, and language ability. A multiple stepwise regression analysis was performed to examine the phonological awareness subtasks that predict language ability. In the TD group, the syllable categorization task significantly predicted the receptive vocabulary and the performance of the Grammaticality Judgment Task. The LD group showed that the syllable counting task significantly predicted the receptive vocabulary, the expressive vocabulary, and the performance of the Grammaticality Judgment Task. The results showed that the phonological awareness performance was significantly different between the two groups. Further, correlation analysis and regression analysis showed different results for each group. The result of the phonological awareness performance predicted the language ability of each group significantly, suggesting the importance of the meta-linguistic awareness ability of phonology.

• Proceedings of the Korean Operations and Management Science Society Conference
• /
• 2003.05a
• /
• pp.208-216
• /
• 2003

ISO 14649 (data model for STEP-NC) is a new interface scheme or language for CAD-CAM-CNC chain under established by ISO TC184 SCI. Up to this point, the new language is mainly made for milling and turning, and other processes such as EDM will be completed in the future. Upon completion, it will be used as the international standard language for e-manufacturing paradigm by replacing the old machine-level language, so called M&G code used since 1950's. With the rich information contents included in the new language, various intelligent functions can be made by the CNC as the CNC knows what-to-make and how-to-make. In particular, On-Machine Inspection required for quality assurance in the machine level, can be done based on the information of feature­based tolerance graph. Previously, On-Machine inspection has been investigated mainly for milling operation, and only a few researches were made for turning operation without addressing the data model. In this thesis, we present a feature-based on-machine inspection process by the 4 Tasks: 1) proposing a new schema for STEP-NC data model, 2) converting the conventional tolerance scheme into that of STEP-NC, 3) modifying the tolerance graph such that the tolerance can be effectively measured by the touch probe on the machine, and 4) generating collision-free tool path for actual measurement. Task 1 is required for the incorporation of the presented method in the ISO 14649, whose current version does not much include the detailed schema for tolerance. Based on the presented schema, the tolerance represented in the conventional drafting can be changed to that of STEP-NC (Task 2). A special emphasis was given to Task 3 to make the represented tolerance accurately measurable by the touch probe on the machine even if the part setup is changed. Finally, Task 4 is converting the result of Task into the motion of touch probe. The developed schema and algorithms were illustrated by several examples including that of ISO 14649 Part 12.

• Journal of the Korea Convergence Society
• /
• v.12 no.7
• /
• pp.37-44
• /
• 2021

Quality estimation (QE) can evaluate the quality of machine translation output even for those who do not know the target language, and its high utilization highlights the need for QE. QE shared task is held every year at Conference on Machine Translation (WMT), and recently, researches applying Pretrained Language Model (PLM) are mainly being conducted. In this paper, we conduct a survey on the QE task and research trends, and we summarize the features of PLM. In addition, we used a multilingual BART model that has not yet been utilized and performed comparative analysis with the existing studies such as XLM, multilingual BERT, and XLM-RoBERTa. As a result of the experiment, we confirmed which PLM was most effective when applied to QE, and saw the possibility of applying the multilingual BART model to the QE task.