• Korean Journal of Cognitive Science
• /
• v.20 no.3
• /
• pp.243-274
• /
• 2009

The association, word frequency and imagery have been considered as the main factors that affect the word recognition. The present study aimed to examine the imagery effect and the interaction of the association effect while controlling the frequency effect. To explain the imagery effect, we compared the two theories (dual-coding theory, context availability model). The lexical decision task using priming paradigm was administered. The duration of prime words was manipulated as 20ms, 50ms, and 450ms in experiments 1, 2, and 3, respectively. The association and imagery of prime words were manipulated as the main factors in each of the three experiments. In experiment 1, the duration of prime words (20ms) which is expected to not activate the semantic context enough to affects the word recognition was used. As a result, only imagery effect was statically significant. In experiment 2, the duration of prime word was 50ms, which we expected to activate the semantic context without perceptual awareness. The result showed both the association and imagery effects. The interaction between the two effects was also significant. In experiment 3, to activate the semantic context with perceptual awareness, the prime words were presented for 450ms. Only association effect was statically significant in this experimental condition. The results of the three experiments suggest that the influence of the imagery was at the early stages of word recognition, while the association effect appeared rather later than the imagery. These results implied that the two theories are not contrary to each other. The dual-coding theory just concerned imagery effect which affects the early stage of word recognition, and context-availability model is more for the semantic context effect which affects rather later stage of word recognition. To explain the word recognition process more completely, some integrated model need to be developed considering not only the main 3 effects but also the stages which extends along the time course of the process.

• Annual Conference on Human and Language Technology
• /
• 2021.10a
• /
• pp.518-523
• /
• 2021

챗봇과 같은 대화형 에이전트 사용이 증가하면서 채팅에서의 혐오 표현 사용도 더불어 증가하고 있다. 혐오 표현을 자동으로 탐지하려는 노력은 다양하게 시도되어 왔으나, 챗봇 데이터를 대상으로 한 혐오 표현 탐지 연구는 여전히 부족한 실정이다. 이 연구는 혐오 표현을 포함한 챗봇-사용자 대화 데이터 35만 개에 한국어 말뭉치로 학습된 KoELETRA 기반 혐오 탐지 모델을 적용하여, 챗봇-사람 데이터셋에서의 혐오 표현 탐지의 성능과 한계점을 검토하였다. KoELECTRA 혐오 표현 분류 모델은 챗봇 데이터셋에 대해 가중 평균 F1-score 0.66의 성능을 보였으며, 오탈자에 대한 취약성, 맥락 미반영으로 인한 편향 강화, 가용한 데이터의 정확도 문제가 주요한 한계로 포착되었다. 이 연구에서는 실험 결과에 기반해 성능 향상을 위한 방향성을 제시한다.