نوع مقاله : مقاله پژوهشی
نویسندگان
1 بخش زبانهای خارجی و زبانشناسی، دانشکده ادبیات و علوم انسانی، دانشگاه شیراز، شیراز، ایران
2 عضو هیات علمی گروه زبانهای خارجی و زبانشناسی دانشگاه شیراز
3 بخش مهندسی علم کامپیوتر و فناوری اطلاعات، دانشکده مهندسی برق و کامپیوتر، دانشگاه شیراز، شیراز، ایران
چکیده
کلیدواژهها
عنوان مقاله [English]
نویسندگان [English]
Automatic detection of conceptual metaphors in Persian language is a major challenge in natural language processing, mainly due to their crucial role in deep semantic interpretation and the lack of sufficiently annotated resources. In this study, to address this challenge, sentences containing keywords associated with five emotion domains—anger, joy, fear, disgust, and sadness—were extracted from the Hamshahri corpus. These sentences were manually annotated to construct a new dataset with binary labels (metaphorical vs. literal), and for metaphorical instances, an additional label specifying the target emotion domain was assigned. Subsequently, a two-stage hybrid framework based on the pre-trained ParsBERT language model was proposed. In the first stage, a binary classification model was fine-tuned to identify metaphorical expressions. In the second stage, a separate ParsBERT-based multi-class classification model was employed to determine the target emotion domain for the sentences identified as metaphorical. Experimental results indicate that the metaphor detection model achieved an F1-score of 95.9%, while the emotion domain classification model obtained an overall accuracy of 87.5%. These results confirm the effectiveness of the proposed architecture for conceptual metaphor processing in Persian language. Moreover, the construction and introduction of this annotated dataset provide a valuable resource for future research in this field.
کلیدواژهها [English]
)