Study on Metaphorical Speech Acts in Arabic Conversations: Implications for Multimodal Understanding of Large Language Models

This article interprets a study on metaphorical speech acts in Arabic conversations in Saudi Arabia, analyzes the functions of metaphors in daily digital communication, and explores the implications of such research for large language models' understanding of multimodal and cross-cultural metaphorical expressions. The study covers core content such as types of metaphors, the role of multimodal elements, and challenges for AI models.

Metaphor is not only a rhetorical device but also a core mechanism of human cognition and expression. From the perspective of cognitive linguistics, daily language is largely based on metaphors. However, large language models face comprehension barriers when dealing with cross-lingual and cross-cultural metaphors, and the study of Arabic conversations in Saudi Arabia provides an important perspective on the evolution of metaphors in the digital age.

This study uses qualitative research methods, collecting natural conversation data from the WhatsApp platform in Saudi Arabia, covering participants of different age groups and genders, and focusing on daily social informal communication (contexts such as family gatherings, festival celebrations, etc.). The data includes multimodal elements such as text content, emojis, and stickers, reflecting the real face of contemporary digital communication.

The study identifies main types of metaphors: expressive (conveying emotional attitudes, common among female participants), directive (indirectly making requests or suggestions, reflecting the emphasis on politeness and face in Arab culture), commissive (implied expressions of future actions), and proverbial (carrying cultural traditions and wisdom inheritance). The younger generation tends to use visual elements (emojis, stickers) to strengthen or replace textual metaphors, forming a unique multimodal metaphorical expression. The digital environment has spawned new forms of metaphors: emojis/stickers have become carriers of conceptual metaphors, and visual metaphors, along with the weakening of traditional textual metaphors, are reshaping expression habits.

Current mainstream models have a certain ability to handle pure textual metaphors, but perform poorly in complex scenarios that integrate emojis, culture-specific expressions, and cross-generational language variations. Key issues to solve include: identifying culture-specific conceptual metaphors, handling the interaction between text and visual elements, and capturing language differences among different user groups. Arabic metaphors are rooted in cultural experiences such as desert life and tribal traditions; without background knowledge, misunderstandings are easy. Building a cross-cultural language understanding system requires integrating in-depth cultural corpora and contextual information.

Future research directions can include: building a cross-cultural corpus of multimodal metaphors, developing evaluation benchmarks for metaphor understanding, and integrating cultural anthropology knowledge into model training. Metaphor research originates from the humanities and is closely related to the development of large language models. Understanding human metaphor use is a necessary path to building AI systems that truly understand language. Progress in language technology needs to draw insights from linguistics and anthropology, balancing scale and deep cultural understanding.