Visual and Language Information Fusion Algorithm in Multimodal Discourse Analysis

With the rapid development of information technology (IT), the Internet is full of multimodal data, including text, images, audio and video, which enriches information expression and challenges information processing technology. Multimodal discourse analysis, as an emerging field in this context, transcends a single text analysis dimension and delves into complex systems involving multiple sensory interactions such as vision and hearing. For machine translation (MT), traditional text-based methods are inadequate when faced with content containing rich visual elements. For this reason, multimodal MT has emerged, integrating visual and linguistic information, improving translation quality, and being close to the original context and emotions. This article innovatively proposes a visual and language information fusion algorithm in multimodal discourse analysis, which deeply integrates deep learning (DL) technology and can automatically capture key information in images as auxiliary clues in the translation process to optimize language generation models. The experimental results show that the algorithm integrates deep visual meanings while retaining the essence of language, significantly improving the naturalness and accuracy of translation, bringing new breakthroughs to the MT field.