[Paper Review] Emerging Trends of Multimodal Research in Vision and Language.

Shagun Uppal, Sarthak Bhagat+5 more|arXiv (Cornell University)|Oct 19, 2020

Multimodal Machine Learning Applications325 references4 citations

TL;DR

This paper provides a comprehensive survey of emerging trends in multimodal vision-and-language research, analyzing task formulations, evaluation strategies, and challenges in semantic perception and content generation. It identifies key directions toward modular and transparent AI systems through insights from recent literature.

ABSTRACT

Deep Learning and its applications have cascaded impactful research and development with a diverse range of modalities present in the real-world data. More recently, this has enhanced research interests in the intersection of the Vision and Language arena with its numerous applications and fast-paced growth. In this paper, we present a detailed overview of the latest trends in research pertaining to visual and language modalities. We look at its applications in their task formulations and how to solve various problems related to semantic perception and content generation. We also address task-specific trends, along with their evaluation strategies and upcoming challenges. Moreover, we shed some light on multi-disciplinary patterns and insights that have emerged in the recent past, directing this field towards more modular and transparent intelligent systems. This survey identifies key trends gravitating recent literature in VisLang research and attempts to unearth directions that the field is heading towards.

Motivation & Objective

• To provide a detailed overview of recent advancements in vision-and-language multimodal research.
• To analyze task-specific trends in semantic perception and content generation across vision and language modalities.
• To examine evaluation strategies and identify open challenges in the field.
• To highlight multi-disciplinary patterns driving the development of more modular and interpretable intelligent systems.
• To map the trajectory of the field by identifying key research directions from recent literature.

Proposed method

• Systematic review of recent literature in vision-and-language research to identify dominant trends and patterns.
• Categorization of research tasks into semantic perception and content generation with focus on problem formulation.
• Analysis of evaluation methodologies used across different vision-and-language tasks.
• Identification of recurring challenges such as modality alignment, generalization, and interpretability.
• Synthesis of insights from interdisciplinary research to guide the design of modular and transparent AI systems.
• Use of thematic analysis to uncover emerging research directions and technological shifts in the field.

Experimental results

Research questions

• RQ1 What are the dominant task formulations in contemporary vision-and-language research?
• RQ2 How are evaluation strategies evolving to meet the demands of complex multimodal tasks?
• RQ3 What are the key challenges impeding progress in vision-and-language understanding and generation?
• RQ4 How are multi-disciplinary insights contributing to the development of more transparent and modular AI systems?
• RQ5 What emerging research directions are shaping the future of vision-and-language research?

Key findings

• Recent research is increasingly focused on joint understanding and generation of visual and linguistic content, with strong emphasis on semantic perception.
• Evaluation strategies are becoming more task-specific and nuanced, reflecting the complexity of multimodal reasoning.
• A growing trend toward modular and interpretable systems is emerging from interdisciplinary insights.
• Challenges related to modality alignment, robustness, and generalization remain central to ongoing research.
• The field is shifting toward more transparent and explainable AI through structured, component-based system designs.
• Content generation tasks are advancing rapidly, particularly in image captioning and visual question answering, driven by transformer-based architectures.