From Concrete to Abstract: A Multimodal Generative Approach to Abstract Concept Learning

Haodong Xie, Rahul Singh Maharjan, Federico Tavella, Angelo Cangelosi

Abstract: Understanding and manipulating concrete and abstract concepts is fundamental to human intelligence, yet this capability remains a significant challenge for artificial agents. This paper introduces an approach to high-order abstract concept learning through a multimodal generative model that integrates visual and categorical linguistic information from concrete ones. Our model initially grounds concrete subordinate level concepts, progresses to basic-level concepts through the combination of concrete concepts, and finally abstracts to superordinate level concepts via the grounding of basic-level concepts. To evaluate the model’s language learning ability, we conducted language-to-visual and visual-to-language tests with high-order abstract concepts. The experimental results demonstrate the model’s proficiency in both language understanding and language naming tasks.

External IDs:doi:10.1007/978-981-96-7033-8_6