On the Role of Entity and Event Level Conceptualization in Generalizable Reasoning: A Survey of Tasks, Methods, Applications, and Future Directions

Weiqi Wang,Tianqing Fang,Haochen Shi,Baixuan Xu,Wenxuan Ding,Liyu Zhang,Wei Fan,Jiaxin Bai,Haoran Li,Xin Liu,Yangqiu Song

2024-06-16

Abstract:Entity- and event-level conceptualization, as fundamental elements of human cognition, plays a pivotal role in generalizable reasoning. This process involves abstracting specific instances into higher-level concepts and forming abstract knowledge that can be applied in unfamiliar or novel situations, which can enhance models' inferential capabilities and support the effective transfer of knowledge across various domains. Despite its significance, there is currently a lack of a systematic overview that comprehensively examines existing works in the definition, execution, and application of conceptualization to enhance reasoning tasks. In this paper, we address this gap by presenting the first comprehensive survey of 150+ papers, categorizing various definitions, resources, methods, and downstream applications related to conceptualization into a unified taxonomy, with a focus on the entity and event levels. Furthermore, we shed light on potential future directions in this field and hope to garner more attention from the community.

Computation and Language

What problem does this paper attempt to address?

The paper mainly addresses the following issues:

1. **The Role of Conceptualization in Generalizable Reasoning**: The paper explores how conceptualization at the entity and event levels plays a key role in general reasoning by abstracting specific instances into higher-level concepts, forming abstract knowledge that can be applied to new situations.

2. **Lack of Systematic Review**: Despite the importance of conceptualization, there is currently a lack of a systematic review to comprehensively examine the definition, execution, and application of conceptualization in existing work to enhance reasoning tasks. The authors fill this gap by providing a systematic review of over 150 related papers and categorizing them into a unified classification system.

3. **Inconsistency in Terminology**: There is inconsistency in the definition of "conceptualization" across different studies, leading to difficulties in understanding. The paper aims to address this issue by providing a unified definition and classification system.

4. **Unclear Methods**: The methods for conceptualizing different types of instances in a scalable and accurate manner are not clear. The paper summarizes various methods and categorizes them to help researchers better understand and apply these methods.

5. **Summary of Applications in Downstream Tasks**: The paper also discusses how conceptualization can enhance the performance of downstream tasks, particularly in applications such as commonsense reasoning.

By discussing the above aspects, the paper hopes to draw more attention from the academic community and provide guidance for future research directions.