A Mīmāṃsā Inspired Framework Towards Temporal Reasoning in Large Language Models
摘要
Mīmāṃsā, one of the systems of Indian Philosophy, deals with the interpretation of the Vedas. The Brāhmaṅas, a division of the Vedas, include precise instructions (Vidhi) for the execution of rituals. Interpreting these immediately can be confusing. In order to fully understand this, the interpretive processes from Mīmāṃsā are utilized. The procedures encompass various components, such as linguistic proficiency, grammatical comprehension, the individual’s capacity to execute the ritual, and logical attributes. In addition, Mīmāṃsā incorporates a mention of diverse sequencing systems known as krama, which precisely outlines the specified sequence in which rituals should be performed. This paper takes inspiration from these sequential techniques and proposes a framework for temporal reasoning from a Logical perspective. This approach is incorporated into the existing MIRA (Mīmāṃsā Inspired Representation of Actions) work, resulting in activity sequencing. Subsequently, it is utilized in the Large Language Models to autonomously produce a sequence of instructions in real-life situations.