Using Large Language Models for the Interpretation of Building Regulations

Overview

• Compliance checking is crucial in construction projects.
• The increasing use of building information models (BIM) in construction has created opportunities for automated compliance checking (ACC).
• Translating building regulations into a computable format for ACC is complex and time-consuming.
• Large language models (LLMs) like GPT-3.5 and GPT-4 have the potential to automate this translation process.

Plain English Explanation

Compliance checking is an important step in construction projects, ensuring that buildings are designed and built according to the relevant rules and regulations. The growing use of building information models (BIM) in the construction industry has created more opportunities for automated compliance checking (ACC). BIM allows for the digital sharing of building design data, which can be used to check compliance with legal requirements.

However, translating these legal requirements, which are typically written in natural language, into a format that can be processed by machines is a complex, costly, and time-consuming task. This is where large language models (LLMs) like GPT-3.5 and GPT-4 come into play. These models have the capability to generate logically coherent text and source code in response to user prompts, which could be leveraged to automate the conversion of building regulations into a semantic and computable representation.

Technical Explanation

This paper evaluates the performance of LLMs, specifically GPT-3.5, in translating building regulations into LegalRuleML, a format suitable for ACC. The researchers used a few-shot learning setup, where they provided GPT-3.5 with only a few example translations to help the model learn the basic structure of the LegalRuleML format.

The researchers also used a system prompt to further specify the LegalRuleML representation and explore the existence of expert domain knowledge in the model. This domain knowledge might be ingrained in GPT-3.5 through its broad pre-training, but it needs to be brought forth through careful contextualization.

Finally, the researchers investigated whether strategies like chain-of-thought reasoning and self-consistency could be applied to this use case. As LLMs become more sophisticated, the increased common sense, logical coherence, and means to domain adaptation can significantly support ACC, leading to more efficient and effective checking processes.

Critical Analysis

The paper provides a promising approach to automating the translation of building regulations into a computable format, which could streamline compliance checking in construction projects. However, the researchers acknowledge that their evaluation was limited to a few-shot learning setup, and further research is needed to assess the model's performance on a larger and more diverse set of building regulations.

Additionally, the researchers note that the quality and accuracy of the LegalRuleML translations produced by the LLM may still require human review and validation, as the model may not fully capture the nuances and complexities of legal language. Ensuring the reliability and trustworthiness of the automated translation process will be crucial for its widespread adoption in the construction industry.

Conclusion

This research demonstrates the potential of large language models to automate the conversion of building regulations into a computable format, which could significantly improve the efficiency and effectiveness of compliance checking in construction projects. As LLMs continue to evolve, incorporating more domain-specific knowledge and reasoning capabilities, they may become increasingly valuable tools for streamlining various regulatory and compliance-related tasks across different industries.