Deontic Modeling and Natural Language Processing for Automated Environmental and Green Compliance Checking

这笔赠款提供资金，用于开发一种计算方法，以检查文本建筑计划（如雨水污染预防计划）是否符合环境法规和合同要求。这旨在自动检测不合规情况，并提供相对丰富的不合规分析，例如违反了哪项法规，违反的原因，违反的可能后果等。为了实现文本处理和自动推理的深层次，将开发三种算法并将其组合到一个计算平台中：1）用于对文本进行分类的基于机器学习的算法，2）用于信息提取的混合句法-语义（其利用文本的语法和含义描述性特征）算法，以及3）用于顺应性推理的基于逻辑的算法。自动化分析将促进规范推理的知识模型（道义模型）在建设领域。这些算法将在概念原型软件的证明中实现。将与建筑承包商合作，制定基于真实的生活的测试案例场景，并用于测试和验证开发的算法。如果成功，这项研究的结果将导致建筑领域的自动监管和合同合规性检查的改进。这项研究的长期目标是为建筑业专业人士和监管机构提供一种方法，可以帮助减少合规性评估错误，以及减少与合规性检查相关的成本和时间。拟议的工作还将有助于为深度自然语言理解/处理和语义推理提供计算工具和方法。该项目还将涉及教育和外联活动，以促进教学，吸引本科生和研究生参与，并接触女高中生。这包括开发一个项目网站、课程模块和关于环保建筑的实践推广材料。

This grant provides funding for the development of a computational approach for checking the compliance of textual construction plans (such as storm-water pollution prevention plans) with environmental regulations and contractual requirements. This is intended to automatically detect non-compliance instances and provide a relatively rich analysis of the non-compliance such as which regulation was violated, the reason for the violation, the possible consequences of the violation, etc. To achieve such deep levels of text processing and automated reasoning, three algorithms will be developed and combined into one computational platform: 1) a machine learning-based algorithm for classifying text, 2) a hybrid syntactic-semantic (that utilizes grammatical and meaning-descriptive features of the text) algorithm for information extraction, and 3) a logic-based algorithm for compliance reasoning. The automated analysis will be facilitated by a knowledge model of normative reasoning (deontic model) in the construction domain. The algorithms will be implemented in a proof of concept prototype software. Real life-based test case scenarios will be formulated - in collaboration with construction contractors - and used to test and validate the algorithms developed.If successful, the results of this research will lead to improvements in the area of automated regulatory and contractual compliance checking in construction. The long-term goal of this research is to provide construction industry professionals and regulators with an approach that can help reduce compliance assessment errors, as well as reduce the cost and time associated with compliance checking. The proposed work will also contribute to the computational tools and methodologies available for both deep natural language understanding/processing and semantic reasoning. This project will also involve educational and outreach activities to promote teaching and learning, engage undergraduate and graduate students, and reach out to female high-school students. This includes the development of a project website, course modules, and hands-on outreach material about environmentally-conscious construction.