SPaR.txt, a Cheap Shallow Parsing Approach for Regulatory Texts

Abstract

Automated Compliance Checking (ACC) systems aim to semantically parse building regulations to a set of rules. However, semantic parsing is known to be hard and requires large amounts of training data. The complexity of creating such training data has led to research that focuses on small sub-tasks, such as shallow parsing or the extraction of a limited subset of rules. This study introduces a shallow parsing task for which training data is relatively cheap to create, with the aim of learning a lexicon for ACC. We annotate a small domain-specific dataset of 200 sentences, SPAR.txt 1 , and train a sequence tagger that achieves 79,93 F1-score on the test set. We then show through manual evaluation that the model identifies most (89,84%) defined terms in a set of building regulation documents, and that both contiguous and discontiguous Multi-Word Expressions (MWE) are discovered with reasonable accuracy (70,3%).

BRISE-Plandok: a German legal corpus of building regulations

“…Recent approaches to the modeling and extraction of regulations in the construction domain vary greatly both in their choice of semantic representation and their methods for mapping text to such representations. Kruiper et al (2021) create the ScotReg corpus of Scottish building regulations, deőne a sequence labeling task that is a combination of shallow parsing (chunking) and semantic role labeling, assigning labels such as Action and Object to spans of text that are also syntactic constituents, and annotate 200 sentences using this representation to create the SPaR.txt dataset, which they use to train a standard deep learning architecture consisting of BERT embeddings, bidirectional Long Short-Term Memory (bi-LSTM) and Conditional Random Fields (CRFs). On the test portion of the dataset their models achieve precision, recall, and F1 scores around 80%.…”