This Element introduces a usage-based computational approach to Construction Grammar that draws on techniques from natural language processing and unsupervised machine learning. The computational experiments are important for testing the learnability, variability, and confirmability of Construction Grammar as a theory of language.
This Element introduces a usage-based computational approach to Construction Grammar that draws on techniques from natural language processing and unsupervised machine learning. This work explores how to represent constructions, how to learn constructions from a corpus, and how to arrange the constructions in a grammar as a network. From a theoretical perspective, this work examines how construction grammars emerge from usage alone as complex systems, with slot-constraints learned at the same time that constructions are learned. From a practical perspective, this work is accompanied by a Python package which enables linguists to incorporate construction grammars into their own corpus-based work. The computational experiments in this work are important for testing the learnability, variability, and confirmability of Construction Grammar as a theory of language.
