The primary aim of this project is to investigate how construction
grammars can be learned by a computational system, based on
either semantically annotated corpora or communicative interactions
in situated tutor-learner scenarios. If successful, the outcome of this
project will constitute a major breakthrough in the field of construction
grammar, as it would for the first time provide a computational model
of how conventionalised form-meaning pairings (constructions) that
support language comprehension and production can be constructed
in a usage-based fashion.
My hypothesis is that a broad spectrum of constructions, ranging from fully idiomatic to completely abstract, can be learned by a combination of storing linguistic observations as holophrase constructions, and generalising and specialising already
learned constructions with respect to novel linguistic observations. I
will design and implement a set of learning operators that facilitate
these learning processes and release them in the form of a
construction grammar learning toolkit. The outcome of this project
has the potential to significantly enhance the performance of a wide
variety of language technology applications and meaning-based AI
systems. It would also be highly valuable as a methodological tool for
usage-based linguistic research, as it would facilitate the automatic
annotation of constructions in text corpora, and provide novel insights
into the compositional and non-compositional aspects of language
use.