Construction Detection in a Conventional NLP Pipeline

TitleConstruction Detection in a Conventional NLP Pipeline
Publication TypeConference Paper
Year of Publication2017
AuthorsDunietz, J., Levin L., & Petruck M. R. L.
Published inProceedings of the AAAI 2017 Spring Symposium on Computational Construction Grammar and Natural Language Understanding
Page(s)178-184
PublisherAAAI
Abstract

This paper presents an approach to detecting constructions based on a conventional NLP pipeline: the “constructions on top” approach to integrating constructions into NLP, as opposed to “constructions all the way down.” The approach is illustrated with the BECauSE corpus of causal language, the BECauSE constructicon, and the Causeway causal language detector, described elsewhere. We argue here that although BECauSE is not a full construction grammar, its lightweight design and compatibility with conventional NLP tools have facilitated progress on and insights into issues related to construction detection in news corpora. The issues we discuss are (1) individuating families of constructions, and (2) dealing with co-present, non-prototypical meanings that may be present alongside the prototypical meaning of a construction. Particularly significant is the observation that the BECauSE constructicon highlights the importance of integrating frame-evoking constructions into frame semantic resources such as FrameNet.

URLhttp://aaai.org/ocs/index.php/SSS/SSS17/paper/view/15288/14532
ICSI Research Group

AI