An AI using Construction Grammar: Automatic Acquisition of Knowledge about Words

Denis Kiselev

2020

Abstract

This paper deals with an AI implementation that uses knowledge in an original Construction Grammar (CG) format for deep understanding of text. CG is a means of processing knowledge pieces—aka constructions— that describe the form and meaning of text parts. Understanding consists in automatically finding in the text the knowledge that constructions contain and in creating knowledge networks that reflect the text information structure. Deeper understanding is achieved by propagating knowledge within the network, i.e. some constructions can share with others information about syntax, semantics, pragmatics and other text properties. A shortcoming of this information-rich method is limited coverage: only text for which a CG database is available can be understood; that database due to its complexity most often needs to be made up manually. The author attempts to increase the coverage by implementing automatic acquisition of word knowledge from sources such as an external (non-CG) knowledge base and formatting the knowledge as CG constructions. The resulting CG database has been used in evaluation experiments to understand the Winograd Schema (WS)—a test for AI. A 28% increase in accurate coverage and opportunities for further improvement are observed.

Paper Citation

in Harvard Style

Kiselev D. (2020). An AI using Construction Grammar: Automatic Acquisition of Knowledge about Words. In Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART, ISBN 978-989-758-395-7, pages 289-296. DOI: 10.5220/0008865902890296

in Bibtex Style

@conference{icaart20,
author={Denis Kiselev},
title={An AI using Construction Grammar: Automatic Acquisition of Knowledge about Words},
booktitle={Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,},
year={2020},
pages={289-296},
publisher={SciTePress},
organization={INSTICC},
doi={10.5220/0008865902890296},
isbn={978-989-758-395-7},
}

in EndNote Style

TY - CONF

JO - Proceedings of the 12th International Conference on Agents and Artificial Intelligence - Volume 2: ICAART,
TI - An AI using Construction Grammar: Automatic Acquisition of Knowledge about Words
SN - 978-989-758-395-7
AU - Kiselev D.
PY - 2020
SP - 289
EP - 296
DO - 10.5220/0008865902890296