Modeling Semantic Plausibility by Injecting World Knowledge
S. Wang, G. Durrett, and K. Erk. (2018)cite arxiv:1804.00619Comment: camera-ready draft (with link to data), Published at NAACL 2018 as a conference paper (oral).
Abstract
Distributional data tells us that a man can swallow candy, but not that a man
can swallow a paintball, since this is never attested. However both are
physically plausible events. This paper introduces the task of semantic
plausibility: recognizing plausible but possibly novel events. We present a new
crowdsourced dataset of semantic plausibility judgments of single events such
as "man swallow paintball". Simple models based on distributional
representations perform poorly on this task, despite doing well on selection
preference, but injecting manually elicited knowledge about entity properties
provides a substantial performance boost. Our error analysis shows that our new
dataset is a great testbed for semantic plausibility models: more sophisticated
knowledge representation and propagation could address many of the remaining
errors.
Description
[1804.00619] Modeling Semantic Plausibility by Injecting World Knowledge
%0 Generic
%1 wang2018modeling
%A Wang, Su
%A Durrett, Greg
%A Erk, Katrin
%D 2018
%K naacl2018 semantics session5 wordknowledge
%T Modeling Semantic Plausibility by Injecting World Knowledge
%U http://arxiv.org/abs/1804.00619
%X Distributional data tells us that a man can swallow candy, but not that a man
can swallow a paintball, since this is never attested. However both are
physically plausible events. This paper introduces the task of semantic
plausibility: recognizing plausible but possibly novel events. We present a new
crowdsourced dataset of semantic plausibility judgments of single events such
as "man swallow paintball". Simple models based on distributional
representations perform poorly on this task, despite doing well on selection
preference, but injecting manually elicited knowledge about entity properties
provides a substantial performance boost. Our error analysis shows that our new
dataset is a great testbed for semantic plausibility models: more sophisticated
knowledge representation and propagation could address many of the remaining
errors.
@misc{wang2018modeling,
abstract = {Distributional data tells us that a man can swallow candy, but not that a man
can swallow a paintball, since this is never attested. However both are
physically plausible events. This paper introduces the task of semantic
plausibility: recognizing plausible but possibly novel events. We present a new
crowdsourced dataset of semantic plausibility judgments of single events such
as "man swallow paintball". Simple models based on distributional
representations perform poorly on this task, despite doing well on selection
preference, but injecting manually elicited knowledge about entity properties
provides a substantial performance boost. Our error analysis shows that our new
dataset is a great testbed for semantic plausibility models: more sophisticated
knowledge representation and propagation could address many of the remaining
errors.},
added-at = {2018-06-03T18:05:47.000+0200},
author = {Wang, Su and Durrett, Greg and Erk, Katrin},
biburl = {https://www.bibsonomy.org/bibtex/2a31843453c24bebd2f197d3bc3c30b4a/albinzehe},
description = {[1804.00619] Modeling Semantic Plausibility by Injecting World Knowledge},
interhash = {03e8415aa777462fa67216315b4d3376},
intrahash = {a31843453c24bebd2f197d3bc3c30b4a},
keywords = {naacl2018 semantics session5 wordknowledge},
note = {cite arxiv:1804.00619Comment: camera-ready draft (with link to data), Published at NAACL 2018 as a conference paper (oral)},
timestamp = {2018-06-03T18:06:15.000+0200},
title = {Modeling Semantic Plausibility by Injecting World Knowledge},
url = {http://arxiv.org/abs/1804.00619},
year = 2018
}