2005.01854.pdf (1.43 MB)
Data augmentation for hypernymy detection
conference contribution
posted on 2023-06-09, 22:57 authored by Thomas Kober, Julie WeedsJulie Weeds, Lorenzo Scott Bertolini, Weir DavidThe automatic detection of hypernymy relationships represents a challenging problem in NLP. The successful application of state-of-the-art supervised approaches using distributed representations has generally been impeded by the limited availability of high quality training data. We have developed two novel data augmentation techniques which generate new training examples from existing ones. First, we combine the linguistic principles of hypernym transitivity and intersective modifier-noun composition to generate additional pairs of vectors, such as “small dog - dog” or “small dog - animal”, for which a hypernymy relationship can be assumed. Second, we use generative adversarial networks (GANs) to generate pairs of vectors for which the hypernymy relation can also be assumed. We furthermore present two complementary strategies for extending an existing dataset by leveraging linguistic resources such as WordNet. Using an evaluation across 3 different datasets for hypernymy detection and 2 different vector spaces, we demonstrate that both of the proposed automatic data augmentation and dataset extension strategies substantially improve classifier performance.
History
Publication status
- Published
File Version
- Accepted version
Journal
Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main VolumePublisher
Association for Computational LinguisticsPublisher URL
Page range
1034-1048Event name
16th conference of the European Chapter of the Association for Computational Linguistics (EACL)Event location
Kyiv / onlineEvent type
conferenceEvent date
19 - 23 April, 2021Department affiliated with
- Informatics Publications
Full text available
- Yes
Peer reviewed?
- Yes
Legacy Posted Date
2021-02-03First Open Access (FOA) Date
2021-04-27First Compliant Deposit (FCD) Date
2021-02-02Usage metrics
Categories
No categories selectedKeywords
Licence
Exports
RefWorks
BibTeX
Ref. manager
Endnote
DataCite
NLM
DC