Uncovering Divergent Linguistic Information in Word Embeddings with Lessons for Intrinsic and Extrinsic Evaluation

Mikel Artetxe, Gorka Labaka, Iñigo Lopez-Gazpio, Eneko Agirre


Abstract
Following the recent success of word embeddings, it has been argued that there is no such thing as an ideal representation for words, as different models tend to capture divergent and often mutually incompatible aspects like semantics/syntax and similarity/relatedness. In this paper, we show that each embedding model captures more information than directly apparent. A linear transformation that adjusts the similarity order of the model without any external resource can tailor it to achieve better results in those aspects, providing a new perspective on how embeddings encode divergent linguistic information. In addition, we explore the relation between intrinsic and extrinsic evaluation, as the effect of our transformations in downstream tasks is higher for unsupervised systems than for supervised ones.
Anthology ID:
K18-1028
Volume:
Proceedings of the 22nd Conference on Computational Natural Language Learning
Month:
October
Year:
2018
Address:
Brussels, Belgium
Venue:
CoNLL
SIG:
SIGNLL
Publisher:
Association for Computational Linguistics
Note:
Pages:
282–291
Language:
URL:
https://aclanthology.org/K18-1028
DOI:
10.18653/v1/K18-1028
Bibkey:
Cite (ACL):
Mikel Artetxe, Gorka Labaka, Iñigo Lopez-Gazpio, and Eneko Agirre. 2018. Uncovering Divergent Linguistic Information in Word Embeddings with Lessons for Intrinsic and Extrinsic Evaluation. In Proceedings of the 22nd Conference on Computational Natural Language Learning, pages 282–291, Brussels, Belgium. Association for Computational Linguistics.
Cite (Informal):
Uncovering Divergent Linguistic Information in Word Embeddings with Lessons for Intrinsic and Extrinsic Evaluation (Artetxe et al., CoNLL 2018)
Copy Citation:
PDF:
https://aclanthology.org/K18-1028.pdf
Code
 artetxem/uncovec +  additional community code