Published May 10, 2022
| Version v1
Conference paper
Open
Dictionary files used in the paper 'Combating the Curse of Multilinguality in Cross-Lingual WSD by Aligning Sparse Contextualized Word Representations'
Description
The uploaded numpy ndarrays were used to obtain the results in the NAACL 2022 publication entitled 'Combating the Curse of Multilinguality in Cross-Lingual WSD by Aligning Sparse Contextualized Word Representations'.
The four different files correspond to the 1024-by-3000 dictionary matrices that were obtained using sparse coding for the last four layers of the bert-large-cased model over the hidden representations derived from the SemCor dataset (as described in the paper).
Files
Files
(49.2 MB)
Name | Size | Download all |
---|---|---|
md5:6e0f093b7e076d6a88959d841827d4f1
|
12.3 MB | Download |
md5:08bc4555fd41a20f77c333f3c21eca4a
|
12.3 MB | Download |
md5:f80b03ba1c9e8ef80a721c885ff11cc5
|
12.3 MB | Download |
md5:8fc003132555013b8c3a10480b8e3ae6
|
12.3 MB | Download |
Additional details
Related works
- Is published in
- Conference paper: https://openreview.net/forum?id=r4xz0gnWBbc (URL)
- Is referenced by
- Software: https://github.com/begab/sparsity_makes_sense (URL)