Shared Task on Monolingual Word Sense Alignment
Monolingual Word Sense alignment (MWSA) is the task of aligning word senses across resources in the same language.
A word can be defined in different ways in different resources. Finding out which ones are somehow connected together is the task of word sense alignment.
This task was recently the focus of the 1st “Monolingual Word Sense Alignment” Shared Task.
As part of the GlobaLex 2020 Workshop at the Language Resource and Evaluation Conference (LREC 2020), we [John P. McCrae and Sina Ahmadi, NUI Galway] organized a shared task on Monolingual Word Sense Alignment (MWSA).
While the workshop was not able to take place due to COVID-19, the papers and the results are available.
The goal of this task was to find senses in two monolingual dictionaries (in the same language), that describe the same concept. For example, ‘bank’ may be described as ‘A financial establishment that uses money deposited by customers for investment, pays it out when required, makes loans at interest, and exchanges currency’ or ‘a financial institution that accepts deposits and channels the money into lending activities’ and the goal would be to recognize these as similar and not related to other definitions such as ‘sloping land (especially the slope beside a body of water)’.
The MWSA task was organized with data from any participating members and observers on the ELEXIS infrastructure and as such, we were able to provide data in 15 languages:
-
- Basque
- Bulgarian
- Danish
- Dutch
- English
- Estonian
- German
- Hungarian
- Irish
- Italian
- Portuguese
- Serbian
- Slovenian
- Spanish
- Russian
In total, we had four teams participating from the Research Institute for Artificial Intelligence at the Romanian Academy, the Austrian Academy of Sciences, the University of Naples L’Orientale and the University of Cambridge.
The participants developed strong systems with the overall best system scoring 84% accuracy in sense alignment.
On the basis of this strong result, we hope to organize a future version of the task.
About the author
John P. McCrae
organized the 1st Monolingual Word Sense Alignment Shared Task.
His work is focused on linguistic linked data, reproducible research, ontologies and NLP for minority languages.
He is leader of workpackage 2, Interoperability and Linked (Open) Data and represents NUI Galway in the ELEXIS consortium.