This repository contains data for CluBERT: A Cluster-Based Approach forLearning Sense Distributions in Multiple Languages ACL paper.
CluBERT is a multilingual approach (tested on English, French, German, Italian and Spanish) for automatically producing distribution of senses from a corpus of raw text. We use BabelNet as inventory of meanings that is shared across languages.
Within the package clubert_v1.0.tar.gz/.zip you will find the sense distributions for all the lexemes tagged within English, Italian, Spanish, French and German Senseval and SemEval competitions for Word Sense Disambiguation (Senseval-2, Senseval-3, SemEval2007 task 17, SemEval2013 task 12, SemEval 2015 task 13). Please refer to the WSD English Evaluation Framework for the test sets in English and to the latest release of multilingual WSD tasks of SemEval2013 and SemEval2015 made available by Sapienza NLP group at https://github.com/SapienzaNLP/mwsd-datasets.
The package is structured as follows:
clubert_v1.0
├── answers
│ ├── de
│ │ └── semeval2013-de.answers.txt
│ ├── en [0/1954]
│ │ ├── ALL.answers.txt
│ │ ├── semeval2007.answers.txt
│ │ ├── semeval2013.answers.txt
│ │ ├── semeval2015.answers.txt
│ │ ├── senseval2.answers.txt
│ │ └── senseval3.answers.txt
│ ├── es
│ │ ├── semeval2013-es.answers.txt
│ │ └── semeval2015-es.answers.txt
│ ├── fr
│ │ └── semeval2013-fr.answers.txt
│ └── it
│ ├── semeval2013-it.answers.txt
│ └── semeval2015-it.answers.txt
├── de
│ ├── lexemes_distributions.bnid.txt
│ └── lexemes_distributions.wnid.txt
├── en
│ ├── lexemes_distributions.bnid.txt
│ ├── lexemes_distributions.wnid.txt
│ └── lexemes_distributions.wnkey.txt
├── es
│ ├── lexemes_distributions.bnid.txt
│ └── lexemes_distributions.wnid.txt
├── fr
│ ├── lexemes_distributions.bnid.txt
│ └── lexemes_distributions.wnid.txt
└── it
├── lexemes_distributions.bnid.txt
└── lexemes_distributions.wnid.txt
For each language, we provide a text file containing rows structured as follows:
lemma#POS[TAB]sense_1:score_1[TAB]sense_2:score_2[TAB] ... [TAB]sense_n:score_n
and [TAB]
is the tab character \t
.
The bnid
files represent senses as BabelNet synset ids whilst wnid
as WordNet synset offsets (WordNet version 3.0).
As for English, we also provide a wnkey
file representing senses with WordNet sense keys.
The folder answers
contains the answers for the gold standard datasets' instances.
We divided the answers by language and provide an answer
file for each dataset. We note that if the system was not able
to provide an answer for a given dataset instance, that instance will be missing in the answer file. This is required by
the scorer released in the English WSD Evaluation framework to properly compute Precision, Recall and F1.
Please refer to https://lcl.uniroma1.it/wsdeval/evaluation-data for the English evaluation datasets and to https://github.com/SapienzaNLP/mwsd-datasets
for the multilingual evaluation datasets (WordNet split (wn) of the latest version (0.1.6)).
When using the data in this repository, please reference the following paper.
@inproceedings{pasini-etal-2020-clubert,
title = "{C}lu{BERT}: {A} Cluster-Based Approach for Learning Sense Distributions in Multiple Languages",
author = "Pasini, Tommaso and
Scozzafava, Federico and
Scarlini, Bianca",
booktitle = "Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics",
month = jul,
year = "2020",
address = "Online",
publisher = "Association for Computational Linguistics",
url = "https://www.aclweb.org/anthology/2020.acl-main.369",
pages = "4008--4018",
}
All data and codes provided in this repository are subject to the Attribution-Non Commercial-ShareAlike 4.0 International license (CC BY-NC 4.0).