Home

Awesome

TSAR-2022-Shared-Task Datasets and Evaluation Scripts

TSAR2022 Shared Task on Lexical Simplification for English (en), Spanish (es) and Portuguese (pt) - Datasets and Evaluation scripts

Please look at the website of the Shared Task for more details about the Evaluation Benchmark, Guidelines, Registration Form, etc... <br/>TSAR-2022 Shared-Task website

Datasets

There is no training dataset for the TSAR-2022 Shared Task. However, a sample of 10 or 12 instances with gold standard annotations is provided here as a trial/sample dataset. <br/> <br/> Format of the files:

Trial dataset

The trial dataset consists of a set of 10 instances (for English and Portuguese) and 12 instances (for Spanish) of a sentence, a target complex word. The trial_none files contain only the instances and the trial_gold files contain the instances and set of gold annotations.

<br/>

Test dataset

The test_none files (used for the evaluation benchmark) contain the instances with the sentences and target complex words.

The test_gold files contain the sentences, target complex words, and gold annotations<br/>

Results of the Evaluation Benchmark

The official results for each language (en, es, and pt) can be found in this directory:<br/> /results/official

The following 10 metrics are reported in the official results:

The extended results for each language (en, es, and pt) can be found in this directory:<br/> /results/extended<br/>

The following metrics are reported in the extended results:

Evaluation Scripts

tsar_eval.py

This script evaluates the following metric:

-  MAP@1/Potential@1/Precision@1
-  MAP@3
-  MAP@5
-  MAP@10
-  Potential@3
-  Potential@5
-  Potential@10
-  Accuracy@1@top_gold_1
-  Accuracy@2@top_gold_1
-  Accuracy@3@top_gold_1  
  

Script options and help

Evaluation Script for the TSAR-2022 Lexical Simplification Shared Task

Usage: tsar_eval.py <options>

Options:
  -h, --help            show this help message and exit
  --gold_file=<PATH>    The path to the file with the gold annotated instances
  --predictions_file=<PATH>
                        The path to file with the predictions
  --output_file=<PATH>  path to the output file
  --verbose             Verbose output mode

usage example

python3 ./tsar_eval.py --gold_file ./gold/tsar_es_gold.tsv --predictions_file ./predicted/TEAMNAME_TRACKNAME_RUNNAME.tsv  --output_file ./results/TEAMNAME_TRACKNAME_RUNNAME.tsv.eval

Dataset Compilation and Baselines

A paper describing the compilation of the datasets for English, Portuguese and Spanish that includes several experiments with two state-of-the-art approaches for Lexical Simplification has been published at this link: https://www.frontiersin.org/articles/10.3389/frai.2022.991242

Lexical Simplification Benchmarks for English, Portuguese, and Spanish.<br/> Sanja Štajner, Daniel Ferrés, Matthew Shardlow, Kai North, Marcos Zampieri and Horacio Saggion.<br/> Front. Artif. Intell. Sec. Natural Language Processing. <br/> doi: 10.3389/frai.2022.991242

Preprint available at ArXiV: https://arxiv.org/abs/2209.05301

License

The python scripts follow AGPL 3.0v license. The datasets (under the /datasets directory) are licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International License CC-BY-NC-SA-4.0.

Contact

https://taln.upf.edu/pages/tsar2022-st/#contact