On Explainable Features for Translatorship Attribution: Unveiling the Translator's Style with Causality
Code for reproducing results presented in the paper On Explainable Features for Translatorship Attribution: Unveiling the Translator's Style with Causality published in IEEE Access, an Open Access Journal, on June 29, 2021.
Cite as:
@ARTICLE{9467290,
author={Caballero, Christian and Calvo, Hiram and Batyrshin, Ildar},
journal={IEEE Access},
title={On Explainable Features for Translatorship Attribution: Unveiling the Translator’s Style With Causality},
year={2021},
volume={9},
number={},
pages={93195-93208},
doi={10.1109/ACCESS.2021.3093370}}
The results can be reproduced in two ways: locally and on the cloud.
For this, you must have Anaconda (or at least miniconda) installed.
- Download or clone this repository locally.
- Open a conda prompt inside the directory and type the following command:
This will create a new environment called
$ conda env create -n translator-attribution -f environment.yml
translator-attibution
with the exact versions of the libraries used to produce the results. - From any Anaconda environment, open
jupyter lab
and you can start running the Notebooks in order (select the recently created kernel—translator-attribution
).
For this other option, you need a Google Drive account.
-
Download the contents of this repository in a folder named
translator-attribution
in your Google Drive root folder. -
You can start running the Notebooks in Colab from this repository (allow Drive to synch between steps) by clicking on the Colab badge on the top of each notebook.
Notebook Description 1 Processing Preprocessing and processing of text files 2 Feature Extraction Extraction of features and saving to disk in JSON format 3 Experiments All experiments and the results are saved to disk 4 Most Relevant Features Extraction and saving to disk of most relevant features ("translator fingerprints") for each translator and for each feature set
The results are saved as LaTeX tables and SVG images on the results folder. They are under version control and can be browsed on this repository.