=Paper=
{{Paper
|id=Vol-3180/paper-145
|storemode=property
|title=How good can an automatic translation of Pokémon names be?
|pdfUrl=https://ceur-ws.org/Vol-3180/paper-145.pdf
|volume=Vol-3180
|authors=Léa Talec-Bernard
|dblpUrl=https://dblp.org/rec/conf/clef/Talec-Bernard22
}}
==How good can an automatic translation of Pokémon names be?==
https://ceur-ws.org/Vol-3180/paper-145.pdf
How good can an automatic translation of Pokémon names be?
Léa Talec-Bernarda
a
University of Western Brittany (UBO), 20 Duquesne Street, Brest, 29490, France
Abstract
For those of you who are not familiar with the successful franchise of Pokémon, it revolves
around imaginary creatures often representing animals mixed with objects, plants, etc.
Their names reflect these characteristics and are, most of the time, wordplays.
For this paper some Pokémon names were automatically translated by using the translation
model T5 with the use of Python. The aim of this paper is to evaluate the overall quality of
such translation. The results were very diverse, most source sentences stayed unchanged
but some others were surprising and interesting to point out for diverse reasons.
Keywords
Automatic translation; Wordplay; Humour; Pokémon names
1. Introduction
For those of you who are not familiar with the successful franchise of Pokémon, it revolves around
imaginary creatures often representing animals mixed with objects, plants, etc. Their names reflect these
characteristics and are, most of the time, wordplays. For this paper the T5 model [1] was used to
automatically translate short puns, many of which were Pokémon names.
Wordplays and puns are amongst the trickiest things to translate as they often revolve on linguistic
and cultural aspects specific to the source language. The translator has to make sure that the translation
is understandable to the target language all the while keeping its humoristic value. [2] Therefore, as one
can imagine, automatically translating puns can prove very difficult as machines lack the human
discernment of humor. This paper aims at analyzing the quality of automatic translations made with the
use of the T5 model.
The JOKER projet [3] aims at unifying the scientific community working on the automatic
translation of humor and puns. The Handbook of Translation Studies [4] also features a section about
humor in translation in which the author addresses the challenges faced with translating humor.
2. Task
The task performed in order to analyze the use of the T5 model in translating humor is the JOKER
lab [5] task 2 named “Translate a given punning construction in a proper noun or a neologism from
English into French.”. This lab was organized as a part of the CLEF-2022 conference. The goal of this
task is to automatically translate single-word or short puns from English to French, which is particularly
tricky due to the lack of context surrounding each word for the model to use. The data includes 1,161
examples of names of videogames or comics characters containing puns. Many of these names being
Pokémon names, this paper will focus on these examples.
3. Methodology description
CLEF 2022 – Conference and Labs of the Evaluation Forum, September 5–8, 2022, Bologna, Italy
EMAIL: Lea.Talec-Bernard@etudiant.univ-brest.fr
©️ 2022 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR Workshop Proceedings (CEUR-WS.org) Proceedings
The Google T5 Model was executed with the use of Python to perform these tasks with the help of
the Simple T5 library1. This model is one of the most performant models existing today. It allows its
users to perform many different NLP (Natural Language Processing) tasks which include translating,
summarizing or simplifying documents amongst others. Its performance comes from the fact that it is
pre-trained with a very diverse corpus called C4 (Colossal Clean Crawled Corpus) which gathers data
from the Common Crawl, an open corpus created from many documents found on the net especially for
projects requiring a lot of data. The data from the C4 corpus has been filtered in order to obtain the best
result possible. In addition to this data, it is possible to train the T5 model with our own data similar to
the document we want to modify. The user also has the possibility to change some of its parameters,
allowing a more personalized result. Some of these parameters include the number of times the model
will train on the training data or the creativity of the model.
4. Results
The final results were not submitted to the CLEF organizers and were, therefore, not evaluated by
them. Some results were interesting enough to point out.
The Pokémon named “Wartorlte” in English was automatically translated into “Brutadou” in
French. The English name is composed of the word “war” and the word “turtle”. We can see that the
T5 model translated the idea of war by the prefix “brut”, which keeps the idea of brutality. However,
the “turtle” aspect was lost. One interesting thing to note is the end of the name “Brutadou” which makes
it fit well with the universe of Pokémon by sounding similar to some other Pokémon names.
Another clever translation was that of “Morelull”, a Pokémon name mixing the words “morel” and
“lullaby”. This name was translated into “Mélulli”. Similarly to the case of “Brutadou”, the translation
only kept one of the two aspects of the original name. The first half of the name “mél”, refers to the
French word “mélodie” which we could translate by “melody” and conveys a similar idea to “lullaby”
in the original name. The second half of the name also conveys this idea but was not translated into
French, instead, the “lull” part was kept but and “i” was added at the end, making the name sound more
like French Pokémon name.
The T5 model was however not as successful for every translation. In fact, in many cases, the names
were not translated at all. In some other cases, the model misinterpreted some aspects of a name. For
example, the Pokémon named “Zangoose” was translated into “Gazouille” a French word evoking the
song of a bird. The model surely interpreted the end of the name “zangoose” as the bird “goose”, in
which case the translation is clever, however, “Zangoose” is a pun using the words “zankon” (“Scar” in
Japanese) and “mangoose” instead of “goose”.
5. Conclusion
Although the model T5 often did not translate the names it was given to translate, it still created
some very interesting translations. Some of which could be used as such, some other could spark hints
of better ideas in the corrector. With some tweaking and improvement, the T5 model could become, in
the future, a great tool in helping translators with their work.
6. Acknowledgements
This group project was done during a week-long intensive course about Artificial Intelligence hosted
by Liana Ermakova and organized by the SEA-EU in April 2022. I would like to thank Liana Ermakova,
1
https://github.com/Shivanandroy/simpleT5
[Tapez ici]
from the University of Western Brittany, for her precious help in setting up the T5 model, for mentioning
this CLEF event as well as for hosting the SEA EU. I would also like to thank my classmates Nina
Španović (University of Split, Croatia), Julliette Le Berrigot (University of Western Brittany, France)
and Mikołaj Bondaryk (University of Gdańsk, Poland) for their collaboration on this project during the
SEA EU class.
7. References
[1] Delia Chiaro, Translation, Humour and Literature, Volume 1, Continuum Advances in Translation,
2010.
[2] Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi
Zhou, Wei Li, Peter J. Liu, (2019), Exploring the Limits of Transfer Learning with a Unified Text-
to-Text Transformer
[3] CLEF Workshop JOKER: Automatic Wordplay and Humour Translation, Liana Ermakova, Tristan
Miller, Orlane Puchalski, Fabio Regattin, Élise Mathurin, Sílvia Araújo, Anne-Gwenn Bosser,
Claudine Borg, Monika Bokiniec, Gaelle Le Corre, Benoît Jeanjean, Radia Hannachi, ̇Gor ̇g Mallia,
Gordan Matas, and Mohamed Saki, 2022
[4] Jeroen Vandaele, Humor in translation, in: Handbook of Translation Studies Volume 1, Edited by
Yves Gambier and Luc van Doorslaer (p.157-162)
[5] Ermakova, L., Miller, T., Regattin, F., Bosser, A.-G., Mathurin, É., Corre, G. L., Araújo, S., Boccou,
J., Digue, A., Damoy, A., Campen, P., & Jeanjean, B. (2022). Overview of JOKER@CLEF 2022:
Automatic Wordplay and Humour Translation workshop. In A. Barrón-Cedeño, G. Da San Martino,
M. Degli Esposti, F. Sebastiani, C. Macdonald, G. Pasi, A. Hanbury, M. Potthast, G. Faggioli, & N.
Ferro (Eds.), Experimental IR Meets Multilinguality, Multimodality, and Interaction. Proceedings
of the Thirteenth International Conference of the CLEF Association (CLEF 2022) (p. 25).
[Tapez ici]