=Paper=
{{Paper
|id=Vol-3180/paper-143
|storemode=property
|title=Automatic Translation and Wordplay: An Amateur’s (Playful) Thoughts
|pdfUrl=https://ceur-ws.org/Vol-3180/paper-143.pdf
|volume=Vol-3180
|authors=Fabio Regattin
|dblpUrl=https://dblp.org/rec/conf/clef/Regattin22
}}
==Automatic Translation and Wordplay: An Amateur’s (Playful) Thoughts==
https://ceur-ws.org/Vol-3180/paper-143.pdf
Automatic Translation and Wordplay:
An Amateur’s (Playful) Thoughts
Fabio Regattin a
a
Dipartimento DILL, Università degli Studi di Udine, Via Petracco 8, 33100 Udine, Italy
Abstract
While translating/adapting wordplay is certainly not, for the time being, within the reach of
machine translation (be it knowledge-based, statistical or neural), the same may hold true for
quite a few human translators as well. In my contribution, I invite readers to take part in a sort
of homemade Turing test, by asking them to uncover the translations of three open access MT
systems (DeepL, Google Translate, and Yandex), mixed with renderings of the same passages
made by human translators. Our corpus consists of English to French translations of some
excerpts taken from Lewis Carroll’s Alice’s Adventures in Wonderland.
Keywords 1
Automatic Translation, Wordplay and Pun Translation, Machine Translation
1. Introduction
A good – albeit unscientific – starting point for dealing with the problem that interests me could be
the following:
(1) Anything that is in the world when you’re born is normal and ordinary and is just a natural part
of the way the world works. (2) Anything that’s invented between when you’re fifteen and thirty-
five is new and exciting and revolutionary and you can probably get a career in it. (3) Anything
invented after you’re thirty-five is against the natural order of things. [1]
Unfortunately for me, and for my ability to rely on it, machine translation really showed its full
potential just after I turned thirty-five. I belong to a generation that has had many good laughs while
trying to make sense of the gibberish of the first Babelfish automatic translations (decidedly, Douglas
Adams is everywhere!) and who had lots of fun reading the description of Umberto Eco’s retranslation
experiments [2] carried out using that same service. I guess that this is the reason why, even if I try my
best to get rid of my deepest convictions, I find it very difficult to fully accept AT – and that I like to
see it having trouble. What could be better then, than to put it to the test in one of the areas which oppose
it the fiercest resistance? By now, those who have not already stopped reading (thank you!) will have
understood at least one thing: these pages do not claim to be exhaustive or even scientific – they are
just a sort of personal reflection by someone who knows quite little of the broad field of AT.
2. Automatic translation is definitely interesting!
To show that I am really doing all that I can to get rid of my preconceived ideas – and that I am
trying my best to raise awareness about the usefulness of machine translation to the ones who want to
hear me – I am going to ask my readers to bear an anecdote.
1
CLEF 2022 – Conference and Labs of the Evaluation Forum, September 5–8, 2022, Bologna, Italy
EMAIL: fabio.regattin@uniud.it
ORCID: 0000-0003-3000-3360
© 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
I teach French into Italian translation in a master’s degree at the University of Udine, in northeast
Italy. In order to be admitted to the program, all future students have to pass a short translation test. For
the last three years, at my first meeting with a new class, I have proposed the following activity. I take
the same text that the students translated during the exam, and I ask them to evaluate a series of versions:
mine, those of two students, those of three free machine translation services (DeepL, Google Translation
and Yandex2). I then ask two things to the class: first, to tell me if in their opinion some of the texts are
the result of machine translation (and, if so, which ones); and, second, to rank them from best to worst.
For the past three years, the resulting little poll has systematically placed my version in the first position
(a fact which, I must admit, is quite gratifying); however, the second place has almost always gone to
the text produced by DeepL (which is only seldom recognized as a machine translation, by the way).
Usually, the two human translations by my students come in third and fourth place, followed by the
versions offered by Google Translate and Yandex. Admittedly, the situation in which the human
translations are produced is far from ideal for my future students. The test is a very old-fashioned one:
it is done in a short time (1h30 for two short translations, from and into French), written by hand, and
without access to any electronic resource whatsoever. I am quite certain that the results would be very
different (although much less fun!) if I gave the same texts to translate at home. That said, I still find
that this ranking forces every translator and every translation teacher to consider machine translation as
a reality which we cannot do without. At least, this is what I tell my students.
For certain textual typologies at least, we can indeed consider that the machine output is not too far
away from human quality.
3. Reasons for hope: we, humans, still have something to say
That said, I think that humans will not disappear from the horizon of translation. Not immediately,
at least: as Nicolas Froeliger points out, “It is human translations that feed the memories used in
statistical machine translation and in CAT tools. If human translations stop, then these two techniques
can no longer work”3 [3].
It is possible to add two other touches of hope (for us, human translators, of course!) to Froeliger’s
words. First: the risk for AT to turn its virtuous circle of learning into a vicious one. The impression of
correctness that texts produced by neural MT can give, despite objective translation errors – a problem
that has been named “fluent inadequacy” by Silvia Bernardini and Federico Garcea [4] – can deceive
human readers in many different and dangerous ways. In a paper which opened a discussion on
Academia in 2019, and which seems to have been removed afterwards, Anthony Pym mentioned the
same problem, in these terms: “The output can sound so good that it tends to be convincing to a fast
reading monolingual reader, who will not know how many errors have been concealed along the
way” [5]. This is a very serious flaw, since users can then validate faulty translations which will be re-
entered into the system. As Pym again pointed out,
The problem […] is not that the algorithms are faulty; it is that many of the people who use them
are often stupid. Once the machine gives a rendition that looks fluent and convincing, users accept
the translation as valid and then put it on a website, where it can be picked up by a web crawler and
fed back into a database. Rubbish out, rubbish in, then more rubbish out – the ideally virtuous circle
turns into vicious circle and the promises of perpetual improvement come to naught [5].
Of course, human translators are far from perfect; as any practitioner knows, the risk of translation
errors is always present. However, human translators can pay more attention exactly to those passages
that, for some reason, deserve it. I will need to quote one last time, on this point, Pym’s paper:
There is some evidence that a human translator expends greater effort on the more high-risk passages
of a text, reducing the risk of error in them […]. Machine translation processes, on the other hand,
invest effort uniformly over the whole text, which means that their probability of error per sentence
2
Accessed at the following addresses: https://www.deepl.com/fr/translator; https://translate.google.com; https://translate.yandex.com.
3
“Ce sont les traductions humaines qui viennent nourrir les mémoires employées en traduction automatique statistique et en TAO. Si les
traductions humaines cessent, alors ces deux techniques ne peuvent plus fonctionner.” All translations for the quotations are mine.
remains theoretically constant. That is, if a page of human translation and a page of machine-
translation output both have three errors in them, the human translation will probably not have the
errors in the high-risk passages but the machine translation might. So when the end-user receives
the translations, they can assume all the parity they like, but they will not know exactly where those
three errors lie [5].
In its English page, DeepL claims to deliver “Fast, accurate, and secure translations”. As we saw,
the service is certainly fast and probably secure; on the other hand, it only seems to be quite accurate,
but it could give results that deviate from the source text in particularly sensitive places – those same
places where the human translator would have been particularly attentive.4 Of course, wordplay poses
a significant challenge to all these issues.
4. Wordplay and wordplay translation
Before we delve into the texts, we will need one last digression. It will concern the very definition
of the term “wordplay”, which is far from unanimous among those who have tackled the problem in
different languages. I have proposed elsewhere [7][8] a classification of this phenomenon – and of its
translation – that I will quickly summarize here. I stated (and I will refer to the cited works for the
details) that the only element which seems to unite interlinguistically the different manifestations of this
object of study is indeed the couple of terms “play”/“game”, which are usually translated into a single
word (gioco, jeu, juego, Spiel…) in most European languages. One just has to think about expressions
such as gioco di parole, jeu de mots, juego de palabras, Wortspiel… And, I added, it is possible to
distinguish at least three different semantic cores relating to those terms: a gratuitous and free activity;
a set of well-defined rules; and, finally, the activity that takes place within this set of rules. According
to Umberto Eco – who examines the question in his introduction to the Italian version of Homo ludens,
by Johan Huizinga [9] – we are not dealing here with a case of polysemy, but of homonymy: a
homonymy which is only partially unveiled by the two terms used in English. I will argue, then, that if
we want to fully understand wordplay we will first have to borrow the one and only term that so many
European languages share: the term gioco, jeu, and so on.5 In its extended meaning, covering the
meaning of the term “game”, too, “play” is both free, gratuitous, and regulated, it is at the same time a
system of rules and an activity carried out in accordance with this system; it cannot do without one of
these aspects, with an exclusive concentration on some others. And (that was my proposition at the
time, and I would tend to confirm it today) wordplay behaves the same way. It would therefore also be
three things at once:
1. “A linguistic expression containing one or more elements of identical form whose semantic bi- or
plurivalence has been consciously exploited by the user”6 [10]. Here we have a first form of play –
the gratuitous, free one, depending on the hazard of the linguistic form;
2. “A text of small dimensions whose construction obeys an explicit rule, preferably concerning the
signifier. This definition comprises three elements of unequal importance: the explicit rule, the small
dimensions, the level of the signifier”7 [11]. This is the other form of play, the one we refer to when
we use the expression “playing a game”;
3. We must finally add to these two meanings the system of rules. A football match is a jeu (this is
meaning number 2, above), but football is a jeu, too: it is a game. If we relate this third meaning to
wordplay, we could think of the Oulipian creation of new constraints.
4
A recent article by Perrine Schumacher [6] points out some other problems relating to neural MT for the English-French language pair (non-
idiomaticity, lack of coherence, abusive corrections…). We refer directly to Schumacher’s contribution for more details.
5
From now on, I will refer to the French, German, Italian, and Spanish terms using only one of them – the French jeu (plural: jeux).
6
“Une expression linguistique contenant un élément ou plusieurs éléments de forme identique dont la bi- ou plurivalence sémantique a été
exploitée consciemment par l’usager.”
7
“Un texte de petite dimension dont la construction obéit à une règle explicite, concernant de préférence le signifiant. Cette définition comporte
trois éléments d’inégale importance : la règle explicite, les petites dimensions, le niveau du signifiant.” I maintain that, talking about the
“unequal importance” of these different elements, Todorov opens up to the existence of jeux which do not play on the signifier (they could
play on the signified, as happens with some figures of thought), or of larger jeux (as, for instance, a book-length constrained text).
In its extended meaning, wordplay seems therefore to be, just like jeu, a difficult concept to pin
down, and which has at least three aspects to consider: (1) free, gratuitous jeu (play); (2) jeu as the play
of a game; (3) finally, jeu as the system of rules, i.e. as a game.
Translating wordplay is, too, a threefold operation. Jeu-1 is the one that allows the greatest latitude
of solutions: if the current trend is to translate a SL8 play with a TL play [7][12] (even at the cost of a
significant deviation from the letter of the ST), things have not always been this way. Jeu-2 seems to
require a translation which reproduces it, following the same set of rules which originated the ST.9
Talking about his Italian translation of Raymond Queneau’s Exercices de Style, Umberto Eco describes
this need when he states that he had to “understand the rules of the game, respect them and then play a
new game in as many moves”10 [13]. Nevertheless, sometimes this may not be possible, because of the
different forms that languages (or their graphic representations) can take. Some constraints are simply
not repeatable when the language changes. How could we “play the game” of the lipogram in Georges
Perec’s La Disparition, for instance, if we switched to a non-alphabetic writing system? In these cases,
we will need to “translate”, so to say, jeu-3 (the game), by creating a new rule adapted to the new
linguistic or graphic system.
5. Machines and wordplay translation
This last type of translation does not seem to me (for the moment?) within reach for AI; as far as the
translation of jeu-2 is concerned, it would perhaps be possible to obtain some results by superimposing
explicit rules on the translation system – this could be relatively easy to do as far as the game would
only be concerned with the form of the TL. Thus, an exercise in style like the following one,
“Lipogramme” [14], written without using the letter e…
Au stop, l’autobus stoppa. Y monta un zazou au cou trop long, qui avait sur son caillou un galurin
au ruban mou. Il s’attaqua aux panards d’un quidam dont arpions, cors, durillons sont avachis du
coup ; puis il bondit sur un banc et s’assoit sur un strapontin où nul n’y figurait.
…should be relatively easy to translate for a properly educated machine. On the other hand,
wordplays where a rule is mechanically applied to the source text seem to me totally out of reach for
any MT system. Just think of an exercise of the “Contre-petteries” type [14], of which I will also quote
only an extract:
Un mour vers jidi, sur la fate-plorme autière d’un arrobus, je his un vomme au fou lort cong et à
l’entapeau chouré d’une tricelle fessée. Toudain, ce sype verpelle un intoisin qui lui parchait sur les
mieds. Cuis il pourut vers une vlace pibre.
Here, the system would first need to reconstitute the source text by means of syllabic inversion, then
to translate it into the TL (this would probably be the simplest part) and finally to reverse the syllables
of the words which form the translated text, taking care at the same time to obtain “words” which,
although non-existent, follow the morphological and phonological rules of neonymy for the TL. As for
now, then, the concept of constraint linked to jeu-2 seems a very problematic one to me, as far as
machine translation is concerned. This is why I will concentrate, after this digression and these less-
than-exemplary examples, on jeu-1, the gratuitous form of play.11 As we have seen, its main
characteristics are, first, ambiguity; second, the concentration on the signifiant side of the sign (in the
Saussurean sense of the term), that is to say on the form; third – but this last trait is perhaps less central
– a relatively small size (we are quite near to the prototypical definition of “pun”).
8
From now on, I will use the acronyms SL and TL in order to refer to source language and target language; ST and TT will stand for source
text and target text.
9
This is true at least for the cases where the rule functions as a generative principle of a text – where it is the main, if not the only, reason for
its existence. If, on the other hand, the rule constitutes one element among others, the possibility of a translation that ignores it (for example,
the translation into prose or free verse of a poem written according to a precise metrical form) is not to be excluded.
10
“Capire le regole del gioco, rispettarle, e poi giocare una nuova partita con lo stesso numero di mosse.”
11
While being aware that in this field, and for the good reason that wordplay aims above all to break the rules of the linguistic system, to play
on its very weaknesses, “there can never be a fully automatic, one-size-fits-all approach” [15].
Having already described some human wordplay translations in the past, [16] in the last part of my
paper I will take some examples on which I have already worked, all drawn from Chapter IX of Alice’s
Adventures in Wonderland, and I will propose that readers do the same exercise that my students have
to do at the beginning of my classes. I will reproduce three puns in English to French translation and I
will – perhaps – slip some versions produced by MT systems in the middle of some human translations.
All readers are invited to recognize them. Of course, this sort of little homemade Turing test is of no
scientific value whatsoever: I hope it will at least be fun! That said, I am quite certain – and I write this
before realizing my “experiment” – that we will be able to distinguish the good translators from the
machine. At the same time, I am not so sure that the difference between a bad human translator and the
machine is so great: that is why I expect a number of false positives!12 The three puns chosen for the
experiment, and which will be rapidly described below, are the following:
1. Take care of the sense, and the sounds will take care of themselves [17].
2. –– Have you seen the Mock Turtle yet?
–– No – said Alice – I don’t even know what a Mock Turtle is.
–– It’s the thing Mock Turtle Soup is made from, said the Queen [17].
3. That’s the reason they’re called lessons, the Gryphon remarked, because they lessen from day to
day [17].
Let us now move on to their translations. At first, I will just list them without revealing their author.
I will then provide a short commentary, in which I will unveil the identity of the authors of each version,
including those that have been proposed by one of the translation systems that I have used. Once you
have read the different versions of these wordplays, then, I suggest that you do not read any further if
you want to give a try to the experiment. Take your time and try to discover the automatic translations!
5.1. Take care of the sense…
The wise advice of the ST plays on a paronymy, with is allusion to the proverb “Take care of the
pence, and the pounds will take care of themselves”. At the cost of a modest modification (if we read it
out loud, [s] replaces [p] on two occasions), the saying is thus diverted into a kind of ars oratoria. Here
are some French versions of the passage:
1. Prenez soin du sens, et les sons s’occuperont d’eux-mêmes.
2. Occupez-vous du sens, et les mots s’occuperont d’eux-mêmes.
3. Prenez soin du sens et les sons prendront soin d’eux-mêmes.
4. Occupons-nous du sens, et laissons les sons s’occuper d’eux-mêmes.
5. Prenez soin du sens, les sons prendront soin d’eux-mêmes.
6. Aide le sens et les sons s’aideront.
1: DeepL; 2: Papy; 3: Google Translate; 4: Parisot; 5: Merle; 6: Riot. In this first case, two versions,
4 and 6, can be attributed with sufficient certainty to humans. The effort to play is indeed more apparent
there, thanks to the homophonic sequence in 4 (laissons les sons) and by the paremiological allusion in
6 (with its reference to Aide-toi et le ciel t’aidera). As for the other versions, they are all more or less
interchangeable; it is certainly possible to notice a little more coherence in the human versions (in
particular, in the structural parallelisms prendre soin/prendre soin or s’occuper/s’occuper) but,
personally, I find that versions 3 and 5, with their protracted alliterations in s, are slightly better than
version 2. That is why I would say that both Magali Merle and Google Translate score a little better
than Jacques Papy.
12
My corpus is made up of the following human translations: Jacques Papy (1961); Henri Parisot (1968); André Bay (1980); Philippe Rouard
(1984); Magali Merle (1990); Elen Riot (2000). To make the game more interesting, only some of these translations will be quoted for any
one wordplay, and the same holds true for the machine translations.
5.2. The Mock Turtle
Lewis Carroll refers here to mock turtle soup, an imitation of the real turtle soup. Usually, the recipe
replaces its pricey ingredient with veal. The wordplay is based on a purposefully ill segmentation of the
expression: mock (turtle soup) is read as if it were mock turtle (soup).
1.
– Avez-vous déjà vu la Tortue Fantaisie ?
– Non, répondit Alice. Je ne sais même pas ce que c’est qu’une Tortue Fantaisie.
– C’est ce avec quoi l’on fait la Soupe à la Tortue « Fantaisie », précisa la Reine.
2.
– As-tu déjà vu la Simili-Tortue ?
– Non, répondit Alice, je ne sais même pas ce que c’est, une Simili-Tortue.
– C’est la chose qui sert à faire le consommé à la Simili-Tortue, repartit la Reine.
3.
– Avez-vous déjà vu la tortue simulée ?
– Non – dit Alice – Je ne sais même pas ce qu’est une tortue simulée.
– C’est la chose dont la soupe de tortue fantaisie est faite, dit la Reine.
4.
– Avez-vous déjà vu la Tortue-à-Tête-de-Veau ?
– Non, dit Alice, je n’ai même pas idée de ce que ça peut être.
– C’est ce qui sert à faire la fausse soupe à la tortue, dit la reine.
5.
– Tu as déjà vu la Fausse Tortue ?
– Non – dit Alice – Je ne sais même pas ce qu’est une Fausse Tortue.
– C’est la soupe de tortue factice, dit la Reine.
1: Parisot; 2: Merle; 3: DeepL; 4: Bay; 5: Yandex. Here, the difference between human and machine
translators seems to be much clearer. The only version that replicates the original play on words is 1,
where Fantaisie can be related to both tortue and soupe à la tortue. The cultural reference is lost, but
this seems a lesser evil to me. Version 2 is consistent, sure, but there’s no wordplay, since the prefix
simili- can only refer to turtle. In 4, the illustrations may have directed the explanation of the first line
(the classic images drawn by John Tenniel indeed show a sort of calf-turtle chimera13); however, the
play on words is lost. The two versions produced by the machine seem to me less acceptable here.
DeepL performs much better than Yandex; the latter completely leaves out the idea of an ingredient
(the fausse tortue turns into a soup in the last line), but version 3 is also quite awkward (by the choice
of the word simulée, as well as by the lack of coherence between this simulée and the fantasie of the
last line).
5.3. Lessons
The false etymology in question is entirely based on the phonological identity of lesson and lessen.
1. C’est bien pour ça qu’on les appelle des cours, fit observer le Griffon, parce qu’ils raccourcissent
d’un jour sur l’autre.
2. C’est la raison pour laquelle on les appelle des leçons, a remarqué le Griffon, parce qu’ils
diminuent de jour en jour.
3. C’est pour cette raison qu’on les appelle des cours, fit remarquer le Griffon : parce qu’ils
deviennent chaque jour un peu plus courts.
4. C’est la raison pour quoi l’on appelle ça des cours, fit observer le Griffon : parce qu’ils deviennent
de jour en jour plus courts.
13
Tenniel’s illustrations were designed for the first edition of the book, and have been taken over by various editions in several languages.
The image I am referring to is available at https://en.wikipedia.org/wiki/Mock_Turtle#/media/File:Alice_par_John_Tenniel_34.png.
5. C’est la raison pour laquelle on les appelle des leçons, a fait remarquer le Gryphon, car elles
s’atténuent de jour en jour.
1: Riot; 2: Yandex; 3: Rouard; 4: Parisot; 5: DeepL. In this case, the readers will be quite confident
in their attributions: the human translators seem to perform much better than the machine. On the human
side, there is convergence on a solution that makes it possible to entirely keep the play of the ST.
Versions 3 and 4 are, from this point of view, almost interchangeable; version 1 seems a little bit less
successful to me, since it does not directly reproduce the homophony but merely alludes to it. As for 2
and 5, they lead to nonsense (no relation between the word leçon and the fact of shortening) and show
other problems, too: the choice of verb tense (but the absence of context did not help) for both versions;
a spelling problem (Gryphon) and a rather absurd lexical choice (s’atténuer) for DeepL; the choice of
the wrong personal pronoun (des leçons, ils diminuent) for Yandex.
6. We were just playing around, so… are there any conclusions?
At the end of this brief journey, no real conclusion seems possible to me. I realize that these lines
have had above all the function of reassuring myself and, perhaps, some of my human colleagues –
those who, like me, see the progress of AT, and its more and more efficient results, with some concern.
Trying to defy some MT engines in the field of wordplay translation was no more than another playful
exercise. That said, I believe that if machine translation wants to have a chance of success when
confronted with wordplay, it will have to set itself, initially at least, a less ambitious objective than
reaching a humanlike translation quality.
As we have seen, linguistic jeu is a multifaceted entity: it is several things at the same time. This is
why AT could begin by addressing only one of the jeu types I highlighted, and, within it, only one or
some of its categories. Perhaps, it would already be great if we could achieve a very good, near-human
automatic recognition of linguistic jeu; and, from there, we could insist on machine-assisted human
translation, along the lines proposed by Miller [15].
As a translator and a translator trainer, I can be quite certain about one thing: if we obtain a help that
allows us to get rid of the most complicated parts of this work, arriving at a solution on our own can be
not only simpler, but also very fun. And why should we, as humans, deprive ourselves of this little
pleasure?
7. References
[1] D. Adams, The Salmon of Doubt, William Heinemann, London, 2002.
[2] U. Eco, Dire quasi la stessa cosa. Esperienze di traduzione, Bompiani, Milano, 2003.
[3] N. Froeliger, Les Noces de l’analogique et du numérique : de la traduction pragmatique, Belles
Lettres, Paris, 2013.
[4] S. Bernardini, F. Garcea, Come funziona, e quanto ci serve, la traduzione automatica,
Linguisticamente.org, 2020, https://bit.ly/3j0jK4S.
[5] A. Pym, How automation through neural machine translation might change the skill sets of
translators, pre-print on Academia.com, 2019. Accessed in September 2019 and removed.
[6] P. Schumacher, La traduction automatique neuronale : technologie révolutionnaire ou poudre de
perlimpinpin ? Compte-rendu d’une expérience pédagogique, Al-Kīmiyā 18, 2020, 67-89,
https://bit.ly/3ayyPqx.
[7] F. Regattin, Le Jeu des mots, Emil, Bologna, 2009.
[8] F. Regattin, Traduire les jeux de mots : une approche intégrée, Atelier de traduction 23, 2015, 129-151.
[9] U. Eco, Homo ludens oggi. In J. Huizinga, Homo ludens, Einaudi, Torino, 1973, vii-xxvii.
[10] R. Landheer, Les règles du jeu de mots en français moderne. In A.G. Sciarone et al. (eds.), Nomen.
Leyden Studies in Linguistics and Phonetics, Mouton, The Hague-Paris, 1969, 81-103.
[11] T. Todorov, Les Genres du discours, Seuil, Paris, 1978.
[12] J. Henry, La Traduction des jeux de mots, Presses de la Sorbonne Nouvelle, Paris, 2003.
[13] U. Eco, Introduzione. In R. Queneau, Esercizi di stile, Einaudi, Torino, 1983, v-xix.
[14] R. Queneau, Exercices de style [1947], Gallimard, Paris, 1997.
[15] T. Miller, The punster’s amanuensis: the proper place of humans and machines in the translation
of wordplay. In I. Temnikova et al. (eds.), Proceedings of the 2nd Workshop on Human-Informed
Translation and Interpreting Technology (HiT-IT 2019), Varna, Bulgaria, 2019, 57-65.
[16] F. Regattin, La traduction des jeux de mots : question de langue, question de traducteur ? In De
l’autre côté du miroir : Translations, déplacements, adaptations. Actes du 10e colloque estudiantin
de la SESDEF, University of Toronto, 2008, https://bit.ly/3MWZ2QJ.
[17] L. Carroll, The Annotated Alice. Alice’s Adventures in Wonderland [1865] and Through the
Looking-Glass [1872], ed. M. Gardner, Penguin, Harmondsworth, 1970.