Content uploaded by Manuel Lardelli
Author content
All content in this area was uploaded by Manuel Lardelli on Oct 31, 2023
Content may be subject to copyright.
Gender-Fair Language in Translation: A Case Study
Angela Balducci Paolucci
University of Vienna, Austria
angelabalducci4@gmail.com
Manuel Lardelli
University of Graz, Austria
manuel.lardelli@uni-graz.at
Dagmar Gromann
University of Vienna, Austria
dagmar.gromann@gmail.com
Abstract
With an increasing visibility of non-
binary individuals, a growing number of
language-specific strategies to linguisti-
cally include all genders or neutralize any
gender references can be observed. Due
to this multiplicity of proposed strate-
gies and gender-specific grammatical dif-
ferences across languages, selecting the
one option to translate gender-fair lan-
guage is challenging for machines and hu-
mans alike. As a first step towards gender-
fair translation, we conducted a survey
with translators to compare four gender-
fair translations from a notional gender
language, English, to a grammatical gen-
der language, German. Proposed transla-
tions were rated by means of best-worst
scaling as well as regarding their readabil-
ity and comprehensibility. Participants ex-
pressed a clear preference for strategies
with gender-inclusive character, i.e., colon.
1 Introduction
Gender in language reflects on an extra-linguistic
reality (Corbett, 1991) in the sense that it reflects
gender associations and stereotypes of a society.
To respect different gender identities, i.e., the sense
of self and “who they are” (Barker and Iantaffi,
2019), it is vital to linguistically acknowledge their
existence within and across languages. Machine
translation (MT) is known to suffer from gender
bias, which is problematic for many reasons. For
instance, machine-translated online contents are
© 2023 The authors. This article is licensed under a Creative
Commons 4.0 licence, no derivative works, attribution, CC-
BY-ND.
consumed without people being aware that they are
MT mediated (Martindale and Carpuat, 2018). In
MT research, the idea to resort to gender-neutral
language to avoid gender issues has been proposed
(Piergentili et al., 2023). However, apart from
information loss, this might not be the preferred
gender-fair strategy by humans. To analyse hu-
man preferences, we propose a first survey1among
language professionals of four distinct gender-fair
translation strategies from English to German.
Translation studies has a long tradition of
considering gender issues, such as in feminist
(Von Flotow, 1997) and queer translation (Baer
and Kaindl, 2017). However, gender beyond the
binary has so far received little scholarly atten-
tion (e.g. Misiek (2020) and L´
opez (2022)). The
same is true for the field of MT, where debias-
ing strategies focus on a binary conception of gen-
der, with some important exceptions (Tomalin et
al., 2021; Saunders and Byrne, 2020). Gender-fair
language, which subsumes gender-inclusive and
gender-neutral strategies, is particularly challeng-
ing in case of grammatical gender languages, i.e.,
several word classes require gender inflections.
In this case study, ten language professionals
rated four gender-fair translations of online mag-
azine articles in direct comparison and regarding
their impact on readability and comprehensibility.
The four German strategies consist of one gender-
neutral neosystem, one gender-inclusive neosys-
tem, a gender-inclusive colon with si:er, and the
same colon with neopronoun xier. Since rating a
translation is in general a highly subjective matter,
the selected method is best-worst scaling, which
allows participants to select and rate their subjec-
tively most (best) and least (worst) preferred trans-
1The survey is made available on Zenodo: https://
zenodo.org/record/7951054
Eva Vanmassenhove, Beatrive Savoldi, Luisa Bentivogli, Joke Daems & Jani¸ca Hackenbuchner
Proceedings of the 1st Workshop on Gender-Inclusive Translation Technologies,p. 13–23
Tampere, Finland, June 2023.
lation. In a previous gender-fair MT workshop
we conducted with translators, non-binary people,
and MT experts (Burtscher et al., 2022), read-
ability and comprehensibility of gender-fair lan-
guage strategies were repeatedly named as impor-
tant factors in the selection process. Thus, we de-
cided to include a rating of these two dimensions
in the present survey. Furthermore, participants
were requested to motivate their choice in the form
of a free text answer. While the perspective of
non-binary individuals and MT experts would be
equally interesting, we believe that preferences and
considerations of language professionals as pro-
ducers of (gender-fair) translations are of vital im-
portance to the field of translation studies as well
as machine translation. The results of this survey
contribute to the discussion on which gender-fair
language strategy is preferred in (machine) trans-
lating to German and which considerations are par-
ticularly important for language professionals.
2 Related Work
Since the focus of this article is on analyzing
gender-fair translation strategies as a first step, this
section focuses on work on gender-fair transla-
tion. In spite of the recent development of queer
translation studies (Baer and Kaindl, 2017), re-
search in the field of translation studies rarely
addresses non-binary genders (Lardelli and Gro-
mann, 2023). Most research focuses on media
translation, e.g. subtitled and dubbed series, and
news articles (L´
opez, 2022; Attig, 2022; Misiek,
2020; ˇ
Sincek, 2020).
L´
opez (2019; 2022) and Attig (2022) analysed
the dubbed and subtitled versions of the Netflix
series One Day at a Time in Spanish and French.
They found that the gender-fair language strategies
used varied between the dubbed and subtitled ver-
sions as well as from European to Latin American
Spanish. The non-binary character was correctly
addressed with non-binary neopronoun elle in the
European Spanish dubbed version only. In the
other cases, they were misgendered with female
forms and/or literal translations of English singu-
lar they. Similarly, in the French dubbed version,
non-binary neopronoun ielle was used whereas in
the subtitles the character was referred to with in-
definite pronoun on (one/we).
In their analysis of English TV series translated
to Polish, Misiek (2020) found a systematic omis-
sion of the non-binary characters’ gender identity.
This phenomenon could also be observed in Croat-
ian movie translations and articles on Sam Smith’s
coming out as non-binary where the third per-
son masculine plural pronoun was generally used
(ˇ
Sincek, 2020). ˇ
Sincek (2020) represents also one
of the few works to include interviews with people,
i.e., non-binary individuals, on the topic.
Recent developments in gender-fair language
strategies have been studied in psycholinguistics
with a focus on binary genders. For instance,
Lindqvist et al. (2019) conducted experiments in
Swedish and English and tested different strate-
gies to reduce male bias in language, i.e., (i) binary
paired forms, (ii) gender-neutral words as well as
(iii) gender-fair pronoun hen and English singular
they. Participants read a description of a candidate
for a job position and were asked to select pho-
tos of men or women corresponding to the said
description. The results suggest that (i) and (iii)
actively reduce male bias.
In German, empirical research concentrated on
the cognitive processing of textual information.
Braun et al. (2007), for example, tested the ef-
fect of male generics and two binary gender-fair
language forms on memory performance and text
intelligibility. No differences in memory perfor-
mance across strategies were found between men
and women. However, as concerns intelligibility,
women indicated no preferences, while men indi-
cated a preference for male generics.
To the best of our knowledge, this is the first
study to consult language professionals regarding
their preferences regarding gender-fair language
strategies. Since language professionals play the
important role of producing gender-fair transla-
tions, needed to fine-tune MT models, we believe
that their perspective is interesting for translation
studies and the field of machine translation.
3 Preliminaries
In order to establish the theoretical foundation of
the present survey, an introduction to the interac-
tion of gender with language and translation is pro-
vided, followed by a brief overview of gender-fair
language strategies in English and German.
3.1 Gender and Language
The relation between gender and language is com-
plex because the term has multiple meanings. In
the field of gender studies, it is defined as a biopsy-
chosocial construct (Barker and Iantaffi, 2019). It
14
hence involves biological, e.g. hormonal, psy-
chological, e.g. a person’s sense of self, and so-
cial, e.g. normative and cultural expectations, fac-
tors. It is commonly used in reference to gen-
der identity, i.e., a person’s sense of their gender,
and not the sex assigned at birth. In linguistics,
the term is generally defined as “classes of nouns
reflected in the behaviour of associated words”
(Hockett, 1958, 231). In other words, associated
word classes are inflected based on the grammati-
cal gender of a specific noun.
Gender is realised differently in natural lan-
guages, which can be classified into (i) grammat-
ical gender, (ii) notional gender, and (iii) gender-
less languages (Stahlberg et al., 2007; McConnell-
Ginet, 2013). In (i), such as German and Italian,
each noun has a gender (Corbett, 1991) and exten-
sive gender marking is required. In (ii), such as En-
glish, third person singular pronouns, i.e., he, she,
it, and specific nouns, e.g. boy/girl, are gender-
specific. In (iii), such as Turkish, gender may be
expressed, e.g. in kinship, but is not grammati-
cally encoded in linguistic structures. Gender as-
signment in the case of human referents is based
on the extra-linguistic reality of a society (Cor-
bett, 1991) and reveals gender associations and
stereotypes as well as connotations (Nissen, 2002;
Jakobson, 1959).
3.2 Gender and Translation
Differences in linguistic structures and gender-
specific connotations impact the translation pro-
cess. In the first case, the translation from no-
tional to grammatical gender languages can re-
quire choices that are not neutral (Nissen, 2002;
Di Sabato and Perri, 2020). In several literary
works, for instance Written on the Body (1993), a
mysterious atmosphere is created by omitting gen-
der markers. However, when translating to another
language, this omission of gender might not be
grammatically feasible, potentially forcing trans-
lators to assign a gender to characters (Di Sabato
and Perri, 2020). This choice is often based on so-
cial gender, i.e., stereotypical associations to gen-
der in a society (Nissen, 2002). In the second case,
gender can be used to convey particular connota-
tions through personifications and metaphors. This
occurs, for example, in marketing texts and/or ad-
vertisement, where an animal, such as a male, fast
tiger, is used to represent a car. Since the same an-
imal can have different or no gender-specific con-
notations in other languages and cultures, transla-
tion choices that deliver the same source text mes-
sage are required (Di Sabato and Perri, 2020).
3.3 Gender-Fair Language
Gender-fair language has a long tradition. Its
development goes back to the 1960s, when dif-
ferences in the linguistic treatment of men and
women gained the attention of second-wave fem-
inists (Kramer, 2016). With an increased visibil-
ity of non-binary people, new gender-fair language
strategies have been accordingly proposed.
In English, singular they has become common
to refer to people whose gender is unknown or
irrelevant to the context of conversation as well
as non-binary people (Apa Style, 2019). Fur-
thermore, gender-neutral alternatives to gendered
words, such as chairperson instead of chairman,
are increasingly used (Weatherall, 2002). In Ger-
man, a grammatical gender language that requires
extensive gender marking, there are mainly four
approaches:
•gender-neutral rewording: sentences are
phrased in order to avoid gendered structures,
e.g. person as gender-neutral word, indefinite
pronouns, passive constructions and particip-
ial forms;
•gender-inclusive characters: typographic
characters, such as gender star (*) or colon (:),
are used to separate male forms from female
endings and include all genders, e.g. Leser*in
(reader). It is also possible to separate the
stem from the noun ending as in Lese*rin,
which should prevent binary thinking.
•gender-neutral characters or endings: for
example xin Lesx (reader) are used to ques-
tion the gender binary.
•neosystems:
–gender-inclusive: a new gender is in-
troduced in the language as in the case
of the Sylvain system (De Sylvain and
Balzer, 2008) with Lesernin (reader).
–gender-neutral: the ens pronoun and
suffix as in Lesens (reader) is intro-
duced as gender-neutral form derived
from Mensch (human) (Hornscheidt and
Sammla, 2021).
15
Furthermore, several neopronouns have been
proposed. For instance, xier is the result of the
combination of third person singular female sie
and male er pronoun and has already been used in
the translation of some English language TV series
(Heger, 2020). Several more detailed overviews
of gender-fair language in German are available
(Hornscheidt, 2012; En et al., 2021; Hornscheidt
and Sammla, 2021).
4 Method
In order to evaluate the perception, readability, and
comprehensibility of gender-fair language strate-
gies, two empirical methods targeted to measure
subjective impressions were selected, i.e., Best-
Worst Scaling (BWS) and the Likert scale. BWS
(Louviere and Woodworth, 1990), a comparative
annotation method, was used to select and evalu-
ate the subjectively best and the worst translation
strategy, whereas the Likert scale (Likert, 1932), a
rating scale, was used to rate the readability and
comprehensibility of the best and worst strategy
chosen by the participants. Readability refers to
whether a text written in a specific gender-fair lan-
guage strategy is easy and enjoyable to read for the
participants of this study subjectively. Comprehen-
sibility refers to the ease to understand the mes-
sage of a text written in a specific strategy for the
participants of this study subjectively. The choice
to combine these two methods is based on the de-
sire to limit the granularity and inconsistencies that
can occur when using solely a rating scale (Kir-
itchenko and Mohammad, 2017).
4.1 Data and Strategy Selection
Four English texts containing the use of singu-
lar they were selected from online articles to be
translated using four different gender-fair language
strategies. To be specific, the texts selected were
interviews and reports on non-binary people in En-
tertainment Weekly (Text 1), People (Text 3) and
on the website of the Brown University (Text 2) as
well as a set of instructions on how to support a
non-binary friend published on Sociomix (Text 4).
Due to the fact that German is a grammatical gen-
der language that associates gender with nouns in
addition to pronouns, adjectives, and determiners,
selected texts should allow to reflect this grammat-
ical variety in the translation. For each original
text, four gender-fair translations are provided in
a set, which only differ in the utilized gender-fair
language strategy. All translations were created
manually and checked by three experts on gender-
fair German. As strategies to be employed during
the translation process, the choice fell on:
1. gender-neutral neosystem ens, because of its
simple grammatical structure, where no de-
clension and consequently easy use is ex-
pected;
2. gender-inclusive Sylvain neosystem, follows
the grammatical rules of the German lan-
guage, which is why it is expected to appear
more natural;
3. colon after the word stem in combination
with the pronoun si:er, because the colon
is already widely known and used and with
the two binary German pronouns combined
should least impact readability and compre-
hensibility, and
4. colon after the word stem in combination with
the xier pronoun, for the same reason of the
colon and because the “x” explicitly empha-
sizes the inclusion of all genders, not only bi-
nary genders (Heger, 2013).
To exemplify the type of text and gender-fair
translation strategies that were used in this study,
we provide all four strategies for the sentence Jim
is a fierce pirate who journeys the seas seeking re-
venge on the people that killed their family. of Text
1, an Entertainment Weekly interview with and ar-
ticle on Vico Ortiz who starred as non-binary pirate
Jim in Our Flag Means Death:
1. Jim ist einens grimmig Piratens, dens durch
die Meere reist, um sich an den Personen zu
r¨
achen, die ens Familie get¨
otet haben.
2. Jim ist einin grimmigin Piratnin, din durch
die Meere reist, um nimser an den Perso-
nen zu r¨
achen, welche nimse Familie get¨
otet
haben.
3. Jim ist ei:ne Pira:tin, dier durch die Meere
reist, um sich an den Personen zu r¨
achen,
welche siese Familie get¨
otet haben.
4. Jim ist ei:ne Pira:tin, dier durch die Meere
reist, um sich an den Personen zu r¨
achen,
welche xiese Familie get¨
otet haben.
16
4.2 Participant Selection
For a principled selection of participating language
professionals, a number of criteria had to be spec-
ified. First, their first language had to be German
and they had to have a high command of English,
i.e., C1 to C2 of the Common European Frame of
Reference for Languages (CEFR), in order to be
able to better identify which gender-fair strategy
could be used as a translation for the English sin-
gular they. Second, participants were required to
have completed or be about to complete a profes-
sional education in the field of translation. Finally,
at least some practical translation experience be-
yond exercises during the education was required.
4.3 Survey Design
After introductory instructions and basic questions
in a Google Forms survey, four translations cor-
responding to the four gender-fair strategies were
presented side by side with the English original for
each of the four source texts. For each pair of orig-
inal and translations, participants were asked to se-
lect the best and the worst translation from the set
and rate the former on a scale from +4 (very good)
to 0 (neutral) and the latter from 0 (neutral) to -4
(very bad), a common scale and practice in BWS.
Furthermore, participants were requested to rate
the readability and comprehensibility of the best as
well as worst translation selected on a Likert scale
from 5 (very true) to 1 (not true). For the best strat-
egy, the statements to be rated were that the best
strategy does not impact the readability of the text
and with the best gender-fair strategy the text is
easy to understand. Thus, a rating of 5 means easy
to read and highly comprehensible. For the worst
strategy, the statements to be rated were that the
worst strategy impacts the readability and makes
the text hard to understand. Thus, a rating of
5 means hard to read and low comprehensibility.
The general assumption was that the best strategy
would have little impact on these two dimensions,
while the worst is expected to achieve low ratings
for both. Participants were also requested to op-
tionally motivate their best/worst choices for each
individual set of gender-fair translations as a free
text answer. Furthermore, the demographic and
general answers were analyzed to determine dif-
ferences across participants and gather their prior
experience with gender-fair language and transla-
tion as well as their opinion on the topic. The basic
questions, thus, included participants’ experience
with and impressions on gender-fair language.
4.4 Analysis
The numeric BWS ratings are summed up by strat-
egy across all four sets and all participants and
divided by the number of times the strategy was
rated to obtain the finally best and worst strategy
on average in the survey. The same procedure was
applied to the ratings on readability and compre-
hensibility. Finally, the free text answers and basic
questions were analyzed and annotated for a topic-
wise presentation of the results.
5 Results
After presenting participants’ profiles, their pref-
erences regarding the evaluated strategies, ratings
for readability and comprehensibility, and overall
comments on the topic are detailed.
5.1 Participant Profile
From the ten participants in the survey, nine iden-
tified as woman and one as man. In terms of age,
30% were between 18 and 25, 40% between 26
and 29, and 30% between 30 and 40. As required,
all participants indicated to be professionally edu-
cated, have translation experience, and a high com-
mand of English (C1 or C2). All participants indi-
cated to have prior knowledge of gender-fair lan-
guage strategies, in particular neutral rewording
and inclusive gender star and colon, and 90% indi-
cated to be actively using gender-fair language in
their daily lives. Another binary strategy that was
indicated is to camel case plural endings with I to
include men and women, e.g. LeserInnen instead
of the female Leserinnen or the male Leser.
5.2 Ratings of Gender-Fair Translations
The detailed results of BWS ratings per partici-
pant, text, and gender-fair strategy are presented
in Table 1. Each of the ten participants rated one
translation per set as best and one as worst, result-
ing in a total of 40 positive/neutral and 40 nega-
tive/neutral ratings for four sets. Positive ratings
are marked in green, negative ratings in red, and
neutral ones in gray. The translation strategies in
the columns correspond to the numbered list in
Section 4.1, that is, S1 corresponds to the ens strat-
egy, S2 the Sylvain system, S3 colon + si:er, and
S4 colon + xier.
In Table 2, the counts of how often a strategy
was selected as best or worst as well as the overall
17
Text 1 Text 2 Text 3 Text 4
Part. S1 S2 S3 S4 S1 S2 S3 S4 S1 S2 S3 S4 S1 S2 S3 S4
P1 0 +3 0 +3 0 +3 +3 0
P2 -4 +1 -2 +2 -3 +4 -1 +1
P3 -3 +2 -3 +2 -2 +2 +3 -2
P4 -2 +3 -3 +3 -3 +4 -1 +4
P5 -3 +2 -3 +3 -3 +3 -3 +4
P6 -4 0 -4 0 -4 0 -4 0
P7 -4 0 0 +1 0 +2 0 +3
P8 +3 -2 +2 -1 +2 -2 +2 -2
P9 -3 +3 -2 +3 -4 +3 -2 +3
P10 -3 +2 -2 +2 -3 +2 -3 +2
Sum -7 -18 +11 +5 0 -18 +8 +11 -10 -12 +14 +9 +1 -9 +8 +7
Table 1: Detailed BWS rating results per participant, strategy, and text
count and percentage it was selected are presented,
alongside the sum, average and median BWS rat-
ing. The gender-fair translation strategy S4 ob-
tained 18.75% of all ratings and achieved the best
average rating of 2.13, followed by S3 with 25%
of all ratings and on average 2.05 as can be seen
from Table 2. While the Sylvain system obtained
by far the most ratings, i.e., 36.25% of in total 80
ratings, from the numeric rating distribution and
the color coding in Table 1 it becomes evident that
it obtained mostly negative scores and the worst
overall result with on average -1.97. Finally, S1
obtained 20% of all ratings and on average a fi-
nal score of -1. Interestingly, S3 colon + si:er was
never selected as worst strategy and did not obtain
a single negative rating as can be seen from Ta-
ble 1. Furthermore, it was most frequently selected
as best strategy with 20 (50%) out of 40 positive
rating counts. The overall best strategy S4 colon +
xier was only selected once as worst strategy and
obtained a negative rating by P3. Given that P3
breaks their previous pattern of rating S2 as worst,
it might have been an accidental selection.
S1 S2 S3 S4
Best C. 5 1 20 14
Worst C. 11 28 0 1
Total C. 16 29 20 15
Av. C. (%) 20.00 36.25 25.00 18.75
Sum R. -16 -57 41 32
Av. R. -1.00 -1.97 2.05 2.13
Median R. -2.00 -3.00 3.00 2.00
Table 2: Summary of BWS rating results (C = Count; R =
Rating; Av = Average)
While the decision that the S2 Sylvain system is
the worst strategy was quite unanimous, some par-
ticipants revealed individual preferences as can be
seen in Table 1. Participant P8 showed a strong
preference for the S1 ens strategy for all texts,
while overall S1 obtained more negative than pos-
itive or neutral ratings. In terms of intra-annotator
consistency, participants P2, P8, and P9 are com-
pletely consistent in their selection of strategies
across texts. Other participants, especially P3, P5,
and P7 changed their preferred strategies depend-
ing on the text, in particular with Text 3 and Text
4. This change could be attributed to the fact that
the first two texts are equivalent in type since both
are interviews, while Text 3 is a report on Demi
Lovato and Text 4 represents a set of instructions
of how to support a non-binary friend. Thus, there
is considerable inter-annotator variation, however,
overall the consensus is that colon with xier is the
best and the Sylvain system is the worst gender-fair
translation strategy for this group of participants.
5.3 Readability and Comprehensibility
Since in previous interactions with the target group
of this study readability and comprehensibility
were named as important factors for the choice
of gender-fair language, participants were asked
to rate both dimensions for the selected best and
worst strategy. In Table 3 the average score for
the best and worst strategy for both dimensions is
provided, where for the best strategy 5 means high
readability and comprehensibility and 1 means
low readability and comprehensibility. For the
worst strategy, participants were asked whether
they agree that the strategy negatively influences
readability and comprehensibility, which means
18
Best S1 S2 S3 S4
Readability 3.40 3.00 2.95 3.07
Comprehensibility 3.60 3.00 3.00 3.21
Worst S1 S2 S3 S4
Readability 4.55 3.86 0.00 4.00
Comprehensibility 4.55 3.50 0.00 4.00
Table 3: Average score on readability and comprehensibility
that full agreement (5) indicates low readability
and comprehensibility, while 1 indicates a high rat-
ing for both dimensions.
As is to be expected, the strategies that were se-
lected as worst were also rated as low in readability
and comprehensibility, where S1 was on average
indicated as the strategy with the highest impact
on both dimensions. Table 3 confirms the fact that
S3 was never selected as worst strategy by any par-
ticipant. Ratings for the best strategy are more sur-
prising, since even though participants considered
a strategy the comparatively best from the set, they
still indicated an impact on how easy to read and
comprehend the gender-fair text is. On average,
the ratings are rather neutral around 3, with only
slightly worse ratings for the ens strategy (S1).
Figure 1: Detailed Scores of Best Strategy
To provide a closer look at the ratings of the best
strategy, a detailed overview of scores is depicted
in Fig. 1. S3 colon + si:er was selected most fre-
quently as best strategy, which means it obtained
most ratings for readability and comprehensibility.
From the overall 20 ratings for S3 as best strategy,
9 (45%) ratings were very low with 1 or 2. Com-
prehensibility seems to be less of an issue, since
only 6 (30%) ratings were below 3. In fact, as
can be seen from Fig. 1, S3 is the only strategy
to ever obtain a rating as low as 1 for both dimen-
sions. However, it should be kept in mind here that
with a small sample, single participants have an
impact. Participant P6 consequently rated both di-
mensions as very low for a strategy across all texts,
making up 4 of the 20 ratings for S3 as best strat-
egy and of its readability and comprehensibility.
This is in line with the overall evaluation of BWS,
where P6 would never assign a higher score than
0 to any strategy (see Table 1). For S1 the results
mostly rely on P8, who consistently selected it as
the best strategy and considered both dimensions
as high. P1 selected S1 ens only once as best strat-
egy and provided a low rating for readability. An
increase in scores from Text 1 to later texts could
be explained by an increase in familiarity with the
strategy, as commented by one participant. Overall
readability seems to be a bigger issue than compre-
hensibility.
5.4 Participant Comments
Free text comments on the individual texts as well
as on the survey in general reflected the over-
all negative attitude of participants towards the
S2 Sylvain system. Participants remarked that
texts written with this gender-fair language strat-
egy would not be intelligible without the English
original and especially the meaning of pronouns is
hard to understand even in context, requiring an
unnecessary cognitive effort. This comment rein-
forced our research design choice to provide the
English original alongside the translations. One
participant considered simply omitting possessive
pronouns with this strategy as the best option.
Other comments included that it generates texts
that are perceived as grammatically incorrect, un-
natural, and unnecessarily complicated, inhibiting
the natural flow of the text.
In reference to S1, the ens system, participants
mainly remarked on a detrimental effect on com-
prehensibility in their comments, which is in line
with the fact that S1 obtained the worst overall rat-
ings on comprehensibility in the survey. P5 re-
marked that they had to read the text several times
in order to grasp its meaning and for P4 the text
with this strategy seemed as if written in Dutch,
distracting them from understanding it. P8, the
only and most fervent advocate for S1, stated that
to them it is the simplest strategy that is easy to
use, both in written and spoken communication.
Furthermore, to P8 ens imitates the English they,
making it the ideal strategy for gender-fair transla-
tion from English. On the other hand, P8 remarks
that the lack of noun declensions with this strategy
might not be ideal.
19
In reference to the two best rated strategies with
colon after the word stem, participants remarked
that pronouns at times still seem unfamiliar, espe-
cially possessive pronouns, e.g. sieser, might at
first be confused with demonstrative articles, e.g.
dieser. Nevertheless, participants noted that it is
comparatively easy to familiarize themselves with
this strategy, which has little negative impact on
the readability of the text.
One interesting change of comments from the
first to the last set of English original and transla-
tions could be observed with participant P7, who
on Text 1 commented that all of the proposed
strategies have an entirely negative impact on read-
ability. For Text 2, the remark changes to colon
with xier might not be entirely unreadable, which
progresses to relatively easy to read in Text 4. This
change of heart is reflected in P7’s ratings of the
dimension readability, which progresses from 1 for
the best strategy in Text 1 to 3 for the best strategy
in Text 3 and 4. Comprehensibility never obtains a
higher rating than 2. P7 finally concludes that the
colon is less distracting also for pronouns than xier
and its corresponding declensions.
As an overall evaluation of the entire survey,
P7 provides an explicitly negative attitude towards
the topic as such and an explicitly low opinion of
gender-fair language in general, indicating to not
use any such strategy privately and expressing the
belief that a general public can hardly be expected
to utilize such “creations”. This overall belief is
shared by P6, who provides low ratings for the best
strategy as well as its readability and comprehen-
sibility and at the end of the survey remarks that,
while the topic is interesting, none of the proposed
strategies find their liking and will hardly be used
in everyday communication.
All participants but one considered the topic of
gender-fair language strategies in translation inter-
esting and important. Beyond the proposed strate-
gies, the repetition of names instead of pronouns
or rewording of nouns in the translation were in-
dicated. One important aspect that was mentioned
is the familiarity with and prior knowledge of the
topic. Participants indicated that readability and
comprehensibility improved from Text 1 to Text 4,
highlighting how fast they were able to get more
accustomed to these strategies. This factor of be-
ing accustomed and familiar with the individual
strategies might in the end also change the over-
all evaluation, which for now leans towards S3 and
S4 as the strategies closest to the current German
language use.
6 Discussion
One initial assumption of this survey was that
the gender-inclusive Sylvain system might be pre-
ferred on the basis that it follows the grammatical
rules of German and thus, might seem more nat-
ural than other strategies. However, this strategy
was overwhelmingly rated as the worst in the set,
appearing unnatural, erroneous, confusing, and
overly complex. Its ratings on readability and com-
prehensibility reflect these comments. The overall
correspondence between BWS ratings and Likert
scores indicates a tight link between personal pref-
erences for gender-fair translation strategies and
their subjective readability and comprehensibility,
emphasizing the importance of the two dimensions
chosen for this study. However, in future research
these dimensions should take the specific needs
of people with physical and/or cognitive disabili-
ties into consideration, e.g. by conducting a survey
with a more diverse group of participants.
The gender-neutral ens system is comparatively
easy to use from a grammatical point of view, since
it requires no declensions. However, this gram-
matical simplicity considerably alters the language
with a detrimental effect on readability and com-
prehensibility, as shown by the overall ratings and
participants’ comments. Even its only advocate in
the survey doubted the general applicability of a
language system without declensions in German.
In the set of proposed strategies the colon af-
ter the word stem emerged as the clear winner,
with a slight preference for its use with the neo-
pronoun xier over introducing another colon in the
pronouns as in si:er. Since these two strategies, S3
and S4, are the ones closest to the current language
use, it can be assumed that familiarity with strate-
gies plays a role in the selection of preferences,
which was reflected in a participants’ comment.
Thus, it would be interesting to evaluate whether
a thorough introduction including exercises to the
other strategies would alter the final selection of
preferred strategies. In fact, in a previously con-
ducted workshop (Burtscher et al., 2022), partic-
ipants obtained such a thorough introduction and
then in exercises opted for the ens strategy (S1).
One participant in this case study even remarked
on the fact that familiarity, readability, and com-
prehensibility already increased from the first to
20
the last text, where each of which was very short.
This indicates that familiarizing participants with
different strategies might be feasible in a large-
scale survey or experimental setting and would
be an interesting alternation for future studies on
gender-fair translation.
Due to the multiplicity of proposed gender-fair
language strategies in German, we opted for a
small selection in this survey in order not to over-
whelm participants. This selection was driven
by the intention to compare gender-neutral and
gender-inclusive strategies. However, for the for-
mer we only included one neosystem, where other
strategies, such as rewording, are available. For
the latter category, three strategies were included,
where even for the best strategy a change in ty-
pographical character might already impact the re-
sponses, i.e., underscore or star instead of colon as
well as placement of the character. The colon after
the word stem was explicitly chosen to reduce the
emphasis on male/female endings.
Since we only included one gender-neutral op-
tion, a subselection of gender-inclusive strategies
and a sample limited in size, no conclusions on
the preference of either category can be drawn,
for which future studies with a different setup
are foreseen. However, the importance to be
equipped with gender-fair translation strategy that
finds acceptance by the general public was em-
phasized. Thereby, common translation problems,
such as involuntary or accidental misgendering in
the translation, could be mitigated or solved. For
instance, if the source text is rather vague on the
gender of a character/person in a notional gender
language, some gender-fair translation strategies
enable equal vagueness in a grammatical gender
language. While the best strategy ultimately de-
pends on the context not only in the textual sense
but also in the sense of the translation assignment,
target group, purpose, etc., some strategies might
be easier to use, comprehend, and read than others
and might impact the transfer from the source to
the target text differently.
In terms of implications for translation technolo-
gies and in particular machine translation, we be-
lieve that this survey reveals how complex and
language-specific the topic of gender-fair language
and translation truly is. While overall preferred
strategies could be identified, individual partici-
pants showed different preferences, e.g. one par-
ticipant clearly preferred the ens strategy. Thus,
machine translation might need to be able to ac-
commodate different gender-fair language strate-
gies depending on the language, context, purpose,
and target audience of a translation. These prefer-
ences or requirements might also change with the
domain of texts, where in this case study the de-
gree of domain-specificity of media texts is rather
low. In this case study, the task was also to se-
lect from a set of existing translations. It would be
interesting to evaluate the performance and pref-
erences of professional translators when asked to
perform gender-fair post-editing of machine trans-
lated texts.
7 Conclusion
In order to socially and linguistically include dif-
ferent gender identities, a multiplicity of gender-
fair language strategies, in particular for grammat-
ical gender languages, has been proposed. The
transfer of gender-fair strategies across structurally
different languages is challenging for machines
and humans. Thus, it is interesting to evaluate the
position and preferences of language professionals
on the topic of gender-fair translation from a no-
tional to a grammatical gender language. In the
presented survey results based on best-worst scal-
ing, ten language professionals revealed a prefer-
ence for the gender-inclusive strategy of colon af-
ter the word stem in combination with the neopro-
noun xier over the gender-inclusive Sylvian and
gender-neutral ens neosytem. The alternative of
colon with si:er was rated only slightly lower than
with xier, where participants commented on a pref-
erence for pronouns without typographical charac-
ter that they considered more natural. For both
strategies with colon the overall rating on the di-
mensions of readability and comprehensibility was
neutral to positive, whereas ens was considered to
negatively impact both dimensions the most. A
correspondence between the expression of prefer-
ences and the ratings of readability and compre-
hensibility as well as explicit references to these
two dimensions in free text comments confirmed
their importance within the context of gender-fair
translation strategies.
In the present study, a preference for a gender-
inclusive strategy could be observed, however,
with a limited selection of strategies and a small
number of respondents. To obtain a general prefer-
ence regarding gender-neutral or gender-inclusive
strategies, a large-scale study with a stronger va-
21
riety of gender-fair language strategies across lan-
guages and a larger, potentially more diverse tar-
get group would be required. The results indicate
that preferences might also vary depending on the
participants’ degree of familiarity with individual
strategies, which is a factor worth investigating in
future endeavors. Finally, the impact of the level
of domain specificity and text type would be inter-
esting factors. Nevertheless, with this first study
on gender-fair translation among language profes-
sionals we hope to have provided a methodologi-
cal contribution as well as first results on gender-
fair translation strategies from the perspective of
language professionals, a method that can easily
be transferred to future studies and even evaluat-
ing machine translation results.
References
Apa Style. 2019. Singular “they”.
Attig, Remy. 2022. A call for community-informed
translation: Respecting queer self-determination
across linguistic lines. Translation and Interpreting
Studies.
Baer, Brian James and Klaus Kaindl. 2017. Introduc-
tion: Queer (ing) translation. In Baer, Brian James
and Klaus Kaindl, editors, Queering Translation,
Translating the Queer, pages 1–10. Routledge.
Barker, Meg John and Alex Iantaffi. 2019. Life isn’t
Binary. London, UK: Jessica Kingsley Publishers.
Braun, Friederike, Susanne Oelkers, Karin Rogalski,
Janine Bosak, and Sabine Sczesny. 2007. “aus
gr¨
unden der verst¨
andlichkeit...“: Der einfluss gener-
isch maskuliner und alternativer personenbezeich-
nungen auf die kognitive verarbeitung von texten.
Psychologische Rundschau, 58(3):183–189.
Burtscher, Sabrina, Katta Spiel, Daniel Lukas Klausner,
Manuel Lardelli, and Dagmar Gromann. 2022. “es
geht um respekt, nicht um technologie”: Erkennt-
nisse aus einem interessensgruppen-¨
ubergreifenden
workshop zu genderfairer sprache und sprachtech-
nologie. In Mensch und Computer 2022. ACM.
Corbett, Greville G. 1991. Gender. Cambridge Uni-
versity Press.
De Sylvain, Cabala and Carsten Balzer. 2008. Die
sylvain-konventionen–versuch einer “geschlechterg-
erechten” grammatik-transformation der deutschen
sprache. Liminalis, 2:40–53.
Di Sabato, Bruna and Antonio Perri. 2020. Gram-
matical gender and translation: A cross-linguistic
overview. In von Flotow, Luise and Kamal Hala, ed-
itors, The Routledge Handbook of Translation, Fem-
inism and Gender, pages 363–373. Routledge.
En, Boka, Tobias Humer, Marija Petri ˇ
cevi´
c, Tinou
Ponzer, Claudia Rauch, and Katta Spiel. 2021.
Geschlechtersensible Sprache – Dialog auf Au-
genh¨
ohe.
Heger, Illi Anna. 2013. Version 3.2 : Xier pronomen
ohne geschlecht.
Heger, Illi Anna. 2020. Version 3.3 : Xier pronomen
ohne geschlecht.
Hockett, Charles F. 1958. A Course in Modern Lin-
guistics. New York: Macmillan.
Hornscheidt, Lann and Ja’n Sammla. 2021. Wie
schreibe ich divers? Wie spreche ich genderg-
erecht?: Ein Praxis-Handbuch zu Gender und
Sprache. Insel Hiddensee: w orten & meer.
Hornscheidt, Lann. 2012. feministische w orte: ein
lern-, denk- und handlungsbuch zu sprache und
diskriminierung, gender studies und feministischer
linguistik. Frankfurt: Brandes & Apsel.
Jakobson, Roman. 1959. On linguistic aspects of trans-
lation. In Brower, Reuben Arthur, editor, On Trans-
lation, pages 232–239. Cambridge.
Kiritchenko, Svetlana and Saif M Mohammad. 2017.
Best-worst scaling more reliable than rating scales:
A case study on sentiment intensity annotation.
arXiv preprint arXiv:1712.01765.
Kramer, Elise. 2016. Feminist linguistics and linguis-
tic feminisms. In Lewin, Ellen and Leni M. Silver-
stein, editors, Mapping Feminist Anthropology in the
Twenty-first Century, pages 65–83. Rutgers Univer-
sity Press.
Lardelli, Manuel and Dagmar Gromann. 2023.
Gender-fair (machine) translation. In Proceedings of
the New Trends in Translation and Technology Con-
ference - NeTTT 2022, pages 166–177.
Likert, Rensis. 1932. A technique for the measurement
of attitudes. Archives of psychology.
Lindqvist, Anna, Emma Aurora Renstr¨
om, and
Marie Gustafsson Send´
en. 2019. Reducing a male
bias in language? establishing the efficiency of three
different gender-fair language strategies. Sex Roles,
81:109–117.
L´
opez, ´
Artemis. 2019. T´
u, yo, elle y el lenguaje no
binario. La Linterna del Traductor, 19.
L´
opez, ´
Artemis. 2022. Trans (de) letion: Audio-
visual translations of gender identities for main-
stream audiences. Journal of Language and Sexu-
ality, 11(2):217–239.
Louviere, Jordan J and George G Woodworth. 1990.
Best worst scaling: A model for largest difference
judgments [working paper]. Faculty of Business.
22
Martindale, Marianna J. and Marine Carpuat. 2018.
Fluency over adequacy: A pilot study in measuring
user trust in imperfect MT. CoRR, abs/1802.06041.
McConnell-Ginet, Sally. 2013. Gender and its rela-
tion to sex: The myth of ‘natural’gender. In Corbett,
Greville G, editor, The Expression of Gender, pages
3–38. De Gruyter Mouton.
Misiek, Szymon. 2020. Misgendered in Translation?:
Genderqueerness in Polish Translations of English-
language Television Series. Anglica. An Interna-
tional Journal of English Studies, 29(2):165–185.
Nissen, Uwe Kjær. 2002. Aspects of translating gen-
der. Linguistik online, 11(2):25–37.
Piergentili, Andrea, Dennis Fucci, Beatrice Savoldi,
Luisa Bentivogli, and Matteo Negri. 2023. From
inclusive language to gender-neutral machine trans-
lation. CoRR, abs/2301.10075.
Saunders, Danielle and Bill Byrne. 2020. Re-
ducing gender bias in neural machine translation
as a domain adaptation problem. arXiv preprint
arXiv:2004.04498.
Stahlberg, Dagmar, Friederike Braun, Lisa Irmen, and
Sabine Sczesny. 2007. Representation of the sexes
in language. Social Communication, pages 163–187.
Tomalin, Marcus, Bill Byrne, Shauna Concannon,
Danielle Saunders, and Stefanie Ullmann. 2021.
The practical ethics of bias reduction in machine
translation: Why domain adaptation is better than
data debiasing. Ethics and Information Technology.
Von Flotow, Luise. 1997. Translation and Gender:
Translating in The ’Era of Feminism’. Routledge.
Weatherall, Ann. 2002. Gender, language and dis-
course. Hove: Routledge.
Winterson, Jeanette. 1993. Written on the Body. Ran-
dom House.
ˇ
Sincek, Marijana. 2020. On, ona, ono: Translating
Gender Neutral Pronouns into Croatian. Journal of
the International Symposium of Students of English,
Croatian and Italian Studies, pages 92–112.
23
