Teaching translation in the age of generative AI: New paradigm, new learning?

Teaching ansla ion
in he age o
gene a i e AI
New pa adigm, new lea ning?
Edi ed by
JC Pene
Joss Moo kens
Masa u Yamada
language
science
p ess
T ansla ion and Mul ilingual Na u al
Language P ocessing 25
T ansla ion and Mul ilingual Na u al Language P ocessing
Edi o s: Oli e Czulo (Uni e si ä Leipzig), Sil ia Hansen-Schi a (Johannes Gu enbe g-Uni e si ä
Mainz), Reinha d Rapp (Hochschule Magdebu g-S endal), Ma io Bisiada (Uni e si a Pompeu
Fab a)
In his se ies (see he comple e se ies his o y a h ps://langsci-p ess.o g/ca alog/se ies/ mnlp):
14. Bisiada, Ma io (ed.). Empi ical s udies in ansla ion and discou se.
15. T a&Co G oup (ed.). T ansla ion, in e p e ing, cogni ion: The way ou o he box.
16. Ni zke, Jean & Sil ia Hansen-Schi a. A sho guide o pos -edi ing.
17. Hobe g, Felix. In o ma ionsin eg a ion in meh sp achigen Tex cha s: De Skype T ansla o
im Sp achenpaa Ka alanisch-Deu sch.
18. Kenny, Do o hy (ed.). Machine ansla ion o e e yone: Empowe ing use s in he age o
a i icial in elligence. (*
19. Kajze -Wie zny, Ma a, Ad iano Fe a esi, Ilma i I aska & Sil ia Be na dini. Media ed
discou se a he Eu opean Pa liamen : Empi ical in es iga ions. *)
20. Ma zouk, Shaimaa. Sp achkon olle im Spiegel de Maschinellen Übe se zung:
Un e suchung zu Wechselwi kung ausgewähl e Regeln de Kon ollie en Sp ache mi
e schiedenen Ansä zen de Maschinellen Übe se zung.
21. F i ella, F ancesca Ma ia. Usabili y esea ch o in e p e e -cen ed echnology: The case
s udy o Sma Te p.
22. P andi, Bianca. Compu e -assis ed simul aneous in e p e ing: A cogni i e-expe imen al
s udy on e minology.
23. Kenny, Do o hy (ed.). 机器翻译知识普及: 为人工智能时代的用户赋能.
24. Czulo, Oli e , Ma in Kappus & Felix Hobe g (H sg.). Digi ale T ansla ologie.
25. Pene , JC, Joss Moo kens & Masa u Yamada (eds.). Teaching ansla ion in he age o
gene a i e AI: New pa adigm, new lea ning?
ISSN: 2364-8899
Teaching ansla ion
in he age o
gene a i e AI
New pa adigm, new lea ning?
Edi ed by
JC Pene
Joss Moo kens
Masa u Yamada
language
science
p ess
JC Pene , Joss Moo kens & Masa u Yamada (eds.). 2026. Teaching ansla ion in
he age o gene a i e AI: New pa adigm, new lea ning? (T ansla ion and
Mul ilingual Na u al Language P ocessing 25). Be lin: Language Science P ess.
This i le can be downloaded a :
h p://langsci-p ess.o g/ca alog/book/520
© 2026, he au ho s
Published unde he C ea i e Commons A ibu ion 4.0 Licence (CC BY 4.0):
h p://c ea i ecommons.o g/licenses/by/4.0/
ISBN: 978-3-96110-549-6 (Digi al)
978-3-98554-169-0 (Ha dco e )
ISSN: 2364-8899
DOI: 10.5281/zenodo.17580856
Sou ce code a ailable om www.gi hub.com/langsci/520
E a a: pape hi e.o g/documen s/ emo e? ype=langsci&id=520
Co e and concep o design: Ul ike Ha bo
Typese ing: Sebas ian No dho
P oo eading: Ma hew Ko e
Fon s: Libe inus, A imo, DejaVu Sans Mono, Sou ce Han Se i JA
Typese ing so wa e: XƎL
A
T
EX
Language Science P ess
Scha nwebe s aße 10
10247 Be lin, Ge many
h p://langsci-p ess.o g
[email p o ec ed]
S o age and ca aloguing done by FU Be lin
Con en s
In oduc ion
JC Pene iii
Acknowledgemen s xi
Abb e ia ions xiii
I New pa adigm: New skills & compe ences
1 T ansla ion compe ence in he age o gene a i e AI: Deba es,
dilemmas, di ec ions
Ga y Massey & Mau een Eh ensbe ge -Dow 3
2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o
aining
E ik Angelone 27
3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
Ramon Inglada 49
II New pa adigm: New knowledge
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
Lynne Bowke 67
5 Teaching ansla ion wi h AI: B idging heo y and p ac ice h ough
p omp enginee ing
Masa u Yamada 87
6 Teaching AI e hics o ansla ion s uden s
Joss Moo kens & Gökhan Doğ u 105

JC Pene , Joss Moo kens & Masa u Yamada
III New pa adigm: New eaching app oaches
7 Re-posi ioning he human ansla o as he u u e expe o GenAI
ansla ion in he ansla ion class oom: The esul s o a collabo a i e
s udy
Senem Öne Bulu 125
8 Compu e -assis ed language media ion in eaching human-cen ed
augmen ed ansla ion
Ma ia Zimina-Poi o 147
9 Teaching sub i ling in he imes o gene a i e AI
Da id O ego-Ca mona 167
10 AI in an L2 ansla ion class
Sonia Vandepi e 191
11 Gami ica ion as a pedagogical ins umen in in e p e e aining
Saha O hmani & Ne min Sha man 211
12 Emb acing machine ansla ion in L2 educa ion: B idging heo y and
p ac ice in he AI Age
A sushi Mizumo o 233
Index 249
ii
In oduc ion
JC Pene
Newcas le Uni e si y, Uni ed Kingdom
1 Why his book?
Since OpenAI launched Cha GPT back in No embe 2022, gene a i e a i icial
in elligence (GenAI) has ac ed as a majo dis up o in he ansla ion indus y
and beyond. O cou se, in any gi en indus y dis up o s a e no in insically
good o bad. They me ely ac as game change s, good and bad. Also commonly
associa ed wi h “cha bo s”, GenAI ools ce ainly a e a game change o he
ansla ion indus y and, consequen ly, o ansla o educa ion oo. Using deep
lea ning o ain on as co po a, no only do LLMs ha e he abili y o “gene a e
ex which is o en indis inguishable om ex w i en by a human” (Moo kens
e al. 2025: 188), bu hey can do much mo e han jus ansla e ex s. In his
book, we shall he e o e make a dis inc ion be ween a i icial in elligence (AI)
and GenAI. In ou con ex , AI e e s b oadly o all AI echnologies, including
neu al machine ansla ion (NMT). GenAI, howe e , e e s o machine lea ning
ools ha gene a e media, including cha -based La ge Language Models (LLMs)
like GPT-4. Since his book deals mos ly wi h ex , he e ms GenAI and LLMs
end o appea synonymously.
Unlike eely a ailable machine ansla ion (MT) ools like Google T ansla e o
DeepL, LLMs can also be p omp ed o adap hei ansla ions o ake he sou ce
ex and/o he a ge ex con ex in o conside a ion – o ee (ini ially). In a
ma e o seconds. I is easy o see how GenAI can po en ially con ibu e o he
u he democ a isa ion o ansla ion by making i eely accessible in con ex s
whe e ansla ion would ha e been nei he p ac ical no a o dable un il now.
This dis up o also p o ides ansla o s wi h new esou ces ha can po en ially
help hem no jus o inc ease he speed and quali y o he ansla ion wo k hey
deli e o clien s, bu also wi h he way hey manage hei ansla ion p ojec s o
JC Pene . 2026. In oduc ion. In JC Pene , Joss Moo kens & Masa u Ya-
mada (eds.), Teaching ansla ion in he age o gene a i e AI: New pa adigm,
new lea ning?, iii–ix. Be lin: Language Science P ess. DOI: 10.5281/zenodo.
17641062
JC Pene
go abou inding new clien s. In he 2024 edi ion o he annual ELIS (Eu opean
Language Indus y Su ey) epo , o ins ance, he independen ansla o s who
saw GenAI as a posi i e end explained ha hey le e aged i s possibili ies “as a
ool (e.g. e minology ex ac ion), as a sou ce o edi ing wo k, and as a mo i a o
o clien s o choose human ansla ion due o bad AI expe iences” (ELIS 2024:
24).
O cou se, he use o AI cha bo s o ansla ion wo k also comes wi h some
impo an limi a ions ha need ou a en ion. One o hem is ha hey need as
amoun s o da a o lea n, meaning ha hei abili y o ansla e depends la gely
on he amoun o language da a a ailable o ain on o any gi en language.
E en hough LLMs di e om all o he machine lea ning-based MT sys ems in
ha hey can be ained on monolingual da a ins ead o bilingual co po a, hus
inc easing he amoun o a ailable da a hey can ain on, hey emain mos ly
ained on English da a. Because o his, hei ansla ion pe o mance is likely
o be much less con incing o so-called low- esou ce languages compa ed wi h
high- esou ce languages . In addi ion, AI cha bo s a e ained o iden i y and
eplica e language pa e ns. A na u al co olla y o his is ha hey lack “any o m
o genuine comp ehension o he cogni i e p ocesses ha humans use o unde -
s and language and con ex ” (Moo kens e al. 2025: 189). Finally, when GenAI
lacks he da a i needs o p oduce pa o a ansla ion, i can “hallucina e”, i.e.
make up con en in he a ge language.
These limi a ions no wi hs anding, GenAI has s a ed eshaping wha i means
o wo k as a p o essional ansla o in an indus y ha was al eady la gely
echnology-d i en, and whe e a guably au oma ion has long been used as a
way o inc ease he speed and each o ansla ion while educing i s cos . This
has some imes been o he de imen o wo king condi ions and job sa is ac ion
o human ansla o s (see, o ins ance, Lambe & Walke 2024). This, in ac ,
makes i e en mo e u gen o all o us ansla o educa o s o engage wi h GenAI
so ha we can, in he p ocess, e-examine he ole and agency o human ans-
la o s wi hin he ansla ion p ocess h ough ha lens. In o he wo ds, GenAI
should encou age us o e hink c i ically he alue and alues ha humans b ing
o he ansla ion p ocess. As pa o his, we mus ( e-)in e oga e wha we do on
ou ansla ion p og ammes so ha we empowe he nex gene a ion o p o es-
sional ansla o s o achie e he kind o “human-cen ed augmen ed ansla ion”
ha will bene i bo h indi iduals and socie y (O’B ien 2024: 391).
Admi edly, howe e , o some o us ansla o educa o s he p ospec o en-
gaging wi h GenAI may ini ially ha e el somewha o e whelming. Fi s o all,
he iming may no ha e been ideal. OpenAI’s Cha GPT was eleased sho ly a -
e we came ou blinking om a se ies o pandemic- ela ed lockdowns, du ing
i
In oduc ion
which eaching happened mos ly online. Fo some o us, his mean ha ing o in-
e ac wi h (new) echnologies ha we may ha e el we e no always ailo ed o
ou own needs and/o ha we we en’ always com o able using. As a esul ,
mos o us go o expe ience i s -hand he “ echnos ess” many p o essional
ansla o s expe ience when hey a e asked o use ansla ion ools and ech-
nologies hey a e no en i ely com o able wi h (Pene 2024). This is some hing
we should ake se iously, as echnos ess can “ educ[e] pe o mance and ha [m]
indi idual wellbeing” (Koskinen 2020: 146). Wi h i s pedagogical app oach, his
edi ed olume he e o e comes as an a emp o help alle ia e eelings o s ess
among some o us as ansla o aine s.
Ideally, we should all eel empowe ed o engage wi h GenAI on ou p o-
g ammes, whe e i also ac s as a dis up o . Again, his was e lec ed in he
indings o he la es ELIS epo s. In i s 2024 su ey, uni e si y s a anked
GenAI implemen a ion as he mos widely sha ed challenge, wi h close o 90%
o hem seeing i as an issue. This p omp ed he epo ’s au ho s o commen
ha : “Gene a i e AI and how o implemen i in he uni e si ies’ p og ammes
has aken he challenge cha o uni e si y s a by s o m, opping e en hei
conce ns abou he isibili y o he p o ession and he e e nal lack o ime” (ELIS
2024: 27). I , he ollowing yea , isibili y o he p o ession had eclaimed op
spo in he lis o challenges, i was s ill closely ollowed by GenAI implemen-
a ion, which emained a conce n o o e 80% o uni e si y s a (ELIS 2025:
25). Ye , adap and implemen we mus ! This is because, s ill acco ding o he
epo , in 2025 “[a]c ual MT use by language companies has inc eased […] and
eaches now he magic ma k o 50% o handled p ojec s. AI makes i s en y wi h
an imp essi e 34%” (ELIS 2025: 35). In a simila ein, Sla o , an online po al o
language indus y news and esea ch, ecen ly made he ollowing poin :
‘Ska e o whe e he puck is going, no o whe e i is’ has always been good
ad ice o ice hockey. Bu i is ema kably p escien o he language se -
ices indus y as we p epa e o he sunami o dis up ion b ough abou by
gene a i e a i icial in elligence (AI) and la ge language models, o LLMs.
The speed o de elopmen has been ene ic since he elease o OpenAI’s
Cha GPT in No embe 2022. […] I would be a b a e o ecas e o p edic
how he co po a e landscape will look i e yea s om now—bu ope a ing
om whe e he puck is oday isks asking he w ong ques ions and missing
he boa (Welocalize 2023).
E en hough his message was clea ly in ended o language se ice p o ide s,
i ce ainly holds some ele ance o us educa o s. Wha e e we may hink o eel

Abb e ia ions
Abb e ia ions
4EA Embodied, Embedded,
Enac ed, Ex ended, and
A ec i e
AI a i icial in elligence
ANT ac o -ne wo k heo y
ASR au oma ic speech
ecogni ion
AVT audio isual ansla ion
CAT compu e -aided
ansla ion
CoT chain o hough
(p omp ing)
DDL da a-d i en lea ning
DTM dynamic ansla ion
memo y
EFL English as a o eign
language
EMT Eu opean mas e s in
ansla ion
ESL English as a second
language
FSL ew-sho lea ning
GBL game-based lea ning
GenAI gene a i e AI
GPT gene a i e p e- ained
ans o me
GUMT guided use o MT
HAI human-agen in e ac ion
HCAI human-cen ed AI
HMI human-machine
in e ac ion
ICL in con ex lea ning
IT in o ma ion echnology
L1 i s language
L2 second language
LLM la ge language model
LM language model
LSC language se ice company
LSP language se ice p o ide
MKO mo e knowledgeable o he
MRU me acogni i e esou ce
use
MT machine ansla ion
MTPE machine ansla ion pos
edi ing
NLP na u al language
p ocessing
NMT neu al machine
ansla ion
PE pos edi ing
PSF public se ice in e p e ing
QC quali y con ol
QE quali y es ima ion
RLHF ein o cemen lea ning
using human eedback
SDL sel -di ec ed lea ning
SERF social ecology o
esponsibili y amewo k
SL sou ce language
ST sou ce ex
STE simpli ied echnical
English
TC ansla ion compe ence
T&I ansla ion and
in e p e ing
TL a ge language
TM ansla ion memo y
TMS ansla ion managemen
sys em
TS ansla ion s udies
TT a ge ex
XAI explainable AI
xi
Pa I
New pa adigm: New skills &
compe ences
Chap e 1
T ansla ion compe ence in he age o
gene a i e AI: Deba es, dilemmas,
di ec ions
Ga y Masseya& Mau een Eh ensbe ge -Dowa
aZHAW Zu ich Uni e si y o Applied Sciences, Swi ze land ( e .)
Un il ecen ly, he desc ip ion and modelling o he compe ences and skills needed
o ansla e success ully, and he ways in which hey de elop, ha e seen a s eady
e olu ion and p edic able expansion, la gely o accommoda e echnological ad-
ances and an inc easing awa eness o si ua edness. Howe e , he impac o neu al
MT and, now, gene a i e AI (GenAI) has been unp eceden ed in apidly ans o m-
ing he co e asks o ansla o s. Toge he wi h a p oli e a ion o c ea i e oles in
a di e si ying language indus y, he ola ili y indica es a pa adigm shi ha is
beginning o supplan e en he once s able epi he “ ansla o ”. I also ques ions
he e icacy o cu en desc ip ions o skills and con on s educa o s wi h dilem-
mas o balancing specialisa ion and gene alisa ion, ou inisa ion and adap i i y,
co e and ans e able skills. This chap e conside s ele an aspec s o modelling
compe ences and hei de elopmen and examines ela ed deba es and dilemmas
engaging educa o s and employe s in he cu en and o eseen language-indus y
clima e. I ou lines di ec ions o aining in he age o ansla ing wi h(ou ) GenAI,
p oposing an app oach ha , alongside co e ex ual, in e lingual ansla ion and
digi al skills, combines ans e able skills wi h human-machine/human-agen in-
e ac ion (HMI/HAI) compe ence.
Ga y Massey & Mau een Eh ensbe ge -Dow. 2026. T ansla ion compe ence in he
age o gene a i e AI: Deba es, dilemmas, di ec ions. In JC Pene , Joss Moo kens &
Masa u Yamada (eds.), Teaching ansla ion in he age o gene a i e AI: New pa adigm,
new lea ning?, 3–26. Be lin: Language Science P ess. DOI: 10.5281/zenodo.17641064

Ga y Massey & Mau een Eh ensbe ge -Dow
1 In oduc ion
1.1 Gene al
Wha elemen s cons i u e ansla ion compe ence (TC)1and how hey can be
de eloped ha e ep esen ed pi o al issues in applied T ansla ion S udies since
i s incep ion (Holmes 1988: 77). They con inue o igge conside able deba e
amongs esea che s, educa o s and p ac i ione s, and hey p esen educa o s
and hei o ganisa ions wi h dilemmas abou how and whe e o a ge esou ces.
In his chap e , we explo e some deba es and dilemmas igge ed by changes o
p o essional and educa ional p ac ices b ough abou by gene a i e a i icial in-
elligence (GenAI). We also ou line po en ial di ec ions o human agen s in he
age o ansla ing wi h and wi hou GenAI. In ou iew, his necessi a es an ap-
p oach combining ans e able skills wi h human-machine/human-agen in e -
ac ion (HMI/HAI) compe ence, alongside “old-school” ansla ion skills. Be o e
p oceeding o he deba es, dilemmas and di ec ions, howe e , we should de ine,
desc ibe and dema ca e wha exac ly we mean by p o essional in e lingual TC
in he age o GenAI.
1.2 De ini ions, desc ip ions and dema ca ion
The wo d in e lingual is used in T ansla ion S udies and he language indus y
o designa e a pa icula ype o language media ion ac i i y, namely ha which
akes place be ween na u al languages and he pa icula cul u es hey ep esen .
Alongside in e p e ing, ansla ion ep esen s a p o o ypical o m o p o essional
in e lingual language media ion. In a language indus y cha ac e ised by apidly
di e si ying job i les and asks (Bond 2018, Sla o 2020: 11–17), he p o o ypical
concep ualisa ion o wha so-called (p o essional) ansla o s do, i.e. ansla e
con en w i en in a sou ce-language (SL) documen in o a a ge -language (TL)
documen , appea s inc easingly ou moded. Language p o essionals adap i ely
engage in a whole ange o se ice p o ision as he lines be ween co e and adja-
cen se ices blu (Angelone 2023, Angelone e al. 2024: 3–5).
The majo d i e behind his s a e o a ai s has been he echnologisa ion o
he indus y as language se ice p o ide s (LSPs) ope a e mo e and mo e down-
s eam and ups eam o p e iously co e ansla ion, localisa ion and in e p e -
1This chap e applies he de ini ions o “skill” and “compe ence” om he la es EMT (2022: 3)
compe ence amewo k. A skill is he “abili y o apply knowledge and use know-how o com-
ple e asks and sol e p oblems”, while compe ence e e s o “ he p o en abili y o use knowl-
edge, social and/o me hodological abili ies, in wo k o s udy si ua ions and in p o essional
and pe sonal de elopmen ”.
4
1 T ansla ion compe ence in he age o gene a i e AI
ing ac i i ies. Indeed, he indus y i sel p e e s he gene ic e m linguis , and
wi h good eason. Since 2016, he highe accu acy and luency o neu al machine
ansla ion (NMT) engines ha e led o mo e ex ensi e au oma ion o he ansla-
ion p ocess. Linguis s ha e hus seen mo e deploymen downs eam o language
ans e (o con e sion) pe se in pos -edi ing (PE), ou pu e alua ion and quali y
assu ance asks. Mos ecen ly, he in oduc ion o au oma ed MT quali y es ima-
ion (QE) o de e mine how much human PE is equi ed is i sel eplacing mo e
cos ly human MT quali y e alua ion (Sla o 2024: 50–51). Ups eam o ans e
p ope , linguis s ha e been expanding hei epe oi e o pe o m asks ha in-
clude linguis ic consul ancy, MT p e-edi ing and mul ilingual con en c ea ion,
o example in ma ke ing, PR and co po a e communica ions (o en e e ed o
as ansc ea ion). Ne e heless, we shall e ain he e m ansla ion, bo h in line
wi h he i le o he p esen olume and in ecogni ion o he con incing a gu-
men made by do Ca mo & Moo kens (2021) ha he ul ima e e hos o ansla ion
is o help communica ion low be ween humans, ega dless o he new echnolo-
gies deployed and o he oles and ac i i ies hese necessi a e.
GenAI has now en e ed he echnological mix, and i s applica ions clea ly go
a beyond in e lingual media ion. Oppo uni ies and isks ha e been iden i ied
in highe educa ion in gene al (A las 2023, Gimpel e al. 2023, OECD 2023), whe e
he po en ial o GenAI as an educa ional ool ex ends o eache s and s uden s
alike. P ope ly used, i can suppo cou se design, ma e ials de elopmen , as-
sessmen , ex and image p oduc ion, coding, c i ical hinking and indi idualised
lea ning. Imp ope ly used, i can acili a e academic dishones y, inc ease echno-
logical dependency, a ophy human skills and agency, unde mine da a p o ec-
ion and in ellec ual p ope y igh s, ein o ce disc imina ion (due o da a bias)
and inc ease inequali y (due o unequal access).
Wi hin he na owe con ines o in e lingual ansla ion, GenAI di e s de-
cisi ely om AI-based language echnology such as NMT in being capable o
gene a ing mul ilingual ex , images and o he con en in esponse o p omp s,
including he p omp o ansla e. These p omp s elici s a is ically p obable ou -
pu gene a ed by deep-lea ning models able o p ocess and gene a e na u al lan-
guages, known as la ge language models (LLMs). The p omp - esponse cycle
can be i e a i e, allowing use s o ask o mo e in o ma ion o o p o ide mo e
de ailed p omp s should he esul s equi e imp o emen (Gimpel e al. 2023:
22). Adequa ely s uc u ing he ins uc ions ha p omp a GenAI model – also
known as p omp design o , a he mo e echnical le el, p omp enginee ing –
hus becomes cen al o i s e ec i e use. The skills and compe ences (see Foo -
no e 1) necessa y o emb ace GenAI in he educa ion and p ac ice o p o essional
5
Ga y Massey & Mau een Eh ensbe ge -Dow
ansla o s hus di e in majo espec s om hose needed o mas e CAT ools
and o he ansla ion echnologies.
I is indispu able ha such skills and compe ences a e essen ial o a language
indus y al eady ansi ioning om NMT o GenAI. Unbabel co- ounde and
CEO João G aça poin s ou ha LLMs a e a ac ing mo e esea ch and de el-
opmen , and can handle mo e complex asks han NMT, such as au oma ic PE,
sou ce co ec ion and cul u al adap a ion.2Lionb idge, one o he wo ld’s lead-
ing LSPs, claims GenAI is “ aking au oma ed ansla ion and localisa ion o new
heigh s”.3Sla o (2024: 52) epo s ha ansla ion managemen sys ems (TMS)
p o ide s a e al eady in eg a ing GPT models o le e age LLMs’ capaci y o
eph ase, summa ise and sugges new ansla ions. LLMs can be used o inco po-
a e QE in o wo k lows and gauge PE e o , do PE o e en p o ide linguis ic in-
sigh s h oughou he ansla ion wo k low. Sla o ’s esul s also show wo- hi ds
o MT p o ide s o e ing ine- uned LLMs and o e 80% o e ing MT-LLM hy-
b id solu ions. LSPs ha e he po en ial o become majo p o ide s and enable s
o mul ilingual, mul imodal GenAI con en h ough con en c ea ion, alida ion,
localisa ion, ansmission and managemen . In keeping wi h he b oadening po -
olios o LSPs, mul ilingual ex gene a ion is cu en ly “ he mos in-demand lan-
guage AI applica ion, a e machine ansla ion” (Sla o 2024: 74). Fine- uning
and p omp ing LLMs o c ea e and ansla e/localise con en a e eme ging ac i -
i ies o language p o essionals o complemen longe -es ablished PE and quali y
assu ance oles.
Compe en – o ideally expe – le e age o GenAI is a mus o all hose
aiming o wo k in he language indus y. Bu which skills and compe ences does
and will his equi e? And in a dawning age o (language) wo k inc easingly
domina ed by AI, wha alue can human in elligence and skills b ing o bea ? I
is ime o conside he deba es a ound p o essional in e lingual ansla ion and
he dilemmas ha educa o s a e acing.
2 Deba es and dilemmas
2.1 Agency
Assche (2023) aised he issue o whe he MT can ac ually be ega ded as a
p ope poin o in e es in T ansla ion S udies om a de ini ional pe spec i e.
He concludes ha i is indeed compa ible wi h bo h p esc ip i e, equi alence-
o ien ed de ini ions o wha ansla ion is as well as desc ip i e de ini ions based
2Sla o pod #216, 12 July 2024 (h ps://www.you ube.com/wa ch? =6smTEp3CXwQ)
3h ps://www.lionb idge.com/gene a i e-ai/
6
1 T ansla ion compe ence in he age o gene a i e AI
la gely on how ansla ions a e ecei ed. Bu he also poin s ou ha he pe cei ed
h ea o AI o he social and p o essional s a us o ansla o s, and o hose
who educa e hem (see also ELIS 2024: 24–25), may well ed aw de ini ional
bounda ies and dis inguish be ween MT and human ansla ion on he basis o
he pe cei ed c ea i e and mo al au hen ici y o conscious agency (Assche 2023:
14–16).
Assche ’s a gumen d aws on cu en posi ions on AI om he b oade hu-
mani ies and social sciences, such as mo al philosophy (e.g. Sebas ián & Rudy-
Hille 2021). Howe e , he e is e idence o simila p eoccupa ions in educa ion
policy and science, wi h emphasis alling on he impo ance o human agency,
including eache agency, and associa ed ans e able skills – esponsible ac ion,
c i ical hinking, sys ems hinking, logical easoning, cul u al agili y, p oblem
sol ing and emo ion egula ion (Gimpel e al. 2023, OECD 2023).
Wi hin he na owe con ines o T ansla ion S udies, i appea s ha agency
is al eady eme ging as he ulc um a ound which he alue o p o essional hu-
man ansla ion can be measu ed. Fo example, Viei a (2020: 327–329) p esen s
PE as a spec um anging om an MT-cen ed p ocess ypi ied by au oma ic PE
o human-cen ed PE whe e humans ha e ull con ol o ou pu . Human agency
mo es back and o h along he spec um as clien , commission and employe de-
mands equi e. The human-cen ed pole is mo e pala able o p o essional ansla-
o s (Viei a 2020: 327), whose esis ance o PE seems o de i e om anxie y o e
a pe cei ed loss o agency (Cadwell e al. 2018, Sakamo o 2019). In he discou se
su ounding PE, agency has hus become a ouchs one o p o essionalism and
p o essional sel -concep .
Simila ly, Rico & González Pas o (2022: 188) ound in hei p e-GenAI s udy
o a i udes o eaching MT ha he ansla ion educa o s made much o he “hu-
man ac o ”. Though he da a a e unclea on p ecisely whe e hey belie e ha
agency esides, he pa icipan s place he ansla o a he cen e o ansla ion
p oduc ion. In e es ingly, his p omp s he esea che s o posi ha “MT has o e -
come i s condi ion o ool”, and o claim ha eaching MT should adop a holis ic
app oach “beyond an ins umen alis agenda ha concen a es on he echnical
p ope ies o he echnologies” and which “e idences how he human ac o is
key o he ansla ion p ocess” (Rico & González Pas o 2022: 190, 193). Re e ing
o o he esea ch, hey also en a i ely conside whe he complemen a y e ec s
o cogni i e impai men o augmen a ion could in luence compe ence de elop-
men (Rico & González Pas o 2022: 192). One wide implica ion is ha , e en
be o e he ad en o GenAI, he pa adigma ic human agency o he ool-use
is being sub e ed by a mo e disce ning and sub le app ecia ion o he apidly
e ol ing in e ac i i y be ween ansla ion echnologies and hei use s.
7
Ga y Massey & Mau een Eh ensbe ge -Dow
o e ed o he ansla o ’s app o al o e ision. Some imes e en pe ec ma ches
a e no app op ia e o he pa icula TT being p oduced (because o e minology,
egis e , clien s yle guidelines, e c.), so he ansla o has he choice o edi ing
hem o dele ing hem o ansla e om sc a ch. Many CAT ools now also ha e
he op ion o accessing MT sugges ions o segmen s o pa s o segmen s ha
do no ha e good ma ches in he TM, ei he au oma ically o a he command o
he ansla o .10
Usually much less yping is in ol ed han in ansla ing om sc a ch, bu o h-
e wise using a CAT ool o p oduce a TT equi es a e y simila se o skills.
Howe e , he e a e addi ional challenges in using a CAT ool o p oduce a TT.
One is he ocus on ex segmen s (e.g. sen ences o pa s o sen ences), which
can make i mo e di icul o he ansla o o e ain a sense o he whole ex .
Al hough CAT ools can p o ide cogni i e elie by making i easie o be con-
sis en , hey can also inc ease he load by making i ha de o he ansla o o
main ain cohesion and cohe ence (K üge 2016). Ano he challenge ela es o he
phenomenon o p iming, in which “cogni i e acili a ion […] is igge ed by lin-
guis ic o e lap be ween ea lie and cu en p ocessing” (Vandepi e e al. 2018:
362). Fo example, sugges ions ha a e no sui able can se e as inspi a ions o
ansla o s o p oduce e sions ha a e (Fa ell 2023). Howe e , he shee p es-
ence o a TM ma ch o MT sugges ion can also block cogni i e p ocessing and
o ce he ansla o o exe ex a e o o mi iga e he e ec s o p iming (Kolb
2024).
3.1.3 Using MT o GenAI o i s d a s
A ansla ion p ocess ha includes p oducing a d a wi h an MT engine o GenAI
is di e en om ansla ing wi h a CAT ool o om sc a ch in ce ain ways, ye
qui e simila in o he s. The simila i ies lie in e alua ing he ansla ion solu ions
h ough a ca e ul bilingual e iew and adap ing he TT o be mo e app op ia e.
In all ypes o ansla ion p ocesses, ho oughly unde s anding he ST and ecog-
nising e o s, mis ep esen a ions and unmo i a ed omissions o addi ions in he
d a a e jus as impo an as being able o edi he la e o p oduce a luen ,
cohesi e and cohe en TT.
T ansla ion p ocesses ha inco po a e GenAI shi he ocus om w i ing o
eading, p omp ing, e alua ing and edi ing. Using a GenAI sys em wi h a simple
p omp o ansla e he ST can be expec ed o p oduce a d a wi h ea u es ypi-
cal o MT ou pu , he quali y o which will depend on he pa icula engine being
10See Kappus (2024) o an o e iew o he de elopmen and con e gence o ansla ion ech-
nology.
14

1 T ansla ion compe ence in he age o gene a i e AI
accessed. The p ocess o imp o ing such ou pu h ough PE di e s in many e-
spec s om edi ing o e ision (do Ca mo & Moo kens 2021). As well as a ca e ul
monolingual eading o he ou pu , PE should always in ol e a bilingual e iew
o elimina e any bias and inapp op ia e addi ions, omissions o “hallucina ions”
andomly in oduced by he engine (see Dale e al. 2023).
Wo king wi h GenAI in a mo e in o med way necessi a es a p omp ing p ocess
ha p o ides de ailed in o ma ion abou he pu pose o he ansla ion, clien
guidelines, a ge gen e and audience. An i e a i e p ocess o p omp ing, e al-
ua ion and ep omp ing o ob ain sui able ou pu could p ecede a p omp ing
sequence o elimina e any p oblems such as e minological inconsis encies and
unnecessa y epe i ion. Simila issues o hose men ioned abo e o MT ou pu
(i.e. bias, inapp op ia e addi ions and omissions, hallucina ions) also need o be
ec i ied.
3.2 Desc ibing ansla ion skills in he GenAI e a
I has been con incingly a gued elsewhe e (e.g. Ni zke & Hansen-Schi a 2021,
O’B ien 2021) ha using MT o p oduce high-quali y TTs equi es a special se
o skills, and hose a gumen s also hold when GenAI is simply used o p oduce
a ansla ion o he ST. Howe e , in o med use o GenAI includes he awa eness
ha he p e-d a ing and d a ing phase o c ea ing a TT equi es well-conside ed
p omp ing, and pos -d a ing demands ca e ul edi ing. In he ollowing, we con-
side ex ual and digi al skills ha should be o eg ounded when aining s u-
den s o wo k wi h GenAI o p oduce TTs as well as he in e lingual compe ence
ha needs o be de eloped o his.
3.2.1 Tex ual skills
As La ebe (2023: 43) has epo ed, p o essional ansla o s should be able o “ac-
qui e subjec -ma e knowledge quickly […] unde s and complex opics, igu e
ou obscu e meaning, app ecia e he au ho s’ in en ions and he eade s’ needs”.
This high le el o li e acy sugges s ha a solid ounda ion in ex analysis o
ansla ion (e.g. No d 2005) is needed as well as he abili y o ecognise one’s
own gaps in knowledge. Ra he han hoping ha an MT engine o GenAI will
p oduce ou pu ha is easie o unde s and, s uden s should be ained o iden i y
di icul ST passages and o do he esea ch needed o make sense o hem. Gim-
pel e al. (2023: 37) poin ou ha “[l]ea ne s mus possess adequa e knowledge
o he subjec unde sc u iny o achie e sa is ac o y ou comes”. In he con ex
o ansla ion, his means ha s uden s mus ha e he skill o quickly sea ch o
15
Ga y Massey & Mau een Eh ensbe ge -Dow
esou ces (e.g. e e ence wo ds, pa allel ex s, ideos, images) o help hem un-
de s and he ST. They also need o ha e e icien , pu pose ul eading skills o
e alua e he use ulness o he espec i e esou ce o hei own comp ehension
and i s po en ial as a GenAI p omp .
Since GenAI can p oduce luen , g amma ically co ec ex s, he os e ing o
ex p oduc ion skills should ocus on e ising, edi ing and p oo eading (see
also Koponen e al. 2021). This ep esen s a signi ican shi om he eaching o
o eign languages and basic w i ing skills wo k ha is s ill done in many en y-
le el deg ee p og ammes, especially o ansla ion in o he L2 (Ce ezo He e o
e al. 2021) and consis en wi h ecen p oposals o cu iculum de elopmen (e.g.
Sawye e al. 2019).
3.2.2 Technological and digi al skills
The echnology compe ence desc ibed in he EMT (2022: 9) amewo k po en-
ially co e s GenAI in s a ing ha “s uden s know how o use he mos ele-
an IT applica ions […] and adap apidly o new ools and IT esou ces”. The
need o a basic unde s anding o MT and da a li e acy a e also e e ed o. How-
e e , GenAI only became widely a ailable o ansla ion a e he la es i e a-
ion o he EMT amewo k appea ed, so e en he mos ecen desc ip ions se-
iously unde speci y wha ype o knowledge is needed and which skills should
be os e ed o achie e compe ence when wo king wi h his ool. I is only since
hen ha mo e de ailed amewo ks speci ically a ge ing p o essional MT li e -
acy, ansla ion-o ien ed da a li e acy and AI li e acy ha e begun o appea (e.g.
K üge 2024, K üge & Hackenbuchne 2024), building on p io gene ic MT and
da a li e acy amewo ks (e.g. Bowke & Bui ago Ci o 2019). Especially ele an
o he p esen con ex is he AI li e acy amewo k o ansla ion, in e p e ing,
and specialised communica ion p oposed by K üge (2024). I con ains a ious
elemen s ela ed o he unc ions and use o LLMs, some o which con e ge wi h
he skills ha we lis in he inal sec ion.
Indeed, schola s ou side T ansla ion S udies ha e al eady s a ed a guing ha
all s uden s need o acqui e speci ic GenAI li e acy. Fo example, P e o ius (2023)
poin s ou isks and challenges such as inaccu a e esul s, unequal access o
he echnology, bias p opaga ion and e hical conce ns abou sensi i e da a. We
would add ha s uden s need o unde s and he consequences o he da a sou ces
used o he LLMs and he ime lags associa ed wi h he la e ’s knowledge up-
da es, which can be in he o de o mon hs i no yea s.
Pe haps he mos no iceable change in echnological and digi al skills in
an age o GenAI conce ns he c i ical hinking equi ed o o mula e e ec i e
16
1 T ansla ion compe ence in he age o gene a i e AI
p omp s, e alua e in e media e e sions and adjus subsequen p omp s acco d-
ingly. This also includes he abili y o ecognise he diminishing e u ns o
con inued p omp ing and he need o use one’s own ex ual skills o polish he
TT. Wo king wi h GenAI is di e en om simply using language echnology
in ha he in e media e e sions can be conside ed a ype o p iming o mu-
ual p omp ing be ween he conscious s uden / ansla o and he non-conscious
AI agen . Only con inuous c i ical e lec ion can coun e ac po en ially o e -
us ing GenAI ou pu du ing he ansla ion p ocess. Technological and digi al
ansla ion skills in an age o GenAI migh he e o e be mo e b oadly concep u-
alised as key componen s o an HMI/HAI compe ence capable o accoun ing o
he pa icula ecip oci ies o human and AI agency.
3.2.3 In e lingual compe ence
P e equisi es o p o essional ansla ion ha e always included high eading p o-
iciency in he SL as well as amilia i y wi h he sou ce cul u e and domain. No
ma e which aids a e being used, he ansla o also needs high p o iciency in
he TL – bo h eading and e ision skills – as well as in-dep h knowledge o he
domain and a ge cul u e. In e lingual ans e migh be called o less o en on
he pa o he p o essional ansla o deploying MT o GenAI, al hough hey a e
always engaged in in e lingual ac ion.
As La ebe (2023: 43) ound in he su ey o ins i u ional ansla o s, p o es-
sionals a e expec ed o “achie e high le els o accu acy in hei ansla ions, con-
eying no only nuances bu also in ended e ec […] d a well in hei TL, com-
pensa ing o poo wo ding in he o iginal when app op ia e while adhe ing o
in-house con en ions”. Jus as ac ua ies a ain high deg ees o nume acy du ing
hei aining bu a ely do calcula ions by hand in hei jobs, s uden s should be
gi en ample oppo uni y o ansla e wi h and wi hou aids. The esul ing in e -
lingual compe ence in combina ion wi h HMI/HAI compe ence will con ibu e
o hei ansla ion li e acy (see also Massey 2021), e ec i ely p epa ing hem o
wo k in he language indus y in an age o GenAI.
3.3 Eme ging oles in he language indus y
T ansla ion aining ins i u ions a e assumed o p epa e s uden s o he lan-
guage indus y, ye up o wo- hi ds o g adua es may no ac ually wo k as ans-
la o s (Hao & Pym 2023: 223). Howe e , acco ding o he same au ho s, many
o hem wo k in language- ela ed p o essions. This sugges s ha p og ammes
would do well o also p epa e hei s uden s o jobs wi hou “ ansla o ” in he
17
Ga y Massey & Mau een Eh ensbe ge -Dow
i le. E en when ansla o s a e explici ly being ec ui ed, hough, a b oad ange
o skills seems o be expec ed. In hei analysis o job no ices om 2005–2020,
P ie o Ramos & Guzmán (2023: 53–55) ound ha he du ies o ansla o s a
sup ana ional and in e go e nmen al o ganisa ions also included (in o de o
o e all a e age men ion): assis ance wi h o he asks, e minology wo k, e i-
sion, CAT- ool managemen and edi ing.
On he opic o how au oma ion is changing he ansla ion p o ession, Pym
& To es-Simón (2021) discuss a ious ecommenda ions o ansla o s o ocus
on wha machines canno (ye ) do, such as language se ice ad ice, se ice p o-
ision, language consul ing and high-s akes communica ion. This is consis en
wi h commen s om indus y obse e s ha ex p oduc ion wi h GenAI is less
sui able o “con en ha has a egula o y o echnical pu pose” (Sla o 2024:
75). The same epo also p edic s inc eased demand o mul ilingual expe s o
e ise co po a e-gene a ed ou pu o mee expec ed quali y le els (Sla o 2024:
63).
LSPs al eady o e a wide ange o AI- ela ed se ices (Sla o 2024: 101), many
o which equi e linguis ic expe ise ha may s ill be di icul o ind (c . Faes
e al. 2024). The la es Eu opean language indus y su ey (ELIS 2025: 34) shows
he only inc ease in echnology implemen a ion o e 2024 o be in GenAI. This
ep esen s an oppo uni y o bo h aining ins i u ions and hei g adua es and,
encou agingly, 64% o he s uden s pa icipa ing in he same su ey epo ed
using GenAI in hei aining, 19% egula ly (ELIS 2025: 37).
This aises one las impo an ques ion abou hose who ain ansla ion s u-
den s. The eache s hemsel es pa en ly equi e he (Gen)AI li e acy needed o
unde s and AI echniques, c i ically assess AI p oduc ions and ecommenda ions,
and use AI c ea i ely in hei eaching (OECD 2023: 401).11 And ins i u ions need
o ha e he s a de elopmen p ocedu es in place ha empowe hei eache s
o do so.
4 Final ake-aways
Whe he employed as ansla o s, linguis s, ansc ea o s, localise s, consul an s
o language p o essionals, ou g adua es ha e an impo an ole o play in he
language indus y despi e o because o he in oduc ion o GenAI. Bu hey will
need app op ia e skills o do so. And hose skills will, abo e all, ha e o accom-
moda e he dynamic in e ac ions o human and AI agency. Though deg ees o
domain specialisa ion a e open o deba e and will need o be de e mined by local
11Ideas o cou se design, ma e ials and eaching scena ios ha align well wi h he skills and
compe ences ou lined he e can be ound in Pym & Hao (2025).
18
1 T ansla ion compe ence in he age o gene a i e AI
condi ions, cu en TC models al eady adequa ely co e he old-school ansla-
ion and ex ual skills ha indus y demands. Bu hei ans e able and ech-
nological and digi al skills a e unde speci ied, and he in oduc ion o GenAI
calls o g ea e p ecision. K üge ’s (2024) AI li e acy amewo k is a esponse,
ou lining he digi al compe ences needed o ha ness AI e icien ly, e hically and
sus ainably. I co e s echnical ounda ions (ope a ing p inciples, aining, e c.),
assessing AI’s use ulness o (domain-)speci ic asks, in e ac ing wi h (Gen)AI,
implemen ing AI in wo k lows, and unde s anding e hical and socie al aspec s.
Pa ially o e lapping wi h elemen s o K üge ’s amewo k, ou own b ie ecom-
menda ions ep esen a se o conc e e p io i ies o emb acing GenAI in ans-
la o educa ion.
We ecommend ha collabo a i e expe ien ial lea ning should always in e-
g a e decisions abou whe he and how o deploy GenAI, oge he wi h c i ical
e alua ions o he esul an p ocesses and p oduc s. Ou comes should ocus on
de eloping speci ic ans e able skills:
• c i ical hinking
• adap i i y
• c ea i e p oblem-sol ing
• cogni i e and emo ional sel - egula ion
• sel -e icacy
• accoun abili y and a sense o e hics
• collabo a i e abili y in human and AI in e ac ions.
The ex ension o he las poin aims o de elop wha we ha e e med HMI/
HAI compe ence, which includes:
• designing e ec i e p omp s ( ansla ion pu pose, clien speci ica ions, a -
ge gen e, audience, e c.), e alua ing GenAI esponses wi h supplemen a y
esea ch (whe e app op ia e) and ep omp ing
• ecognising p iming e ec s and mi iga ing nega i e ones
• iden i ying and elimina ing bias, addi ions, omissions, hallucina ions
• unde s anding LLM da a sou ces and knowledge ime-lags.
None o he abo e should be ega ded as op ional add-ons bu mus ecei e
a leas equal weigh o so-called co e skills. Only hen will ou g adua es be
p ope ly equipped o he GenAI age, wha e e oles hey ake on.
19

Ga y Massey & Mau een Eh ensbe ge -Dow
Re e ences
Angelone, E ik. 2023. Wea ing adap i e expe ise in o ansla o aining. In
Ga y Massey, Elsa Hue as-Ba os & Da id Ka an (eds.), The human ansla o
in he 2020s, 60–73. London, UK: Rou ledge.
Angelone, E ik, Ga y Massey & Mau een Eh ensbe ge -Dow. 2024. In oduc-
ion: Con ex ualizing language indus y s udies. In Ga y Massey, Mau een
Eh ensbe ge -Dow & E ik Angelone (eds.), Handbook o he language indus y:
Con ex s, esou ces and p o iles, 1–13. Be lin, Ge many: De G uy e Mou on.
Assche , Om i. 2023. The posi ion o machine ansla ion in ansla ion s udies:
A de ini ional pe spec i e. T ansla ion Spaces 12(1). 1–20. DOI: 10.1075/ s.22035.
ass.
A las, S ephan. 2023. Cha GPT o highe educa ion and p o essional de elopmen :
A guide o con e sa ional AI. Tech. ep. Kings on: Uni e si y o Rhode Island.
h ps://digi alcommons.u i.edu/cba_ acpubs/548.
Bond, Es he . 2018. The s unning a ie y o job i les in he language indus y. h ps:
//sla o .com/ he-s unning- a ie y-o -job- i les-in- he-language-indus y/.
Bowke , Lynne & Jai o Bui ago Ci o. 2019. Machine ansla ion and global e-
sea ch: Towa ds imp o ed machine ansla ion li e acy in he schola ly commu-
ni y. Leeds, UK: Eme ald Publishing. DOI: 10.1108/9781787567214.
B adshaw, Je ey M., Paul J. Fel o ich & Ma hew Johnson. 2011. Human-agen
in e ac ion. In Guy A. Boy (ed.), The handbook o human-machine in e ac ion,
283–300. Boca Ra on, USA: CRC P ess. h ps : / / www . aylo ancis . com /
chap e s/edi /10.1201/9781315557380-14/human%E2%80%93agen -in e ac ion-
je ey-b adshaw-paul- el o ich-ma hew-johnson.
B and zaeg, Pe e Bae, Ma i a Skju e & Asbjø n Føls ad. 2022. My AI iend:
How use s o a social cha bo unde s and hei human–AI iendship. Human
Communica ion Resea ch 48. 404–429. DOI: 10.1093/hc /hqac008.
Cadwell, Pa ick, Sha on O’B ien & Ca los S. C. Teixei a. 2018. Resis ance and
accommoda ion: Fac o s o he (non-) adop ion o machine ansla ion among
p o essional ansla o s. Pe spec i es 26(3). 301–321. DOI: 10.1080/0907676X.
2017.1337210.
Ce ezo He e o, En ique, As id Schmidho e & Meli a Kole nik. 2021. An LSP
amewo k o ansla ion and in e p e ing pedagogy. Re is a de lenguas pa a
ines especi icos 37(2). 140–156. h ps://ojsspdc.ulpgc.es/ojs/index.php/LFE/
a icle/ iew/1406.
Cha u edi, Rkjul, Sanjee Ve ma, Ronnie Das & Yogesh K. Dwi edi. 2023. Social
companionship wi h a i icial in elligence: Recen ends and u u e a enues.
20
1 T ansla ion compe ence in he age o gene a i e AI
Technological Fo ecas ing & Social Change 193. 122634. DOI: 10.1016/j. ech o e.
2023.122634.
Ch is ensen, Tina Paulsen, Ma ian Flanagan & Anne Schjoldage . 2017. Mapping
ansla ion echnology esea ch in T ansla ion S udies. An in oduc ion o he
hema ic sec ion. He mes 56. 7–20. DOI: 10.7146/hjlcb. 0i56.97199.
Dale, Da id, Elena Voi a, Loic Ba aul & Ma a R. Cos a-jussà. 2023. De ec ing
and mi iga ing hallucina ions in machine ansla ion: Model in e nal wo kings
alone do well, sen ence simila i y e en be e . In P oceedings o he 61s Annual
Mee ing o he Associa ion o Compu a ional Linguis ics, ol. 1, 36–50. h ps:
//aclan hology.o g/2023.acl-long.3.pd .
do Ca mo, Félix & Joss Moo kens. 2021. Di e en ia ing edi ing, pos -edi ing and
e ision. In Maa i Koponen, B ian Mossop, Isabelle S. Robe & Gio anna
Scocche a (eds.), T ansla ion e ision and pos -edi ing: Indus y p ac ices and
cogni i e p ocesses, 35–49. London, UK: Rou ledge.
Eapen, Tojin T., Daniel J. Finkens ad , Josh Folk & Lokesh Venka aswamy. 2023.
How gene a i e AI can augmen human c ea i i y. Ha a d Business Re iew
(July-Augus ). h ps://hb .o g/2023/07/how-gene a i e-ai-can-augmen -
human-c ea i i y.
Eh ensbe ge -Dow, Mau een, And ea Hunzike Heeb, Ga y Massey, U sula Mei-
de , Silke Neumann & Heid un Becke . 2016. An in e na ional su ey o he
e gonomics o p o essional ansla ion. ILCEA 27. 1–23. h p://ilcea. e ues.
o g/4004.
ELIS. 2024. ELIS 2024 Eu opean language indus y su ey. Tech. ep. ELIS Re-
sea ch. 1–57. h ps://elis- su ey.o g/wp- con en /uploads/2024/03/ELIS-
2024-Repo .pd .
ELIS. 2025. ELIS 2025 Eu opean language indus y su ey. Tech. ep. ELIS Re-
sea ch. 1–53. h p://elis- su ey.o g/wp- con en /uploads/2025/03/ELIS-
2025_Repo .pd .
EMT. 2017. Eu opean mas e ’s in ansla ion compe ence amewo k 2017. Tech. ep.
B ussels: Eu opean Commission. 1–12.
EMT. 2022. Eu opean mas e ’s in ansla ion compe ence amewo k 2022. Tech.
ep. B ussels: Eu opean Commission. 1–12. h ps://commission.eu opa.eu/
sys em/ iles/2022-11/em _compe ence_ wk_2022_en.pd .
Es andia i, Mohammad Reza, Nas in Shok pou & Fo ough Rahimi. 2019. An
e alua ion o he EMT: Compa ibili y wi h he p o essional ansla o ’s needs.
Cogen A s & Humani ies 6(1: 1601055). 1–17. DOI: 10 . 1080 / 23311983 . 2019 .
1601055.
Faes, Flo ian, Alex Edwa ds & Anna Wyndham. 2024. Sla o Q1 2024 esea ch
b ie ing. online. h ps://sla o .com/e en /sla o -q1-2024- esea ch-b ie ing/.
21
Ga y Massey & Mau een Eh ensbe ge -Dow
Faes, Flo ian & Ga y Massey. 2024. Cha ing he language indus y: In e iew
wi h an indus y obse e . In Handbook o he language indus y, 17–31. Be lin,
Ge many: De G uy e Mou on. DOI: 10.1515/9783110716047-002.
Fa ell, Michael. 2023. Do ansla o s use machine ansla ion and i so, how?
Resul s o a su ey held among p o essional ansla o s. In Joss Moo kens &
Vilelmini Sosoni (eds.), P oceedings o T ansla ing and he Compu e 44, 49–60.
Gene a, Swi ze land: Edi ions T adulex. h ps://www. adulex.com/ a ia/
TC44-luxembou g2022.pd .
Fügene , And eas, Jö n G ahl, Alok Gup a & Wol gang Ke e . 2021. Will humans-
in- he-loop become bo gs? Me i s and pi alls o wo king wi h AI. MIS Qua -
e ly 45(3). 1527–1556. DOI: 10.25300/MISQ/2021/16553.
Gimpel, Henne , Ca oline Ruine , Man ed Schoch, Ma eike Schoop, Luis Läm-
me mann, Nils U bach, K is ina Hall, To s en Eymann, Maximilian Röglinge ,
S e en Vandi k, Alexande Mädche & S e an Decke . 2023. Unlocking he
powe o gene a i e AI models and sys ems such as GPT-4 and Cha GPT o highe
educa ion: A guide o s uden s and lec u e s. Wo king pape 02-2023. Hohen-
heim, Ge many: Uni e si y o Hohenheim. 4–46. h ps://hdl.handle.ne /10419/
270970.
Guing ich, Rose E. & Michael S. A. G aziano. 2024. Cha bo s as social companions:
how people pe cei e consciousness, human likeness, and social heal h bene i s in
machines. h ps://a xi .o g/abs/2311.10599.
Hao, Yu & An hony Pym. 2023. Whe e do ansla ion s uden s go? A s udy o he
employmen and mobili y o mas e g adua es. The In e p e e and T ansla o
T aine 17(2). 211–229. DOI: 10.1080/1750399X.2022.2084595.
Holmes, James S. 1988. T ansla ed! Pape s on li e a y ansla ion and ansla ion
s udies. Raymond an den B oeck (ed.), ol. 7 (App oaches o T ansla ion S ud-
ies). Ams e dam, Ne he lands: Rodopi.
Hu ado Albi , Ampa o (ed.). 2017. Resea ching ansla ion compe ence by PACTE
g oup. Ams e dam, Ne he lands: John Benjamins.
Kappus, Ma in. 2024. Language echnology de elope s. In Ga y Massey, Mau-
een Eh ensbe ge -Dow & E ik Angelone (eds.), Handbook o he language in-
dus y: Con ex s, esou ces and p o iles, 121–141. Be lin, Ge many: De G uy e
Mou on.
Ki aly, Don. 2000. A social cons uc i is app oach o ansla o educa ion: Empow-
e men om heo y o p ac ice. London, UK: Rou ledge.
Ki aly, Don. 2013. Towa ds a iew o ansla o compe ence as an eme gen phe-
nomenon: Thinking ou side he box(es) in ansla o educa ion. In Don Ki aly,
Sil ia Hansen-Schi a & Ka in Maksymski (eds.), New p ospec s and pe spec-
22
1 T ansla ion compe ence in he age o gene a i e AI
i es o educa ing language media o s, 197–224. Tübingen, Ge many: Na Ve -
lag.
Kolb, Wal aud. 2024. ‘I am a bi su p ised’. Li e a y ansla ion and pos -edi ing
p ocesses compa ed. In And ew Ro hwell, Andy Way & Roy Youdale (eds.),
Compu e -assis ed li e a y ansla ion, 53–68. London, UK: Rou ledge.
Koponen, Maa i , B ian Mossop, Isabelle S. Robe & Gio anna Scocche a. 2021.
T ansla ion e ision and pos -edi ing: Indus y p ac ices and cogni i e p ocesses.
London, UK: Rou ledge.
K üge , Ralph. 2016. Con ex ualising compu e -assis ed ansla ion ools and
modelling hei usabili y. T ans-kom 9(1). 114–148. h ps : / / d - nb . in o /
1105491919/34.
K üge , Ralph. 2024. Ou line o an a i icial in elligence li e acy amewo k
o ansla ion, in e p e ing and specialised communica ion. Lublin S udies in
Mode n Languages and Li e a u e 48(3). 11–23. DOI: 10.17951/lsmll.2024.48.3.11-
23.
K üge , Ralph & Janiça Hackenbuchne . 2024. A compe ence ma ix o machine
ansla ion-o ien ed da a li e acy eaching. Ta ge 36(2). 245–275. DOI: 10.1075/
a ge .22127.k u.
La ebe , Anne. 2023. Skills and knowledge equi ed o ansla o s in ins i u ional
se ings. In Tomáš S oboda, Łucja Biel & Vilelmini Sosoni (eds.), Ins i u ional
ansla o aining, 30–48. London, UK: Rou ledge.
La ou , B uno. 2005. Reassembling he social - An in oduc ion o ac o -ne wo k-
heo y. Ox o d, UK: Ox o d Uni e si y P ess.
Massey, Ga y. 2021. T ansla ion li e acy in addi ional language lea ning: Closing
he concep ual di ide be ween ansla ion and language educa ion. Bulle in
suisse de linguis ique appliquée 114. 31–45. DOI: 10.5169/seals-1030136.
Mellinge , Ch is ophe D. 2017. T ansla o s and machine ansla ion: Knowledge
and skills gaps in ansla o pedagogy. The In e p e e and T ansla o T aine
11(4). 280–293. DOI: 10.1080/1750399X.2017.1359760.
Moo kens, Joss & Ana Gue be o A enas. 2024. A i icial in elligence, au oma-
ion and he language indus y. In Ga y Massey, Mau een Eh ensbe ge -Dow
& E ik Angelone (eds.), Handbook o he language indus y: Con ex s, esou ces
and p o iles, 71–97. Be lin, Ge many: De G uy e Mou on.
Nah, Fiona Fui-Hoon, Ruilin Zheng, Keng Siau & Lang ao Chen. 2023. Gene a-
i e AI and Cha GPT: Applica ions, challenges, and AI-human collabo a ion.
Jou nal o In o ma ion Technology Case and Applica ion Resea ch 25(3). 277–304.
DOI: 10.1080/15228053.2023.2233814.
Ni zke, Jean & Sil ia Hansen-Schi a. 2021. A sho guide o pos -edi ing. Be lin,
Ge many: Language Science P ess. DOI: 10.5281/zenodo.5646896.
23
E ik Angelone
Delibe a e p ac ice hinges on he lea ne ’s in insic mo i a ion, ano he co e
condi ion. T ansla o s looking o ad ance along an expe ise ajec o y need o
emb ace wo king ou side o hei com o zones and ha e a i m belie ha de-
libe a e p ac ice will ul ima ely imp o e hei pe o mance. Delibe a e p ac ice
akes immense ime and e o . The o -deba ed 10,000-hou ‘ ule’ o becoming
an ‘expe ’ (Gladwell 2011) is o en ci ed in a emp s o quan i y jus how much
ime and e o a e equi ed. This would amoun o “ wo and a hal yea s o sus-
ained e o ul p ac ice e e y day o i e hou s” (Sh e e 2019: 173–174), aising
ques ions ega ding easibili y o his ‘ ule’ and i s place in he delibe a e p ac ice
model.
Delibe a e p ac ice calls o he lea ne ’s commi men o “conscious pe o -
mance moni o ing” (Ho n & Masunaga 2006: 601) and engaging in ull concen-
a ion, as opposed o “mindless, ou ine pe o mance” (E icsson 2006: 692). This
akes sel -discipline and lea ne dedica ion o honing me acogni i e capaci ies.
S a egic sca olding by he aine , along wi h aine / ainee dedica ion o he
cen al ideas o cogni i e cons uc i ism (Piage 1952), a e ins umen al in bol-
s e ing lea ne me acogni ion and pe o mance moni o ing p ocesses.
Figu e 1 p o ides an o e iew o he co e condi ions o delibe a e p ac ice ou -
lined in his chap e . I does no ep esen an exhaus i e lis o all condi ions
men ioned in he Expe ise S udies li e a u e, bu a he ocuses on hose condi-
ions aken up in T ansla ion S udies o da e.
3 Challenges o he implemen a ion o delibe a e p ac ice
I is wo h no ing he e ha empi ical esea ch on he bene i s o delibe a e p ac-
ice on ansla ion pe o mance is s ill qui e scan . This dea h can a leas be
pa ly explained by a se ies o inhe en challenges associa ed wi h i s implemen-
a ion in o mal aining con ex s. As men ioned in he p e ious sec ion, se e al
p oponen s o delibe a e p ac ice sugges ha i s bene i s a e bes ealised in con-
ex s in ol ing lea ne s wo king one on one wi h indi idual aine s and in he
absence o a p e-se cu iculum (E icsson e al. 1993: 367). This ype o design
does no eadily align wi h he ashion in which ansla o s a e usually ained
o a numbe o di e en easons, s a ing wi h he inancial cons ain o need-
ing o hi e a pe sonal aine .
Ano he cons ain po en ially s anding in he way o mee ing se e al o he
co e condi ions o delibe a e p ac ice is he deg ee o lea ne he e ogenei y com-
monly ound in class oom-based aining con ex s. S uden s o ansla ion o en
ha e widely a ying le els o compe ence and expe ience, no o men ion di-
30

2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
immedia e,
in o ma i e
eedback
e o
co ec ion
oppo uni ies
" aine "
p esence
app op ia e
di icul y
le el
in insic
mo i a ion
conscious
pe o mance
moni o ing
Co e condi ions
o delibe a e
p ac ice
Figu e 1: Co e condi ions o delibe a e p ac ice
e se lea ning needs and in e es s. This can make i di icul o aine s o es-
ablish uly indi idualised lea ning objec i es and co esponding asks o mee
hem. The p oblem is exace ba ed when s uden en olmen s a e high. Feedback,
a guably he mos impo an dimension o delibe a e p ac ice, o en becomes
less de ailed and, ou o necessi y, much less immedia e. Pee eedback and sel -
eedback ac i i ies can help add ess his gap, bu immediacy, as a c i e ion o
p ac ice o be delibe a e, o en emains e y di icul o ob ain.
Beyond ime cons ain s, he delibe a e p ac ice condi ion o in o ma i e, im-
media e eedback migh no be me i he kind o eedback being p o ided is no
ans e able in he sense o applying o u u e ansla ion asks. O en imes, as-
sessmen ub ics a e used o ma k up e o s in acco dance wi h a ious ex ual
le els, such as g amma , wo d choice, and syn ax. I he eedback gi en pe ains
o pa e ns along hese lines ha a e applicable ac oss asks, i could be ega ded
as uly in o ma i e. I , on he o he hand, eedback simply consis s o ma king
up one-o e o s pe aining o i ems ha he ansla o may ne e encoun e
31
E ik Angelone
again in u u e ansla ions, such as an isola ed colloca ion e o , i is ela i ely
shallow and no pa icula ly in o ma i e.
Lea ne he e ogenei y in la ge coho se ings also p esen s challenges when
i comes o mee ing he delibe a e p ac ice condi ion o making su e asks a e be-
ing unde aken a an app op ia e le el o di icul y. T ansla ion p ac ice cou ses
a e o en in o med by aine in ui ion and p edic ion o app op ia e di icul y
le el, in u n based on such ace s as le el o s udy, in ake examina ion esul s,
a ious cogni i e p ocess me ics (Sun & Sh e e 2014), o p e ious s uden pe -
o mance and co esponding ‘ ich poin s’ (PACTE 2011). Howe e , s udies ha e
shown a dange o misalignmen be ween pe cei ed o p edic ed p oblems and
wha ac ually p o es o be p oblema ic when ansla ing (Angelone 2018). Wha
is assumed o be di icul is ac ually no necessa ily so, making a emp s a p e-
dic ing and se ing an app op ia e di icul y le el challenging, pa icula ly ac oss
a wide ange o s uden s wi h di e ging needs. Fu he mo e, he idea o ha ing
s uden s wo k ou side o hei com o zone, o a he pe iphe y o wha hey
can ealis ically accomplish, o he pu pose o ad ancing expe ise may con a-
dic app oaches o ansla o aining ha emb ace he p edic abili y o s aying
wi hin he lea ne ’s com o zone.
Acco ding o delibe a e p ac ice guidelines, he lea ne ’s in insic mo i a ion
needs o be cons an . Ha ing ansla ion s uden s wo k ou side o hei com o
zone in pedagogical con ex s, pa icula ly when g ades a e in ol ed, uns he
isk o hampe ing such mo i a ion. The ou comes o delibe a e p ac ice would
need o be gauged using me ics beyond o mal g ades, wi h a ocus on help-
ing lea ne s become mo e sel - e lec i e ansla o s. In insic mo i a ion would
come no so much om ge ing good g ades o doing well in a gi en cou se, bu
a he om seeing he bene i s o pu ing in ha d wo k and emb acing di icul y
in o de o become a be e ansla o .
4 Gene a i e AI as a acili a o o delibe a e p ac ice
Agains he backd op o he condi ions o delibe a e p ac ice p esen ed in Sec-
ion 2 and he cons ain s po en ially s anding in i s way, as ou lined in Sec ion 3,
he ques ion emains: how can we bes go abou acili a ing delibe a e p ac ice
in ansla o aining o pu poses o expe ise acquisi ion and ad ancemen ?
In pa icula , how can we es ablish he equisi e highly indi idualised, ‘coach’-
o ien ed app oach a he hea o delibe a e p ac ice? The pedagogical ea u es
o gene a i e AI, in p o iding eal- ime eedback and p omp -d i en in e ac ion
in a use -cen ed ashion, would seem o hold po en ial in his ega d, as will be
illus a ed h ough a se ies o conc e e scena ios in Sec ion 6.
32
2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
A he ime o w i ing, we a e s ill wi nessing he dawn o gene a i e AI as a
ehicle o op imising ansla ion, bo h in p o essional and pedagogical con ex s.
I s inc easingly ubiqui ous in eg a ion by LSP companies in p ojec wo k lows
has d awn a en ion o a need o a i icial in elligence (AI) li e acy (K üge 2023),
alongside MT li e acy and da a li e acy in a b oad sense. AI li e acy can be de-
ined as “a se o compe encies ha enables indi iduals o c i ically e alua e AI
echnologies, communica e and collabo a e e ec i ely wi h AI, and use AI as a
ool online, a home, and in he wo kplace” (Long & Mage ko 2020:2). In con-
junc ion wi h discussions o ansla o pe o mance, much o he discou se on
AI in elligence o da e has ocused on how gene a i e AI can be le e aged o a-
cili a e human-in- he-loop ansla ion, wi h an emphasis on he ansla ion p od-
uc . Ou side o se e al o he chap e s in his olume, ela i ely li le discussion
has been dedica ed o he po en ial bene i s o gene a i e AI as a con e sa ional
agen , ocusing less on gene a ing ansla ed con en , and mo e so on enabling
ansla o s o e lec on hei pe o mance and engage in ansla ion asks o a-
cili a e delibe a e p ac ice.
Ex ending on sociocul u al lea ning heo ies (Vygo sky 1965), some ha e come
o ega d a gene a i e AI ool like Cha GPT as a ‘mo e knowledgeable o he ’
(MKO), in essence aking on he ole o a pe sonalised aine ha can “lead
he lea ne om he zone o cu en de elopmen o he zone o p oximal de-
elopmen – he space whe e one canno qui e mas e a con en / ask o hei
own, bu hey can wi h he help o an expe ” (S ojano 2023: 2). As an MKO,
Cha GPT can add ess he a o emen ioned condi ions o delibe a e p ac ice, pa -
icula ly he p esence o a one-on-one pe sonalised aine . Th ough eal- ime
esponses o lea ne p omp s, Cha GPT ensu es he immediacy o eedback ha
is so di icul o ob ain in a la ge en olmen class oom-based ansla o aining
en i onmen . The lea ne ’s in insic mo i a ion is likely o be heigh ened when
aining is pe sonalised and sel -d i en, oo ed in immedia e, in o ma i e eed-
back, and in e ac i e, d i en by hei own p omp s in ela ion o aspec s o hei
own pe o mance.
5 Gene a i e AI as a sca old o sel -di ec ed lea ning
The u ilisa ion o gene a i e AI o pu poses o delibe a e p ac ice calls o he
lea ne o pa ake in sel -di ec ed lea ning (SDL), whe e “indi iduals ake he ini-
ia i e, wi h o wi hou he help o o he s, in diagnosing hei lea ning needs, se -
ing lea ning goals, iden i ying esou ces, choosing app op ia e lea ning s a e-
gies, and e alua ing hei lea ning ou comes” (Knowles 1975: 18). In his case,
33
E ik Angelone
gene a i e AI helps sca old lea ning in line wi h lea ne p omp s. T ansla ion
aine s well- e sed in he condi ions o delibe a e p ac ice can p o ide lea ne s
wi h aluable in o ma ion on he na u e o p omp s hey should en e . Howe e ,
sel -di ec ed lea ne s need o be “able, eady, and willing o p epa e, execu e
and comple e lea ning independen ly” (Jossbe ge e al. 2010: 419). The need o
lea ne independence does no make he aine supe luous, bu a he shi s he
ocus o assignmen s unde aken and how hey a e assessed.
Models o sel -di ec ed lea ning bea e y close esemblance o he delibe a e
p ac ice model. One such model ha is widely ci ed in he li e a u e consis s
o h ee closely in e ela ed dimensions: 1) sel -managemen , 2) sel -moni o ing,
and 3) mo i a ion (Ga ison 1997). Sel -managemen in ol es he lea ne es ab-
lishing conc e e lea ning goals and managing lea ning esou ces o achie e hese
goals. In o he wo ds, hey ake con ol, deciding on he asks in which hey
will engage. F om a delibe a e p ac ice pe spec i e, h ough s a egic p omp s,
he ansla o can le e age gene a i e AI o anno a e e o s in hei ansla ions.
Gene a ed anno a ions could hen se e as a amewo k o he ansla o o sel -
disco e he na u e o he e o s. Gene a i e AI could hen be p omp ed o p o-
ide simila ansla ion asks, wi h he goal o engaging he lea ne in delibe a e
p ac ice cen ed a ound a ce ain e o pa e n (such as a oiding alse cogna es,
e oneous li e al ansla ion, o p oblema ic ansla ionese a a syn ac ic le el).
Beyond e o de ec ion and mi iga ion, he ansla o can also use gene a i e AI
p omp ing o sel -manage he di icul y le el o he asks hey a e unde aking.
Sec ion 6 p o ides mo e conc e e scena ios and desc ip ions along hese lines.
Sel -moni o ing, he second componen o Ga ison’s model, pe ains o he
lea ne ’s me acogni i e p ocesses. As an impo an dimension o sel -di ec ed
lea ning, sel -moni o ing “ equi es lea ne s o ake esponsibili y o cons uc
meanings” (Ga ison 1997: 24). Th ough in e ac i e eedback, gene a i e AI can
shed aluable ligh o help lea ne s iden i y salien ea u es o he ansla ion ask
on which o ocus hei a en ion. Fo example, ansla o s can en e a p omp
asking gene a i e AI o anno a e sou ce con en ha could be an icipa ed o p e-
dic ed o p esen challenges in ansla ion. O e he pas decade, sc een eco ding
has ound a place in p ocess-o ien ed ansla o aining o pu poses o os e ing
sel -moni o ing and o enhance lea ne me acogni ion based on documen a ion
o ansla ion beha io s sugges ing p oblems, including pausing, in o ma ion e-
ie al, and e ision (Angelone 2019). A p esen , gene a i e AI ools like Cha -
GPT do no o e unc ionali y whe e ansla o s can upload sc een eco dings o
hei wo k o pu poses o ecei ing analy ic eedback a a g anula le el. Gi en
ecen ad ancemen s in his echnology howe e , such as APIs ha can p o ide
34
2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
au oma ed ideo summa isa ion, i is qui e likely ha u ilisa ion o gene a i e
AI ools o such pu poses is no oo a away.
The hi d componen o Ga ison’s sel -di ec ed lea ning model, mo i a-
ion, di ec ly aligns wi h mo i a ion as a co e condi ion o delibe a e p ac ice.
Whe eas delibe a e p ac ice ega ds mo i a ion as he lea ne emb acing chal-
lenge h ough an inhe en desi e o become be e , Ga ison d aws a en ion o
he impo ance o mo i a ion o pu poses o s aying on ask. The con e sa ional
in e ace o gene a i e AI ools such as Cha GPT equi es ac i e pa icipa ion
on behal o he lea ne , and, he eby, a heigh ened need o s ay on ask. How-
e e , wi hou he physical p esence o ac ual aine s o pee s, s aying on ask,
and mo i a ion in gene al, is no a gi en. Cha GPT is no inclined o openly
p aise he pe o mance o he lea ne o pique in e es and mo i a ion. Indeed,
a p esen , he Cha GPT in e ace i sel is qui e basic, lacking any s uc u al o
discou se elemen s ha migh lend hemsel es well o inhe en ly acili a ing
lea ne mo i a ion. I will be in e es ing o see i his changes o e ime, pe haps
in line wi h empi ical use expe ience s udies.
6 Applica ion scena ios
This chap e will now p o ide a se ies o conc e e scena ios o illus a e how
gene a i e AI can be used o acili a e each o he co e dimensions o delibe -
a e p ac ice pu o wa d in Sec ion 2 and o help ansla o s ad ance along an
expe ise ajec o y. The examples will be based on in e ac ion wi h Cha GPT
based on GPT-4o,1gi en he ela i e ubiqui y and popula i y o his pa icula
gene a i e AI ool a he ime o w i ing. This ocus on Cha GPT o illus a-
i e pu poses does no disc edi he g owing ange o o he gene a i e AI ools
a ailable o use in a simila ashion. The ansla ion scena io being used o
pu poses o con ex ualisa ion is he Ge man-English ansla ion o web con en
om a Ge man p i a e heal h insu ance company,2 ansla ed o an in e na-
ional, English-speaking audience o in o ma i e pu poses. The English ansla-
ion was gene a ed using DeepL.3
Figu e 2 p o ides side-by-side alignmen o he sou ce and a ge con en .
As a poin o depa u e, and o unde sco e he impo ance o s a egic p omp -
ing when i comes o u ilising GenAI o acili a ing delibe a e p ac ice, (2) de-
1h ps://openai.com/index/gp -4/
2h ps://www.huk.de/gesundhei - o so ge- e moegen/k anken e siche ung/
k anken oll e siche ung.h ml#huk
3h ps://www.deepl.com/en/ ansla o
35

E ik Angelone
Figu e 2: Ge man-English ansla ion o p i a e heal h insu ance con-
en
pic s he ela i ely shallow ou pu p oposed by Cha GPT based on he ollowing
gene ic p omp :
(1) “Se up ansla ion aining asks wi h he goal o acili a ing delibe a e
p ac ice”.
(2) Cha GPT ou pu o se ing up ansla ion asks o acili a e delibe a e
p ac ice
Se ing up ansla ion aining asks o delibe a e p ac ice
in ol es ocusing on speci ic skills, ecei ing immedia e
eedback, and g adually inc easing di icul y.
1. Iden i y key skills and goals
Skills: Vocabula y, g amma , idioma ic exp essions, cul u al
con ex , speed, accu acy, s yle, and one.
Goals: Imp o e comp ehension, enhance ansla ion quali y,
inc ease speed wi hou losing accu acy, and de elop a nuanced
unde s anding o bo h sou ce and a ge languages.
2. Feedback and e lec ion
Immedia e Feedback: P o ide p omp eedback on ansla ions,
highligh ing e o s and explaining co ec ions.
Sel -Assessmen : Encou age lea ne s o e lec on hei wo k
and iden i y a eas o imp o emen .
3. Delibe a e p ac ice cycle
Plan: Iden i y speci ic skills o ocus on in each session.
P ac ice: Pe o m a ge ed ansla ion asks.
Feedback: Recei e and e iew eedback om ins uc o s o pee s.
36
2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
Re lec : Re lec on he eedback and iden i y a eas o
imp o emen .
Repea : Repea he cycle wi h new asks, g adually inc easing
di icul y.
(3) depic s co esponding aining asks p oposed by Cha GPT in pa allel o
he con en in (2).
(3) Week 1: Basic sen ences
T ansla e 10 simple sen ences om he sou ce language o he
a ge language, wi h a ocus on ocabula y and basic g amma .
Week 2: Themed pa ag aphs
T ansla e a pa ag aph ela ed o a speci ic heme (e.g.,
echnology), wi h a ocus on hema ic ocabula y and cohe ence.
Week 3: Idioma ic exp essions
T ansla e sen ences ich in idioma ic exp essions, wi h a ocus
on cul u al nuances and idioma ic accu acy.
Week 4: Time ansla ions
T ansla e a 200-wo d a icle wi hin 20 minu es, wi h a ocus on
speed and main aining accu acy unde ime cons ain s.
Week 5: Re e se ansla ion
Pe o m a back ansla ion exe cise, wi h a ocus on
unde s anding equi alence and iden i ying sub le changes in
meaning.
While apping in o he co e condi ions o delibe a e p ac ice pu o wa d in
he li e a u e, his gene ic amewo k p oposed by Cha GPT seems o ocus i s
and o emos on g oup-based ( a he han uly pe sonalised) aining and also
seems o make he assump ion ha lea ne s a e s a ing ‘ om sc a ch’ (mo ing
om basic ocabula y o ield-speci ic e minology, o cul u al and idioma ic
exp essions, o example). Gi en he inhe en lea ne he e ogenei y desc ibed
in Sec ion 3, many ansla o s would no di ec ly bene i om such a locks ep,
linea app oach. As a as di icul y is conce ned, i seems as i he e is less ocus
on wo king a an app op ia e di icul y le el pe se and mo e o an emphasis on
inco po a ing a a ie y o po en ial sou ces o di icul y in a mo e agmen ed
sense ( hema ic ocabula y, idioma ic accu acy, ime p essu e, e c.). This is no
o sugges ha Cha GPT canno pe sonalise delibe a e p ac ice and ca e o he
needs o he indi idual use , bu a he highligh s he impo an , ac i e ole ha
he indi idual lea ne needs o play in o de o bene i om gene a i e AI. This
37
E ik Angelone
s a s wi h u ilising mo e g anula p omp s in acco dance wi h each speci ic
condi ion o delibe a e p ac ice a he han aking a b oad-b ush app oach such
as ha associa ed wi h he gene ic p omp ha gene a ed he ou pu he e.
7 Immedia e, in o ma i e eedback
A a e y basic le el, ansla o s can u ilise Cha GPT o ge immedia e, in o ma-
i e eedback on hei pe o mance by en e ing sou ce con en and hei co e-
sponding ansla ions, along wi h co ela ing p ojec speci ica ions o he ask
(such as audience and pu pose), in conjunc ion wi h he ollowing p omp :
(4) “P o ide me wi h eedback on he ollowing ansla ion, Ge man o
English, o con en om a p i a e heal h insu ance company in Ge many
o an in e na ional English-speaking audience, wi h he pu pose o
p o iding in o ma ion”.
(5) depic s he eedback ob ained om Cha GPT in ela ion o his p omp .
(5) Cha GPT eedback on a ansla ion
You ansla ion is clea and accu a e, bu I ha e a ew
sugges ions o make i sound mo e na u al o an in e na ional
English-speaking audience. He e a e he changes made:
1.Changed "ad an ages" o "bene i s" o a mo e common e m in
heal h insu ance.
2. Changed " a i s" o "plans" o be e i he con ex .
3. Added a colon a e "You bene i s a a glance" o be e
eadabili y.
4. Capi alized "P i a e Heal h Insu e " o consis ency wi h
he i le awa ded.
This eedback was gene a ed in a ma e o a ew seconds, o e coming he ime
cons ain o en s anding in he way o eedback being immedia e in a ypical
ansla o aining en i onmen . As men ioned in Sec ion 3, eedback becomes
uly in o ma i e when i is applicable ac oss ansla ion asks unde simila
condi ions. Wi h his in mind, he ansla o could en e he ollowing p omp in
esponse o he ob ained eedback:
(6) “Which o his eedback would be applicable when ansla ing o he
p i a e heal h insu ance con en om Ge man in o English in he
u u e”?
38
2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
(7) p o ides an o e iew o he esponse p o ided by Cha GPT, illus a ing
how eedback can become uly in o ma i e and ans e able, as opposed o iso-
la ed o he one ansla ion ask a hand.
(7) In o ma i e, ans e able eedback
When ansla ing con en o o he p i a e heal h insu ance
companies in he u u e, you can apply he ollowing eedback
o ensu e cla i y, accu acy, and app op ia eness o an
in e na ional English-speaking audience:
1. Use Clea and Common Te ms:
Replace e ms like "ad an ages" wi h "bene i s" o mo e common
unde s anding.
2. Ensu e Readabili y and Na u al Flow:
Add colons o o he punc ua ion o imp o e eadabili y (e.g.,
"You bene i s a a glance:").
3. Main ain Consis en Fo ma ing:
Use bulle poin s o lis bene i s o easy eading and be e
p esen a ion.
In e es ingly, Cha GPT seems o a ibu e ans e abili y o each piece o gen-
e al eedback p o ided in esponse o he ini ial p omp , as depic ed in (1). This
sugges s a po en ial endency o Cha GPT o con la e ask-speci ic and ans-
e able eedback. Howe e , his did no hold ue in he con ex o a di e en
simula ed ansla ion ask, whe e some o he ask-speci ic eedback was delibe -
a ely il e ed ou and no inhe en ly ega ded by Cha GPT as also being ans e -
able. Ne e heless, om a p omp ing s andpoin , o pu poses o ob aining uly
in o ma i e eedback, he ansla o may wan o ask: “Which o his eedback
is likely speci ic only o his one ansla ion ask and which o he eedback is
ans e able ac oss ansla ions in he same domain?”, wi h he la e being mo e
bene icial om a delibe a e p ac ice pe spec i e.
7.1 E o co ec ion oppo uni ies
Immedia e, in o ma i e eedback, as a co e condi ion o delibe a e p ac ice, p o-
ides he ansla o wi h g anula insigh in o he na u e o e o s. In hopes o
ad ancing along an expe ise ajec o y, i is jus as impo an o he ansla o
o ha e oppo uni ies o co ec e o s hey ha e made by i ue o ha ing hem
anno a ed bu no spelled ou , and o ansla e addi ional sou ce con en in which
39
E ik Angelone
E icsson, K. Ande s & And eas C. Lehmann. 1996. Expe and excep ional pe o -
mance: E idence o maximal adap a ion o ask cons ain s. Annual Re iew o
Psychology 47. 273–305. DOI: 10.1146/annu e .psych.47.1.273.
Ga ison, D. R. 1997. Sel -di ec ed lea ning: Towa d a comp ehensi e model.
Adul Educa ion Qua e ly 48(1). 18–33. DOI: 10.1177/074171369704800103.
Gladwell, Malcolm. 2011. Ou lie s: The s o y o success. New Yo k, USA: Back Bay
Books.
Ha ano, Giyoo & Kayako Inagaki. 1986. Two cou ses o expe ise. In Ha old
S e enson, Hi oshi Azuma & Kenji Haku a (eds.), Child de elopmen and ed-
uca ion in Japan, 262–272. New Yo k, USA: W. Y. F eeman & Co.
Ho n, John & Hi omi Masunaga. 2006. A me ging heo y o expe ise and in elli-
gence. In The Camb idge handbook o expe ise and expe pe o mance, 587–611.
Camb idge, UK: Camb idge Uni e si y P ess. DOI: 10.1017/CBO9780511816796.
034.
Jossbe ge , Helen, Saskia B and-G uwel, Henny Boshuizen & Ma gje an de Wiel.
2010. The challenge o sel -di ec ed and sel - egula ed lea ning in oca ional
educa ion: A heo e ical analysis and syn hesis o equi emen s. Jou nal o
Voca ional Educa ion & T aining 62(4). 415–440. DOI: 10.1080/13636820.2010.
523479.
Knowles, Malcolm. 1975. Sel -di ec ed lea ning: A guide o lea ne s and eache s.
Chicago, USA: Folle .
K üge , Ralph. 2023. A i icial in elligence li e acy o he language indus y
– wi h pa icula emphasis on ecen la ge language models such as GPT-4.
Lebende Sp achen 68(2). 283–330. DOI: 10.1515/les-2023-0024.
Long, Du i & B ian Mage ko. 2020. Wha is AI li e acy? Compe encies and design
conside a ions. In P oceedings o he 2020 CHI Con e ence on Human Fac o s in
Compu ing Sys ems (CHI ’20), 1–16. Honolulu, USA: Associa ion o Compu -
ing Machine y. DOI: 10.1145/3313831.3376727. h ps://doi.o g/10.1145/3313831.
3376727.
Mille , Sco D., Da yl Chow, B uce E. Wampold, Ma k A. Hubble, A. C. Del
Re, Cyn hia Maeschalck & Susanne Ba gmann. 2020. To be o no o be (an
expe )? Re isi ing he ole o delibe a e p ac ice in imp o ing pe o mance.
High Abili y S udies 31(1). 5–15. DOI: 10.1080/13598139.2018.1519410. h ps:
//doi.o g/10.1080/13598139.2018.1519410.
PACTE. 2011. Resul s o he alida ion o he PACTE ansla ion compe ence
model: T ansla ion p oblems and ansla ion compe ence. In Cecilia Al s-
ad, Adelina Hild & Elisabe Tiselius (eds.), Me hods and s a egies o p ocess
esea ch: In eg a i e app oaches o ansla ion s udies, 317–343. Ams e dam,
Ne he lands: John Benjamins.
46

2 Gene a i e AI as a acili a o o delibe a e p ac ice in ansla o aining
Piage , Jean. 1952. The o igins o in elligence in child en. New Yo k, USA: In e na-
ional Uni e si ies P ess.
Sh e e, G ego y. 2006. The delibe a e p ac ice: T ansla ion and expe ise. Jou nal
o T ansla ion S udies 9(1). 27–42.
Sh e e, G ego y. 2018. Le els o explana ion and ansla ion expe ise. He mes 57.
97–108.
Sh e e, G ego y. 2019. P o essional ansla o de elopmen om an expe ise
pe spec i e. In E ik Angelone, Mau een Eh ensbe ge -Dow & Ga y Massey
(eds.), The Bloomsbu y companion o language indus y s udies. London, UK:
Bloomsbu y Academic.
S ojano , Ana. 2023. Lea ning wi h Cha GPT 3.5 as a mo e knowledgeable o he :
An au oe hnog aphic s udy. In e na ional Jou nal o Educa ional Technology in
Highe Educa ion 20(1). 35. DOI: 10.1186/s41239-023-00404-7.
Sun, Sanjun & G ego y Sh e e. 2014. Measu ing ansla ion di icul y: An empi -
ical s udy. Ta ge 26(1). 98–127.
Vygo sky, Le . 1965. Though and language. Camb idge, USA: MIT P ess.
47
Chap e 3
AI Li e acy: The concep o sui abili y
and co e ansla ion skills
Ramon Inglada
He io -Wa Uni e si y, Uni ed Kingdom
The elease o Cha GPT in No embe 2022 eigni ed he deba e on he u u e o he
ansla ion p o ession and, consequen ly, on how ansla o aining p og ammes
should change and adap . Simila o discussions ha occu ed wi h he in oduc-
ion o neu al machine ansla ion a ew yea s ea lie , some a gued ha co e ans-
la ion skills, such as language knowledge, ansla ion abili y and cul u al expe ise
ha ha e long been essen ial componen s o many ansla o aining p og ammes,
had now been ende ed obsole e by he a i al o gene a i e AI (GenAI).
Inspi ed by he concep o machine ansla ion li e acy, a case will be made ha
hese co e skills ( oge he wi h some o he complemen a y skills, such as selec ion
and assessmen ) a e absolu ely essen ial in o de o asce ain whe he he con en
p oduced by GenAI ools is no simply accu a e o inaccu a e, bu , mo e impo -
an ly, sui able o he equi emen s o any gi en ansla ion p ojec , as speci ied
in he ele an ansla ion b ie .
Examples will be gi en o speci ic AI-based asks (such as ansla ion o con en , e -
minology ex ac ion, mul ilingual glossa y c ea ion and machine ansla ion pos -
edi ing) in which he sui abili y o he esul s canno be de e mined wi hou e-
cou se o so-called adi ional co e ansla ion skills.
1 Machine ansla ion li e acy and a i icial in elligence
li e acy
The concep o machine ansla ion (MT) li e acy, in oduced by Bowke &
Bui ago Ci o (2019), has a ac ed a lo o in e es , bo h in academia and in
Ramon Inglada. 2026. AI Li e acy: The concep o sui abili y and co e ansla ion
skills. In JC Pene , Joss Moo kens & Masa u Yamada (eds.), Teaching ansla ion in he
age o gene a i e AI: New pa adigm, new lea ning?, 49–64. Be lin: Language Science
P ess. DOI: 10.5281/zenodo.17641068
Ramon Inglada
he language indus y, as well as in o he sec o s. Fu he mo e, Bowke ’s MT
Li e acy p ojec 1aims o educa e use s on he dos and don’ s o using MT ou pu
and i s in og aphics ha e been e y popula on social media. I also highligh s
he impo ance o con iden iali y and p i acy when using p og ams such as
DeepL and Google T ansla e, wi h an emphasis on di e en use cases o ee
online MT sys ems. The p ojec ’s ocus on enhancing digi al li e acy skills and
he esponsible, e hical, and sus ainable use o MT sys ems can be used as he
basis o u he de elop and adap he concep o A i icial In elligence (AI) li -
e acy. Long & Mage ko (2020: 2) de ined AI li e acy as “a se o compe encies
ha enables indi iduals o c i ically e alua e AI echnologies; communica e and
collabo a e e ec i ely wi h AI; and use AI as a ool online, a home, and in
he wo kplace”. In his chap e , his de ini ion will be used as he basis o a
di e en de ini ion speci ically c ea ed wi h he ansla o aining con ex in
mind. This app oach aligns wi h (K üge 2024) amewo k o AI li e acy in
ansla ion, which emphasises he need o ansla o s o de elop compe encies
in unde s anding and e ec i ely using AI echnologies in hei wo k.
I is undeniable ha he elease o Cha GPT in No embe 2022 gene a ed a lo
o hype. I also eigni ed he deba e on he u u e o he ansla ion p o ession
and, consequen ly, on how ansla o aining p og ammes should change in
o de o adap o his la es echnological ad ancemen . Simila o discussions
ha occu ed wi h he in oduc ion o neu al machine ansla ion (NMT) sys ems
a ew yea s ea lie , some a gued ha co e ansla ion skills ha ha e long been
essen ial componen s o ansla o aining p og ammes had now been ende ed
obsole e by he a i al o GenAI. Howe e , i could be a gued ha his is an
o e simplis ic app oach.
This chap e ad oca es o echnological ad ancemen , as i can b ing many
bene i s o socie y and o e solu ions o some o he g ea es challenges human-
i y cu en ly aces. In he ield o ansla ion, when used sensibly as pa o well-
designed wo k lows, compu e -assis ed ansla ion (CAT) ools and MT can be
a g ea asse o p o essional ansla o s. The same applies o GenAI cha bo s
based on la ge language models (LLMs), such as OpenAI’s Cha GPT, Mic oso
Copilo and Google Gemini. Howe e , gi en how ecen his echnology is and
he pace a which i e ol es, i could be a gued ha he ansla ion indus y
migh s uggle o ag ee on wha a ‘sensible’ and ‘well-designed’ AI-based wo k-
low cons i u es in he ield o p o essional ansla ion. In any case, i is always
wo h keeping in mind ha no echnology-based solu ion is in allible. This ob-
iously applies o Cha GPT (and o he o he LLMs) and i could be said ha , o
1h ps://si es.google.com/ iew/machine ansla ionli e acy/
50
3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
a ce ain ex en , e en Cha GPT i sel eminds us o be igilan and use ou own
judgmen and common sense. A he ime o w i ing, his is he message ha
appea s unde nea h Cha GPT’s cha in e ace:
Cha GPT can make mis akes. Conside checking impo an in o ma ion.
2 Co e ansla ion skills
The pu pose o his chap e is no o p o ide an exhaus i e lis o co e ans-
la ion skills, desc ibing wha hey a e and why hey can be conside ed as key
skills all ansla o s should ha e, as many o he s ha e al eady done his e y
e ec i ely in he pas . Ins ead, i will ocus on h ee e y gene al skills which,
a guably, mos ansla ion schola s would gene ally ag ee a e essen ial o all
p o essional ansla o s (and which, he e o e, should ea u e p ominen ly in all
ansla o aining p og ammes). This is also done o he sake o simplici y as,
ul ima ely, hese skills a e me e examples. This selec ion o co e skills has been
based on he e e ence s anda ds o ansla o aining se ou in he Eu opean
Mas e ’s in T ansla ion (EMT) Compe ence F amewo k (2022) and on he ans-
la ion compe ence model c ea ed by PACTE (2003). Di e en ansla o aining
p og ammes, ins i u ions and se ings could na u ally selec o he skills ha hey
would conside as essen ial.
These chosen h ee co e skills a e:
• Linguis ic knowledge
• T ansla ion abili y
• Cul u al expe ise
Co e ansla ion skills such as he ones men ioned abo e should no be allowed
o disappea om ansla o aining p og ammes, bu a he should now be
conside ed mo e impo an han e e , p ecisely because o he ad en o GenAI.
This is no o say ha ansla o aining cu icula should no be adap ed: hey
should, pa icula ly as concep s such as augmen ed ansla ion and human in he
loop a e becoming inc easingly impo an , and discussions a ound ansla o s
now becoming ‘language specialis s’ o ‘language expe s’ a e happening wi h
inc easing equency. Howe e , he deba e should no be abou hese co e skills
becoming obsole e, bu abou how essen ial hey a e when GenAI is used in
ansla ion.
51

Ramon Inglada
3 The concep o sui abili y o e co ec ness
When ansla o s o ansla ion s uden s use Cha GPT o o he simila ools o
assis hem in hei ansla ion asks, he ocus should no be on whe he he AI-
gene a ed ou pu is ‘co ec ’ (some hing ha could also en ail a deg ee o subjec-
i i y), bu a he on whe he such ou pu is sui able (o i - o -pu pose) o any
gi en se o ansla ion equi emen s, as desc ibed in he ansla ion b ie , o in-
s ance in e ms o one, egis e , a ge audience, ocabula y and ex ype. The
p io i y should be placed on he concep o ‘sui abili y’ o AI-gene a ed con en ,
a he han on i s ‘co ec ness’. This canno be success ully achie ed i he a o e-
men ioned co e skills (as a minimum, linguis ic knowledge, ansla ion abili y
and cul u al expe ise) a e no p esen and well honed. T ansla o s would s ug-
gle o decide whe he AI-gene a ed con en is sui able o any gi en ansla ion
p ojec (and i s se o unique needs) i hey do no know how o ansla e (and
hey ha e no de eloped hei co e ansla ion skills). Using GenAI cha bo s is
easy; using hei ou pu c i ically equi es hough .
4 New and complemen a y skills
I could also be a gued ha no only a e co e ansla ion skills s ill essen ial, bu
ha hey should also be complemen ed wi h he ‘new’ skills o selec ion and as-
sessmen . These a e ob iously no new skills — a e all, many ansla o s ha e
been al eady selec ing and assessing ansla ion memo y ma ches o decades —
bu hey ha e now become undamen al when using GenAI in ansla ion. P o-
essional and ainee ansla o s need o be able o selec among he al e na i es
o e ed by Cha GPT and o he simila ools. They also need o be able o c i i-
cally assess he sui abili y o he ou pu being p esen ed o hem. The e o e, a
enewed emphasis should be gi en o selec ion and assessmen as co e skills o
be in eg a ed in o and de eloped in ansla o aining p og ammes.
The easons why ‘ adi ional’ co e language, ansla ion and cul u al expe -
ise skills should s ill be a pi o al pa o ansla o aining p og ammes ha e
al eady been discussed, in combina ion wi h he easons why hey should also be
complemen ed by he no -so-new skills o selec ion and assessmen . Howe e , is
he e any addi ional new skill ha should be added o his lis o co e skills o he
aining o ansla o s in he new e a o GenAI? The answe is a esounding yes.
This skill is p omp ing (also known as p omp enginee ing). P omp enginee -
ing can be de ined as he p ocess o designing, c a ing, and e ining inpu s o
elici speci ic esponses om a GenAI model, aiming o op imize in e ac ion ou -
comes h ough ca e ul conside a ion o he p omp s (Bozku 2024). The model
52
3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
hen gene a es a esponse based on he inpu i ecei es. E ec i e p omp ing is
c ucial o ob aining ele an , cohe en and sui able answe s. T ansla o ain-
ing p og ammes should seek o in eg a e p omp ing skills in o hei cu icula,
so ha u u e ansla o s can ensu e a g ea e deg ee o sui abili y in he ou pu
gene a ed by GenAI. When adap ing exis ing ansla o aining p og ammes
(o when designing new o e ings), an emphasis should be placed on p ocesses
aimed a he ca e ul c ea ion o e ec i e p omp s (o sequences o p omp s) wi h
di e en oles/pe sonali ies, di e en le els o complexi y o e en di e en de-
g ees o c ea i i y.
5 AI Li e acy in ansla ion: a simple de ini ion
Based on all he p inciples discussed be o e, AI Li e acy o T ansla ion could
be de ined as he combined applica ion o a de ini e se o basic co e skills in
o de o maximise he use ulness and ele ance o GenAI ou pu and ensu e i s
sui abili y in ela ion o he equi emen s se ou by he ansla ion b ie in any
gi en ansla ion ask. The basic co e skills a e linguis ic knowledge, ansla ion
abili y and cul u al expe ise, complemen ed by selec ion and assessmen . All
hese skills should be applied o GenAI ou pu p oduced as a esul o e ec i e
p omp ing. The o e a ching no ion ha unde pins he applica ion o AI Li e acy
in ansla ion is ha o sui abili y, a he han ‘co ec ness’, o he gene a ed
ou pu .
6 AI Li e acy applied o p ac ical language and
ansla ion asks
Fi e examples will now be p o ided o ansla ion o ansla ion- ela ed asks,
chosen o ep esen some o he asks ha p o essional ansla o s ca y ou
on a egula basis (and which, as such, a e also commonly obse ed wi hin he
con ines o he ansla ion class oom). The concep o AI Li e acy will be applied
o hese examples. This means ha , in all cases, an a gumen will be p esen ed o
emphasise he impo ance o elying on he h ee gene ic co e skills o linguis ic
knowledge, ansla ion abili y and cul u al expe ise (p esen ed ea lie in his
chap e ) in o de o selec he mos app op ia e GenAI ou pu and assess he
sui abili y o his ou pu . An example o he p omp (o p omp s) used o eques
he comple ion o he ask in hand by he GenAI cha bo (Cha GPT based on GPT
3.5) will also be p o ided o each example.
53
Ramon Inglada
In he i s example, Cha GPT i sel was used o eques he c ea ion o a sen-
ence ela ed o enewableene gy con aining some g amma issues and awkwa d
wo d g oupings. In he o he ou examples, he h ee ini ial pa ag aphs om he
English e sion o he Wikipedia a icle on he opic o enewable ene gy ( om
July 2024) we e used as he e e ence sou ce ex .
These i e examples a e: sen ence e o mula ion, ansla ion o con en , e mi-
nology ex ac ion, mul ilingual glossa y c ea ion and MT pos -edi ing (MTPE).
6.1 Example 1: Sen ence e o mula ion
Fo ou i s example, I will s a wi h one o he simples asks LLMs can be
used o . This ask has also been chosen as an in oduc o y example as anecdo al
e idence seems o sugges ha his ep esen s one o he p ima y ways in which
GenAI cha bo s a e commonly u ilised, especially by non-na i e speake s o he
language hey a e engaging wi h.
Cha GPT can be a aluable ool o e o mula ing badly w i en sen ences.
By p o iding a poo ly cons uc ed sen ence o Cha GPT, use s can ecei e sug-
ges ed e isions ha add ess spelling mis akes, g amma e o s, imp o e cla i y
and enhance o e all eadabili y. Cha GPT le e ages i s model o language pa -
e ns and g amma ules o gene a e al e na i e ph asing ha is mo e cohe en
and idioma ic. Use s can inpu sen ences om a ious ields, such as enewable
ene gy in ou example, o ecei e ailo ed sugges ions o imp o emen . Wi h
Cha GPT’s assis ance, indi iduals can e ine hei w i ing skills, p oduce clea e
communica ion and con ey hei messages e ec i ely ac oss di e se con ex s.
In his example, he ollowing p omp was used o ask Cha GPT o c ea e he
sen ence I would use as a basis o e o mula ion (see (1)):
(1) P omp used o elici a poo ly cons uc ed sen ence om Cha GPT 3.5.
(C ea ed by Ramon Inglada)
This was he sen ence esul ing om he p omp abo e:
(2) ‘The sun is gi ing us muchly ene gy, so we should pu many sola panels
o ca ch i all’.
The ollowing p omp in (3) was hen used o eques an imp o ed e sion o
he sen ence, co ec ing any g amma issues and changing any unna u al wo d
colloca ions:
54
3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
(3) Sen ence e o mula ion p omp in Cha GPT. (C ea ed by Ramon Inglada)
This was Cha GPT’s eply o ou eques :
(4) ‘The sun p o ides us wi h abundan ene gy, so we should ins all
nume ous sola panels o cap u e i e icien ly’.
I could ce ainly be a gued ha his e sion co ec s exis ing g amma is-
sues and uses mo e idioma ic colloca ions o con ey he in ended meaning mo e
clea ly. Howe e , i would only be possible o do ha i ou knowledge o he
English language is ad anced enough. I I now y o apply he concep o AI
Li e acy in his i s example, i is immedia ely e iden ha I would need o ely
on one o he a o emen ioned co e skills (in his case, knowledge o he English
language) i s ly o ealise ha he o iginal sen ence con ains some issues, and
secondly o assess he sui abili y o he sugges ed ‘imp o ed’ e sion p o ided
by he LLM. Fu he mo e, in his speci ic example, Cha GPT decided ha h ee
elemen s in he o iginal sen ence needed o be changed. These a e ‘muchly en-
e gy’, ‘many sola panels’ and ‘ca ch i all’. I my English language skills we e
insu icien , I could simply accep all h ee sugges ed imp o emen s, assuming
(o e en hoping) ha he esul ing sen ence is now much mo e sui able o my
needs. Howe e , I could also decide o selec only some o hese changes, so I
would only keep hose ha I (and no Cha GPT) conside as sui able. I could
e en accep hem all and hen go on o u he modi y hem o end up wi h a
inal sen ence which would be he collabo a i e esul be ween me and he LLM.
This p ocess exempli ies he concep o “cen au asks” as desc ibed by Mollick
(2023), whe e he e is a clea di ision o labou be ween human and AI, le e ag-
ing he s eng hs o each. Once mo e, he decision on whe he his inal sen ence
is he mos sui able op ion o ou needs is some hing ha can only be achie ed
i ou co e skills ha e been de eloped enough.
6.2 Example 2: T ansla ion o con en
This second example is in ended o co e he use o LLMs as MT ools. While
LLMs we e no in p inciple designed o be used as MT p o ide s, his is ce ainly
one o he many asks hese ools can pe o m. I is also p obably one o he i s
uses ha comes o mind when one hinks abou he po en ial uses o GenAI in
55
Ramon Inglada
a emp o o e come hese high le els o hype. This objec i e is c i ical because
i helps o ensu e ha he expec a ions o AI a e ealis ic. I could be said ha ,
in some cases, he hype su ounding AI has led o un ealis ic expec a ions. Yes,
GenAI can be e y use ul in he ansla ion ield and i has many po en ially
use ul applica ions (as we ha e seen in his chap e and as many o he s ha e
explained elsewhe e). Howe e , i is s ill essen ial o ansla ion s uden s o ac-
qui e co e ansla ion skills and o p o essional ansla o s o con inue de el-
oping and enhancing hem h oughou hei ca ee s. By applying he p inciples
o AI Li e acy, i is hoped ha we can con ibu e o ensu ing ha expec a ions
linked o he use o GenAI in ansla ion a e ealis ic and ha he echnology is
used in a way ha is bene icial o all s akeholde s (including ansla o s). I is
equally impo an o no e ha AI is no a panacea o all ansla ion ‘p oblems’,
and i is essen ial o use i in conjunc ion wi h o he echnologies and app oaches.
GenAI should become ano he ool in he ex ensi e ansla o ’s oolbox, bu i
should no aim a eplacing ansla o s al oge he .
I is also impo an o men ion ha he concep o AI Li e acy and i s key p in-
ciples ha e been adap ed wi h lexibili y i mly in mind. They a e no necessa ily
mean o be aken li e ally as hey ha e been p esen ed in his chap e (al hough
i is hoped ha his would indeed be possible), bu a he as guiding p inciples
open o in e p e a ion and adap a ion based on speci ic con ex s and e ol ing
equi emen s and pe spec i es. As Long & Mage ko (2020) discuss, AI Li e acy
encompasses a ange o compe encies ha can be ailo ed o di e en needs and
con ex s. Simila ly, K üge (2024) ou lines an AI Li e acy amewo k ha em-
phasises he adap abili y o se e al skills. The co e skills (linguis ic knowledge,
ansla ion abili y, and cul u al expe ise), complemen a y skills (selec ion and
assessmen ), and he new skill (p omp ing) ha ha e been p esen ed in his chap-
e as unde pinning he concep o AI Li e acy, ha e been included as examples.
The numbe o skills is lexible and can be changed based on equi emen s. In
addi ion, hese skills can be ailo ed o modi ied o align wi h di e se needs. AI
Li e acy is no mean o be an unmo able o ixed en i y, bu a he a dynamic
and e ol ing concep ha adap s o changing ci cums ances, con ex s, and p i-
o i ies.
A poin ha wa an s ei e a ed emphasis is ha GenAI should no be iewed
as a eplacemen o human ansla o s. Ins ead, i should be conside ed as a ool
ha could be used o augmen he wo k o ansla o s. The concep o AI Li e -
acy and i s unde lying p inciples emphasise he need o use co e ansla ion skills
when using GenAI o ansla ion asks. These skills include no only linguis ic
knowledge bu also he abili y o unde s and he con ex , cul u e, and nuances o
he language. T ansla o s can add alue by le e aging hese skills o imp o e he
62

3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
quali y and sui abili y o he ou pu p oduced by GenAI. Mo eo e , he po en ial
o AI o hallucina e should ne e be unde es ima ed. Waldo & Boussa d (2024)
analysed he easons why LLMs hallucina e and concluded ha hese models
encoun e ed challenges wi h opics o which limi ed da a was a ailable online,
o en gene a ing inaccu a e esponses p esen ed in a ealis ic o ma wi hou ac-
knowledging he inaccu acies. The esponsibili y o ind issues and ix hem, and
o make he ou pu p oduced by GenAI sui able o any se o speci ic needs a
any gi en ime, is ou inely placed on ansla o s. Wi hou he skills he con o m
he basis o AI Li e acy, his would be an impossible endea ou .
Deba es a ound he eme gence, e olu ion and implemen a ion o AI in ou
mode n socie ies should be based on he o e a ching p inciples o e hics, sus ain-
abili y and da a p i acy. Discussions a ound AI Li e acy should no be an excep-
ion and, as he applica ion o GenAI in ansla ion con inues o e ol e, a en ion
should be de o ed o conce ns abou i s e hical implica ions, en i onmen al im-
pac , and po en ial impac on he ansla ion p o ession. The de elopmen and
deploymen o AI in he ield o ansla ion aise a wide ange o e hical conce ns,
including hose ela ed o accu acy, bias, anspa ency, au ho ship, p i acy and
pe sonal da a. GenAI in ansla ion has many po en ially use ul applica ions and
his has been widely discussed in his chap e . Howe e , i s e hical, sus ainabil-
i y, and p o essional implica ions mus be ca e ully conside ed and add essed o
ensu e ha AI-based ansla ion- ela ed asks and wo k lows a e de eloped and
used esponsibly and sus ainably. I is hoped ha p omo ing he no ion o AI
Li e acy (and in eg a ing his concep in ansla o aining cu icula) will also
con ibu e o be e decision-making and o mo e ma u e deba e (and sc u iny)
among all s akeholde s in he ansla ion indus y.
As men ioned ea lie in his chap e , using GenAI cha bo s is easy, bu us-
ing hei ou pu c i ically equi es hough . Ul ima ely, AI Li e acy in he ield
o ansla ion (and, mo e speci ically, in he ield o ansla ion aining) aims
a equipping u u e ansla o s —and o he language p o essionals— wi h his
undamen al abili y o c i ical hinking.
Re e ences
Bowke , Lynne & Jai o Bui ago Ci o. 2019. Machine ansla ion and global e-
sea ch: Towa ds imp o ed machine ansla ion li e acy in he schola ly commu-
ni y. Leeds, UK: Eme ald Publishing. DOI: 10.1108/9781787567214.
Bozku , A as. 2024. Tell me you p omp s and I will make hem ue: The
alchemy o p omp enginee ing and gene a i e AI. Open P axis 16(2). 111–118.
DOI: 10.55982/openp axis.16.2.661.
63
Ramon Inglada
EMT. 2022. Eu opean mas e ’s in ansla ion compe ence amewo k 2022. Tech.
ep. B ussels: Eu opean Commission. 1–12. h ps://commission.eu opa.eu/
sys em/ iles/2022-11/em _compe ence_ wk_2022_en.pd .
K üge , Ralph. 2024. Ou line o an a i icial in elligence li e acy amewo k
o ansla ion, in e p e ing and specialised communica ion. Lublin S udies in
Mode n Languages and Li e a u e 48(3). 11–23. DOI: 10.17951/lsmll.2024.48.3.11-
23.
Long, Du i & B ian Mage ko. 2020. Wha is AI li e acy? Compe encies and design
conside a ions. In P oceedings o he 2020 CHI Con e ence on Human Fac o s in
Compu ing Sys ems (CHI ’20), 1–16. Honolulu, USA: Associa ion o Compu -
ing Machine y. DOI: 10.1145/3313831.3376727. h ps://doi.o g/10.1145/3313831.
3376727.
Mollick, E han. 2023. On-boa ding you AI In e n. (25 No embe , 2024). h ps:
//www.oneuse ul hing.o g/p/on-boa ding-you -ai-in e n.
PACTE. 2003. Building a ansla ion compe ence model. In Fabio Al es (ed.), T i-
angula ing ansla ion: Pe spec i es in p ocess o ien ed esea ch (45), 43–66. Am-
s e dam, Ne he lands: John Benjamins. DOI: 10.1075/b l.45.06pac.
Waldo, Jim & Soline Boussa d. 2024. GPTs and hallucina ion: Why do la ge lan-
guage models hallucina e? Queue 22(4). 19–33. DOI: 10.1145/3688007.
Yamada, Masa u. 2023. Op imizing machine ansla ion h ough p omp engi-
nee ing: An in es iga ion in o Cha GPT’s cus omizabili y. In Masa u Yamada
& Félix do Ca mo (eds.), P oceedings o Machine T ansla ion Summi XIX, Vol.
2: Use s T ack, 195–204. Macau, China: Asia-Paci ic Associa ion o Machine
T ansla ion. h ps://aclan hology.o g/2023.m summi -use s.19.
64
Pa II
New pa adigm: New knowledge
Chap e 4
Teaching ansla ion s uden s abou da a
in he age o gene a i e AI
Lynne Bowke
Uni e si é La al, Canada
AI ansla ion ools a e now a key pa o ansla ion educa ion, bu many edu-
ca o s a e sea ching o e ec i e ways o each he essen ials o hese ools o
s uden s wi h no backg ound in compu e science. This chap e explains why co -
po a make a good en y poin o lea ning abou AI ansla ion ools, and i explo es
how science communica ion echniques such as aming, analogies and isualiza-
ion can be used o help ansla ion educa o s and s uden s come o g ips wi h da a
and machine lea ning.
1 In oduc ion
Though ansla o educa ion o en akes place in an a s o humani ies acul y,
echnology has been a i al pa o he ansla ion p o ession, and hence o ans-
la o educa ion p og ammes, o a leas hi y yea s. O e his pe iod, ansla o s
and ansla ion s uden s ha e shown ema kable esilience as hey adap o new
ools and new echnology-based ways o wo king. Howe e , his does no mean
ha eaching echnologies o ansla ion s uden s is easy, and educa ing he ed-
uca o s can be a pa icula challenge (Bowke 2023, Kenny 2020). As he pace o
new ool eleases ge s as e , i can be di icul o ansla o educa o s o know
whe e o begin. This chap e p oposes ha eaching ansla ion s uden s abou
da a is one o he key building blocks in p epa ing hem o use ansla ion ech-
nology e ec i ely. In he case o ansla ion, da a equen ly akes he o m o
ex s o ganised in o co po a. This chap e he e o e begins wi h a b ie e iew o
co po a and co pus-based ools, no ing how AI ansla ion ools ha e in luenced
Lynne Bowke . 2026. Teaching ansla ion s uden s abou da a in he age o gene a-
i e AI. in JC Pene , Joss Moo kens & Masa u Yamada (eds.), Teaching ansla ion in
he age o gene a i e AI: New pa adigm, new lea ning?, 67–85. Be lin: Language Sci-
ence P ess. DOI: 10.5281/zenodo.17641070

Lynne Bowke
he na u e o co po a. Nex , he chap e ou lines why a science communica ion
app oach, a he han adi ional scien i ic communica ion, could be use ul in he
con ex o ansla ion echnology educa ion. Th ee science communica ion ech-
niques – aming, analogies and isualisa ion – a e combined wi h examples o
demons a e how science communica ion could be adap ed o each ansla ion
s uden s abou da a- ela ed opics in he age o AI.
2 Co po a and co pus-based ansla ion ools
2.1 The changing cha ac e is ics o co po a in he age o AI
O e he pas ew decades, ansla o s ha e seen he in oduc ion and in eg a ion
o a ange o di e en ools, including conco dance s (Zane in 2023), e m ex-
ac o s (Ko kon zelos & Ananiadou 2022), ansla ion memo y sys ems (Melby
& W igh 2023), machine ansla ion sys ems (Way 2020), and mos ecen ly,
gene a i e AI (GenAI) sys ems (Siu 2024). The a ious ools ha e become in-
c easingly sophis ica ed wi h ega d o hei capabili ies and hei unde lying
a chi ec u e, bu a common ea u e o all o he ools named abo e is ha hei
co e unc ionali y e ol es a ound p ocessing da a in he o m o ex s. Collec-
ions o ex s a e usually e e ed o as co po a, and hese can ake di e en
o ms depending on he na u e o he ex s and he way ha hey a e o ganised
(McEne y 2022). T ansla ion ools ha p ocess co po a a e equen ly desc ibed
as co pus-based o da a-d i en ools (Wang e al. 2022).
As poin ed ou by Isabelle e al. (1993: 205), “exis ing ansla ions con ain mo e
solu ions o mo e ansla ion p oblems han any o he a ailable esou ce”. The e-
o e, one ype o co pus ha has been used e y o en by ansla ion ools is
he bilingual pa allel co pus (Sima d 2020). In his ype o co pus, a collec ion
o sou ce ex s a e aligned – usually a sen ence le el – wi h hei coun e pa
a ge ex s. In o he wo ds, each sen ence in he sou ce ex is linked o i s co -
esponding ansla ion in he a ge ex . Some ansla ion ools migh also use
monolingual co po a o o iginal ex s in he sou ce and/o a ge language o ac
as a linguis ic model o ha language.
Fo s uden s who ha e al eady lea ned abou conco dance s, e m ex ac o s
o ansla ion memo y sys ems, he no ion o a co pus is al eady amilia because
he co pus is a e y isible esou ce in such ools. When using hese ools, he
ansla o o ansla ion s uden o en has a hand in c ea ing he co pus o may
need o upload he co pus ha has been p o ided by he clien o educa o . Owing
o hei amilia i y, co po a make a good s a ing poin o lea ning abou neu al
machine ansla ion o GenAI ools.
68
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
Acco ding o Bowke & Pea son (2002: 9),
A co pus can be desc ibed as a la ge collec ion o au hen ic ex s ha
ha e been ga he ed in elec onic o m acco ding o a speci ic se o c i-
e ia. The e a e ou impo an cha ac e is ics o no e he e: ‘au hen ic’,
‘elec onic’, ‘la ge’ and ‘speci ic c i e ia’.
While he gene al no ion o a co pus emained ela i ely s able in he pe iod
be o e AI ansla ion ools appea ed, he in oduc ion o hese ools has ushe ed
in some changes wi h ega d o he ea u es o co po a. The e o e, i is impo an
o ansla ion s uden s o unde s and how AI ools ha e in luenced and al e ed
he na u e o co po a.
2.2 Machine- eadable o m
The need o he co pus o be in elec onic o machine- eadable o m has no
changed. Indeed, we could say i is mo e impo an han e e since AI ans-
la ion ools ake on an e en g ea e deg ee o co pus p ocessing han do ools
such as conco dance s o ansla ion memo y sys ems. While conco dance s and
ansla ion memo y sys ems conduc pa e n ma ching and hen so and display
in o ma ion o he ool use s o in e p e , AI ansla ion ools go u he by a -
emp ing o in e p e he esul s and p esen ully o med ansla ion solu ions.
2.3 Size
Co po a a e used o e eal linguis ic pa e ns, which only become appa en when
he e a e mul iple examples o a gi en linguis ic phenomenon. The e o e, an-
o he ea u e o co po a is ha hey a e usually e y la ge collec ions o ex .
Howe e , ou unde s anding o wha cons i u es “la ge” has e ol ed o e ime.
The i s gene a ion o co po a c ea ed in he 1960s con ained hund eds o hou-
sands o wo ds and we e mainly consul ed by linguis s (McEne y 2022). These
linguis s used co pus analysis ools (e.g. conco dance s) o help hem so and
display he ex da a, bu he linguis s we e s ill esponsible o in e p e ing i .
Today, he co po a used o powe AI ools con ain hund eds o billions o wo ds
(Hughes 2023). This is in la ge pa because hese AI ools do no unde s and ex
in he way ha people do, and hey need a much la ge numbe o examples in
o de o p edic pa e ns wi h con idence. Howe e , his need o ex emely la ge
co po a is in luencing o he cha ac e is ics o co po a, such as he ex s selec ed
o inclusion (see Sec ion 2.4) and some imes e en he au hen ici y o he ex s
(see Sec ion 2.5).
69
Lynne Bowke
The a ailabili y o machine- eadable ex s can di e d ama ically om one lan-
guage o ano he , c ea ing dispa i ies wi h ega d o an AI ool’s pe o mance
in di e en languages. A high- esou ce si ua ion occu s when i is ela i ely
s aigh o wa d o ga he a la ge numbe o high-quali y esou ces o a gi en
language o language pai . Fo ins ance, bo h English and F ench a e widely used
languages, and he e is a lo o ansla ion ac i i y be ween hese wo languages.
As a esul , i is no oo di icul o compile monolingual and bilingual pa al-
lel co po a o hese languages, and hey a e hus e e ed o as high- esou ce
languages. In con as , a low- esou ce si ua ion can occu when languages (o
language a ie ies) a e less widely used, such as some o he Indigenous lan-
guages o he Ame icas o he a ie y o F ench used in Canada. Fo languages
o language a ie ies o limi ed di usion, i can be mo e challenging o build a
la ge co pus. Mo eo e , e en i wo languages ha e a la ge numbe o speake s
(e.g. Russian and Hindi), he e may no be a lo o ansla ion ac i i y be ween
hem, making i ha d o c ea e bilingual pa allel co po a o his language pai .
The e o e, languages, language a ie ies o language pai s o which he e a e
ew co po a a ailable a e desc ibed as being low esou ce.
2.4 Speci ic c i e ia
As emphasised by McEne y (2022), in o de o be mos use ul, a co pus canno
consis o ex s ha ha e been ga he ed a andom o in a pu ely oppo unis ic
way. Ra he , he ex s in a co pus a e selec ed because hey co espond o speci ic
c i e ia and a e ep esen a i e o a la ge se o ex s wi h hose cha ac e is ics.
One clea c i e ion in he con ex o ansla ion is ha he ex s should be o
high quali y. Beyond his, he e a e many di e en op ions o designing a co pus
depending on i s in ended pu pose, bu he key poin he e is ha he choice o
which ex s o include is mo i a ed. Fo example, in he con ex o a co pus o
be used o ansla ion, i could be impo an o selec ex s ha a e on a gi en
opic, o a ce ain ex ype o egis e , o om a pa icula ime pe iod.
E idence o he impo ance o co pus design can be seen in he way ha
ansla o s cons uc ansla ion memo y da abases, such as by c ea ing di e -
en da abases (o adding ele an me ada a) o di e en domains o o di e en
clien s (e.g. o espec hei p e e ed e minology o house s yle). In his way,
hey can es ic a sea ch o ex s ha ha e speci ic ea u es (Melby & W igh
2023). Likewise, neu al machine ansla ion ools a e known o achie e be e
quali y when he co pus is adap ed o a speci ic domain (Chu & Wang 2018).
Howe e , as no ed in sec ion Sec ion 2.3, AI ansla ion ools need o ha e an
eno mous numbe o ex s in he co pus. As a esul , i can be challenging o
70
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
achie e he necessa y size while ying o be selec i e abou he con en . The
low- esou ce si ua ion desc ibed in sec ion Sec ion 2.3 can be u he compli-
ca ed when i comes o inding ce ain ex ypes o ex s on e y specialised
opics in less widely used o ansla ed languages and language a ie ies. A con-
sequence o no ha ing a la ge enough high-quali y co pus is ha he AI ool
does no ha e enough good examples o d aw on, and so he ool’s pe o mance
may be poo e in low- esou ce si ua ions (Way 2025). I lowe quali y ex s a e
included in he co pus in o de o inc ease he size, hen he ool may gene a e
low-quali y ansla ions. The implica ions o insu icien quan i y and quali y o
ex s in aining co po a used by AI ansla ion ools is discussed in mo e de ail
in sec ions Sec ion 4.2 and Sec ion 4.3.
2.5 Au hen ici y
Finally, he need o ha e au hen ic ex s in he co pus used o be sac osanc .
Fo ins ance, McEne y (2022) desc ibes a co pus as “a la ge body o linguis ic
e idence composed o a es ed language use” (494) and “a collec ion o na u ally
occu ing language da a” (495). In he case o ansla ion, he e is a desi e o ha e
high-quali y da a, which means using ex s ha ha e been ansla ed by language
p o essionals. As obse ed by Kenny (2011: 2), he eason ha he de elope s
o ansla ion ools use co po a o human ansla ions o ain hei sys ems
is because such co po a a e assumed o con ain good answe s o ansla ion
p oblems; and hey a e assumed o con ain good answe s p ecisely because hey
con ain ansla ions pe o med by human beings.
One consequence o he need o bigge and bigge bilingual pa allel co po a
o aining AI ansla ion ools has been ha such au hen ic high-quali y human
ansla ed ex s ha e become inc easingly aluable commodi ies. This in u n has
aised many e hical ques ions abou owne ship o ansla ion da a and pe mis-
sion o use i , p omp ing an explici need o discuss hese issues wi h ansla ion
s uden s. Moo kens (2022) con ains a de ailed examina ion o such e hical issues,
along wi h sugges ions o how hese can be in eg a ed in o ansla o educa ion
(see Sec ion 4.1).
Ano he esponse o he po en ial sho age o ex needed o cons uc ing
e y la ge co po a o use wi h AI ansla ion ools has been o se aside he long-
es ablished adi ion o using au hen ic da a and o explo e he use o syn he ic
da a. In he con ex o ansla ion, syn he ic da a is c ea ed by using a machine
ansla ion ool o ansla e addi ional ex s, and hen adding hese machine-
ansla ed ex s o he co pus (Senn ich e al. 2016). The quali y o syn he ic da a
can a y, and while he need o addi ional da a ends o be o low- esou ce
71
Lynne Bowke
To see examples o images o bananas, ansla ion educa o s and s uden s can
na iga e o open access image si es such as Pixabay.com o Unsplash.com and
use he sea ch e m “bananas”. To pe o m well, an AI ool would need o be
p o ided wi h many, many examples o pic u es o bananas in o de o be able
o gene a e a new image based on hese examples. People ypically need ewe
examples om which o lea n.
A un exe cise o do in class is o ask he s uden s o look a a dozen pic u es
o bananas on an open access si e, and hen o imagine ha hey a e an AI image
gene a ion ool ha has been asked o gene a e (i.e., d aw) hei own image in
esponse o he ollowing p omp : “D aw a banana.”
This will no likely pose oo much o a challenge o he s uden s – e en
hose who a e no a is ically inclined! As a ollow-up, sha e wi h s uden s he
esul s o he Daniel Hook (2023) expe imen s in which he asked he GenAI
ool Midjou ney o gene a e images o a banana. Wha Hook ound was ha he
AI ool only gene a ed images wi h wo o mo e bananas, a he han a single
banana. Hook e ined he p omp se e al imes asking o “a banana”, “a single
banana”, “one banana”, and so on, bu he ool con inued o gene a e images
con aining mul iple bananas.
This example, which Hook (2023) dubs “The lone banana p oblem”, can be used
as a simple ye e ec i e way o explaining he issue o da a bias. I he aining
da a does no con ain a su icien numbe o images o indi idual bananas, hen
he AI ool will no be able o gene a e an image o a single banana. O cou se,
he p oblem can be ixed by adding mo e da a (i.e., images o lone bananas), bu
as Hook wa ns, we do no always ealise wha gaps a e he e. Hook (2023) also
s esses ha i is impo an o ecognise ha AI ools do no unde s and he wo ld
he way ha people do: hey only unde s and commonly occu ing pa e ns.
While humans a e amazing pa e n ma che s, ha skill is augmen ed by com-
mon sense (in many bu no all cases), con ex and an e ol ed and sub le unde -
s anding o he physical wo ld a ound us. AIs don’ ye ha e hose augmen a-
ions – hey a e pu e pa e n ma ching powe . And hence, hey a e only as good
as he da a ha we inpu in o he aining se and hence can be no mo e han
he s a is ical a e age o hose inpu s. In he lone banana p oblem, he s a is ics
sugges ed ha bananas only appea in wos (o mo e) and so he AI could no
imagine a single banana. (Hook 2023, n.p.)
4.3.3 Example 2: The snow de ec o
As a ollow-up o he lone banana p oblem, i may be use ul o sha e he example
o an AI ool whose ask is image classi ica ion. This example comes om Ribei o
78

4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
e al. (2016), who ained an AI ool using wo di e en se s o images – en
images o husky dogs and en images o wol es. A e he aining session, he
esea che s es ed he AI ool’s abili y o classi y images co ec ly by showing
i en new images ( i e huskies and i e wol es) ha had no been pa o he
ini ial aining da a. The AI ool classi ied mos o he images co ec ly, bu i
made wo e o s. In in es iga ing he cause o he e o s, he esea che s ound
ha he AI ool was making decisions no based on he animals in he pic u es
bu based on he backg ound. In he aining da a, all he pic u es o he wol es
con ained snow, bu none o he pic u es o huskies con ained snow. The e o e,
in he es phase, i he image showed an animal agains a snowy backg ound,
he AI ool classi ied he image as a wol , bu i he e was no snow, hen he AI
ool classi ied he image as a husky. In o he wo ds, he AI ool was looking o
pa e ns, bu he pa e n ha i used as a basis o i s decision-making was no a
sensible one.
The snow de ec o is a unny example, bu he medical li e a u e con ains
mo e se ious examples o simila phenomena. In Na la e al. (2018), esea che s
ied o ain an AI ool o dis inguish be ween images o cance ous and non-
cance ous skin lesions. Howe e , in he aining co pus, he majo i y o images o
cance ous lesions also con ained a ule o measu e he size o he lesion. Du ing
he es ing phase, i became clea ha he AI ool was equa ing he absence o a
ule wi h he absence o cance . In ano he example, an AI sys em was ained
o dis inguish lung X- ays o people wi h pneumonia om lung X- ays o people
wi h COVID-19 (Robe s e al. 2021). Un o una ely, he aining da a con aining
examples o pneumonia consis ed mainly o pedia ic pa ien s, while he aining
da a wi h examples o COVID-19 was om mainly adul pa ien s. As a esul , he
AI ool lea ned o dis inguish be ween child en and adul s a he han be ween
pneumonia and COVID-19.
4.3.4 Connec ing he do s: F om images o ansla ion
Though humo ous, he lone banana and snow de ec o examples a e in ended
o d i e home he e y se ious poin ha AI ools do no hink, unde s and
o e en p ocess in o ma ion he way ha people do. By ocusing on ela i ely
s aigh o wa d, conc e e and isual asks such as image gene a ion and image
classi ica ion, hese examples a e easy o s uden s o ela e o, and hey can hen
be connec ed o ansla ion- ela ed issues. Fo example, he lone banana p oblem
can lead in o a discussion o da a bias and o wha happens when a ansla ion
co pus has gaps in he aining da a. Fo ins ance, Vanmassenho e e al. (2018)
ha e demons a ed how da a bias can lead o issues o gende bias in ansla ion,
79
Lynne Bowke
while Bowke & Blain (2022) poin ou how a lack o da a in one language a ie y
can lead o con en om ano he a ie y being used e oneously in a ansla ion.
In addi ion o ein o cing he lesson abou da a bias, he snow de ec o exam-
ple, along wi h he mo e se ious examples o he ule de ec o and he child
de ec o , can be used o p omo e discussions on opaci y and wha can go w ong
when i is no clea wha pa e ns he AI ool is using as he basis o decisions.
These examples can also eed discussions abou isk assessmen – an impo an
elemen o any decision abou whe he o no o use an AI ansla ion ool (Ko-
ponen & Nu minen 2024). In some cases, he ansla ion ask a hand migh be
ela i ely low-s akes (e.g. ansla ing a ex o he pu poses o en e ainmen ),
and so using an AI ool could be a good op ion. Howe e , in o he ci cums ances,
he ansla ion ask migh be one wi h highe s akes, whe e he consequences o
a poo ansla ion could nega i ely a ec someone’s heal h o li e (Way 2013).
Ano he ype o in o ma ion ha can be impo an as pa o isk assessmen
is unde s anding whe he he si ua ion is mo e likely o be a high- esou ce o
a low- esou ce si ua ion, gi en ha he pe o mance o he AI ool is likely o
be be e o high- esou ce languages, language a ie ies, language pai s and do-
mains, and poo e in low- esou ce si ua ions (see Sec ion 2.3).
5 Conclusion
AI ansla ion ools a e apidly inse ing hemsel es in o he ansla ion indus-
y and mus also be add essed in ansla o educa ion. The echnology behind
hese ools is e y sophis ica ed and may be in imida ing o ansla ion educa-
o s and s uden s alike, gi en ha mos do no ha e a backg ound in compu e
science. Howe e , mos language p o essionals do no need o unde s and all
he de ails o how AI ansla ion ools wo k in o de o use hem in an in o med
and esponsible way. In con as , i is ex emely use ul o language p o ession-
als o unde s and how hese ools in e ac wi h da a, and how he da a used o
machine lea ning can a ec ool pe o mance and ansla ion quali y. In he con-
ex o AI ansla ion ools, he da a is ypically a co pus. T ansla ion s uden s
a e al eady amilia wi h co po a, which a e also used wi h o he ypes o ans-
la ion ools (e.g. conco dance s, e m ex ac o s, ansla ion memo y sys ems).
Howe e , s uden s need o lea n how AI ools a e impac ing he cha ac e is ics
o co po a.
Ra he han app oaching he eaching o AI ansla ion ools om he mo e
adi ional pe spec i e o scien i ic o expe - o-expe communica ion, ansla-
ion educa o s may ind ha science communica ion echniques (e.g. aming,
80
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
analogies, isualisa ion) a e mo e use ul o eaching s uden s abou da a and
machine lea ning.
In he spi i o ans e sal lea ning, ansla ion educa o s may e en ind ha
de eloping e ec i e echniques o each hei ansla ion s uden s abou da a and
i s ole in AI ansla ion ools can pay o in o he ways. As Bo owiec (2023: 1)
lamen s,
Despi e he e being many excellen easons o scien is s o engage in sci-
ence communica ion, hey o en lack he ools o do so. […] Scien is s e-
quen ly ci e lack o aining and/o con idence in hei science communica-
ion skills as a ba ie o hei pa icipa ion in public- acing ac i i ies.
Since many ansla o educa o s a e also esea che s, hey may be able o ap-
ply hese science communica ion echniques o hei esea ch-o ien ed ac i i-
ies. Mo eo e , Eh ensbe ge -Dow e al. (2023) sugges ha ansla o aine s
a e well placed o ake on a ole such as machine ansla ion li e acy consul an ,
which could in ol e helping clien s (who a e unlikely o be compu e scien is s)
o unde s and how AI ools p ocess da a o how da a esou ces can be cu a ed
o imp o e he ools’ pe o mance.
Re e ences
Bo owiec, B i ney G. 2023. Ten simple ules o scien is s engaging in science
communica ion. PLoS Compu a ional Biology 19(7). e1011251. DOI: 10 . 1371 /
jou nal.pcbi.1011251.
Bowke , Lynne. 2023. T ansla ion echnologies. In Chan Sin-wai (ed.), Rou ledge
encyclopedia o ansla ion echnology, 2nd edn., 95–113. London, UK: Rou -
ledge.
Bowke , Lynne & F édé ic Blain. 2022. When F ench becomes Canadian F ench:
The cu ious case o localizing COVID-19 e ms wi h Mic oso T ansla o . The
Jou nal o In e na ionaliza ion and Localiza ion 9(1). 1–37. DOI: 10.1075/jial.
22007.bow.
Bowke , Lynne & Jenni e Pea son. 2002. Wo king wi h specialized language:
A p ac ical guide o using co po a. London, UK: Rou ledge. DOI: 10 . 4324 /
9780203469255.
B i a-Iglesias, Vicen & Sha on O’B ien. 2022. The language enginee : A
ans e sal, eme ging ole o he au oma ion age. Quade ns de Filologia: Es-
udis Lingüís ics XXVII. 17–48.
81
Lynne Bowke
Bu ns, T. W., D. J. O’Conno & S. M. S ocklmaye . 2003. Science communica ion:
A con empo a y de ini ion. Public Unde s anding o Science 12(2). 183–202. DOI:
10.1177/09636625030122004.
Chu, Chenhui & Rui Wang. 2018. A su ey o domain adap a ion o neu al ma-
chine ansla ion. In P oceedings o he 27 h in e na ional con e ence on compu-
a ional linguis ics, 1304–1319. San a Fe, USA: Associa ion o Compu a ional
Linguis ics.
Cooke, S e en J., Aus in J. Gallaghe , Na alie M. Sopinka, Vi ian M. Nguyen,
Rachel A. Skubel, Neil Hamme schlag, Sa ah Boon, Na han Young & Andy J.
Danylchuk. 2017. Conside a ions o e ec i e science communica ion. Face s
2. 233–248. DOI: 10.1139/ ace s-2016-0055.
Eh ensbe ge -Dow, Mau een, Alice Delo me Beni es & Ca oline Leh . 2023. A
new ole o ansla o s and aine s: MT li e acy consul an s. The In e p e e
and T ansla o T aine 17(3). 393–411. DOI: 10.1080/1750399X.2023.2237328.
Flach, Pe e . 2012. Machine lea ning. Camb idge, UK: Camb idge Uni e si y P ess.
Hackenbuchne , Janiça & Ralph K üge . 2023. Da aLi MT – eaching da a li e acy
in he con ex o machine ansla ion li e acy. In Ma y Nu minen, Judi h B en-
ne , Maa i Koponen, Si ku La omaa, Mikhail Mikhailo , F ede ike Schie l,
Tha indu Ranasinghe, E a Vanmassenho e, Se gi Al a ez Vidal, No a A an-
be i, Ma a Nunzia ini, Ca la Pa a Esca ín, Mikel Fo cada, Maja Popo ic,
Ca olina Sca on & Helena Moníz (eds.), P oceedings o he 24 h annual con-
e ence o he Eu opean Associa ion o Machine T ansla ion, 285–293. Tampe e,
Finland: Eu opean Associa ion o Machine T ansla ion.
Hanauska, Monika. 2019. His o ical aspec s o ex e nal science communica ion.
In Anne e Leßmöllmann, Ma celo Dascal & Thomas Gloning (eds.), Science
communica ion, 585–600. Be lin, Ge many: De G uy e Mou on. DOI: 10.1515/
9783110255522-027.
Hen ille, Le i ia. 2020. E ec i e science communica ion decons uc ed. Uni e -
si y A ai s. h ps://uni e si ya ai s.ca/ca ee -ad ice/ask-d -edi o /e ec i e-
science-communica ion-decons uc ed/.
Hook, Daniel. 2023. The lone banana p oblem. Digi al Science. h ps:/ /www.
digi al- science.com/ ld /a icle/ he- lone- banana- p oblem- o - he- new-
p og amming-speaking-ai/.
Hughes, Alex. 2023. Cha GPT: e e y hing you need o know abou OpenAI’s GPT-4
ool. h ps://www.science ocus.com/ u u e- echnology/gp -3.
Isabelle, Pie e, Ma c Dyme man, Geo ge Fos e , Jean-Ma c Ju as, Ellio
Macklo i ch, F ançois Pe aul , Xiaobo Ren & Michel Sima d. 1993. T ansla-
ion analysis and ansla ion au oma ion. In P oceedings o he i h in e na-
82
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
ional con e ence on heo e ical and me hodological issues in machine ansla ion
o na u al languages: Associa ion o compu a ional linguis ics, 201–217.
Kenny, Do o hy. 2007. T ansla ion memo ies and pa allel co po a: Challenges
o he ansla ion aine . In Do o hy Kenny & Kyongjoo Ryou (eds.),
Ac oss bounda ies: In e na ional pe spec i es on ansla ion s udies, 192–208.
Newcas le-Upon-Tyne, UK: Camb idge Schola s Publishing.
Kenny, Do o hy. 2011. The e hics o machine ansla ion. In Sybille Fe ne (ed.),
P oceedings o he 20 h New Zealand Socie y o T ansla o s and In e p e e s na-
ional con e ence. Auckland, New Zealand: New Zealand Socie y o T ansla o s
& In e p e e s.
Kenny, Do o hy. 2018. Machine ansla ion. In Pie s Rawling & Philip Wilson
(eds.), The Rou ledge handbook o ansla ion and philosophy, 428–445. London,
UK: Rou ledge.
Kenny, Do o hy. 2020. Technology and ansla o aining. In Minako O’Hagan
(ed.), The Rou ledge handbook o ansla ion and echnology, 498–515. London,
UK: Rou ledge.
Koponen, Maa i & Ma y Nu minen. 2024. Risk managemen o con en deli e y
ia aw machine ansla ion. In Ma ion Win e s, Sha on Deane-Cox & U sula
Böse (eds.), T ansla ion, in e p e a ion and echnological change, 111–135. Lon-
don, UK: Bloomsbu y.
Ko kon zelos, Ioannis & Sophia Ananiadou. 2022. Te m ex ac ion. In Ruslan
Mi ko (ed.), The Ox o d handbook o compu a ional linguis ics, 2nd edn., 991–
1012. Ox o d, UK: Ox o d Uni e si y P ess.
Lank o d, Séamus, Hai hem A li & Andy Way. 2023. adap NMT: An open-sou ce,
language-agnos ic de elopmen en i onmen o neu al machine ansla ion.
Language Resou ces and E alua ion 57. 1671–1696. DOI: 10.1007/s10579-023-
09671-2.
Ma shman, Elizabe h & Lynne Bowke . 2012. T ansla ion echnologies as seen
h ough he eyes o educa o s and s uden s: Ha monizing iews wi h he
help o a cen alized eaching and lea ning esou ce. In Sé e ine Hubsche -
Da idson & Michał Bo odo (eds.), Global ends in ansla o and in e p e e
aining, 69–95. London, UK: Con inuum.
McEne y, Tony. 2022. Co po a. In Ruslan Mi ko (ed.), The Ox o d handbook o
compu a ional linguis ics, 2nd edn., 494–507. Ox o d, UK: Ox o d Uni e si y
P ess.
Melby, Alan K. & Sue Ellen W igh . 2023. T ansla ion memo y. In Chan Sin-
wai (ed.), Rou ledge encyclopedia o ansla ion echnology, 2nd edn., 766–781.
London, UK: Rou ledge.
83

Lynne Bowke
Moo kens, Joss. 2022. E hics and machine ansla ion. In Do o hy Kenny (ed.),
Machine ansla ion o e e yone, 121–140. Be lin, Ge many: Language Science
P ess.
Na la, Akhila, B e Kup el, Ka i a Sa in, Robe o No oa & Jus in Ko. 2018. Au-
oma ed classi ica ion o skin lesions: F om pixels o p ac ice. Jou nal o In es-
iga i e De ma ology 138(10). 2108–2110. DOI: 10.1016/j.jid.2018.06.175.
Na ional Academies o Sciences, Enginee ing, and Medicine. 2017. Communica -
ing science e ec i ely: A esea ch agenda. Washing on DC, USA: The Na ional
Academies P ess. DOI: 10.17226/23674. h ps://nap.na ionalacademies.o g/
ca alog/23674/communica ing-science-e ec i ely-a- esea ch-agenda.
Ribei o, Ma co T., Samee Singh & Ca los Gues in. 2016. Why should I us you?
Explaining he p edic ions o any classi ie . In P oceedings o he 22nd ACM
SIGKDD In e na ional Con e ence on Knowledge Disco e y and Da a Mining:
Associa ion o compu ing machine y, 1135–1144. New Yo k, USA. DOI: 10.1145/
2939672.2939778.
Ridley, Michael. 2025. Human-cen e ed explainable a i icial in elligence: An an-
nual e iew o in o ma ion science and echnology (ARIST) pape . Jou nal
o he Associa ion o In o ma ion Science and Technology 76(1). 98–120. DOI:
10.1002/asi.24889.
Robe s, Michael, De ek D iggs, Ma hew Tho pe, Julian Gilbey, Michael Ye-
ung, S ephan U sp ung, Angelica I. A iles-Ri e o, Ch is ian E mann, Ca hal
McCague, Lucian Bee , Jona han R. Wei -McCall, Zhongzhao Teng, Aix-
Co ne E ossyni Gk ania-Klos as, James H. F. Rudd, E is Sala & Ca ola-
Bibiane Schönlieb. 2021. Common pi alls and ecommenda ions o using ma-
chine lea ning o de ec and p ognos ica e o COVID-19 using ches adio-
g aphs and CT scans. Na u e Machine In elligence 3. 199–217. DOI: 10.1038/
s42256-021-00307-0.
Senn ich, Rico, Ba y Haddow & Alexand a Bi ch. 2016. Imp o ing neu al ma-
chine ansla ion models wi h monolingual da a. In P oceedings o he 54 h
annual mee ing o he Associa ion o Compu a ional Linguis ics ( olume 1): As-
socia ion o compu a ional linguis ics, 86–96.
Sima d, Michel. 2020. Building and using pa allel ex o ansla ion. In Minako
O’Hagan (ed.), The Rou ledge handbook o ansla ion and echnology, 78–90.
London, UK: Rou ledge.
Siu, Sai Cheong. 2024. Re olu ionising ansla ion wi h AI: Un a elling neu-
al machine ansla ion and gene a i e p e- ained la ge language models. In
Yuhong Peng, Huihui Huang & De eng Li (eds.), New ad ances in ansla ion
echnology, 29–54. Singapo e: Sp inge . DOI: 10.1007/978-981-97-2958-6_3.
84
4 Teaching ansla ion s uden s abou da a in he age o gene a i e AI
Vanmassenho e, E a, Ch is ian Ha dmeie & Andy Way. 2018. Ge ing gende
igh in neu al machine ansla ion. In P oceedings o he 2018 Con e ence on
Empi ical Me hods in Na u al Language P ocessing, 3003–3008. B ussels, Bel-
gium: Associa ion o Compu a ional Linguis ics.
Wang, Hai eng, Hua Wu, Zhongjun He, Liang Huang & Kenne h W. Chu ch. 2022.
P og ess in machine ansla ion. Enginee ing 18. 143–153. DOI: 10.1016/j.eng.
2021.03.023.
Way, Andy. 2013. T adi ional and eme ging use-cases o machine ansla ion.
In P oceedings o T ansla ing and he Compu e 35. London, UK: ASLIB. h ps:
//aclan hology.o g/2013. c-1.12/.
Way, Andy. 2020. Machine ansla ion: Whe e a e we a oday? In E ik Angelone,
Mau een Eh ensbe ge -Dow & Ga y Massey (eds.), The Bloomsbu y companion
o language indus y s udies, 311–332. London, UK: Bloomsbu y.
Way, Andy. 2025. Wha does he u u e hold o ansla ion echnologies? In S e-
an Baumga en & Michael Tiebe (eds.), The Rou ledge handbook o ansla ion
echnology and socie y, 448–461. London, UK: Rou ledge.
Zane in, Fede ico. 2023. Conco dancing. In Chan Sin-wai (ed.), Rou ledge ency-
clopedia o ansla ion echnology, 2nd edn., 498–511. London, UK: Rou ledge.
85
Chap e 5
Teaching ansla ion wi h AI: B idging
heo y and p ac ice h ough p omp
enginee ing
Masa u Yamada
Rikkyo Uni e si y, Japan
This chap e explo es he inno a i e applica ion o la ge language models (LLMs)
in ansla o aining, ocusing on he use o ew-sho p omp s and chain-o -
hough p omp ing. I p oposes a no el app oach ha in eg a es me alanguages
and concep s o T ansla ion S udies in o p omp enginee ing, mo ing beyond
adi ional na u al language p ocessing goals o imp o ing machine ansla ion
quali y. The chap e demons a es how his me hod can c ea e in e ac i e and
engaging lea ning expe iences o ansla ion s uden s, allowing hem o explo e
a ious ansla ion s a egies and de elop c i ical hinking skills. Th ough con-
c e e examples, he chap e illus a es he po en ial o LLMs o gene a e di e se
ansla ion a ia ions and p o ide insigh ul analyses o ansla ion p ocesses.
While acknowledging limi a ions and he need o c i ical e alua ion, he esea ch
emphasises he posi i e and p oac i e possibili ies o LLMs in ansla o aining.
This app oach no only b idges he gap be ween ansla ion heo y and p ac ice
bu also opens new a enues o au onomous lea ning and he de elopmen o
essen ial skills o u u e ansla o s in he AI e a.
1 In oduc ion
Recen ad ancemen s in a i icial in elligence (AI), pa icula ly he eme gence o
La ge Language Models (LLMs), ha e gene a ed ex ensi e deba e ega ding hei
bene i s and d awbacks. In esponse, he UK’s Ins i u e o T ansla ion and In e -
p e ing (ITI) has a icula ed he Slow T ansla ion Mani es o (ITI 2024). D awing
Masa u Yamada. 2026. Teaching ansla ion wi h AI: B idging heo y and p ac ice
h ough p omp enginee ing. In JC Pene , Joss Moo kens & Masa u Yamada (eds.),
Teaching ansla ion in he age o gene a i e AI: New pa adigm, new lea ning?, 87–104.
Be lin: Language Science P ess. DOI: 10.5281/zenodo.17641072
Masa u Yamada
eme and may lack ealism, i can be likened o he speci ic s yle and oice used
in a pa icula company’s use manuals in a eal-wo ld con ex .
The pu pose o his app oach is wo old: o obse e how he LLM adap s i s
ou pu based on he p o ided ansla ion memo y, and o encou age lea ne s o
c i ically analyse how hey can lea n om and apply speci ic linguis ic ea u es,
ones, and s yles in hei own ansla ion wo k. This exe cise, while using an
exagge a ed example, aims o heigh en awa eness o he impo ance o adap ing
language o speci ic con ex s and audiences in ansla ion p ac ice.
This p epa ed sample se was hen gi en as a ew-sho p omp o he LLM,
ins uc ing i o ansla e a new sou ce ex ollowing he co pus. The LLM used
was Claude 3.5 Sonne . To p o ide a compa ison, he new sou ce ex was also
ansla ed using he MT se ice DeepL. Subsequen ly, he LLM ansla ions and
he MT ansla ions we e compa ed agains he o iginal co pus. Addi ionally,
e alua ions we e conduc ed using he au oma ic e alua ion ool COMET2The
esul s o hese compa isons and e alua ions a e p esen ed below.
LLM P omp 5.1:
Below a e p o ided he English ex [sou ce ex ], in which an elde ly
indi idual li ing in con empo a y Ame ica e lec s on hei li e, and i s
Japanese ansla ion [ a ge ex ]. Based on his co pus, please ansla e
he [new sou ce ex ] in o Japanese.
[sou ce ex ] “When I was young, li e was so di e en . We didn’ ha e
all hese gadge s and echnology. We had o wo k ha d o e e y hing
we had. Bu , hose we e good imes oo. I emembe walking miles o
school, playing ou side un il dusk, and e e yone knew each o he in he
communi y.”
[ a ge ex ] 「若い頃はのう、生活はまったく違っとったんじゃ。今みたい
なガジェットやテクノロジーはなかったんじゃ。何でも手に入れるために一生
懸命働かなければならんかったんじゃ。でも、それもまた良い時代だったんじ
ゃよ。学校まで何マイルも歩いて、夕暮れまで外で遊んで、みんながコミュニ
ティの中でお互いを知っとったんじゃ。」
2We used WMT20-COMET o his e alua ion.
94

5 Teaching ansla ion wi h AI
[new sou ce ex ] “I miss hose simple days. We didn’ ha e much, bu
we we e happy. I spen my summe s helping my a he on he a m and
lea ning abou li e om him. Sundays we e always special, wi h amily
ga he ings and big dinne s. As I g ew olde , he wo ld changed apidly,
and some imes i ’s ha d o keep up. Bu I che ish hose memo ies, hey
keep me g ounded.”
Figu e 1: Claude s. DeepL.
Acco ding o he COMET sco es in Figu e 1, Claude’s sco es ange om 0.360
o 0.995, while DeepL’s sco es ange om 0.633 o 1.051. These sco es indica e
gene ally high-quali y ansla ions, as sco es close o 1.0 ypically e lec s ong
alignmen wi h he e e ence ex . On a e age, DeepL achie es highe sco es a
0.626 compa ed o Claude a 0.499.
Howe e , when e alua ed by a human ansla o , a s a k s ylis ic di e ence
becomes e iden be ween he wo. Claude T ansla ion skil ully cap u es he dis-
inc i e elde ly speech pa e ns ound in he e e ence ansla ions. Using a small
co pus, i c ea es he imp ession ha he same elde ly pe son is con inuing he
con e sa ion seamlessly. In con as , DeepL T ansla ion employs a comple ely
s anda d Japanese one, making i appea as hough a di e en pe son is speak-
ing, he eby dis up ing he low o he monologue. This disc epancy highligh s a
limi a ion o he COMET sco ing sys em: i ails o accoun o s ylis ic ea u es
o cul u al nuances such as hose p esen in Japanese “elde ly speech pa e ns”.
95
Masa u Yamada
In an educa ional con ex , i is c ucial o human ansla o s o i s an ici-
pa e he linguis ic ea u es ha can be lea ned om he co pus p o ided in he
p omp , and hen compa e hese expec a ions wi h he LLM’s esponse. Addi-
ionally, compa ing wi h MT allows e lec ion on how hese p edic ions di e
om ac ual ou pu s. When COMET sco es a e a ailable, i may be bene icial o
c i ically analyse whe he hese sco es accu a ely e lec ou p edic ions. As men-
ioned abo e, in his case, i appea ed ha COMET did no adequa ely e lec he
si ua ion, so we decided o ques ion an LLM (in his ins ance, Cha GPT) abou
why his migh be he case. The esponse we ecei ed is as ollows:
LLM Response 5.1:
You obse a ion abou he one and manne o speech in ansla ions
highligh s an impo an aspec o ansla ion quali y ha COMET migh
no ully cap u e. He e’s why:
1. Li e al Accu acy s. S ylis ic Fideli y: COMET is p ima ily designed
o e alua e he li e al accu acy and luency o ansla ions. I does well
in judging how close he MT is o he e e ence in e ms o meaning and
linguis ic quali y.
2. S ylis ic Nuance: While COMET can cap u e some aspec s o luency
and na u alness, i migh no ully app ecia e s ylis ic nuances such as
he speci ic one, oice, o manne o speech ha a e p esen in he
e e ence ansla ion. Fo example, an elde ly man’s dis inc i e speech
pa e ns migh no be ully alued in COMET’s sco ing i he e e ence
doesn’ explici ly indica e he impo ance o hese cha ac e is ics.
3. Con ex ual and Cul u al Sensi i i y: Human e alua o s can ecog-
nise and app ecia e sub le ies such as one, cul u al e e ences, and
s ylis ic consis ency. Au oma ed me ics like COMET a e s ill limi ed in
hei abili y o assess hese elemen s comp ehensi ely.
(I should be no ed ha such ca ego ies a e analy ical cons uc s used
by human e alua o s; he LLM i sel does no “unde s and” o delibe a ely
apply hem.)
96
5 Teaching ansla ion wi h AI
This ype o ac i i y can p o ide a no el ansla o aining expe ience. Fi s ,
lea ne s unde s and he ela ionship be ween ew-sho p omp s and pa allel co -
po a, lea ning ha LLMs can gene a e simila ansla ions om hem. While he
abo e example ocuses on one and manne , i could also be in e es ing o in es-
iga e whe he LLMs can eplica e aspec s such as e minology and g amma ical
s uc u es. As no ed in ela ed wo k, simila a emp s ha e been made in he
con ex o me apho (Do s 2024). To achie e his, i migh be necessa y no
only o show pa allel co po a bu also o add speci ic p omp s ha supplemen
wha aspec s o he language should be lea ned om hem. He e, he p ac ical
applica ion o TS concep s and p o essional ansla ion conside a ions can be
implemen ed in he o m o p omp s.
Mo eo e , his app oach encou ages c i ical hinking abou ansla ion qual-
i y assessmen . By compa ing human e alua ions wi h au oma ed me ics such
as COMET, lea ne s can de elop a deepe unde s anding o he complexi ies
in ol ed in ansla ion e alua ion. This speci ic exe cise highligh s he impo -
ance o conside ing bo h quan i a i e me ics and quali a i e aspec s such as
s yle, one, and cul u al nuances in assessing ansla ion quali y. I also unde -
sco es he cu en limi a ions o au oma ed e alua ion ools and he con inued
ele ance o human expe ise in ansla ion assessmen . Fu he mo e, engaging
wi h LLMs o analyse disc epancies be ween human and au oma ed e alua ions
can p o ide aluable insigh s in o he s eng hs and limi a ions o di e en e al-
ua ion me hods, os e ing a mo e comp ehensi e and nuanced app oach o ans-
la ion quali y assessmen in bo h educa ional and p o essional con ex s.
6 CoT p omp ing and he ansla ion p ocess
In his sec ion, a simple CoT p omp was used, employing a axonomy o ans-
la ion s a egies. Ra he han ec ea ing he ansla ion p ocess h ough a CoT
p omp , in o he wo ds, no a s ic s ep-by-s ep p omp , i acili a ed a ela i ely
high- esolu ion cogni i e p ocess conce ning ansla ion. Speci ically, he well-
known p ocedu es desc ibed by Vinay & Da belne (1958/1995) we e p o ided
o he LLM. The de ini ions we e gene a ed by Cha GPT-4o. Subsequen ly, he
model was ins uc ed o ansla e a Japanese sen ence (p e iously used in an ex-
ample) by applying all he p ocedu es om he p o ided axonomy. Le us i s
examine he ollowing p omp :
97
Masa u Yamada
LLM P omp 5.2:
T ansla e he English sen ence “When I was young, li e was so di e en ”
in o Japanese, adhe ing o he p o ided axonomy below.
Di ec T ansla ion Me hods
1. Bo owing: Di ec ly aking wo ds om he sou ce language in o he
a ge language wi hou ansla ion (e.g., “pizza”).
2. Calque: T ansla ing a o eign wo d o ph ase li e ally, c ea ing a new
exp ession in he a ge language (e.g., “skysc ape ” as “g a e-ciel” in
F ench).
3. Li e al T ansla ion: Wo d- o -wo d ansla ion ha adhe es closely o
he o iginal syn ax and meaning, app op ia e when he languages sha e
simila s uc u es and cul u al con ex s.
Oblique T ansla ion Me hods
4. T ansposi ion: Changing he g amma ical ca ego y o a wo d wi hou
al e ing i s meaning (e.g., ans o ming a noun in o a e b).
5. Modula ion: Changing he o m o he message, in oducing a shi in
poin o iew o cogni i e ca ego y o i he na u al exp ession in he
a ge language (e.g., “I ’s no di icul ” ins ead o “I ’s easy”).
6. Equi alence: Using an en i ely di e en exp ession o con ey he same
si ua ion, o en applied in idioma ic exp essions o p o e bs (e.g., “Ou o
sigh , ou o mind” ansla ed as “Loin des yeux, loin du cœu ” in F ench).
7. Adap a ion: Modi ying he cul u al e e ence when he sou ce lan-
guage si ua ion is unknown o un ecognisable in he a ge cul u e (e.g.,
changing a e e ence o Thanksgi ing in he U.S. o a ele an local
holiday in he a ge cul u e).
While such axonomies o ansla ion s a egies a e o en lec u ed on in in o-
duc o y TS cou ses, i was no ealis ic o use classi ica ion ables o discussing
ansla ions in p ac ical ansla ion classes. Especially when eaching ansla ion
in o Japanese, Eu opean-de eloped classi ica ions o en did no di ec ly apply,
necessi a ing he use o cus omised li e a u e o adap a ion o p ac ical use. In
essence, his s a egy classi ica ion axonomy exempli ies how ca ego isa ions
and accumula ions om TS esea ch we e seldom u ilised in p ac ical eaching
classes. F ankly, i was cumbe some. Howe e , his is no o say ha human
98
5 Teaching ansla ion wi h AI
ansla o s do no men ally engage in a p ocess simila o his axonomy, c ea -
ing and selec ing om mul iple ansla ion op ions. No is i o sugges ha such
axonomies a e en i ely wi hou me i . To ei e a e, he p ima y issue was he
ime-consuming na u e o he p ocess.
Gi en his con ex , we p o ided he a o emen ioned p omp and eques ed a
ansla ion. The esul is as ollows. Fo he ansla ion, Claude 3.5 Sonne was
u ilised, consis en wi h he ea lie example.
LLM Response 5.2:
Bo owing: ヤングの頃、ライフはとても違っていた。
Calque: 若かった時、生命は非常に異なっていた。
Li e al T ansla ion: 若かった頃、人生はとても違っていた。
T ansposi ion: 若い時代には、人生は大きく異なっていた。
Modula ion: 若かった頃は、今とは全く違う人生だった。
Equi alence: 若い頃は、今とは別世界だった。
Adap a ion: 青春時代は、今とは天と地ほど違っていた。
As men ioned ea lie , Vinay and Da belne ’s ansla ion axonomy is chal-
lenging o apply o Japanese ansla ion. The e o e, I was imp essed ha he
LLM could ins an ly ‘unde s and’ his axonomy and c ea e examples adap ed
o Japanese ansla ions. Howe e , unde s anding his classi ica ion in Japanese
emains di icul , so I supplemen ed he p omp by asking o speci ic explana-
ions o “which pa s o he ansla ion each s a egy [ his] is applied o”. Con-
sequen ly, he ollowing de ailed explana ions we e ob ained. Due o space con-
s ain s, only he i s h ee esul s a e p esen ed he e.
LLM Response 5.3:
1. Bo owing: “Yangu no ko o, ai u wa o emo chiga e i a.”(ヤングの頃、
ライフはとても違っていた。) - The English wo ds “young” and “li e” a e
bo owed di ec ly in o Japanese as “yangu” and “ ai u” espec i ely.
2. Calque: “Wakaka a oki, seimei wa hijou ni ko ona e i a.”(若かった
時、生命は非常に異なっていた。) - “Li e” is ansla ed as “seimei” (生命),
99

Masa u Yamada
which is a mo e li e al ansla ion o “li e” as in “li ing being”, a he
han he mo e common “jinsei” (人生) meaning “li e ime” o “cou se o
li e”. - “So di e en ” is ansla ed mo e li e ally as “hijou ni ko ona e i a”
(非常に異なっていた).
3. Li e al T ansla ion: “Wakaka a ko o, jinsei wa o emo chiga e i a.”(若
かった頃、人生はとても違っていた。) - The sen ence s uc u e o he o igi-
nal is main ained, wi h each wo d ansla ed o i s Japanese equi alen .
…
In ac ual ansla ion p ac ice, hese s a egies would ypically be com-
bined o c ea e a na u al and e ec i e a ge ex . Some o hese exam-
ples may appea unna u al as hey a e designed o highligh he speci ic
ea u es o each ansla ion s a egy.
While i may seem epe i i e, I mus emphasise how ema kably clea , accu-
a e, and well-explained hese examples a e. The p ecision and co ec ness o
bo h he ins ances and hei desc ip ions a e no ewo hy.
This e i ica ion sugges s he po en ial o LLMs o suppo human ansla-
o s h oughou hei ansla ion p ocess using CoT p omp ing. Howe e , mo e
immedia ely appa en is how such ac i i ies could o e signi ican insigh s o
ansla ion educa ion. The example p esen ed he e, applying a well-known ax-
onomy o ansla ion s a egies o c ea e ansla ion examples, is jus one illus-
a ion o his po en ial.
Beyond his, we migh also conside he applica ion o o he TS asse s, such
as lea ning abou ansla ion issue ypologies. These exe cises p o ide a angible
sense o how heo e ical concep s om TS can be p ac ically applied.
In essence, his app oach no only demons a es he imp essi e capabili ies o
LLMs in unde s anding and applying complex ansla ion heo ies bu also opens
up new possibili ies o ansla ion educa ion. By b idging he gap be ween heo-
e ical knowledge and p ac ical applica ion, i o e s a no el way o engage wi h
TS concep s ha ha e adi ionally been challenging o implemen in p ac ical
aining se ings.
100
5 Teaching ansla ion wi h AI
7 Discussion and concluding ema ks
Fi s ly, i is c ucial o unde s and ha p omp enginee ing is no solely aimed a
imp o ing au oma ic ansla ion quali y based on NLP echnologies o uni ied
alue sys ems. We o en pe cei e AI and LLMs as ad e sa ial o human ansla-
o s. While impo an , his pe spec i e, which leads o a ocus on cau ious edu-
ca ional o p ac ical use – such as copy igh conce ns o e hical conside a ions
– may be o e ly conse a i e. To d aw an ex eme analogy, i is akin o ocus-
ing solely on a oiding acciden s o p oblems when planning an enjoyable ip,
a he han conside ing how o make he jou ney uly ewa ding. This chap e
has speci ically explo ed he posi i e and p oac i e possibili ies o u ilising LLMs
in ansla o aining.
To elucida e hese posi i e applica ions, we ha e examined he use o accumu-
la ed TS asse s and me alanguages as p omp s o LLMs. Howe e , I mus ei e -
a e ha he p ima y aim o his chap e was no o apply TS concep s o p omp s
o he simple goal o imp o ing LLM-gene a ed ansla ion quali y, as is o en
he ocus in NLP. Ins ead, he objec i e was o demons a e how ansla ion me -
alanguages can be used o engage in dialogue wi h AI, po en ially enhancing
ansla ion educa ion.3
Speci ically, we ha e p esen ed conc e e examples o lea ning scena ios us-
ing ansla ion concep s wi hin he amewo ks o ew-sho p omp s and CoT
p omp ing. The p omp u ilising p ocess axonomy o gene a e a ious ansla-
ion a ia ions was pa icula ly su p ising and insigh ul. C i ical analysis and
AI consul a ion on his app oach yielded highly accu a e esponses. While hese
examples a e limi ed and he high accu acy may be pa ly a ibu ed o he use o
he well-known Vinay and Da belne axonomy, hey ep esen jus he i s s ep
in explo ing a as po en ial. As an educa o , I am bo h in igued and obliga ed
o u he in es iga e hese possibili ies.
While hese esul s a e p omising, i is impo an o no e ha his is jus he be-
ginning. Fu he esea ch and expe imen a ion could e eal e en mo e ways in
which LLMs can enhance ansla ion educa ion and suppo p o essional ans-
la ion p ac ices. Mo eo e , his me hod could po en ially make he lea ning p o-
cess mo e in e ac i e and engaging o s uden s, allowing hem o explo e a i-
ous ansla ion s a egies and hei implica ions in a mo e hands-on manne . I
also p o ides a pla o m o discussing he nuances o ansla ion choices, which
is c ucial o de eloping c i ical hinking skills in aspi ing ansla o s. This ap-
3I should be emphasised ha such sys ems do no in ac “unde s and” ansla ion concep s;
a he , hey gene a e ou pu based on s a is ical p edic ions o likely wo d sequences.
101
Masa u Yamada
p oach may o e oppo uni ies o os e au onomous lea ning and de elop es-
sen ial quali ies o independen ansla o s.
I is also impo an o acknowledge some limi a ions. No all esponses om
LLMs we e accu a e, and some p omp s we e less success ul han o he s. Fo
ins ance, while he LLM co ec ly explained why COMET could no p o ide a
ai e alua ion in he ew-sho p omp example, i ailed o gi e a easonable
answe when asked which ansla ion (Claude o DeepL) was close o he e -
e ence ansla ion. Such e o s and limi a ions o LLMs become mo e appa en
wi h inc eased use. Howe e , gi en ha ansla o aining inhe en ly equi es
main aining a c i ical pe spec i e, I belie e explo ing he possibili ies o using
LLMs is as impo an as conside ing he isks.
In conclusion, his chap e has demons a ed conc e e me hods o explo ing
he po en ial o LLMs in ansla ion educa ion. By le e aging he concep s and
me alanguages o TS in p omp enginee ing, we can c ea e mo e engaging, in-
e ac i e, and e ec i e lea ning expe iences o ansla ion s uden s. While chal-
lenges and limi a ions exis , he po en ial bene i s o in eg a ing LLMs in o ans-
la o aining a e signi ican and wa an u he in es iga ion and de elopmen .
Re e ences
Ame ican T ansla o s Associa ion. 2024. ATA s a emen on a i icial in elligence.
h ps://www.a ane .o g/ad ocacy- ou each/a a- s a emen - on- a i icial-
in elligence/.
Do s , Ale a G. 2024. Me apho in li e a y machine ansla ion: S yle, c ea i -
i y and li e a iness. In And ew Ro hwell, Andy Way & Roy Youdale (eds.),
Compu e -assis ed li e a y ansla ion, 173–186. New Yo k, USA: Rou ledge.
DOI: 10.4324/9781003357391-9.
ELIS. 2025. ELIS 2025 Eu opean language indus y su ey. Tech. ep. ELIS Re-
sea ch. 1–53. h p://elis- su ey.o g/wp- con en /uploads/2025/03/ELIS-
2025_Repo .pd .
Eu opean Council o Li e a y T ansla o s’ Associa ions. 2024. No one le behind,
no language le behind, no book le behind. h ps://www.cea l.eu/no-one-le -
behind-no-language-le -behind-no-book-le -behind.
Gao, Yuan, Ruili Wang & Feng Hou. 2023. How o design ansla ion p omp s o
Cha GPT: an empi ical s udy. h ps://a xi .o g/abs/2304.02182.
He, Sui. 2024. P omp ing Cha GPT o ansla ion: a compa a i e analysis o ans-
la ion b ie and pe sona p omp s. h ps://a xi .o g/abs/2403.00127.
102
5 Teaching ansla ion wi h AI
Hendy, Am , Mohamed Abdel ehim, Am Sha a , Vikas Raunak, Mohamed
Gab , Hi okazu Ma sushi a, Young Jin Kim, Mohamed A i y & Hany Hassan
Awadalla. 2023. How good a e GPT models a machine ansla ion? A comp e-
hensi e e alua ion.h ps://a xi .o g/abs/2302.09210.
House, Juliane. 1981. A model o ansla ion quali y assessmen . 2nd edn. Tübin-
gen, Ge many: Gun e Na .
ITI. 2024. Slow ansla ion mani es o. h ps://www.i i.o g.uk/disco e /policy/
slow- ansla ion-mani es o.h ml.
Japan Associa ion o T ansla o s. 2024. S a emen on he public and p i a e sec o
ini ia i e o use AI o high- olume ansla ion and expo o manga. h ps://
p imes.jp/main/h ml/ d/p/000000001.000143535.h ml.
Jiao, Wenxiang, Wenxuan Wang, Jen- se Huang, Xing Wang, Shuming Shi &
Zhaopeng Tu. 2023. Is Cha GPT a good ansla o ? Yes wi h GPT-4 as he engine.
h ps://a xi .o g/abs/2301.08745.
LITHME P ojec . 2021. Language in he human-machine e a. h ps://li hme.eu/.
Miya a, Rei, Masa u Yamada & Kyo Kageu a. 2023. Me alanguages o dissec ing
ansla ion p ocesses: Theo e ical de elopmen and p ac ical applica ions. Lon-
don, UK: Rou ledge. h ps : / / www . ou ledge . com / Me alanguages - o -
Dissec ing-T ansla ion-P ocesses-Theo e ical-De elopmen -and-P ac ical-
Applica ions/Miya a-Yamada-Kageu a/p/book/9781032168951.
Nida, Eugene A. & Cha les R. Tabe . 1969/2003. The heo y and p ac ice o ans-
la ion. Leiden, Ne he lands: B ill.
Papineni, Kisho e, Salim Roukos, Todd Wa d & Wei-Jing Zhu. 2002. BLEU: A
me hod o au oma ic e alua ion o machine ansla ion. In P oceedings o
he 40 h Annual Mee ing on Associa ion o Compu a ional Linguis ics, 311–318.
Philadelphia, USA: Associa ion o Compu a ional Linguis ics. DOI: 10.3115/
1073083.1073135. h ps://www.aclweb.o g/an hology/P02-1040.
Rei, Rica do, C aig S ewa , Ana C. Fa inha & Alon La ie. 2020. COMET: A neu-
al amewo k o m e alua ion. In P oceedings o he 2020 Con e ence on Em-
pi ical Me hods in Na u al Language P ocessing (EMNLP 2020), 2685–2702. On-
line: Associa ion o Compu a ional Linguis ics. DOI: 10.18653/ 1/2020.emnlp-
main.213. h ps://www.aclweb.o g/an hology/2020.emnlp-main.213.
Saye s, Da e, Rui Sousa-Sil a, S ia lana Höhn, Lule Ahmedi, Kais Allki i-
Me soja, Dimi a Anas asiou, Š e an Beňuš, Lynne Bowke , Elio By yçi, Ale-
jand o Ca ala, Anila Çepani, Rubén Chacón-Bel án, Sami Dadi, Fisnik Dalipi,
Vladimi Despo o ic, Agnieszka Doczekalska, Sebas ian D ude, Ka ën Fo ,
Robe Fuchs, Ch is ian Galinski, Fede ico Gobbo, Tunga Gungo , Siwen Guo,
Klaus Höckne , Pe alea Láncos, Tome Libal, Tommi Jan unen, Dewi Jones,
Blanka Klimo a, Emine kan Ko kmaz, Sepesy Maučec Mi jam, Miguel Melo,
103
Joss Moo kens & Gökhan Doğ u
wi hd awn wi h li le wa ning due o a change o ci cums ances (o geopoli ical
expediencies). A he ime o w i ing, he e is a push o LLM so e eign y, wi h
in e na ional o ganiza ions like he Eu opean Union, na ional go e nmen s and
some companies in es ing in supe compu e s and building hei own LLMs. O
cou se, no all LMs ha e o be huge, bu he ocus o US big ech p o ide s
con inues o be scale, c ea ing la ge and la ge models ha ake longe o ain,
wi h knock-on e ec s ega ding en i onmen al sus ainabili y.
The p e aining and deploymen o ounda ional LLMs is a esou ce-in ensi e
p ocess ha equi es signi ican capi al in es men o acqui ing he necessa y
ha dwa e, aining da a and human alen as well as ope a ing cos s, and each
la ge-scale p e- aining cycle has signi ican en i onmen al impac s due o high
elec ici y and wa e consump ion. Combined, hese ac o s allow only a ew
companies o ha e e y powe ul LLMs. Since many na ional go e nmen s, la ge
co po a ions and public o ganiza ions conside LLMs and AI in gene al as s a e-
gic enable s o hei u u e, hey a e inc easing hei in es men in his a ea as
well and se ing big a ge s o b oade AI use. Howe e , he impac o his AI
‘a ms ace’ on he en i onmen is likely o be ca as ophic on ou plane i e e y
en i y ies o de elop hei own p op ie a y LLM. In ecen yea s, AI sys ems
ha e in gene al become mo e e icien , bu he end o scale means quali y im-
p o emen is based on mo e da a and pa ame e s, equi ing mo e aining ime.
Luccioni e al. (2025) a gue ha inc eased e iciency in AI aining and in ime
sa ed by using AI will end o spu mo e use o AI, con inuing an upwa d end
no only in emissions, bu in wa e use o da a cen es, a e ea h me als used
o ICT, and mo e ha m ul elec onic was e.
These en i onmen al ha ms and he impac s di ec ly om AI a e di icul o
measu e exac ly. Emission le els may di e depending on whe he ene gy comes
om ossil uel o enewable sou ces (Sh e iono & Vanmassenho e 2022) o
depending on he ime o day and da a cen e loca ion (Dodge e al. 2022). Big
ech i ms a e looking o buy up enewables and nuclea powe sou ces as hey
come online, in o de o maximise hei ne -ze o c eden ials. Wa e oo p in s
a e also likely o di e depending on ime and egion (Li e al. 2025). P esen ly,
GenAI does no equi e a la ge p opo ion o esou ces, bu p ojec ions, such as
hose om he head o he UK Na ional G id p edic ing an AI-d i en six- old
inc ease in powe equi emen s o da a cen es in he nex decade (BBC 2024),
a e wo ying. Fo now, ene gy and wa e use a ibu able o GenAI a e e y small
in compa ison o he huge equi emen s o wa ching s eaming media o joining
a Zoom mee ing (My on e al. 2024).
110

6 Teaching AI e hics o ansla ion s uden s
4 Possible solu ions
Posi ioned wi hin he in e na ional AI ‘a ms ace’, wi h poli ically-mo i a ed
suppo o some de elope s leading o p e e ed companies ecei ing p e e en-
ial access o go e nmen s and cu ing-edge echnologies, he e a e also mo e-
men s o use ‘AI o Good’. One pla o m by ha name is led by he In e na ional
Telecommunica ion Union o he Uni ed Na ions (UN) o use AI o achie e UN
sus ainable de elopmen goals, and he e a e smalle ini ia i es such as he Dis-
ibu ed A i icial In elligence Resea ch Ins i u e ha looks o push back agains
he in luence o big ech on AI esea ch, de elopmen and deploymen . Many
people seek o use (Gen) AI o educe ha ms and inequali y. The e ha e been
p oposed uses o AI o ou e powe use o maximise he use o enewable ene gy
and o imp o e ene gy e iciency in he design, building, and use o comme cial
buildings (Ding e al. 2024).
Van Wynsbe ghe (2021) eels ha he e a e wo mo i a ions behind ‘AI o
sus ainabili y’ and ‘sus ainabili y o AI’ ha ough o be combined. The o me
seeks o do good, ye migh en ail nega i e en i onmen al impac s, whe eas he
la e acknowledges ha o AI o be sus ainable, he e needs o be lowe en i-
onmen al cos s o AI aining, uning and in e ence. She de ines sus ainable AI
as a necessa y mo emen o “ os e change in he en i e li ecycle o AI p oduc s
(i.e. idea gene a ion, aining, e- uning, implemen a ion, go e nance) owa ds
g ea e ecological in eg i y and social jus ice” (Van Wynsbe ghe 2021: 217).
Na ional and in e na ional legisla ion, mos no ably he EU AI Ac , seek o
limi ha m ul uses o AI. The AI Ac de ines a ypology o ie ed uses o AI based
on isk, wi h ‘unaccep able isk’ uses o bidden and high- isk uses, such as e-
c ui men decision-making and job alloca ion, subjec o special egula ion. This
ende s algo i hmic job alloca ion in he EU illegal, al hough i ’s likely ha he
ecommenda ion o an au oma ed p ojec managemen ool will s ill be ollowed.
In a posi ion pape by Moo kens e al. (2024) we bo owed he idea o a
iple bo om line om Business E hics (Elking on 1997) o p opose ha ans-
la ion echnologies and LLMs be e alua ed no jus ocusing on pe o mance,
bu a he gi ing equal weigh o people, plane and pe o mance. This is nec-
essa ily a heu is ic a he han an exac me ic, bu ollows on om c i icisms
o a ocus pu ely on pe o mance pushing AI de elopmen in he w ong di ec-
ion by Schwa z e al. (2020) and o he s. Fo people, we migh conside how
he LLM impac s anno a o s, ansla o s, pla o m wo ke s, and hose who ha e
been dispossessed o hei da a, balancing hese agains bene i s o people. Fo
he plane , we migh look a ene gy cos s/CO2, e icien models, and ICT cos
111
Joss Moo kens & Gökhan Doğ u
and disposal. Finally, o pe o mance, we should use ask-app op ia e and com-
p ehensi e s anda ds.
A p e ious sugges ion o ansla ion da a om Moo kens & Lewis (2019) was
a communi y-owned and managed digi al commons, ollowing he ideas o Os-
om (2011), wi h ie ed access a ailable o a cos . We a gued ha his would
help o “sus ain he occupa ion o ansla ion and o minimise he po en ial isks
and ha ms o ansla o s and he public” (Moo kens & Lewis 2019: 17). This idea
seems o be simila o he in ended implemen a ion o a Eu opean Language Da a
Space as a decen alised ma ke place o ex , ideo and audio da a in di e en
languages (Rehm e al. 2024).
The ad ice om some esea che s, such as Rudin (2019), is o en i ely a oid
black-box, subsymbolic sys ems o high- isks uses. This is because opaci y is
la gely baked in o subsymbolic AI sys ems, as desc ibed p e iously, al hough
Rudin and o he s belie e ha , in many cases, compa able esul s may be achie ed
wi h mo e anspa en , hyb id sys ems.
Fo e y la ge, closed sou ce sys ems, we do no know wha aining da a has
been used o wha he RLHF gua d ails a e. Howe e , no all LLMs a e closed
sou ce – o e en ha la ge. We men ioned he al e na i e op ions o small-scale
LMs in Sec ion 2. These can be use ul o na owly-de ined asks, wi h bene-
i s o low cos , low en i onmen al impac , and cus omisabili y. Fo echnically
con iden s uden s, guidelines o building a cus om small LM a e p o ided by
Moo kens e al. (2025).
In academic esea ch e hics, he c edo has mo ed on om ‘do no ha m’ o
he need o ac ually bene i esea ch pa icipan s. Rela edly, bes p ac ice o en-
gaged esea ch in ol es a pa icipa o y app oach, wo king coope a i ely, pa -
icula ly wi h ma ginalised g oups, as co-c ea o s o knowledge a he han im-
posing na a i es o pu ing wo ds in o hei mou hs. Bi hane e al. (2022) p o-
pose his app oach o building AI sys ems, o e ing examples o pa icipa o y
app oaches ha aim o lessen exis ing imbalances o powe . In his way, de elop-
e s “acknowledge ha he communi ies and publics beyond echnical designe s
ha e knowledge, expe ise and in e es s ha a e essen ial o he de elopmen o
AI ha aims o s eng hen jus ice and p ospe i y” (Bi hane e al. 2022: 7).
5 T ansla ion and AI e hics in he class oom
5.1 Class oom discussion ac i i ies
As s uden use s o ansla ion echnology – and mos likely use s o ela ed
AI ools and cloud se ices mo e b oadly – s uden s in a ansla ion class oom
112
6 Teaching AI e hics o ansla ion s uden s
a e al eady pa o he in e connec ed web o e hical issues om he p e ious
sec ion. They may no ha e gi en hem much hough , as so much o he hype
abou GenAI ocuses on i s posi i e po en ial a he han e hical issues. GenAI
is also en angled wi h wha B and & Wissen (2021) call he ‘impe ial mode o
li ing’, h ough which public and p i a e o ganisa ional s a egies and indi idual
li es yles and p ac ices in he Global No h ely on he unlimi ed app op ia ion
o esou ces, a disp opo iona e claim o global and local ecosys ems, and cheap
labou , ideally om dis an loca ions. B yan (2022: 330) concep ualises indi idual
ela ions wi h he clima e c isis as a “ o m o ‘di icul knowledge’, pa icula ly as
i ela es o lea ne s’ sel -implica ion in he condi ions ha a e being add essed”.
We can easonably b oaden his o AI e hics o ainee ansla o s and educa o s.
T ansmission-based lec u es alone seem inapp op ia e o his opic, as di icul
knowledge may aise sensi i i ies; nobody likes o be hec o ed o o eel ha
hei pe sonal e hics a e in ques ion.
In his sec ion, we in oduce wo me hods o s imula ing discussion and e lec-
ion abou ansla ion and AI e hics in he class oom. The i s uses scena ios o
case s udies, placing e hical dilemmas in o amilia con ex s o discussion. The
second d aws om B yan’s (2022) ‘pedagogy o he implica ed’, which seeks o
p omp c i ical e lec ion abou ou own posi ioning as ‘implica ed subjec s’ and
o os e agency o change. B yan uses he no ion o he implica ed subjec om
Ro hbe g (2019) o look beyond dicho omies o indi idual e sus ins i u ional e-
sponsibili y o injus ices o a discussion o how we a e enmeshed wi h sys ems
in many ways ac oss his o ical (diach onic) and con empo a y social-s uc u e
(synch onic) lines.
5.2 Case s udies
Acco ding o Benbunan-Fich (1998), a combina ion o lec u es and discussion
a e complemen a y ways o in oducing e hical issues in he class oom. To be-
gin wi h, lec u es abou “e hical concep s can lay he heo e ical ounda ions”
so ha s uden s can “p ac ice e hical analyses” he ea e using case s udies
(Benbunan-Fich 1998: 20). Case s udies ha e p o ed o be a use ul ool, pa ic-
ula ly in business schools, o many yea s. Acco ding o Ba nes e al. (1994),
case s udies ex end lea ning beyond each class, s imula ing deepe insigh s ha
link ac oss classes and modules. Led by ins uc o s wi h app op ia e case s ud-
ies, s uden s will engage and can de elop and a icula e c i ical insigh s. These
come h ough ou pa icula ac o s: si ua ional analysis, ac i e s uden in ol e-
men , a non- adi ional ins uc o ole and he need o ela e analysis and ac ion.
Si ua ional analysis means ha e hical issues a e applied in si u a he han in
113
Joss Moo kens & Gökhan Doğ u
heo y. S uden s a e ac i e, engaged and lea ning wi hou di ec eaching, ene -
gised and challenged by discussions. This pu s he ins uc o in a non- adi ional
ole o acili a o wi hou domina ing he class oom. Ra he han only building
knowledge, case s udies combine analysis and ac ion: he “impo ance o ac ion
in luences he en i e case discussion, which ocuses on he p ac ical and doable”
(Ba nes e al. 1994: 48).
The case s udy chosen mus be app op ia e – Zhou (2022: 400), o example,
ecommends inding “ ep esen a i e ansla ion-e hical cases”. The e a e wo
ha may be used o adap ed o GenAI om Moo kens (2022) in he open ac-
cess book Machine T ansla ion o E e yone. Case s udies may be accompanied
wi h a gene al ques ion o mo i a e discussion (e.g. ‘Wha e hical issues do you
see in his scena io?’) o a case wo kshee ea u ing a numbe o ques ions o
asks. Some examples ha could be included a e:
• Desc ibe he key e hical issues and he s akeholde s in ol ed.
• Wha e hical heo ies migh p o ide guidance in his scena io?
• Can you hink o guidelines om a ansla o code o e hics ha could be
use ul in his scena io?
• A e he e any legal issues in ol ed in his scena io?
• Wi h he abo e ques ions in mind, a e he e any ac ions ha you p opose
o esol e his scena io in an e hical manne ?
S uden s may add ess hese ques ions in g oups o as pa o a plena y discus-
sion mode a ed by he ins uc o . A inal ecap, making no e o any new o un-
expec ed issues ha we e b ough up by s uden s, can e ec i ely summa ise a
class and ein o ce key lessons.
5.3 Pedagogy o he implica ed
The app oach p oposed by B yan (2022) in he a icle on ‘pedagogy o he impli-
ca ed’ could wo k in combina ion wi h o ins ead o case s udies. As men ioned,
B yan d aws on Ro hbe g (2019) o he desc ip ion o (mos o ) us as “implica ed
subjec s”, who can be “bene icia ies and pe pe ua o s o sys ems ha a e no o
[ou ] own making o ha [we] ha e no di ec ances al a achmen o” (B yan
2022: 338, i alics in o iginal). A p eamble explaining his complex en anglemen
o implica ion is a aluable p ecu so o discussion in he class oom. The p omp
114
6 Teaching AI e hics o ansla ion s uden s
o his discussion could come om ca e ully chosen images o a mapping exe -
cise o isualise hese en angled ela ionships.
B yan’s (2022: 341) a icle is ocused on clima e change educa ion, o which
he e is a endency owa ds wha she calls an “apocalyp ic sublime aes he ic”.
This aes he ic, she a gues, “posi ions iewe s as me e oyeu s o passi e spec a-
o s – a he han ac i e agen s o implica ed subjec s – in he un olding chaos”.
Images o AI a e pe haps gene ally less apocalyp ic, bu a e simila ly unin ol -
ing. These science- ic ion-inspi ed images will be amilia o hose who ha e
seen many isualisa ions o AI ea u ing s ylised obo s wi h inge s o hei
chins among gigan ic ci cui boa ds, concealing any human in ol emen , e hi-
cal dilemmas, o e en any ques ion abou he smoo h ansi ion owa ds wise
and sen ien in elligen obo s.
Dihal & Dua e (2023) summa ise how ypical s ock images o AI a e mislead-
ing o non-expe s, in ha hey hide socie al and en i onmen al impac s, p o-
mo e un ealis ic expec a ions o AI capabili ies, hide human accoun abili y, and
o en p omo e s e eo ypical assump ions abou gende , e hnici y and eligion.
This led hem o s a he websi e Be e Images o AI1 ea u ing a is -c ea ed
images, collages and illus a ions ha o eg ound he issues hidden in egula
s ock images. These images a e highly conside ed and hus make an excellen
s a ing poin o class oom discussion and e lec ion.
B yan p oposes a mapping exe cise as pa o a discussion o sel -implica ion
in a complex in e connec ed consume - ocused sys em, such as ha behind
GenAI, o p omp c i ical sel - e lec ion. The main example ha she uses is wha
she calls he Social Ecology o Responsibili y F amewo k (SERF), inspi ed by
B on enb enne ’s (2009) heo y o he ecology o human de elopmen . B on en-
b enne was “conce ned wi h he mechanisms, p ocesses and condi ions ha
shape indi iduals’ de elopmen and de ised a model ha heo ises he ecip o-
cal in e ac ions ha occu wi hin and be ween di e en nes ed en i onmen s
(o sys ems) which in u n a ec de elopmen al ou comes” (B yan 2022: 338).
The SERF can be a use ul ool o isualise he in e connec ions be ween he
indi idual ac ions o he ‘implica ed subjec ’ wi hin hei immedia e con ex and
la ge social, na ional and in e na ional sys ems. These a e ep esen ed by con-
cen ic ci cles, expanding om he indi idual’s mic osys em (ou immedia e con-
ex ) o ou mac osys em o socie al, poli ical, and cul u al no ms and con ex .
Wi hin hese mic o- and mac osys ems a e he mesosys em, desc ibing poin s o
connec ion be ween di e en le els, and he exosys em o ins i u ions and o gan-
isa ions ha exe in luence wi hou any di ec con ac wi h us as indi iduals.
1h ps://be e imageso ai.o g
115

Joss Moo kens & Gökhan Doğ u
Na a o & Tudge (2023) sugges ha ou mic osys ems should be subdi ided
in o physical and i ual ca ego ies, as ou day- o-day ac i i ies, social oles,
and in e pe sonal ela ions may di e (o blu ) be ween physical and digi ally-
media ed domains. The mesosys em g oups ou mul iple mic osys ems oge he .
We migh conside he ema kably consis en messaging abou echnology and
GenAI coming om he exo- and mac osys ems, why ha migh be, and he
po en ial impac s on ou mesosys em.
Figu e 1 ea u es an adap a ion o B yan’s (2022) SERF, using a laye ed unnel
design and b inging in he idea o he iple bo om line. This could o m a use ul
s a ing poin o class oom discussions o an example o a mapping exe cise.
The in en ion is no o blame indi iduals – he ch onosys em pa o he SERF
makes clea he his o ical con ex o cu en sys ems – bu o empowe he
indi iduals a he cen e, which o cou se include he educa o s as elemen s and
pe pe ua o s wi hin he o e all sys em.
Figu e 1: An adap ed e sion o B yan’s (2022) Social Ecology o Re-
sponsibili y F amewo k wi h some ideas o use o class oom discus-
sion
Ano he use ul image in ended o s imula e e lec ion and discussion o he
116
6 Teaching AI e hics o ansla ion s uden s
ansla o ’s esponsibili ies, wi h he po en ially con lic ing p essu es om e h-
ical codes, business e hics, pe sonal in e es s, and social esponsibili y comes
om Joseph Lambe ’s (2023) book T ansla ion E hics. The sec ion on esponsi-
bili ies and ela ed ques ions place many o he poin s om his chap e wi hin
he con ex o ansla ion wo k.
6 Conclusion
Many cu icula now inco po a e AI li e acy as an impo an compe ence. Chan
& Collo on (2024) ecommend a ine-g ained in e p e a ion o AI li e acy ha is
ailo ed o subjec o p o essional a eas, and a majo pa o his o ansla ion
s uden s is AI e hics in ansla ion. In his chap e , we b ie ly in oduce some o
he issues and deba es ega ding AI e hics, along wi h some posi i e ini ia i es
ha a e ongoing.
We p opose in e ac i e, social cons uc i is me hods o eaching s uden s
abou AI e hics o ansla ion, highligh ing wo po en ial me hods in pa icula .
The i s o hese is he use o ca e ully chosen case s udies wi h accompany-
ing suppo ma e ial o help s uden s alk h ough e hical issues. Benbunan-Fich
(1998: 21) eels ha i is “c ucial o s uden s o p ac ice e hical decision mak-
ing, so when hey ha e e hical decisions o make in he eal business wo ld hey
ha e a amewo k o ollow”. The second sugges ed me hod is an adap a ion o
B yan’s (2022) pedagogy o he implica ed, using images and mindmapping o
isualise and unde s and ou own place wi hin he angle o bene icia ies and
pe pe a o s o he sys em wi hin which GenAI is de eloped and ma ke ed. A
disingenuous endency o highligh indi idual esponsibili y o issues o e hics
and sus ainabili y le s la ge o ganisa ions o he hook, bu hese me hods show
how indi idual and collec i e esponsibili ies a e in e linked. Ou s a us as im-
plica ed subjec s means ha we mus also ake esponsibili y o ou ac ions and
e hical decisions. We hope ha hese me hods will be e ec i e beyond he indi-
idual class, leading o deepe knowledge in s uden s ha can come o hei own
indi idual e hical posi ion on he deba es and issues in AI e hics o ansla ion.
Acknowledgemen
This was w i en wi h he inancial suppo o Resea ch I eland a ADAPT, he
Resea ch I elandCen e o AI-D i en Digi al Con en Technology a Dublin Ci y
Uni e si y [13/RC/2106_P2].
117
Joss Moo kens & Gökhan Doğ u
Re e ences
Abdallah, K is iina. 2011. Towa ds empowe men : S uden s’ e hical e lec ions on
ansla ing in p oduc ion ne wo ks. The In e p e e and T ansla o T aine 5(1).
129–154. DOI: 10.1080/13556509.2011.10798815.
Ba nes, Louis B., C. Roland Ch is ensen & Abby J. Hansen. 1994. Teaching and he
case me hod: Tex , cases, and eadings. 3 d edn. Bos on, USA: Ha a d Business
School P ess.
BBC. 2024. Da a cen e powe use ‘ o su ge six- old in 10 yea s’. h ps://www.bbc.
com/news/ echnology-68664182.
Benbunan-Fich, Raquel. 1998. Guidelines o using case scena ios o each com-
pu e e hics. ACM SIGCAS Compu e s and Socie y 28(3). 20–24. DOI: 10.1145/
298972.298982. h ps://dl.acm.o g/doi/10.1145/298972.298982.
Bende , Emily M. 2011. On achie ing and e alua ing language independence in
NLP. Linguis ic Issues in Language Technology 6. DOI: 10.33011/lil . 6i.1239.
Bi hane, Abeba, William Isaac, Vinodkuma P abhaka an, Ma k Diaz, Madeleine
Cla e Elish, Iason Gab iel & Shaki Mohamed. 2022. Powe o he people? Op-
po uni ies and challenges o pa icipa o y AI. In P oceedings o he 2nd ACM
Con e ence on Equi y and Access in Algo i hms, Mechanisms, and Op imiza ion
(EAAMO ’22). A ling on, USA: Associa ion o Compu ing Machine y. DOI:
10.1145/3551624.3555290.
Boyle, Siân. 2025. Re ealed: Google acili a ed Russia and China’s censo ship
eques s. The Obse e . h ps://www. hegua dian.com/wo ld/2025/ eb/15/
google-helped- acili a e- ussia-china-censo ship- eques s.
B and, Ul ich & Ma kus Wissen. 2021. The impe ial mode o li ing: E e yday li e
and he ecological c isis o capi alism. New Yo k, USA: Ve so.
B on enb enne , U ie. 2009. Ecology o human de elopmen : Expe imen s by na-
u e and design. Camb idge, USA: Ha a d Uni e si y P ess.
B own, Pe e , John Cocke, S ephen Della Pie a, Vincen J. Della Pie a, F ede ick
Jelinek, Robe L. Me ce & P. Roossin. 1988. A s a is ical app oach o F ench/
English ansla ion. In P oceedings o he Second Con e ence on Theo e ical and
Me hodological Issues in Machine T ansla ion o Na u al Languages. Pi sbu gh,
USA. h ps://aclan hology.o g/1988. mi-1.19/.
B yan, Aud ey. 2022. Pedagogy o he implica ed: Ad ancing a social ecology
o esponsibili y amewo k o p omo e deepe unde s anding o he clima e
c isis. Pedagogy, Cul u e & Socie y 30(3). 329–348. DOI: 10.1080/14681366.2021.
1977979. h ps://www . and online.com/doi / ull/10.1080 / 14681366.2021.
1977979.
118
6 Teaching AI e hics o ansla ion s uden s
Chan, Cecilia Ka Yuk & Tom Collo on. 2024. Gene a i e AI in highe educa ion:
The Cha GPT e ec . London, UK: Rou ledge.
Dihal, Kan a & Tania Dua e. 2023. Be e images o AI: a guide o use s and
c ea o s. Tech. ep. Camb idge, UK: The Le e hulme Cen e o he Fu u e o
In elligence, We & AI. h ps://blog.be e imageso ai.o g/wp-con en /uploads/
2023/02/Be e -Images-o -AI-Guide-Feb-23.pd .
Ding, Chao, Jing Ke, Ma k Le ine & Nan Zhou. 2024. Po en ial o a i icial in el-
ligence in educing ene gy and ca bon emissions o comme cial buildings a
scale. Na u e Communica ions 15(1). 5916. DOI: 10.1038/s41467-024-50088-4.
h ps://www.na u e.com/a icles/s41467-024-50088-4.
Dodge, Jesse, Taylo P ewi , Remi Tache des Combes, E ika Odma k, Roy
Schwa z, Emma S ubell, Alexand a Sasha Luccioni, Noah A. Smi h, Nicole
DeCa io & Will Buchanan. 2022. Measu ing he ca bon in ensi y o AI in cloud
ins ances. In P oceedings o he 2022 ACM Con e ence on Fai ness, Accoun abil-
i y, and T anspa ency (FAccT ’22), 1877–1894. Seoul, Republic o Ko ea: Asso-
cia ion o Compu ing Machine y. DOI: 10.1145/3531146.3533234.
Elking on, John. 1997. Cannibals wi h o ks: The iple bo om line o 21s cen u y
business. S ony C eek, USA: New Socie y Publishe s.
Fe e o, Tania, Leono Pais & Nuno Rebelo dos San os. 2015. Decen wo k: An
aim o all made by all. In e na ional Jou nal o Social Sciences 4(3). 30–42. h ps:
//www.eu ec.o g/ijoss-a icle-193.
Fı a , Gökhan. 2024. T ansla ion wo k in he digi al economy: Wo king condi ions
o ansla o s on pla o ms and in coope a i es. Uni e si y o Su ey. (Doc o al
disse a ion). h ps://open esea ch.su ey.ac.uk/esplo o/p o ile/g%C3%
B6khan_ i a /ou pu /all?ins i u ion=44SUR_INST.
Gi oux, Hen y A. & Pe e McLa en. 1997. Paulo F ei e, pos mode nism and he
u opian imagina ion: A Blochian eading. In Jamie Owen Daniel & Tom Moy-
lan (eds.), No ye : Reconside ing E ns Bloch, 138–162. London, UK: Ve so.
Kellehe , John D., B ian MacNamee & Aoi e D’A cy. 2015. Fundamen als o ma-
chine lea ning o p edic i e da a analy ics: Algo i hms, wo ked examples, and
case s udies. Camb idge, USA: MIT P ess.
Lambe , Joseph. 2023. T ansla ion e hics. London, UK: Rou ledge. DOI: 10.4324/
9781003148265.
Li, Peng ei, Jianyi Yang, Mohammad A. Islam & Shaolei Ren. 2025. Making AI
less “ hi s y”: unco e ing and add essing he sec e wa e oo p in o AI models.
a Xi :2304.03271 [cs]. DOI: 10.48550/a Xi .2304.03271. h p://a xi .o g/abs/
2304.03271.
119
Senem Öne Bulu
1 In oduc ion
The impac o machine ansla ion (MT) on he ansla ion p o ession and ans-
la o educa ion has been p o ound in he las decade, as has he esea ch in o
his impac . The li e a u e has ocused p ominen ly on he ense ela ionship be-
ween he human ansla o and he machine/ echnology, which Picke ing (2008)
concep ualised as a “dance o agency”. Olohan (2011: 354) d ew a en ion o he
“decen ing” o he human agen in his con ex . O’B ien (2012: 119–120) u he
cha ac e ised ansla ion as “human-compu e in e ac ion”, de ining wha was
happening hen as a “shi in pa adigm” and poin ing ou he po en ial o “dia-
logue” be ween human ansla o s and echnology.
Ruokonen & Koskinen (2017: 321) u he elabo a ed on he human-machine
dynamic, wi h a ocus on he emo ional dimension inhe en in his ela ionship.
They ound ha he human agen s we e bo h “willing” and “ eluc an ” in e ms
o assigning agency o he machine and ha he ela ionship be ween he hu-
man and he machine is oo “complex” o be educed o “simpli ied man e sus
machine concep ions”. In a simila ein, Cadwell e al. (2018: 312) obse ed ha
human ansla o s held bo h posi i e and nega i e pe cep ions o MT. Posi i e
pe cep ions ela ed o educed e o and inc eased inspi a ion while he nega i e
we e linked o conce ns abou he nega i e in luence o MT on human c ea i i y.
The au ho s also a gued ha he ‘dance o agency’ be ween he human and he
machine was a symbio ic one, as hey mu ually eed o and shape each o he ;
he “dis inc ion be ween he human and he ma e ial agen ” is blu ed (Cadwell
e al. 2018: 303).
The p ac ical consequences o he pe asi e in eg a ion o MT echnologies
in o he global ansla ion indus y and hei economic impac ha e also been
explo ed. Moo kens (2017), o ins ance, no ed he nega i e consequences o he
eelance employmen model, which has been inc easingly adop ed in he global
ansla ion indus y, in e ms o p ices and isola ion. Acco ding o him, he o-
cus should be on how he human ansla o ac ually wo ks wi h MT. Simila ly,
Viei a (2018: 15) ound ha , a he han eeling an “in insic nega i i y o he
echnology”, he pe cei ed h ea om echnology among ansla o s appea s o
s em om business p ac ices in he indus y.
Ano he ocus in he ela ed li e a u e has been on he need o ede ine he
p o ile o human ansla o s. Sakamo o (2019: 68–69) unde lined he necessi y o
he “ ede ini ion o he concep o ‘ ansla o ’, and consequen ly ‘ ansla ion’ pe
se” on he g ounds ha “ he no ion o ‘ ansla ion’ is now being challenged by
he g ow h o echnologies”. The au ho also no ed ha a en ion should be paid
126

7 Re-posi ioning he human ansla o
o he unce ain ies ela ed o he legal equi emen s in MT use, p icing o pos -
edi ing, and pos -edi o s’ p o iles and skills (Sakamo o 2019: 63–66). Based on a
comp ehensi e analysis o he eme ging needs and equi emen s in he ansla-
ion sec o , which was conduc ed wi h he aim o add essing he “dis ance be-
ween eaching and p o essional p ac ice” in he con ex o MT compe ences,
Gaspa i e al. (2015: 333–334) epo ed “a s ong need o an imp o emen in
quali y assessmen me hods, ools and aining”, unde lining he “g owing up-
ake o MT and he pe cei ed inc ease o i s p e alence in u u e wo k lows”.
The e has been ex ensi e and subs an ial esea ch in o embedding in ans-
la o educa ion he new skills, abili ies and compe encies ha human ansla-
o s should possess in he age o MT, as well as in o he concomi an need o
e-p o ile ansla o educa ion. Fo ins ance, Dohe y & Kenny (2014: 296–297)
designed an SMT syllabus, which o eg ounds he “empowe men ” o he hu-
man ansla o by enabling him/he o in e ene in and add alue o he SMT
wo k lows when acing legal, e hical and echnical obs acles (Kenny & Dohe y
2014: 288). This “empowe ing” app oach has been u he ex ended by Mellinge
(2017: 284), who p oposes a holis ic, “c oss-module o c oss-cu icula in eg a-
ion” o pos -edi ing and MT “ o a oid compa men alisa ion o a ious compe-
ences and skillse s”. Mellinge also sugges s he inclusion o “con olled au ho -
ing, e minology managemen , engine uning, and pos -edi ing” in ansla ion
p ac ice cou ses, as hese a e “ ep esen a i e o many o he skills men ioned
in ecen wo ks on machine ansla ion pedagogy” (Mellinge 2017: 284). Pym
(2019) d ew a en ion o he iden i ica ion o “au oma ion- esis an skills” and
hei in eg a ion in o ansla o educa ion.
The measu es o be aken by ansla o educa o s o empowe s uden s and
help hem become “awa e o hei use ulness in o de o maximise hei agency as
ansla o s” has also been he ocus in a s udy by Moo kens (2018: 375–376), who
designed a ansla ion e alua ion exe cise o enable s uden s o “demys i y NMT
ou pu ”. Based on a econcep ualisa ion o ansla o compe ence in he age o
MT, Öne Bulu (2019: 3) designed a lea ning p ac ice o help s uden s “ aise hei
awa eness o hei p o essional sel -concep as human ansla o s” and sugges ed
he conside a ion o human ansla o compe ence and human ansla ion me a
compe ence. Ni zke e al. (2019: 248) p oposed a no el “pos -edi ing compe ence
model”, which included he co e compe ences o isk assessmen compe ence,
s a egic compe ence, consul ing compe ence and se ice compe ence. Öne &
Öne Bulu (2021: 100) in es iga ed ansla ion s uden s’ pe cei ed di icul ies
and bene i s in he con ex o “pos -edi ing o ien ed neu al machine ansla ion
e o anno a ion and quali y e alua ion”.
127
Senem Öne Bulu
Explo ing he “dynamics o he human-machine dance in he ansla ion class-
oom”, Öne Bulu & Alimen (2023: 377) designed a lea ning expe imen o enable
he s uden s o ca y ou eme gen MT- ela ed asks o pos -edi ing, p e-edi ing
and e o anno a ion. The au ho s epo ed ha he expe imen helped he ma-
jo i y o pa icipa ing s uden s “ aise awa eness o hei sel -concep as human
agen s and o he human added alue hey can c ea e while dancing wi h he
machine” (Öne Bulu & Alimen 2023: 387).
These examples o pas s udies ocusing on he impac o MT on he ansla ion
p o ession and ansla o educa ion show ha ansla ion s udies (TS) esea ch
has indeed come a long way o come o e ms wi h and adap o he pa adigm
shi (O’B ien 2012) induced by he inc easingly all-pe asi e echnologisa ion
o ansla ion, especially since he launch o NMT in 2016.
Ye , he yea 2022 p esen ed bo h ansla o s and ansla o educa o s wi h an-
o he , unp eceden ed challenge. Wi h he launch o Cha GPT in 2022 (OpenAI),
which allows o he p oduc ion o ansla ion h ough a p omp -based dialogue
be ween human and gene a i e a i ical in elligence (GenAI), he ea ly cha ac e -
isa ions o ansla ion as human-compu e in e ac ion (O’B ien 2012) and dance
o agency (Olohan 2011) ha e become eali ies in he ulles sense. This poses new
ques ions and opens new a eas o esea ch o he TS communi y, especially o
educa o s and esea che s o ansla o educa ion.
In hei abo e-ci ed esea ch, Öne Bulu & Alimen (2023: 389) no ed ha , in
he ace o he apid ad ancemen s in MT echnologies and he unce ain ies
conce ning hei in eg a ion in o educa ion, he e-posi ioning o ansla o ed-
uca o s is as necessa y as ha o human ansla o s. I has also been a gued ha
such e-posi ioning in ol es an “awa eness o he ac ha he ansla o educa-
o has he p ima y ask o diagnosing and p omo ing eme gen a eas o human
added alue c ea ion in he MT age”, and ha his can only be achie ed by e-
s uc u ing he ansla ion class oom as a “pla o m o collabo a i e lea ning,
whe e all in ol ed can wo k oge he o disco e wha emains and will emain
human in he MT age” (Öne Bulu & Alimen 2023: 389).
In he ace o he pa adigm shi induced by he in oduc ion o Cha GPT and
o he GenAI ools, ansla o educa o s should once again wo k in collabo a-
ion wi h s uden s o explo e he dynamics o in eg a ing hese ools in o hei
wo k low. Mo i a ed by his e y need, he p esen chap e p esen s he esul s
o a collabo a i e esea ch expe imen designed o explo e he dynamics o he
human-GenAI dialogue and ind he pa hways o be ollowed in inco po a ing
he insigh s gained in o ansla o educa ion.
The emainde o his chap e is s uc u ed as ollows. Fi s , he unp eceden ed
p ac ical and heo e ical challenges posed by GenAI echnology in e ms o he
128
7 Re-posi ioning he human ansla o
e y aison d’ê e o he human ansla ion a e explo ed. Second, in o ma ion
on he design and implemen a ion o he expe imen is gi en. Thi d, he da a
ob ained om he expe imen is analysed. Fou h, a discussion o he esul s is
p esen ed oge he wi h conclusions and sugges ions o u u e esea ch.
2 Human added alue and he human ansla o ’s
sel -concep in he age o GenAI: Re- isi ing he
in e p e i e ask and esponsibili y o he human
ansla o
Cha GPT is a GenAI ool powe ed by a la ge language model (LLM) and has
no been de eloped speci ically o in e lingual ansla ion. Howe e , he e -
olu ion i has b ough o ansla ion is o g ea signi icance as he ool akes
p omp s om human agen s and hence allows hem o p oduce a ansla ion by
p o iding he ool wi h con ex ual in o ma ion and guidance, i.e. by pe o ming
p omp enginee ing (Yamada 2023 and Chap e 5 o his book). This was ce ainly
no he case wi h MT, which makes GenAI echnology a eal b eak h ough ha
now equi es ansla o educa o s o ask and y o answe new ques ions abou
human- echnology in e ac ion.
In my opinion, in his endea ou we should no a emp o ein en he wheel.
Ins ead, we should e isi he ele an exis ing insigh s in o MT, especially hose
ha deal wi h he in eg a ion o MT in o ansla o educa ion, in a way ha
empowe s human ansla o s o c ea e human added alue, as ou lined in he
p e ious sec ion (see In oduc ion). This should also in ol e ying o answe
he challenging ques ion posed by GenAI: wha is human in ansla ion? This
is because, unlike MT, GenAI c ea es he illusion o ha ing an ac ual dialogue
wi h a non-human agen which p e ends o unde s and, in e p e and p oduce a
linguis ic message.
This new o m o challenge should u ge us o e isi he in e p e i e ask and
esponsibili y o he human ansla o . Acco dingly, he app oach p esen ed he e
is pe o ming a backwa d eading. Ra he han gi ing p io i y o he in es iga-
ion o he pe o mance, po en ial and/o limi a ions o GenAI, wha he p esen
s udy p oposes is he ins umen alisa ion o he human-GenAI dialogue in o de
o y o econside wha is/will emain human in ansla ion in a new, b igh e
ligh and ind ways o inco po a e i in o ansla o educa ion in he age o GenAI.
In a seminal con ibu ion by Massey & Ki aly (2019) i led “The Fu u e o T ans-
la o Educa ion: A Dialogue”, Ki aly s a es ha “ anscoding ( he mechanical e-
placemen o linguis ic uni s om a lis wi h co esponding uni s om a pa allel
129
Senem Öne Bulu
lis ) is no a he hea o ansla ion a all” (16, emphasis mine) and ha , al hough
in he u u e he ansla ion p o ession will be “di e en ”, i is “ce ainly no go-
ing o disappea – unless a some poin in ou e olu ion we no longe need o
in e p e ex s” (16, emphasis mine).
I is no coincidence ha Ki aly highligh s he ac o in e p e ing as he aison
d’ê e o he human ansla o , which echoes he he meneu ical app oach applied
o ansla ion by Schleie mache (1813/1977). Acco ding o He mans (2015: 101–
102), Schleie mache belie ed ha he sole op ion o a ansla o is “ o ac as
he he meneu icis does: o wo k o a ain he bes possible unde s anding o he
o eign ex which ne e heless emains o eign, and p esen o he eade […]
exac ly ha unde s anding”. Acco dingly, he meaning o a ex is no encoded in
a ex o be decoded by he ansla o , a he i should be in e p e ed by he ans-
la o o a ain he bes possible unde s anding, which cons i u es he he meneu-
ical ask o he ansla o . The ansla o is hen equi ed o p esen his unde -
s anding o he eade , which, acco ding o He man’s eading o Schleie mache ,
“exceeds he meneu ics” (He mans 2015: 99). In a simila ein, Şe ban & La isa
(2016: 295) unde line ha in Schleie mache ’s app oach, he mee ing be ween
he au ho and he eade is “media ed and o ches a ed by he ansla o ” and
he ansla o assumes “ he mos ac i e media ing ole”.
Al hough he he meneu ical app oach is o c i ical impo ance in ha i allows
he ansla o oom o ac i e media ion, a speci ic concep ualisa ion by A ojo
(1997: 18) o he “ine i able in e e ence” o he ansla o in he con ex o ans-
la ion e hics could u he he a emp o econside wha is human in ansla ion.
A ojo iews ansla ion e hics in e ms o he ela ionship be ween he “au ho-
ial powe ” o he ansla o and his/he “e hical esponsibili y” (A ojo 1997: 18).
Based on a pos mode n iew o “language and subjec ”, he au ho unde lines
ha “[as] no eading can e e aspi e o epea o p o ec someone else’s ex ,
ansla o s necessa ily ha e he igh o exe cise hei au ho ial powe ‘as long
as hei game is played up on ’” (Be man 1995: 93, ci ed in Simon 1996: 36). In
he iew o he au ho , “[s]uch ‘ igh ’ implies, howe e , an e hical esponsibili y
which pa allels ha o ‘o iginal’ au ho s” (A ojo 1997: 18).
In he wake o he eme ging pa adigm b ough abou by GenAI, his line o
hough can be ollowed by esea che s and ansla o educa o s o o eg ound
he human ansla o ’s in e p e i e ask and he accompanying esponsibili y
wi h he aim o explo ing human added alue (Massey 2021) and human ans-
la o s’ sel -concep (Ki aly 1990, 2000b, Ha o-Sole & Ki aly 2019 and Massey &
Ki aly 2019) in he age o GenAI.
D awing ca e ul a en ion o he human aspec o added alue, Massey (2021:
39) cha ac e ises human added alue as being “mani es in ansla ional decision-
130
7 Re-posi ioning he human ansla o
making and p oblem-sol ing on a concep ual le el ha anscends he su ace
lexical ealisa ions by which meaning is con eyed in sou ce and a ge ex s”.
He also obse es ha “ he added alue ha human ansla o s can and do b ing
o bea esides in hei socio-cul u al, socio- echnical 4EA (Embodied, Embed-
ded, Enac i e, Ex ended and A ec i e) cogni ion, as opposed o disembodied, de-
con ex ualised a i icial in elligence” (Massey 2021: 39). In doing so, he au ho
e isi s Venu i’s (2019) ad ocacy o he “he meneu ical model o unde s anding
ansla ion no as ‘ he ep oduc ion o ans e o an in a ian ha is con ained
in o caused by he sou ce ex ’ bu as ‘an in e p e i e ac ha ine i ably a ies
sou ce- ex o m, meaning, and e ec ’” and Pym’s (2003) “minimalis de ini ion
o ansla ion compe ence as ‘ he abili y o gene a e a se ies o mo e han one
iable a ge ex (TTI, TT2 ... TTn) o a pe inen sou ce ex (ST); he abili y o
selec only one iable TT om his se ies, quickly and wi h jus i ied con idence’”
(Massey 2021: 51).
The delinea ion by Massey (2021: 39) o he added alue o he human ansla-
o is-à- is AI s ongly cap u es he concep ualisa ion o he human ansla o
as an agen who has he he meneu ical ask and powe o in e p e ing he ex
and who bea s he esponsibili y o hei in e p e i e ac . This is some hing ha
emains qui e impossible o he “disembodied, de-con ex ualised a i icial in el-
ligence”. In he age o GenAI, we mus he e o e pay unp eceden ed a en ion
o he ansla o ’s “sel -concep ” (Ha o-Sole & Ki aly 2019), which includes “ he
image o he ansla o ’s social ole”, “ he ansla o ’s app aisal o his o he com-
pe ency o ansla ing a pa icula ex ” and “an unde s anding o esponsibili y
owa ds he o he ac o s in he ansla ion con ex o si ua ion” (261, emphasis
mine). Consequen ly, we belie e ha he cons uc s o human added alue and
o (human) ansla o ’s sel -concep should cons i u e he p i ileged oci in ex-
plo ing he human-GenAI dialogue and inco po a ing he gained insigh s in o
ansla o educa ion in a way ha helps s uden s aise hei sel -concep s (Ha o-
Sole & Ki aly 2019: 261–262) as human ansla o s whose aison d’ê e is o pe -
o m he he meneu ical ask o in e p e ing and bea ing esponsibili y. I is wi h
his ocus in mind ha he ollowing collabo a i e esea ch expe imen was con-
duc ed.
3 Design and implemen a ion o he collabo a i e
esea ch expe imen
The esea ch expe imen , he esul s o which a e epo ed in he p esen s udy,
was designed ollowing he ene s o he collabo a i e esea ch model, imple-
131

Senem Öne Bulu
men ed by Ha o-Sole & Ki aly (2019). I ollows he social cons uc i is and
eme gen is app oach o ansla o educa ion (Ki aly 2000a: 256), whe eby he
educa o - esea che wo ks collabo a i ely wi h ansla ion s uden s “ o in es-
iga e opics in he domain o T ansla ion Psychology wi h he goal o ha ing
eache - esea che s lea n abou he ansla o ’s psychological ‘sel ’ igh along
wi h hei s uden s”. Acco dingly, he esea ch expe imen p esen ed he e was
collabo a i ely pe o med by g adua e le el ansla ion s uden s, who pa ici-
pa ed in he esea ch on a olun a y basis, and a ansla o educa o (also he
au ho o he p esen chap e ). The aim was o explo e he dynamics o human-
GenAI in e ac ion as human ansla o s wo k wi h GenAI ools as p omp en-
ginee s. The s uden - esea che g oup consis ed o i e s uden s en olled in he
PhD p og amme in ansla ion s udies a Yıldız Technical Uni e si y, Tü kiye.
All had p e ious ansla ion educa ion and expe ience and nea -na i e English
p o iciency. The educa o - esea che was an associa e p o esso o ansla ion
s udies a he same uni e si y.
In he esea ch p ocess, he s uden - esea che s and he educa o - esea che
collec i ely decided on he design o he ask o p omp enginee ing in h ee
sessions held online in June 2024. A e he design was ag eed upon, he s uden -
esea che s pe o med he ask indi idually and epo ed on hei lea ning/
esea ch p ocesses h ough sel - e lexi e accoun s o hei expe iences.
The ini ial decisions made in he eco ded online discussions conce ned he
selec ion o he sou ce ex s and o he GenAI ool o be used o ansla e he
ex s h ough c a ing and, when necessa y, cu a ing p omp s (i.e. p omp engi-
nee ing; Yamada 2023) so as o p e-p ocess, ini ia e, p oduce, edi and e alua e
ansla ions ha achie ed he in ended quali y le el and unc ions. I was collec-
i ely decided o use he ee e sion o OpenAI’s Cha GPT. As o he sou ce
ex s, wo English ex s, an essay and a book desc ip ion, we e selec ed on he
g ounds ha bo h ex s we e o mixed o ms be ween exp essi e and ope a i e
ex ypes, while also displaying he ea u es o an in o ma i e ex ype (Reiss
1981: 124–125). Fu he , he ansla ion o bo h ex s demanded me iculous a en-
ion o cul u al nuances and con ex , idioma ic exp essions, one, in en , and lin-
guis ic s yle. The i s sou ce ex was an essay by Anna Quindlen, published in
he “Li e in he 30s” column in The New Yo k Times (Quindlen 1987). The second
sou ce ex was he book desc ip ion o The Time Regula ion Ins i u e, he English
ansla ion o Ahme Hamdi Tanpına ’s Tu kish no el i led Saa le i Aya lama
Ens i üsü, ansla ed by Alexande Dawe and Mau een F eely and published in
2014 (Tanpina 2014).
Nex , decisions had o be made conce ning he guidelines, namely he wo
ansla ion b ie s and he gene al p inciples o p omp enginee ing. Fi s , he
132
7 Re-posi ioning he human ansla o
educa o - esea che p o ided he s uden - esea che s wi h he exis ing (albei
limi ed, due o he ac ha i is a ela i ely new opic) li e a u e on p omp en-
ginee ing. This led o an exchange o opinions in he i s wo online sessions.
D awing on hose discussions, he educa o - esea che hen d a ed he guide-
lines o he ask and sen hem o he s uden - esea che s o e alua ion. In he
hi d session, he d a guidelines we e discussed and inalised collec i ely.
The educa o - esea che and he s uden - esea che s hen had o decide
whe he o use he inalised ansla ion b ie s as he i s p omp in he dialogue
wi h he GenAI ool. Taking in o conside a ion he esul s o he expe imen s
al eady conduc ed in o he s udies on p omp enginee ing (e.g. He 2024, Peng
e al. 2023, Yamada 2023), i was decided no o include he ansla ion b ie s in
p omp enginee ing and ins ead, o use hem as guidance o human ansla o s-
p omp enginee s only. In his way, he human ansla o s we e en us ed wi h
he esponsibili y o he ansla ion ac despi e pe o ming his ac h ough
p omp enginee ing. The ansla ion b ie s o he i s and second sou ce ex s
we e as ollows:
(1) T ansla ion B ie 1:
“The Name is Mine” is an essay by Anna Quindlen published in he “Li e
in he 30s” column in The New Yo k Times on Ma ch 4, 1987. An online
magazine websi e edi o in Tu key needs a Tu kish ansla ion o he
essay and asks you o ansla e i in o Tu kish o publish i on he
websi e. The in ended audience o he ansla ion is p ima ily adul
eade s who would be in e es ed in he wo ks o Anna Quindlen, who
w i es p ima ily on eminism and amily li e and is a Puli ze
P ize-winning columnis , jou nalis and au ho . The edi o asks you o
ansla e he second and nin h pa ag aphs o he essay as a es .
(2) T ansla ion B ie 2:
The sou ce ex is he book desc ip ion o he no el The Time Regula ion
Ins i u e, which is he English ansla ion o he p ominen Tu kish
no elis and poe Ahme Hamdi Tanpına ’s Tu kish no el i led Saa le i
Aya lama Ens i üsü (1961), ansla ed by Alexande Dawe and Mau een
F eely and published by Penguin Classics in 2014. A Tu kish li e a y
schola is asking you o ansla e he English book desc ip ion in o
Tu kish in o de o examine how he English ansla ion o he book (The
Time Regula ion Ins i u e) is ma ke ed o he English-speaking audience
by Penguin Classics.
133
Senem Öne Bulu
As o he gene al p inciples o p omp ing, “pe sona p omp ” (He 2024) was
selec ed as p omp ing me hod. This mean ha p omp ing began wi h he s a e-
men “You a e a p o essional ansla o ” wi hou speci ying any speci ic domain.
I was also decided ha p omp enginee ing would include he ollowing s eps:
1) p omp ing he ool o analyse he sou ce ex o he pu pose o ansla ion, 2)
p omp ing he ool o ansla e he sou ce ex , 3) p omp ing he ool o edi i s
ansla ion h ough cu a ing p omp s whe e deemed necessa y, and 4) p omp -
ing he ool o e alua e he quali y o , and commen on, i s ansla ion. In o de o
o eg ound he esponsibili y o he human ansla o , a inal s ep, no included
wi hin he scope o he p omp enginee ing ask, was added: pos -edi ing he ob-
ained ansla ion ou pu h ough p omp enginee ing o p oduce he ansla ion
end-p oduc . O he han hese, no speci ic p omp s a emen s we e d a ed in o -
de o allow he human ansla o s-p omp enginee s some eedom and c ea i -
i y in hei indi idual dialogue wi h he GenAI ool. As a esul , each sou ce ex
gene a ed i e e sions o p omp enginee ing and i e ansla ion end-p oduc s.
A e comple ing he ask, each s uden - esea che w o e a epo on hei
p omp enginee ing expe ience, he analysis o which is p esen ed below.
4 Analysis o he esul s: Towa ds deepe awa eness o
he powe o he human ansla o is-à- is GenAI
A e he expe imen was comple ed, he s uden - esea che s w o e e ospec i e
p ocess epo s on hei expe iences wo king wi h GenAI by answe ing he ques-
ionnai e d a ed by he educa o - esea che . Globally, hei esponses indica ed
ha he collabo a i e esea ch had allowed hem o bo h e alua e he po en ial
impac o GenAI on he ansla ion p o ession and e-a i m hei ole as human
agen s and expe s in ansla ion.
The answe s o Ques ion 1 (“Wha is you gene al assessmen o he expe i-
men ?”), indica ed ha mos o he pa icipan s came o ecognise ha human
in e en ion is s ill needed in se e al espec s. One pa icipan (P4), o ins ance,
no ed ha du ing he expe imen she “came o he ealisa ion ha human ans-
la o s s ill ha e he inal say”. Likewise, ano he pa icipan (P3) s a ed ha de-
spi e he “gene al concep ion” ha LLMs “a e al eady adequa e o daily use o
e en echnical ansla ion a some poin ”, “ hey would ne e be able o eplace
human ansla o s” in e ms o “li e a y ex s” and “mixed o ms”. Ano he pa ic-
ipan (P1) s a ed ha human ansla o s’ pos -edi ing is “a necessi y a he han a
p e e ence”, especially o “ es uc u ing exp essi e language” and “ ans e ing
134
7 Re-posi ioning he human ansla o
he s yle”. Pa icipan 2 highligh ed he us a ion a he AI ool needing “con-
inuous p omp ing” o edi i s ansla ion “un il a inal ansla ion, which s ill
need[ed] pos -edi ing” was ob ained. Ano he pa icipan (P5) cha ac e ised he
expe imen as being “educa ing”, no ing ha al hough she had used “Cha GPT
on a ious occasions”, he expe imen was he i s ime ha she “ac ually sa
down and ied o gene a e a ansla ion h ough p omp ing”.
The answe s o Ques ion 2 (“Upon comple ing he expe imen , ha e you aced
any emo ional challenges as a human agen collabo a ing wi h GenAI o p o-
duce a ansla ion? I yes, wha we e hese challenges?”) showed ha mos o
he pa icipan s aced emo ional challenges while wo king wi h GenAI. A eel-
ing o us a ion was aised by wo pa icipan s. Pa icipan 1 s a ed ha he go
“ us a ed” when he LLM did no “di ec ly implemen ” his sugges ions and he
needed o “p o ide p omp s again”. Ano he pa icipan (P5) also s a ed ha she
“ ound alking o Cha GPT qui e us a ing” and ha she ealised ha she didn’
“ha e he equi ed pa ience o guide i h ough e e y hing”. One pa icipan (P2)
no ed ha i was “emo ionally exhaus ing o p omp AI o non-li e al ansla-
ion”, especially “i a wo d is used igu a i ely” as i was “challenging o explain
he unc ion o he wo d and design he p omp so as o each an accep able ans-
la ion”. Only one pa icipan (P3) desc ibed his eeling as “cons an puzzlemen ”
a he han “ us a ion o ange ” as he ool equi ed “an in-dep h hough p o-
cess on how o w i e he p omp s” and ha he “ el like he LLM wouldn’ be able
o comple e he asks” he way he “wan ed i o do”. The emo ional challenge
exp essed by one speci ic pa icipan (P4) di e ed om he challenges expe i-
enced by he o he pa icipan s in ha he “ eeling o s ess” expe ienced by his
pa icipan s emmed om he unde es ima ion o he “dep h o AI’s analy ical
capabili ies” and i s “speed”. The pa icipan (P4) also s a ed she “expe ienced a
lack o con idence while p o iding p omp s” in he ansla ion o he i s sou ce
ex and “became mo e com o able” in he ansla ion o he second sou ce ex
as she amilia ised he sel wi h he ool.
The answe s o Ques ion 3 (“Upon comple ing he expe imen , wha ha e you
disco e ed abou you powe is-à- is AI and you sel -concep as a human ans-
la o ?”) e ealed ha he pa icipan s add essed a ious aspec s o he powe o
he human ansla o on he basis o hei expe ience wo king wi h GenAI. One
pa icipan (P3) desc ibed his expe ience by saying ha i el like being “[a] su-
pe iso ying o each a e y in elligen new employee how o conduc he job
and how i can imp o e i sel in ce ain a eas”. Acco ding o him (P3), he ool can
“easily analyse he ex ” and “ ind ou wha kind o s a egies could be needed o
ansla e he ex o he [ a ge ] audience”. Howe e , i ails o “p oduce ans-
la ion s a egies o o e come ansla ion p oblems” as i emains “ oo ‘ ai h ul’
135
A sushi Mizumo o
he la e e e ing o an app oach ha iews lea ne s’ use o mul iple languages
as a esou ce a he han a p oblem.
The de ici -o ien ed pe spec i e, o en aligned wi h adi ional language each-
ing me hodologies, pe cei es L1 o MT use as de imen al o L2 acquisi ion. This
pe spec i e ypically leads o policies discou aging o p ohibi ing MT use, iew-
ing i as an impedimen o au hen ic language lea ning. Wi h his pe spec i e,
i is belie ed ha ha MT use may hinde he de elopmen o c i ical hinking
skills in he a ge language and in e e e wi h imme si e language expe iences.
Con e sely, he anslanguaging iew adop s a mo e inclusi e app oach o lan-
guage lea ning. This pe spec i e alues lea ne s’ en i e linguis ic epe oi e as a
esou ce (Wei 2018). Ad oca es o anslanguaging a gue ha MT can be a alu-
able ool o accessing and le e aging lea ne s’ ull ange o linguis ic knowl-
edge, po en ially enhancing bo h language awa eness and lea ning ou comes.
This iew aligns wi h con empo a y unde s andings o bilingualism and mul ilin-
gualism, which concep ualise languages as pa o an in eg a ed communica ion
sys em a he han as sepa a e en i ies.
These con as ing pe spec i es on MT use in language educa ion a e u he
exempli ied in he app oaches educa o s and ins i u ions ake when add essing
s uden use o MT. Jolley and Maimone’s (2022) comp ehensi e e iew o h ee
decades o MT esea ch in language eaching and lea ning highligh s wo dis-
inc app oaches: he “MT as Chea ing” app oach, which leads o a De ec -Reac -
P e en Response, and he “MT as Resou ce” app oach, which encou ages an
In eg a e-Educa e-Model s a egy.
The “MT as Chea ing” pe spec i e, aligned wi h he de ici -o ien ed iew,
ea s MT use as a o m o academic dishones y. This app oach ocuses on s a e-
gies o de ec unau ho ised MT use, eac puni i ely, and p e en u u e occu -
ences. P oponen s o his iew ecommend implemen ing clea syllabus policies
agains MT use, designing assignmen s esis an o MT use, and educa ing s u-
den s abou he pi alls o elying on MT. This pe spec i e o en leads o policies
ha ban MT use ou igh , iewing i as incompa ible wi h language lea ning
goals.
In con as , he “MT as Resou ce” app oach, mo e closely aligned wi h he
anslanguaging iew, sees MT as a po en ial ool o language lea ning. This
pe spec i e ad oca es in eg a ing MT in o he cu iculum, educa ing s uden s on
i s app op ia e use, and modeling e ec i e s a egies o le e aging MT in lan-
guage lea ning. Resea che s like S aple on & Kin (2019) and Niño (2020) a gue
o accep ing he eali y o MT use and inding ways o inco po a e i meaning-
ully in o language educa ion. This app oach acknowledges he ubiqui y o MT
in mode n li e and seeks o p epa e s uden s o use i c i ically and e ec i ely.
238

12 Emb acing machine ansla ion in L2 educa ion:
The shi om he De ec -Reac -P e en mindse o he In eg a e-Educa e-
Model app oach e lec s a g owing ecogni ion o he ine i abili y o MT use in
language lea ning con ex s. As Duca & Schocke (2018) no e, he key ques ion is
no longe whe he eache s can p e en lea ne s om using MT, bu a he how
o help hem use i e hically and e ec i ely as pa o hei language lea ning
jou ney.
These con as ing app oaches o MT use in language educa ion exempli y he
b oade ideological ensions iden i ied in G ie e e al. (2024) and e lec he ou e -
mos laye o ideological ac o s in Jiang e al.’s (2024) amewo k. They demon-
s a e how deeply held belie s abou language acquisi ion and he ole o ech-
nology can shape educa ional policies, pedagogical p ac ices, and ul ima ely, s u-
den s’ engagemen wi h and pe cep ions o MT in hei L2 de elopmen p ocess.
This in e play be ween ideological s ances and p ac ical app oaches unde sco es
he complexi y o in eg a ing MT in o language educa ion and highligh s he
need o con ex -sensi i e s a egies ha conside bo h he po en ial bene i s and
challenges o MT use in L2 lea ning and eaching.
2 In eg a ing MT in o L2 educa ion: A new pa adigm
2.1 MT as augmen ed L2 compe ence
The anslanguaging pe spec i e and “MT as Resou ce” app oach, implemen ed
h ough he In eg a e-Educa e-Model s a egy, p o ide a heo e ical and p ac-
ical ounda ion o inco po a ing MT in o L2 educa ion. Building upon hese
concep s, we can u he concep ualise MT use in language lea ning h ough he
lens o “MT as Augmen ed L2 Compe ence.” This model o e s a isual ep esen-
a ion o how MT can enhance lea ne s’ language abili ies, pa icula ly in b idg-
ing he gap be ween ecep i e and p oduc i e skills. By iewing MT as a ool
o augmen ing compe ence a he han eplacing language lea ning, we align
wi h he anslanguaging idea o luid language p ac ices and he “MT as Re-
sou ce” app oach. The In eg a e-Educa e-Model s a egy can hen be applied o
help lea ne s e ec i ely u ilise MT o expand hei augmen ed compe ence zone,
while simul aneously de eloping hei own language skills. This in eg a ed pe -
spec i e no only jus i ies he use o MT in language lea ning bu also p o ides
a amewo k o unde s anding i s ole in enhancing o e all L2 p o iciency.
The concep o MT as augmen ed L2 compe ence is illus a ed in Figu e 1,
which p o ides a isual ep esen a ion o how MT, and also GenAI such as Cha -
GPT, can enhance language lea ne s’ abili ies.
239
A sushi Mizumo o
Augmen ed compe ence
wi h MT (AI)
P oduc i e compe ence
Recep i e compe ence
Figu e 1: The concep o MT as augmen ed L2 compe ence
The igu e demons a es he ela ionship be ween ecep i e compe ence, p o-
duc i e compe ence, and he po en ial o augmen ed compe ence h ough MT
use. He e is a b eakdown o he key elemen s:
• Recep i e Compe ence: This is ep esen ed by he la ge , ou e o al. I
e e s o he abili y o unde s and he a ge language (L2), which is ypi-
cally mo e de eloped han p oduc i e skills. Fo mos L2 English language
lea ne s, hei capaci y o comp ehend English exceeds hei abili y o p o-
duce i .
• P oduc i e Compe ence: Shown as he smalle , inne o al, his ep esen s
he lea ne ’s abili y o ac i ely use he language. I is gene ally mo e lim-
i ed han ecep i e compe ence, which aligns wi h heo ies like Swain’s
ou pu hypo hesis (1985), emphasising he impo ance o language p oduc-
ion in second language acquisi ion.
• Augmen ed Compe ence wi h MT (AI): This is depic ed by he da k g ay
a ea ex ending beyond he p oduc i e compe ence o al. I illus a es how
MT can b idge he gap be ween wha lea ne s can ecognise as co ec
( ecep i e knowledge) and wha hey can p oduce on hei own.
Figu e 1 sugges s ha MT can se e as a ool o augmen lea ne s’ compe ence,
pa icula ly in a eas whe e hey can ecognise co ec ness by sigh bu s uggle
o p oduce i accu a ely. This augmen a ion is especially bene icial o mo e p o-
icien lea ne s, as suppo ed by p e ious s udies (Klimo a e al. 2022, Ohashi
240
12 Emb acing machine ansla ion in L2 educa ion:
2024). Highe p o iciency lea ne s end o ha e a la ge gap be ween hei ecep-
i e and p oduc i e skills, p o iding mo e oom o MT o assis in b idging his
di ide.
Impo an ly, his model unde sco es ha he e emains a s ong a ionale o
s udying English (o any L2). The augmen ed compe ence p o ided by MT is
buil upon he ounda ion o he lea ne ’s own language skills. Wi hou de el-
oping one’s own ecep i e and p oduc i e compe encies, he bene i s o MT aug-
men a ion would be limi ed. Fu he mo e, as lea ne s’ p o iciency inc eases, hey
become be e equipped o e ec i ely u ilise MT, maximising i s po en ial as a
lea ning ool.
This concep ualisa ion o MT as augmen ed L2 compe ence aligns wi h he
indings om sys ema ic e iews (Jolley & Maimone 2022, Lee 2023) ha high-
ligh MT’s e ec i eness when used app op ia ely, pa icula ly o mo e ad-
anced lea ne s. I also suppo s he need o p ope guidance and aining in
MT use, as he ool’s e ec i eness is con ingen upon he lea ne ’s abili y o
c i ically e alua e and apply i s ou pu .
In sum, he model p esen ed he e p o ides a amewo k o unde s anding
how MT can be in eg a ed in o language lea ning p ocesses. I emphasises ha
MT is no a eplacemen o language s udy, bu a he a ool ha can en-
hance and ex end lea ne s’ exis ing compe encies, po en ially accele a ing hei
p og ess owa ds highe le els o language p o iciency.
2.2 MT ins uc ion o L2 lea ning
Niño (2009) p oposed ou models o MT use in L2 educa ion: a “bad model”, a
“good model”, oca ional applica ions (pa icula ly in ansla ion- ela ed ields),
and as a compu e -assis ed language lea ning (CALL) ool. Ini ially, MT was em-
ployed as a “bad model”, whe e s uden s iden i ied and co ec ed e o s h ough
pos -edi ing, a p ocess necessi a ed by he limi ed accu acy o ea ly sys ems. In
con as , he “good model” in ol ed using MT ou pu s as exempla s o s uden s.
These models, e lec ing he e olu ion o MT echnology and i s pedagogical ap-
plica ions, illus a e a signi ican shi in ocus. As MT echnology has ad anced,
i s p ima y use has ansi ioned o se ing as a CALL ool, whe e i acili a es
s uden engagemen in sol ing language p oblems independen ly, as e idenced
in ecen s udies (Lee 2020, S aple on & Kin 2019, Tsai 2019).
Lee (2023) ecommends ha eache s should p o ide guidelines o using MT
and explici ly each e ec i e s a egies o s uden s p io o using i , which leads
o enhancing s uden pe o mance, as suppo ed by O’Neill (2016).
241
A sushi Mizumo o
While many esea che s discuss he need o explici MT ins uc ion in lan-
guage class ooms, only a limi ed numbe o s udies p o ide conc e e examples
(Cancino & Panes 2021, Chang e al. 2022, Mi zaeian 2021, O’Neill 2016, Ryu e
al. 2022), all o which epo posi i e esul s. Among hese, Ryu e al. (2022) im-
plemen ed a model called he Guided Use o MT (GUMT). The GUMT model
ac i i ies we e de eloped and implemen ed in an uppe -elemen a y Ko ean as a
o eign language cou se a a U.S. uni e si y. The GUMT model has i e s eps:
• Ins uc ional Session on MT Tools: As Ac i i y 1, S uden s e iew s eng hs
and weaknesses o popula MT esou ces like Google T ansla e and Na e
Papago, discussing di e ences in speech s yles and poli eness le els. As Ac-
i i y 2, s uden s engage in p agma ic and g amma e alua ion exe cises,
assessing he app op ia eness o ansla ions by conside ing con ex s such
as o mali y and speake -lis ene ela ionships.
• P ac ice wi h W i ing Assignmen s: Th oughou he semes e , s uden s ap-
ply lea ned concep s in i e w i ing asks, using MT o aid hei ansla ion
om Ko ean (L2) o English (L1) and ice e sa.
• Re lec ion on MT Use: A e each w i ing assignmen , s uden s e lec on
hei use o MT, assessing i s e ec i eness and making adjus men s based
on hei expe iences.
• Ins uc o Feedback: Ins uc o s p o ide w i en eedback on s uden s’
d a s, ocusing on g amma ical and p agma ic accu acy, and highligh
a eas needing co ec ion.
• In-Class Re iew Sessions: S uden s e ine hei d a s in class based on in-
s uc o s’ eedback, discussing and e ising highligh ed issues o imp o e
hei unde s anding and applica ion o MT in w i ing.
Ryu e al. (2022) epo ed ha implemen ing he GUMT model signi ican ly
con ibu ed o he de elopmen o MT use s a egies and enhanced s uden s’
con idence and sel -assessed luency in L2 w i ing, as e idenced by analyses o
bo h p e- and pos -su eys and s uden e lec ions.
2.3 Lea ne s as me acogni i e agen s
The ole o lea ne s as me acogni i e agen s in using MT and o he language
esou ces is c ucial o e ec i e language lea ning. Mizumo o (2023) p oposed a
242
12 Emb acing machine ansla ion in L2 educa ion:
amewo k called Me acogni i e Resou ce Use (MRU), which p o ides a comp e-
hensi e app oach o unde s anding how lea ne s can s a egically u ilise a ious
language esou ces, including MT and GenAI ools like Cha GPT (Figu e 2).
Figu e 2: F amewo k o me acogni i e esou ce use (adap ed om
Mizumo o 2023).
The MRU amewo k is g ounded in me acogni ion heo y and consis s o
wo main componen s: me acogni i e knowledge and me acogni i e egula-
ion. Me acogni i e knowledge encompasses unde s anding o he pe son (sel -
knowledge), ask ( equi emen s and cons ain s), and s a egy (a ailable ap-
p oaches). Me acogni i e egula ion in ol es he ac i e managemen o his
knowledge h ough planning, moni o ing, and e alua ing one’s use o esou ces
h ough employing lea ning s a egies.
The amewo k isualises a ange o esou ces ha lea ne s can u ilise, includ-
ing online dic iona ies, co pus-based da a-d i en lea ning (DDL) ools like con-
co dance s, web apps, sea ch engines, GenAI sys ems like Cha GPT, MT ools,
and g amma co ec ion ools such as G amma ly. This e lec s he eal-wo ld
usage o ools by lea ne s when engaging wi h language asks.
243

A sushi Mizumo o
Fo ins uc ion on me acogni i e esou ce use, Mizumo o (2023) sugges s
adop ing p inciples om lea ning s a egy ins uc ion (Chamo & Ha is 2019).
This app oach in ol es enhancing lea ne s’ me acogni i e knowledge and im-
pa ing speci ic me acogni i e and ela ed cogni i e s a egies. The ins uc ional
model ocuses on exposing s uden s o a di e se a ay o ools and guiding hem
h ough he p ocesses o planning, execu ing, e alua ing, and adap ing hei
s a egic esou ce use.
I is impo an o no e ha he GUMT model employed by Ryu e al. (2022)
aligns well wi h his app oach. The GUMT model does no ely on a single ool
bu inco po a es a ious esou ces, including di e en MT pla o ms and e lec-
i e p ac ices. Fu he mo e, he ins uc ion model in Ryu e al. (2022) sha es sim-
ila i ies wi h bo h DDL and s a egy ins uc ion app oaches, demons a ing ha
hese me hodologies can be discussed wi hin he same heo e ical amewo k
ha Mizumo o (2023) ad oca es o in eg a ing DDL and GenAI.
This in eg a ed pe spec i e allows o a mo e comp ehensi e app oach o lan-
guage esou ce use. Ins ead o ocusing solely on MT, lea ne s should be encou -
aged o become me acogni i e agen s capable o e ec i ely u ilising a wide ange
o esou ces, including GenAI ools like Cha GPT. By adop ing his app oach, we
can os e he de elopmen o au onomous lea ne s who a e adep a selec ing
and using he mos app op ia e ools o hei language lea ning asks.
The MRU amewo kno only enhances lea ne s’ awa eness and use o MT bu
also ex ends o hei conscious u ilisa ion o GenAI ools. This holis ic app oach
has he po en ial o cul i a e uly au onomous lea ne s who can na iga e he
complex landscape o language lea ning esou ces e ec i ely.
Mo eo e , he esea ch indings accumula ed in he ield o MT should no be
dis ega ded bu a he e-examined and alida ed wi hin his in eg a ed and ex-
ended amewo k. By doing so, we can ensu e ha he aluable insigh s gained
om MT esea ch con inue o in o m and enhance ou unde s anding o lan-
guage lea ning p ocesses in he e a o di e se digi al esou ces.
In conclusion, by posi ioning lea ne s as me acogni i e agen s wi hin he MRU
amewo k, we can c ea e a mo e comp ehensi e and e ec i e app oach o lan-
guage lea ning ha emb aces he ull spec um o a ailable esou ces, om a-
di ional ools o cu ing-edge AI echnologies.
2.4 Dis inguishing MT om GenAI
While his chap e has p ima ily ocused on MT, i is impo an o acknowl-
edge he eme ging ole o GenAI ools like Cha GPT in L2 educa ion. Al hough
bo h MT and GenAI can assis wi h language asks, hey di e undamen ally
244
12 Emb acing machine ansla ion in L2 educa ion:
in hei capabili ies and po en ial applica ions. T adi ional MT sys ems a e de-
signed speci ically o ansla ion be ween languages, ocusing on main aining
seman ic equi alence while adap ing o a ge language con en ions. In con as ,
GenAI sys ems can engage in open-ended language gene a ion, including no
only ansla ion bu also ex ending o explana ion, summa isa ion, pa aph asing,
and in e ac i e dialogue.
The di e ences be ween hese echnologies ex end o hei abili y o handle
con ex and main ain cohe en in e ac ions. While MT p ima ily p ocesses ex a
he sen ence o pa ag aph le el, GenAI can main ain longe con e sa ions, ake
b oade con ex in o accoun , and engage wi h complex p omp s ac oss mul iple
exchanges. This enhanced con ex ual awa eness allows GenAI o se e no jus
as a ool o language con e sion, bu as an in e ac i e pa ne in he language
lea ning p ocess.
These undamen al di e ences lead o dis inc educa ional applica ions. MT
se es p ima ily as a ansla ion ool, hough i can be used pedagogically o
language compa ison and e o analysis. GenAI, howe e , can po en ially unc-
ion as an in e ac i e u o , p o iding explana ions, examples, p ac ice exe cises,
and eedback on language use. This b oade unc ionali y sugges s ha while MT
ypically se es as a s andalone ool o ansla ion asks, GenAI has he po en ial
o in eg a e and po en ially eplace mul iple language lea ning ools, including
dic iona ies, g amma checke s, and co pus esou ces.
3 Concluding ema ks
Looking ahead, he in eg a ion o MT, GenAI, and o he AI-powe ed language
ools in L2 educa ion is likely o con inue e ol ing apidly. Fu u e esea ch
should ocus on de eloping and empi ically es ing comp ehensi e amewo ks
like he MRU, which emb ace a wide ange o digi al esou ces beyond jus MT.
As he capabili ies o GenAI con inue o expand, he e is a pa icula need o un-
de s and how hese ools can complemen o po en ially ans o m adi ional
MT use in language lea ning con ex s. The e is a need o longi udinal s udies o
assess he long- e m impac o hese ools on language acquisi ion and lea ne au-
onomy, speci ically examining how he in e ac i e capabili ies o GenAI migh
di e om adi ional MT in suppo ing language de elopmen . Addi ionally, as
AI language models become mo e sophis ica ed, p ac i ione s and esea che s
will need o con inually adap hei app oaches, balancing he bene i s o hese
ools wi h he co e objec i es o language lea ning. The ield may see a shi o-
wa ds mo e pe sonalised, AI-assis ed language lea ning expe iences, necessi a -
ing new pedagogical s a egies and e hical guidelines o hei implemen a ion
245
A sushi Mizumo o
in L2 lea ning and eaching. Unde s anding he dis inc oles and po en ial syne -
gies be ween MT and GenAI will be c ucial o de eloping e ec i e educa ional
p ac ices in his apidly ans o ming ield.
Re e ences
Cancino, Ma co & Jaime Panes. 2021. The impac o Google T ansla e on L2
w i ing quali y measu es: E idence om Chilean EFL high school lea ne s.
Sys em 98. 102464. DOI: 10.1016/j.sys em.2021.102464.
Chamo , Anna Uhl & Vee Ha is (eds.). 2019. Lea ning s a egy ins uc ion in he
language class oom: Issues and implemen a ion. B is ol, UK: Mul ilingual Ma -
e s.
Chang, Peichin, Pin-Ju Chen & Li-Ling Lai. 2022. Recu si e edi ing wi h Google
T ansla e: The impac on w i ing and e o co ec ion. Compu e Assis ed Lan-
guage Lea ning. 1–26. DOI: 10.1080/09588221.2022.2147192.
Chung, Eun Seon & Soojin Ahn. 2022. The e ec o using machine ansla ion
on linguis ic ea u es in L2 w i ing ac oss p o iciency le els and ex gen es.
Compu e Assis ed Language Lea ning 35(9). 2239–2264. DOI: 10.1080/09588221.
2020.1871029.
Duca , Cyn hia & Debo ah Houk Schocke . 2018. Machine ansla ion and he
L2 class oom: Pedagogical solu ions o making peace wi h Google T ansla e.
Fo eign Language Annals 51(4). 779–795. DOI: 10.1111/ lan.12366.
G ie e, A e il, Ami Rouhshad, Elpida Pe aki, Alan Bechaz & Da id Wei Dai.
2024. Nu sing and midwi e y s uden s’ e hical iews on he accep abili y o us-
ing AI machine ansla ion so wa e o w i e uni e si y assignmen s: A de ici -
o ien ed o anslanguaging pe spec i e? Jou nal o English o Academic Pu -
poses 70. 1475-1585, 101379. DOI: 10.1016/j.jeap.2024.101379.
Hellmich, Emily A. 2023. S uden use and ins uc o belie s: Machine ansla ion
in language educa ion. Language Lea ning & Technology 27(1). h ps ://hdl .
handle.ne /10125/73525.
Jiang, Lianjiang, Rong Yu & Yi Zhao. 2024. Theo e ical pe spec i es and ac o s
in luencing machine ansla ion use in L2 w i ing: A scoping e iew. Jou nal
o Second Language W i ing 64. 101099. DOI: 10.1016/j.jslw.2024.101099.
Jolley, Jason R. & Luciane Maimone. 2022. Thi y yea s o machine ansla ion
in language eaching and lea ning: A e iew o he li e a u e. L2 Jou nal 14(1).
26–44. DOI: 10.5070/L214151760.
246
12 Emb acing machine ansla ion in L2 educa ion:
Klimo a, Blanka, Ma cel Pikha , Alice Delo me Beni es, Ca oline Leh &
Ch is ina Sanchez-S ockhamme . 2022. Neu al machine ansla ion in o eign
language eaching and lea ning: A sys ema ic e iew. Educa ion and In o ma-
ion Technologies 28. 663–682. DOI: 10.1007/s10639-022-11194-2.
Lee, Sangmin-Michelle. 2020. The impac o using machine ansla ion on EFL
s uden s’ w i ing. Compu e Assis ed Language Lea ning 33(3). 1–19. DOI: 10.
1080/09588221.2018.1553186.
Lee, Sangmin-Michelle. 2023. The e ec i eness o machine ansla ion in o -
eign language educa ion: A sys ema ic e iew and me a-analysis. Compu e As-
sis ed Language Lea ning 36(1-2). 103–125. DOI: 10.1080/09588221.2021.1901745.
Lee, Sangmin-Michelle & Nayeon Kang. 2024. E ec s o machine ansla ion on
L2 w i ing p o iciency: The complexi y, accu acy, lexical di e si y, and luency.
Language Lea ning & Technology 28(1). 1–19. h ps://hdl.handle.ne /10125/
73585.
Lo, Siowai. 2024. The e ec s o NMT as a de ac o dic iona y on ocabula y lea n-
ing: A compa ison o h ee look-up condi ions. Compu e Assis ed Language
Lea ning. 1–21. DOI: 10.1080/09588221.2024.2310285.
Lo, Siowai. 2025. Neu al machine ansla ion in EFL class ooms: Lea ne s’ ocab-
ula y imp o emen , immedia e ocabula y e en ion and delayed ocabula y
e en ion. Compu e Assis ed Language Lea ning 38(3). 592–611. DOI: 10.1080/
09588221.2023.2207603.
Mi zaeian, Vahid R. 2021. The e ec o edi ing echniques on machine ansla ion-
in o med academic o eign language w i ing. The EUROCALL Re iew 29(2). 11.
Mizumo o, A sushi. 2023. Da a-d i en lea ning mee s gene a i e AI: In oducing
he amewo k o me acogni i e esou ce use. Applied Co pus Linguis ics 3(3).
100074. DOI: 10.1016/j.aco p.2023.100074.
Niño, Ana. 2009. Machine ansla ion in o eign language lea ning: Language
lea ne s’ and u o s’ pe cep ions o i s ad an ages and disad an ages. ReCALL
21(2). 241–258. DOI: 10.1017/S0958344009000172.
Niño, Ana. 2020. Explo ing he use o online machine ansla ion o independen
language lea ning. Resea ch in Lea ning Technology 28. DOI: 10.25304/ l . 28.
2402.
O’Neill, E ol M. 2016. Measu ing he impac o online ansla ion on FL w i ing
sco es. IALLT Jou nal o Language Lea ning Technologies 46(2). 1–39. DOI: 10.
17161/iall . 46i2.8560.
Ohashi, Louise. 2024. Machine ansla ion in language educa ion: A sys ema ic
e iew o open access a icles. Kenkyu Nenpou: The Annual Collec ion o Es-
says and S udies 70. 105–125. h ps://www.gakushuin.ac.jp/uni /le / op/
publica ion/KE_70/KE_70_008.pd .
247
Name index
Viei a, Lucas Nunes, 7, 126
Vinay, Jean-Paul, 97
Vygo sky, Le , 33
Wadensjö, Cecilia, 214
Waldo, Jim, 63
Walke , Callum, i
Wang, Hai eng, 68
Wang, Longyue, 90
Wang, Lulu, 193
Wang, Rui, 70
Wang, Zi ing, 211, 219, 220
Way, Andy, 68, 71, 80, 88, 90
Wei, Jason, 107
Wei, Li, 236
Welocalize,
Wigge s, Kyle, 109
Windhage , Flo ian, 8
Win e s, James, 148
Wissen, Ma kus, 113
Woo, Jieyeon, 8
W igh , Sue Ellen, 68, 70
Yamada, Masa u, 56, 91, 93, 129, 132,
133
Yang, Jian, 148
Yang, Zhishen, 90
Zane in, Fede ico, 68
Zhang, Biao, 90
Zhong, Linping, 192, 193
Zhou, Meng, 114
Zhu, Yilun, 148
Zhuang, Yan, 146, 148
Zimina, Ma ia, 155, 156
Zimina-Poi o , Ma ia, 149, 155, 157,
158
254

Teaching ansla ion in he age o
gene a i e AI
Since he launch o OpenAI’s Cha GPT in 2022, gene a i e a i icial in elligence (GenAI)
has s a ed eshaping wha i means o wo k as a p o essional ansla o in an indus y
ha is becoming inc easingly au oma ed. This p omp s us o in e oga e, once again, he
ole and agency o human ansla o s in he ansla ion p ocess o , in o he wo ds, he in-
insically human alue and alues hey add o i . A na u al co olla y is ha GenAI o ces
us ansla o educa o s o ( e-)in e oga e wha we do in ou ansla ion p og ammes.
Wha e e we may hink o eel abou GenAI, we owe i o ou s uden s o engage wi h i
in ou p og ammes. Howe e , because GenAI is no jus ano he ool in he ansla o ’s
oolki , we mus also o do so in a way ha aises s uden s’ awa eness o some o he
e hical and sus ainabili y issues a ound i .
This is wha Teaching T ansla ion in he Age o Gene a i e AI: New Pa adigm, New
Lea ning aims o do. A icula ed a ound h ee main pa s, Pa 1 explo es he new skills
and compe ences ansla o educa o s need o help hei s uden s de elop in he age
o GenAI. In Pa 2, he ocus shi s o he new knowledge (such as da a li e acy and
p omp ing) ha s uden s in ansla ion p og ammes need o engage wi h in he age
o GenAI. Finally, Pa 3 pu s some lesh on he bones, as i e iews some o he new
eaching app oaches adop ed by colleagues since he ad en o GenAI. I does so by
in oducing he eade o a se ies o igne es aken om a a ie y o ansla ion- ela ed
disciplines and con ex s.
Th oughou he en i e edi ed olume, he ambi ion is o be as accessible as possible,
so ha his olume can be o help o as many o us in ansla ion educa ion as possible,
as we all lea n o nego ia e he uncha ed e i o y o GenAI.