Chap e 10
AI in an L2 ansla ion class
Sonia Vandepi e
Uni e si ei Gen , Belgium
Teaching ansla ion om he so-called na i e language (L1) in o a second language
(L2) equi es app oaches di e en om eaching ansla ion in o he L1. Wi h he
ad en o some gene a i e AI ools made eely accessible e e ywhe e, a powe ul
esou ce has become a ailable ha can be pu o good use in he L2 ansla ion
class. This con ibu ion will epo on such an explo a o y class ha in ol ed a
ansla ion assignmen in which Cha GPT was used as an edi ing ool. I will ex-
plain he ansla ion se ing and de ail he design o how AI was in eg a ed in o
ansla ion aining om a Eu opean language in o English. I will desc ibe he
s uden s’ lea ning objec i es and he s eps aken o acili a e hei pa hs owa ds
hose goals. By illus a ing he s uden s’ main demons a ed compe ences, he class
design will be e lec ed upon and some ideas o al e na i e u u e classes will be
sugges ed. I will be shown ha in eg a ing AI in o he L2 ansla ion class oom
en iches L2 ansla ion lea ning, and a cons uc i e c i ical a i ude will be called
o .
1 In oduc ion
Au umn 2022 b ough he ool o gene a i e AI (GenAI) o he gene al public. In
a ansla ion aining class, his mean ha s uden s could now p oduce ansla-
ions and edi ex s on he basis o he mul ilingual capabili ies o a use - iendly,
con e sa ional de ice which elied on a huge knowledge da abase. Because i
ook in o accoun con ex ual ea u es, i s ansla ions in well- esou ced language
pai s, and especially wi h he English language, u ned ou o p ese e one and
meaning be e han any au oma ed ool be o e. In addi ion, i s ansla ions o
idioms and cul u e-speci ic concep s in hose languages also p o ed o con ain
ewe ailu es han he p e ious models o neu al machine ansla ion (MT). I
Sonia Vandepi e. 2026. AI in an L2 ansla ion class. 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?, 191–210. Be lin: Language Science P ess. DOI: 10.5281/zenodo.17641082
Sonia Vandepi e
any o ganisa ion could use such a ool, he impo ance o making s uden s amil-
ia wi h i in hei ansla ion aining he e o e also became essen ial.
The nex sec ion will e iew he mos ele an cha ac e is ics o a aining
class eaching ansla ion in o English as he s uden s’ second language, ans-
la ion in a GenAI con ex , and some pionee ing p ojec s ha al eady in eg a ed
GenAI in o a ansla ion class om he ou se .
2 Li e a u e e iew
2.1 Teaching ansla ion in o he L2
T ansla ion in o a second language, i.e. a language which is no he ansla-
o ’s main language, o en abb e ia ed as L2, is o en conside ed o equi e a
highe cogni i e load (K ings 1986, Göp e ich 2017), al hough a ious schola s
ha e demons a ed he high quali y o he ou come o such ansla ion p ocesses
(e.g. G osman e al. 2000). T ansla ion aining p og ammes, oo, o en con ain
modules wi h p ac ice in his ansla ion di ec ion. The eason is simple: in many
coun ies he e is a need o ansla ion in o languages, usually a so-called dom-
inan language o a lingua anca (see e.g. House 2018), o which no ansla o s
a e a ailable.
Hence, aining p og ammes om he EMT L2 T ansla ion wo king g oup
(Vandepi e 2023) ha e collec ed a se o bes p ac ices o eaching L2 ansla ion.
Fi s , i was ag eed ha mos s uden s we e somewha less eage o ansla e in o
an L2, had less con idence in hei abili ies, and we e mo e hesi an abou hei
p og ess han in hei ansla ion in o hei main language.
Fi s , hose conside a ions equi ed app op ia e app oaches o he gene al
cou se design. I was sugges ed ha he inclusion o eal ansla ion p ojec s (i
possible, wi h ex a emune a ion o s uden s) wo ked a ou ably. I hey also
include coope a ion wi h ellow class s uden s who can ac as bilingual e ise s
o hei i s d a s o o he s uden s (ab oad) who can collabo a e as monolin-
gual e iewe s because hei main language is he a ge ed L2, s uden s’ con i-
dence in hei abili y o p oduce a well-w i en ansla ion is boos ed. I s uden s’
ini ial compe ence a ies conside ably, o i classes a e la ge, his disc epancy
may be iden i ied a he beginning o he cou se and g oups o ganised in such a
way ha be e -skilled s uden s suppo o he s (enhancing s uden s’ social skills)
As a as class design is conce ned, he ollowing bes p ac ices we e men-
ioned as sui able o p oduce be e insigh in o he ansla ion ask ahead. In he
i s place, much a en ion should be de o ed o ex choice. Especially, he ele-
ance, leng h and opic o he ex s ha e an impac on s uden s’ willingness o
192
10 AI in an L2 ansla ion class
s a a di icul ask. Tex s commonly equi ed by he ansla ion ma ke o ex s
sha ed by p o essional o ganisa ions a e p e e able. A opic ha is oo speci ic
will ende he ex oo demanding and an oppo uni y o s uden s o obse e
hei own p og ess will be los . The eache can p o ide specialised glossa ies,
in oduce eed-up sessions, in which bo h g amma ical, lexical and cul u al s um-
bling blocks a e poin ed ou , le he s uden s make a c i ical analysis o he com-
munica i e si ua ion ( he ansla ion b ie ), o o ganise a class discussion on he
domain o which he ex opic belongs. S uden s can ecei e pee and eache
eedback when i s and inal d a s a e published anonymously on he lea ning
pla o m, allowing s uden s he oppo uni y o lea n om each o he ’s wo k.
2.2 T ansla ion in a gene a i e AI con ex
Wi h he de elopmen o GenAI (Eloundou e al. 2023, Fel en e al. 2023, Akh a
2024), ansla ion p ac ices a e o e ed a u he s ep in he au oma ion p ocess.
While he use o MT (bo h s a is ical and neu al) has been amply discussed (e.g.
Moo kens e al. 2025), esea ch o he use o a cha bo like Cha GPT (OpenAI n.d.,
Roumelio is & Tselikas 2023) in he ansla ion p ocess is s ill in i s in ancy. I
ocuses mainly on he quali y pe o mance o he gene a i e ool in compa ison
o ha o MT.
Robinson e al. (2023), o ins ance, ound ha he amoun o esou ces in a
pa icula language plays an essen ial ole. Thei s udy in ol ing abou 200 lan-
guages showed ha quali y measu es in e ms o wo di e en me ics o ansla-
ions wi h languages o which he e is a high olume o esou ces o GPT mod-
els we e as high as o equally high as hose o adi ional MT models. Wi h mos
languages ha ing a low olume o esou ces, howe e , GPT model pe o mance
consis en ly lagged behind. Geng e al. (2024), oo, emphasised ha non-English
la ge language models (LLMs) can only be de eloped i a ailable da a se s a e
la ge enough. Mujadia e al. (2023), on he o he hand, epo posi i ely on hei
expe imen s wi h ansla ions in ol ing English and 22 Indian languages using
LLMs based on Me a’s LLaMa.
O he schola ly wo k on ansla ion by Cha GPT has been ca ied ou in di -
e en knowledge domains. Ülkü (2023) ocussed on he hospi ali y indus y se -
ices ha will likely bene i om he use o GPT-4. He p edic ed ha eal- ime
ansla ion by means o he gene a i e model will acili a e e ec i e communica-
ion be ween ho el s a and gues s, bu did no p o ide any c i icism o cau ion,
no any da a o suppo his op imis ic belie in he ansla ion quali y o he
ool. Teng (2024) esea ched he ield o poe y ansla ion and compa a i ely
explo ed ai h ulness, exp essi eness, and elegance in bo h a human and GPT-4
193
Sonia Vandepi e
ansla ion o a Chinese poem. Cao & Zhong (2023) ca ied ou a s a is ical in es-
iga ion o assess he ansla ion quali y o a Chinese ecipe in o English by GPT-
4. D awing on bo h au oma ic and human e alua ion sys ems, hey ound ha
oge he wi h o he LLMs, GPT-4 ou pe o med he ine- uned MT models and
he ine- uned mul ilingual encode -decode models. GPT-4 c ea ed a ansla ion
ha inco po a ed cul u al adap a ions and al e na i e names o ing edien s. No-
ably, i was mo e closely aligned wi h he sou ce ex han he human e e ence
ex , e en hough i p oduced double he numbe o okens o he ansla ions
p oduced by he o he models. Wi hin he legal a ea, Giampie i (2024) submi -
ed an English a bi a ion clause o he GPT cha bo o 2024 and p omp ed i o
ansla e i in o I alian. She epo ed some lexical inaccu acies and inconsis en-
cies ha needed o be pos -edi ed.
In summa y, he quali y as judged by hese p elimina y s udies, which we e
no ye a ailable when he decision was aken o in eg a e GenAI in o he class,
clea ly depends on a numbe o ac o s. Fi s , he numbe o esou ces a ailable
o he gene a i e ool plays a big ole, and languages o which he e a e ew
esou ces yield poo AI- ool ou pu . Howe e , o a language pai like Chinese-
English, Cha GPT ou pe o med o he MT-models and, in pa icula , also showed
success ul ansla ion choices in cul u al con ex s. Secondly, he ex ype also
has an impac on he ansla ion quali y o he gene a i e ool, wi h legal ex s,
o ins ance, equi ing special a en ion o how he sou ce ex meaning is en-
de ed.
2.3 In eg a ing gene a i e AI in o he ansla ion class
Including GenAI such as Cha GPT in educa ion has ecen ly been men ioned in
a ious epo s as a p omising aid o imp o e s uden s’ w i ing (Cao & Zhong
2023, Labadze e al. 2023, Ouyang e al. 2023), and, in pa icula , hei w i ing
in English as a Fo eign o Second Language (EFL/ESL) (Fi ia 2023). One ad an-
age o AI-powe ed cha bo s is clea : Cha GPT can p o ide ins an and pe sonal
eedback a a ime when he lea ne ’s mind is se o lea n, so ha i s ele ance
and impac may be highe han when hey ecei e simila eedback a a la e
junc u e. This will undamen ally change L2 lea ning: he AI ools will be able
o de ec gaps in indi idual s uden s’ pe o mances and e en gi e eedback and
al e na i e exe cises.
The wo ldwide a ailabili y o Cha GPT owa ds he end o 2022 also aised
ansla ion eache s’ awa eness ha he new echnological de elopmen s a e
he in oduc ion o CAT ools and he a ious o ms o MT would b ing abou
ye ano he se o changes o he ansla ion p ocess.
194
10 AI in an L2 ansla ion class
This awa eness led o one o he i s a i udinal examina ions: Saha i e al.’s
s udy (2023) in es iga ed he opinions o bo h ansla ion s uden s and eache s
abou he use ulness o Cha GPT when compa ed wi h Google T ansla e. They
concluded ha s uden s and eache s did no ag ee on which ool was p e e able.
Howe e , he AI ool was ecognised as a acili a o o yping and spelling, bu
he ool’s esul s o ine- uning and double-checking could no be us ed.
Wang e al. (2024) also in es iga ed s uden s’ a i udes o he use o Cha GPT
in ansla ion. They did so wi hin he amewo k o he Uni ied Theo y o Accep-
ance and Use o Technology, whose model poin s a ac o s in luencing pa ic-
ipan s’ a i udes. They ound ha despi e a posi i e a i ude owa ds Cha GPT,
s uden s s ill commonly used MT sys ems.
Gao (2024) desc ibes a pilo p og amme, in which he ex s o poli ical
speeches we e submi ed o Cha GPT 3.5. The ansla ions p oduced we e dis-
cussed by he s uden s and compa ed o ansla ions p oduced by p o essional
UN ansla o s. They ound ha he GPT model was success ul in p oducing
i s d a ansla ions wi h app op ia e lexical and g amma ical in o ma ion.
Howe e , i was no ye capable o su icien ly dealing wi h complex linguis ic,
cul u al, ideological o o he con ex ual nuances.
Cao & Zhong (2023) p oduce s a is ical e idence o he use ulness o Cha GPT
in he con ex o a ansla ion class. They ca ied ou an analysis o ansla ions
submi ed by Chinese ESL/EFL lea ne s and compa ed bo h he o e all quali y
(measu ed using BLEU sco es) and some linguis ic ea u es (Coh-Me ix) o h ee
ypes o s uden ansla ions o a sho p ess elease: some ansla ions only un-
de wen sel - eedback, o he s elici ed eache eedback, and a hi d se ecei ed
eedback by Cha GPT. The esul s o his analysis indica ed highe BLEU sco es
and syn ac ic Coh-Me ix sco es o he ansla ions p oduced a e eache eed-
back and he lowes sco e was gi en o he GPT eedback gene a ed ansla ions.
T ansla ions wi h Cha GPT eedback only sco ed be e in he lexical domain.
They sugges ha Cha GPT “ elies on eplica ing pa e ns om i s aining da a
a he han pe o ming ue syn ac ic analysis, he eby ailing o o e subs an-
i e eedback” (Cao & Zhong 2023:13).
While Gao’s p og am in eg a ed GenAI in o he ansla ion class p ocess as a
ool ha is p omp ed o pe o m he ansla ion i sel , Cao and Zhong designed
a p ocess in which he ool is eques ed o ac like an educa o and o p o ide
eedback o he s uden s’ ansla ions.
The ollowing sec ions will desc ibe and discuss an explo a o y cou se which
in eg a ed Cha GPT in o he s uden s’ ansla ion p ocess in ye ano he way.
This cou se om a Eu opean EMT ansla ion p og amme in Sp ing o 2023
had s uden s ansla e in o an L2, a ask in which s uden s could bene i i hey
195
Sonia Vandepi e
could access he opinions o a a ge language speake . The cou se he e o e had
s uden s use Cha GPT as a monolingual e ise o hei own ansla ion wo k.
3 Cou se design o Cha GPT in an L2 ansla ion class
As men ioned abo e, he cou se epo ed he e had s uden s use a GenAI ool,
mainly o edi ing. The cou se was pa o a Mas e ’s ansla ion p og amme
and equi ed s uden s o be able o ansla e om hei L1 in o English as an L2.
The use o he AI- ool a he e ision/ e iew s age o hei ansla ion p ocess
would he e o e gi e hem mo e con idence abou he choices hey had made in
hei L2 ex p oduc ion. The class included 28 s uden s, and al hough English
is no he only L2 in hei p og amme, hei le el o English p o iciency can be
desc ibed as C1-C2.
A he ime o submi ing cou se desc ip ions, GenAI had no ye become pub-
licly accessible and was he e o e no included in he cou se desc ip ion. Hence,
i was a p ima y e hical and legal equi emen o he cou se design o assess
whe he he se o lea ning objec i es al eady iden i ied in he cou se desc ip-
ion con ac ually allowed o he in oduc ion o AI in o his inal-yea mas e ’s
cou se. I u ned ou ha he mos gene al lea ning objec i e, ac ing in an unp e-
dic able and complex con ex , and he lea ning objec i e including he app op ia e
me hods applied du ing he s uden ’s indi idual cogni i e ansla ion p ocess,
applying he app op ia e digi al ( e e ence) ools, made i possible o include an
exe cise wi h GenAI.
A he s a o he cou se, be o e some o he esea ch e iewed abo e was pub-
lished, ew s uden s had hea d abou GenAI, which hey would become mo e a-
milia wi h in hei compulso y language echnology module o he p og amme.
In o de o in oduce hem e y quickly o he GenAI ools, he s uden s we e
p esen ed a s aigh o wa d lis wi h di ec AI websi e links om which hey
could choose one, including Cha GPT-3.5, Pe plexi y and Jaspe . The e was also
a e e ence o u u e de elopmen s such as Me a’s Llama and Google Gemini.
The s uden ’s use o a gene a i e ool was in eg a ed in o hei ansla ion
p ocess as ollows. Fi s , s uden s we e asked o ansla e a passage om he
sou ce ex and place i on he cou se collabo a ion pla o m. Second, hey we e
asked o e ise and e iew a eam pa ne ’s ansla ion independen ly. Thi d,
he o iginal ansla o would submi he edi ed ansla ion o an AI- ool, asking
i o imp o e he ex o gi ing i ano he p omp o hei choice, and pas ing
he ou come unde nea h he edi ed ansla ion. Finally, hey would compa e he
AI ool e sion wi h he e ised ansla ion and indica e any di e ences wi h
196
10 AI in an L2 ansla ion class
an anno a ion; hey would pu a + sign in he balloon i hey ag eed wi h he
p oposed change o w i e an al e na i e solu ion, i hey did no ag ee wi h he
AI ool e sion.
Simila ly o p e ious non-AI ansla ion p ocesses, s uden s we e eminded o
adop a ame o mind ha is simila o ha o a eade o he a ge ex (in his
case, a o eign s uden , see below) and o ask hemsel es wha expec a ions he
ypical a ge eade migh ha e in e ms o egis e and e minology. They we e
also eminded o in e p e he sou ce ex as well as possible, and o be awa e o
spelling, layou , g amma and e minology.
As a esul , s uden s’ wo k would look as ollows. Figu e 1 shows he i s d a
submission o a s uden and he wo edi ing commen s made by hei pee s, one
o which has been complemen ed wi h a eache commen .
Figu e 1: A s uden ’s i s d a ansla ion wi h pee and eache eed-
back.
Figu e 2 illus a es he GPT-3.5 ou pu o an edi ing p omp gi en by he o ig-
inal ansla ion s uden and he own anno a ed commen s wi h pluses o al e -
na i e solu ions.
Fu he no ewo hy de ails o he ansla ion se ing in his cou se a e he
ollowing. Fi s , a eal-wo ld clien was in ol ed: he uni e si y’s legal o ice
needed a ansla ion o i s Uni e si y Codex om he s uden s’ main language
in o English. This sou ce ex mean ha legal language was in ol ed and hei
197
Sonia Vandepi e
Figu e 2: The GPT-edi ed e sion o he pee - e ised d a ( om Fig-
u e 1) wi h he ansla ing s uden ’s commen s.
a ge audience consis ed o o eign s a and s uden s. Since his ansla ion
se ing included a paying clien , s uden s would also be emune a ed (wi h book
okens) o hei ex a ansla ion wo k. In addi ion, he clien was a ailable o
any ques ions ha migh a ise ela ing o he ansla ion and s uden s would,
hence, be able o p ac ice clien communica ion. Fo his pu pose, a wo king
documen was p epa ed, called ‘Ques ions o he clien ’, in which s uden s could
w i e abou hei ansla ion p oblems. This documen was discussed in class and
ques ions ha we e e ained o e e al o he clien we e collec ed and sen
by he eache . Each communica ion om he clien was hen also p o ided o
he whole class in he same documen . The s uden s could u he consul bo h
a pa allel and compa able co pus and glossa ies, which had been compiled in
p e ious classes o ansla ions o he same clien and he same a ge ex .
Finally, as in o he L2 ansla ion classes, s uden s we e able o wo k on a
s uden collabo a ion pla o m, in which each s uden ’s con ibu ion was isible
ela i ely anonymously o e e yone.
All ma e ials p o ided by he s uden s who pa icipa ed in his p ojec ha e
ecei ed s uden consen o esea ch pu poses and may be used o illus a e
wha he s uden s we e capable o in his pilo p ojec .
4 Findings
Fi s , s uden s did no do much expe imen a ion wi h a a ie y o GenAI ools bu
only applied Cha GPT and only used i in hei e ision p ocess. I also u ned
ou ha some s uden s con used Machine-T ansla ed ex s wi h AI-gene a ed
ex s and e e ed o, o ins ance, Deep-L, which was al eady amilia o hem.
Howe e , s uden s did no p omp Cha GPT o p oduce a ansla ion.
198
10 AI in an L2 ansla ion class
Mo e impo an ly, s uden s we e able o iden i y di e ences be ween he ans-
la ion e sion ha had unde gone monolingual and bilingual e ision by hei
pee s and he ex as edi ed by he GenAI ool used. In pa icula , hey we e no
only able o iden i y be e o mula ions in he L2, bu also unwelcome omis-
sions, addi ions, and e en (some imes mino ) di e ences o meaning be ween
hei own e sions and he ones p o ided by Cha GPT.
The ollowing se o examples shows ha s uden s la gely ag eed wi h Cha -
GPT imp o emen s. They in ol ed he mo e idioma ic use o he Anglo-Saxon
geni i e (less equen ly used in hei own na i e language) (1), he a oidance
o a spli o-in ini i e cons uc ion (2) and he use o he ac i e a he han he
passi e oice (3):
(1) a. Pee - e ised d a
an opinion by he Execu i e Boa d is eques ed
b. GPT-edi ed e sion
he Execu i e Boa d’s opinion is eques ed
c. S uden commen
+
(2) a. Pee - e ised d a
he chai can decide o exclusi ely send he documen o membe s o
he Execu i e Boa d
b. GPT-edi ed e sion
he chai can decide o send he documen exclusi ely o membe s o
he Execu i e Boa d
c. S uden commen
+
(3) a. Pee - e ised d a
he o e iew o decisions is published by he sec e a ia o he Boa d o
Go e no s on he in ane
b. GPT-edi ed e sion
he sec e a ia o he Boa d o Go e no s mus publish an o e iew o
decisions on he in ane
c. S uden commen
+
The nex se o examples (4–6) illus a es s uden s’ ecogni ion o undesi able
addi ion and omission as p o ided by Cha GPT:
199
Sonia Vandepi e
(12) The ansla ion aims o accu a ely con ey he meaning o he o iginal
Du ch ex . Howe e , nuances in language and cul u al con ex migh no
be ully cap u ed. Fo mo e speci ic o legal in e p e a ions, i is always
ecommended o consul wi h a legal p o essional amilia wi h Belgian
law and uni e si y go e nance.
[…]
“Bes uu de s” has been ansla ed as “go e no s” o e lec he ole o
hese indi iduals in he uni e si y’s go e nance.
“Raad an Bes uu ” and “Bes uu scollege” ha e been ansla ed as “Boa d
o Go e no s” and “Execu i e Boa d” espec i ely, based on common
English e ms used in uni e si y go e nance.
The ph ase “maa schappelijk geëngagee de” has been ansla ed as
“socially engaged” o con ey he uni e si y’s commi men o socie al
issues.
(Google 2024)
No did he cou se design include any e alua ion s age o he s uden s’ indi id-
ual use o Cha GPT, which is an impo an s age o he s uden lea ning p ocess.
Ou complex and e e -changing wo ld will need ansla o s o be able o ans-
la e ex s hemsel es, pe o ming all he adi ionally equi ed cogni i e w i ing
p ocesses, as well as o pos -edi a GenAI- ool.
6 Conclusion
To conclude, in eg a ing AI in o he L2 ansla ion class oom wi h a eal-wo ld
ask no only esponds o he call o leade s a all le els o go e nmen o in o-
duce AI in o eaching in o de o main ain global compe i i eness. I also acili-
a es he oad o eaching basic lea ning objec i es, and, he e o e, en iches L2
ansla ion lea ning in a way ha enhances e lec i e hough . The la e may in
i sel inc ease in insic mo i a ion, a ocal poin in much L2 ansla ion aining.
Conside ing he impo ance o he ole o GenAI in educa ion (Labadze e al.
2023), his explo a o y p ojec ecommends he c i ical applica ion o a GenAI-
ool o be added as an addi ional lea ning objec i e in u u e L2 ansla ion
cou ses. I u he shows eache s he di ec ion in which he cou se design could
be imp o ed. In o de o e ie e all he ele an in o ma ion, s uden s can be
gi en a empla e o hei assignmen s wi h space o hei i s d a o he ans-
la ion, he monolingual and bilingual e ision o ha d a , a sc eensho o s u-
den s’ p omp s showing he gene a i e ool used, an AI cha bo ou pu wi h s u-
den s’ commen s, and a inal conclusion on he compa ison be ween he e ised
206
10 AI in an L2 ansla ion class
e sion and he AI cha bo ou pu (including he numbe and ype o changes
accep ed/igno ed and whe he hey would use he ool again). De o ing cou se
ime o explo e di e en ansla ion-gene a i e ools and o discuss hei ou pu
is also ecommended.
While his quali a i e inqui y has only ocused on jus one aspec o using AI
in ansla ion aining, o he eaching implica ions ha e been le undiscussed:
ways in which a cha bo can help he eache se up an examina ion ex on a
opic om ex s seen in class o design a schedule o pee collabo a ion in la ge
g oups.
Howe e , as Gao w i es: “GPT is no , and will no be, an adequa e subs i u e
o he language expe ise, cul u al awa eness, ideological sensi i i y, and c e-
a i e abili ies o ‘ e-w i e’ on he pa o p o essional human ansla o s” (2024:1).
The e o e, he aining o a cons uc i e while a he same ime c i ical a i ude
owa ds GenAI is essen ial, pe haps mo e s ongly han e e be o e. This should
keep he s anda ds o academic in eg i y and inc ease awa eness o local and in-
e na ional e hical and legal issues. Today, cou s ha e decided ha AI sys ems
do no ha e any igh s, no do hey ha e a legal pe sonali y (which is ese ed o
na u al and ju idical pe sons). No will AI sys ems e e show ue in elligence
(Chomsky 2023). The main p oblem wi h hem de i es om he ac ha hey
a e solely based on a olume o da a and algo i hms ha apply o ha olume
and we e p o ided by o en uniden i iable humans. In spi e o GPT’s ecen ly
implemen ed measu es o bias educ ion and con en mode a ion, ha olume
may s ill con ain bias, oxic language o o he ha m ul con en . Be e aining
and mo e eliable da a sou ces will no comple ely s op an AI- ool om yield-
ing hallucina ions o esponses ha do no mi o ue ac s, no will p o ec ion
measu es be oolp oo agains malicious ac o s. In o he wo ds, quali y con ol
will emain an ongoing e o ha use s need o be awa e o .
In o de o build he necessa y cons uc i e c i ical a i ude, a ious oads can
be aken by ansla ion aine s. They can p o ide o a c ea i e skills module
in he ansla ion cu iculum (Gue be o A enas & Asimakoulas 2023), hey can
make hei s uden s awa e o sus ainabili y issues in he e alua ion o he ech-
nology hey use (Moo kens e al. 2024), and/o hey can a ibu e human-cen ed
augmen ed in elligence (O’B ien 2024) a cen al place in hei own cou se.
Re e ences
Akh a , Za i Bin. 2024. Un eiling he e olu ion o gene a i e AI (GAI): A com-
p ehensi e and in es iga i e analysis owa d LLM models (2021–2024) and
207
Sonia Vandepi e
beyond. Jou nal o Elec ical Sys ems and In o ma ion Technology. 22. DOI: 10.
1186/s43067-024-00145-1.
Cao, Siyi & Linping Zhong. 2023. Explo ing he e ec i eness o Cha GPT-based
eedback compa ed wi h eache eedback and sel - eedback: e idence om Chi-
nese o English ansla ion. h ps://a xi .o g/abs/2309.01645.
Chomsky, Noam. 2023. The alse p omise o Cha GPT. The New Yo k Times. h ps:
//www.ny imes.com/2023/03/08/opinion/noam-chomsky-cha gp -ai.h ml.
Eloundou, Tyna, Sam Manning, Pamela Mishkin & Daniel Rock. 2023. GPTs a e
GPTs: an ea ly look a he labo ma ke impac po en ial o la ge language models.
h ps://a xi .o g/abs/2303.10130.
Fel en, Ed, Mana Raj & Robe Seamans. 2023. How will language modele s like
Cha GPT a ec occupa ions and indus ies? h ps://a xi .o g/abs/2303.01157.
Fi ia, Ti a Nu . 2023. A i icial in elligence (AI) echnology in OpenAI Cha GPT
applica ion: A e iew o Cha GPT in w i ing English essays. ELT Fo um: Jou -
nal o English Language Teaching 12(1). 44–58. DOI: 10.15294/el . 12i1.64069.
h ps://jou nal.unnes.ac.id/sju/el /a icle/ iew/64069.
Gao, Fei. 2024. When minds mee machines: A pilo p og amme o using Cha -
GPT o each poli ical speech ansla ion in a Chinese- o-English ansla ion
class oom. In e na ional Jou nal o Chinese and English T ansla ion & In e p e -
ing 5. 11 pp. DOI: 10.56395/ijce i. 3i1.94.
Geng, Xiang, Ming Zhu, Jiahuan Li, Zhejian Lai, Wei Zou, Shuaijie She, Jiaxin
Guo, Xiao eng Zhao, Yinglu Li, Yuang Li, Chang Su, Yanqing Zhao, Xinglin
Lyu, Min Zhang, Jiajun Chen, Hao Yang & Shujian Huang. 2024. Why no
ans o m cha la ge language models o non-English? h ps://a xi .o g/abs/
2405.13923.
Giampie i, Pa izia. 2024. AI and he BoLC: S eamlining legal ansla ion. Com-
pa a i e Legilinguis ics 58. 68–90. h ps://p ess o.amu.edu.pl/index.php/cl/
a icle/ iew/42397.
Google. 2024. Gemini (aug 15 e sion). La ge language model.
Göp e ich, Susanne. 2017. Cogni i e unc ions o ansla ion in L2 w i ing. In
John W. Schwie e & Aline Fe ei a (eds.), The handbook o ansla ion and
cogni ion, 402–422. Quebec, Canada: John Wiley & Sons.
G osman, Me a, Mi a Kad ic, I ena Ko ačič & Ma y Snell-Ho nby (eds.). 2000.
T ansla ion in o non-mo he ongues in p o essional p ac ice and aining. Tübin-
gen, Ge many: S au enbu g Ve lag.
Gue be o A enas, Ana & Dimi i Asimakoulas. 2023. C ea i e skills de elop-
men : T aining ansla o s o w i e in he e a o AI. He mes 63. 227–243. DOI:
10.7146/hjlcb. i63.143078.
208
10 AI in an L2 ansla ion class
House, Juliane. 2018. The impac o English as a global lingua anca on in e cul-
u al communica ion. In Andy Cu is & Roland Sussex (eds.), In e cul u al com-
munica ion in Asia: Educa ion, language and alues, 97–114. Be lin, Ge many:
Sp inge . DOI: 10.1007/978-3-319-69995-0_6.
K ings, Hans-Pe e . 1986. T ansla ion p oblems and ansla ion s a egies o ad-
anced Ge man lea ne s o F ench (L2). In Juliane House & Shoshana Blum-
Kulka (eds.), In e lingual and in e cul u al communica ion: Discou se and cogni-
ion in ansla ion and second language acquisi ion s udies, 263–276. Tübingen,
Ge many: Gun e Na .
Labadze, Lasha, Maya G igolia & Lela Machaidze. 2023. Role o AI cha bo s in
educa ion: Sys ema ic li e a u e e iew. In e na ional Jou nal o Educa ional
Technology in Highe Educa ion 20. 17 pp, 56. DOI: 10.1186/s41239-023-00426-1.
Moo kens, Joss, Sheila Cas ilho, Fede ico Gaspa i, An onio To al & Maja Popo ić.
2024. P oposal o a iple bo om line o ansla ion au oma ion and sus ain-
abili y: An edi o ial posi ion pape . Jou nal o Specialised T ansla ion (41). 2–
25. DOI: 10.26034/cm.jos ans.2024.4706.
Moo kens, Joss, Andy Way & Séamus Lank o d. 2025. Au oma ing ansla ion.
London, UK: Rou ledge.
Mujadia, Vandan, Ashok U lana, Yash Bhaska , Penumalla Adi ya Pa ani, Kukka-
palli Sh a ya, Pa ameswa i K ishnamu hy & Dip i Mis a Sha ma. 2023. As-
sessing ansla ion capabili ies o la ge language models in ol ing English and
Indian languages. h ps://a xi .o g/abs/2311.09216.
O’B ien, Sha on. 2024. Human-cen e ed augmen ed ansla ion: Agains an ag-
onis ic dualisms. Pe spec i es 32(3). 391–406. DOI: 10.1080/0907676X.2023.
2247423.
OpenAI. 2024. Cha GPT (Aug 15 e sion). La ge language model. h ps://cha .
openai.com/cha .
OpenAI. N.d. Tex gene a ion models. h ps://pla o m.openai.com/docs/guides/
ex -gene a ion.
Ouyang, Fan, Tuan Anh Dinh & Weiqi Xu. 2023. A sys ema ic e iew o AI-
d i en educa ional assessmen in STEM educa ion. Jou nal o STEM Educa ion
Resea ch 6. 408–426. DOI: 10.1007/s41979-023-00112-x.
Robe , Isabelle S., Jim J. J. U eel, Aline Remael & Ayla Rigou s Te yn. 2018. Con-
cep ualizing ansla ion e ision compe ence: A pilo s udy on he ‘ ai ness
and ole ance’ a i udinal componen . Pe spec i es 26(1). 2–23. DOI: 10.1080/
0907676X.2017.1330894.
Robinson, Na haniel R., Pe ez Ogayo, Da id R. Mo ensen & G aham Neubig.
2023. Cha GPT MT: Compe i i e o high- (bu no low-) esou ce languages.
209
Sonia Vandepi e
In P oceedings o he Eigh h Con e ence on Machine T ansla ion (WMT), 392–
418. h ps://aclan hology.o g/2023.wm -1.40.
Roumelio is, Kons an inos I. & Nikolaos D. Tselikas. 2023. Cha GPT and OpenAI
models: A p elimina y e iew. Fu u e In e ne 15. 192. DOI: 10.3390/ i15060192.
Saha i, Youse , Abdu M. Talib Al-Kadi & Jamal Kaid Mohammed Ali. 2023. A
c oss sec ional s udy o Cha GPT in ansla ion: Magni ude o use, a i udes,
and unce ain ies. Jou nal o Psycholinguis ic Resea ch 52. 2937–2954. DOI: 10.
1007/s10936-023-10031-y.
Teng, Shuhan. 2024. Explo ing he e ec o Cha GPT on he ansla ion o poe y
in li e a y wo ks: A case s udy o he Da id Hawkes ansla ion o A D eam
o Red Mansions. Dean & F ancis 1. 7. DOI: 10.61173/zg 1q485.
Ülkü, Abdullah. 2023. Cha GPT-4 o hospi ali y: Implica ions. Jou nal o Tou ism
and Gas onomy S udies 11(3). 1727–1743. DOI: 10.21325/jo ags.2023.1263.
Vandepi e, Sonia. 2023. AI in a ‘ ansla ion in o L2’ class. P esen a ion a he
Au umn EMT Ne wo k Mee ing. Salamanca, Spain.
Vidako ic, I ana. 2023. Cha GPT-p omp s oo e aling om de k ach an aal e
on slui en. h ps:// ex co ex.com/nl/pos /cha gp -p omp s- o - ansla ion.
Wang, Lulu, Simin Xu & Kanglong Liu. 2024. Unde s anding s uden s’ accep ance
o Cha GPT as a ansla ion ool: A UTAUT model analysis. h ps://a xi .o g/
abs/2406.06254.
210
