Chap e 9
Teaching sub i ling in he imes o
gene a i e AI
Da id O ego-Ca mona
Uni e si y o Wa wick, Uni ed Kingdom and Uni e si y o he F ee S a e, Sou h
A ica
This chap e explo es he in eg a ion o Gene a i e A i icial In elligence (GenAI)
in o sub i le aining and i s impac on ansla o educa ion. D awing on eaching
expe iences and esea ch insigh s, i p esen s s a egies o designing and deli e -
ing aining cou ses ha p epa e u u e sub i ling p o essionals o wo k e ec i ely
in an AI-enhanced indus y. Sub i ling is a mul i ace ed ansla ion p ac ice ha
equi es echnical, linguis ic and cul u al skills. Recen de elopmen s in GenAI a e
eshaping how hese skills a e augh and applied. Based on ealis ic scena ios and
explo a o y eaching me hods, he chap e examines how La ge Language Models
(LLMs) can suppo di e en s ages o he sub i ling p ocess h ough inquisi i e in-
eg a ion. The discussion encompasses p ac ical app oaches o in eg a ing GenAI
ools in o he class oom, d awing on examples ela ed o he ansla ion o cul u al
e e ences and empla e c ea ion. The chap e adop s a hands-on p oblem-sol ing
app oach o aining ha encou ages s uden s o e alua e echnological possibili-
ies whils de eloping ounda ional knowledge. Th ough examples om eaching
p ac ice, i shows how compa ing di e en AI solu ions and assessing hei sui -
abili y o speci ic asks helps s uden s make in o med decisions abou implemen -
ing au oma ed solu ions. This app oach posi ions s uden s as ac i e agen s in hei
lea ning p ocess while helping hem unde s and he po en ial and limi a ions o
au oma ion. C i ically examining he ole o educa o s in his changing landscape,
he chap e ad oca es o aining ha p epa es adap able p o essionals who can
na iga e echnological de elopmen s whils main aining high s anda ds and ad o-
ca ing o sus ainable wo king condi ions. Mo e b oadly, i con ibu es o discus-
sions abou p o iding s uden s wi h he necessa y ools and knowledge o shape
sus ainable ca ee s in an inc easingly au oma ed media localisa ion indus y.
Da id O ego-Ca mona. 2026. Teaching sub i ling in he imes 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?, 167–190. Be lin: Language Science P ess.
DOI: 10.5281/zenodo.17641080
Da id O ego-Ca mona
1 In oduc ion
This chap e add esses he ques ion o in eg a ing Gene a i e A i icial In-
elligence (GenAI) in o sub i le aining. Audio isual ansla ion (AVT) is a
echnology-d i en ield, and he econ igu a ion o sub i ling p ocesses ap ly
encapsula es how accele a ed GenAI-in luenced changes a ec educa ion.
Sub i le s equi e echnical, linguis ic and ans e skills, as well as ma ke
awa eness, o ope a e e icien ly in he indus y. T aining p o essional sub i le s
who can engage e ec i ely wi h he indus y in ol es “ aking in o accoun he
linguacul u al dimension as well as he echnological possibili ies and he ma -
ke eali y” (Díaz-Cin as & Remael 2021: 62). Un il ela i ely ecen ly, echnolog-
ical ools in sub i ling wo k lows ocused on echnical aspec s such as synch o-
nisa ion and segmen a ion, and some imes au oma ic speech ecogni ion (ASR)
(Geo gakopoulou 2020). The ield is expe iencing exponen ial g ow h in he im-
plemen a ion o machine ansla ion (MT) and GenAI solu ions (Sla o 2024),
pushed o wa d by Language Se ice Companies (LSCs) o speed up p ocesses
and add ess wha hey ha e called a alen c unch (Es opace 2017, Pape cup 2022),
al hough sub i le s a ibu e his so-called sho age o declining a es o emu-
ne a ion a he han an ac ual lack o a ailable p o essionals (O ego-Ca mona
2024). The expansion o he media localisa ion indus y has placed sub i ling as
one o he mos ac i e sec o s o he ma ke , esul ing in g owing in e es om
LSCs, inc easing demands on p oduc ion ne wo ks and p essu e on he wo k-
o ce.
This apid ans o ma ion p esen s unique challenges o sub i ling educa ion.
A su ey conduc ed by Sla o (2024) showed ha di e en ypes o cap ioning
and sub i ling se ices in ol ing AI a e commonly o e ed by Language Se ice
P o ide s. Bo h aine s and ainees mus con inuously adap hei knowledge
and skills as echnologies e ol e. The adi ional app oach o mas e ing es ab-
lished ools and wo k lows needs o gi e way o de eloping adap abili y and
c i ical e alua ion skills o eme ging echnologies.
This chap e explo es how we can s uc u esub i ling aining o add ess hese
challenges while main aining high p o essional s anda ds. D awing on eaching
expe iences and esea ch insigh s, I p esen s a egies o designing and deli e -
ing aining cou ses ha empowe u u e sub i ling p o essionals o wo k e ec-
i ely in an AI-shaped indus y. Unde s anding he condi ions, cons ain s and
oppo uni ies a e e y s age allows u u e p o essionals o make in o med deci-
sions and suppo s hei communica ion and nego ia ion skills. The discussion
encompasses p ac ical app oaches o in eg a ing GenAI ools in o he class oom
168
9 Teaching sub i ling in he imes o gene a i e AI
Figu e 1: Resul s o a su ey by Sla o om May 2024 ha asked 129
Language Se ice P o ide s which language AI se ices hei company
o e s
wi hin a p ojec -based app oach (Mi chell-Schui e oe de 2020), explo ing con-
e sa ional La ge Language Models (LLMs), and implemen ing p oblem-sol ing
ac i i ies o de elop c i ical e alua ion skills.
The chap e adop s a hands-on p oblem-sol ing app oach de eloped h ough
eaching p ac ice, equipping u u e g adua es o espond o he apidly changing
demands o he media localisa ion indus y. The p oposed ac i i ies in en ionally
ocus on non-con en ional sub i le asks o demons a e he wide- anging appli-
ca ion o AI echnologies and encou age a b oade e lec ion on AI in eg a ion.
The chap e examines how educa o s can keep up wi h echnological de elop-
men s and guide s uden s o ha ness AI’s po en ial while os e ing a deep un-
de s anding o linguis ic ea u es, cul u al con ex s, and e hical conside a ions
in insic o sub i ling, au oma ion, and he media indus y. Mo e b oadly, his
chap e con ibu es o discussions abou p o iding s uden s wi h he necessa y
ools and knowledge o shape a sus ainable ca ee .
2 E ol ing landscape o sub i ling aining
T aining sub i le s equi es add essing he echnical equi emen s o he p ac-
ice ela ed o synch onisa ion and spo ing while balancing he linguis ic and
cul u al equi emen s. T adi ional sub i ling wo k lows o single language pai s
ollowed a ela i ely s aigh o wa d p ocess: ansc ip ion o he sou ce con en ,
ansla ion, synch onisa ion (o spo ing) o sub i les wi h he audio isual con-
en , and quali y con ol (QC) (Díaz-Cin as & Remael 2021). Each s age equi ed
speci ic skills and was ypically ca ied ou sequen ially, wi h sub i le s wo king
on s andalone wo ks a ions using sub i ling so wa e.
The globalisa ion o he media indus y has led o he implemen a ion o em-
pla e-based wo k lows (Geo gakopoulou 2019). In his model, LSCs c ea e a mas-
e ile, usually in English, which se es as he basis o pi o ansla ions in o
mul iple a ge languages (Figu e 2) and s eamlines he mul ilingual sub i ling
169
Da id O ego-Ca mona
p ocess (Valdez e al. 2023). Templa es a e essen ial o he mul ilingual ope a-
ions o ansna ional pla o ms. Ne lix (n.d.) de ines sub i le empla es as “an
edi ed, posi ioned, esea ched, anno a ed and checked sub i le ile, imed o sho
and audio, ma ching he sou ce language o he associa ed con en ”. These em-
pla es include no only he ex bu also echnical pa ame e s and no es ha sup-
po ansla ions in o mul iple languages. Fo example, in p oducing sub i les
o Ne lix’s success ul Ko ean show “Squid Game”, an English empla e se es
as he pi o o ansla ions in o mul iple languages. A empla e o his show
would include no only he sc ip and iming bu also essen ial cul u al anno a-
ions. C i ical elemen s ha would equi e anno a ion include Ko ean hono i ic
o ms (indica ing ela ionships be ween cha ac e s), cul u al e e ences (such as
he child en’s games ha s uc u e he plo ), and social hie a chies exp essed
h ough linguis ic choice. Fo ins ance, he ela ionship be ween he wo cha ac-
e s who sha e a childhood iendship equi es ca e ul anno a ion as hei use o
in o mal language e lec s hei amilia i y, while hei in e ac ions wi h o he
cha ac e s ollow s ic social hie a chies e lec ed in o mal language pa e ns.
These anno a ions in he English empla e help ansla o s wo king in o o he
languages unde s and and p ese e hese nuanced ela ionships and cul u al el-
emen s in hei a ge e sions.
In some cases, hese empla es a e locked and do no allow ansla o s o al e
he imings o he sub i les. Thus, subsequen ansla ions need o adap o he
synch onisa ion and eading speed o he pi o language. This app oach s eam-
lines he p oduc ion p ocess bu also ans o ms he sub i le ’s ole om ha ing
con ol o e echnical and linguis ic aspec s o ocusing p ima ily on ansla ion
wi hin p e-es ablished cons ain s.
Templa e
c ea ion
Templa e
QC
T ansla ion
in o a ge
language(s)
T ansla ion
QC
P oduc
deli e y
Figu e 2: Templa e-based ansla ion p ocess
The sub i ling landscape is expe iencing an unp eceden ed ans o ma ion. As
Sza kowska & Jankowska (2024) obse e, wo k lows a e becoming inc easingly
au oma ed. They sugges ha manual spo ing and ansc ip ion may soon be-
come obsole e, eplaced by au oma ic speech- ecogni ion-based spo ing. This
coincides wi h Bolaños Ga cía-Esc ibano’s iew ha hese changes a e “pe -
haps leading o scena ios whe e spo ing will be inc easingly au oma ised and
linguis s will se e as language enginee s” (2025: 19). In cu en pos -edi ing
170
9 Teaching sub i ling in he imes o gene a i e AI
p ocesses, AI ools in he o m o ASR and MT eshape adi ional p ac ices and
embed e ision s ages o accommoda e he human e i ica ion o he au oma ic
ou pu (Figu e 3). These wo k lows ypically in ol e wo dis inc pos -edi ing
phases: ansc ip ion pos -edi ing, whe e sub i le s e i y and co ec au oma -
ically gene a ed ansc ip s om ASR sys ems, and ansla ion pos -edi ing,
whe e hey e ine machine- ansla ed con en . This p ocess ans o ma ion is
u he accele a ed by he shi om s andalone wo ks a ions o cloud-based
en i onmen s ha in eg a e ansla ion memo ies, QC au oma ion and MT and
ans e he con ol o hese esou ces o he LSCs.
Au oma ic
imed an-
sc ip ion
T ansc .
PE MT PE Quali y
con ol Deli e y
Figu e 3: Pos -edi ing (PE) p oduc ion p ocessing using ASR and MT
The implemen a ion o AI ools o echnical aspec s has a long adi ion in
sub i ling so wa e and sub i le s end o be mo e acquain ed wi h hem. Fea-
u es such as impo ing and segmen ing sc ip s, au oma ic iming and synch o-
nisa ion, sho -change iden i ica ion and au oma ed QC a e s anda d in sub i ling
wo k lows. Well-es ablished ools, such as Ooona, EZTi les and ZOOsubs in e-
g a e hese by de aul , wi h some e en allowing cus omisa ion.
Co e compe ences in sub i ling os e he mas e ing o echnical aspec s such
as synch onisa ion and segmen a ion, while de eloping he linguis ic and cul-
u al awa eness needed o p oduce high-quali y ansla ions. Howe e , he cu -
en indus y landscape demands addi ional skills. Sub i le s now need o unde -
s and and c i ically assess he ou pu o mul iple au oma ed sys ems, om ASR
o MT, when engaging wi h pos -edi ing wo k lows. Pos -edi ing AVT equi es
p o essionals o coo dina e an inc easing numbe o sou ces o in o ma ion and
e alua e hem be o e deciding o keep, edi o ully eplace au oma ically gene -
a ed con en .
The indus y’s push owa ds au oma ion esponds o mul iple ac o s (O ego-
Ca mona 2024). The expansion o s eaming pla o ms has inc eased he demand
o sub i led con en , wi h companies equi ing as u na ound imes o mul-
iple language combina ions. LSCs a gue ha a alen sho age cons ains hei
abili y o mee hese g owing demands, leading hem o implemen hyb id and
au oma ed solu ions (Iyuno SDI G oup 2022, Ma king 2022). Howe e , p o es-
sional associa ions disag ee wi h his assessmen , denying ha low paymen and
poo wo king condi ions, a he han a alen sho age, a e he p ima y conce ns
(AVTE 2023a).
171
Da id O ego-Ca mona
The changes in he p oduc ion p ocesses equi e a subs an ial e ision o how
we app oach sub i le aining. The ocus mus shi om eaching speci ic ech-
nical and ansla ion skills o de eloping adap able p o essionals who can c i -
ically engage wi h e ol ing echnologies while main aining high-quali y s an-
da ds. Like in o he a eas o ansla o aining, p og ammes need o add ess
new echnological equi emen s (ASR, MT, au oma ed spo ing, LLMs), indus y
demands, c i ical assessmen skills o AI-gene a ed ou pu s, and p o essional
de elopmen needs in a apidly changing en i onmen .
The challenge o aining p og ammes is o balance ansla ion skills and
echnical p o iciency wi h c i ical awa eness. S uden s need o unde s and no
only how o use new ools bu also how o e alua e hei sui abili y o di e -
en p ojec s, ma ke s and con ex s. This in ol es de eloping wha Tip on (2024)
calls digi al e lexi i y: he abili y o iden i y challenges in he ex e nal en i on-
men , e alua e a ailable esou ces and implemen app op ia e s a egies. Such
an app oach helps s uden s na iga e he changing landscape while main aining
p o essional s anda ds and de eloping sus ainable ca ee s.
3 Designing AI-enhanced aining and digi al e lexi i y
Digi al e lexi i y e e s o s uden s’ abili y o e alua e challenges and esou ces,
and implemen solu ions. This can be illus a ed h ough sub i ling p ojec wo k
whe e s uden s mus make in o med decisions a e e y s age. Fo ins ance, when
c ea ing sub i les, s uden s i s analyse hei sou ce p oduc ’s ansla ion needs,
including he echnical, cinema og aphic, linguis ic and cul u al aspec s. They
hen e alua e a ailable ools and esou ces, om sub i ling so wa e o AI solu-
ions, jus i ying how hese align wi h p ojec equi emen s. Th ough his p ocess,
s uden s lea n o iden i y po en ial challenges in he ex e nal en i onmen (such
as echnical cons ain s o cul u al e e ences), assess hei esou ces (bo h ech-
nological and knowledge-based), and implemen app op ia e s a egies o b idge
any gaps.
This e lec i e p ac ice mani es s in p ac ical decisions such as:
• Selec ing app op ia e so wa e based on p ojec needs
• Deciding which s ages o he wo k low can bene i om au oma ion
• E alua ing when o use AI ools and when human expe ise is c ucial
• Documen ing and jus i ying ansla ion decisions
172
9 Teaching sub i ling in he imes o gene a i e AI
Re lec i e p ac ice ex ends beyond echnical decision-making o include sel -
awa eness abou how echnological changes a ec p o essional iden i y, wo k
sa is ac ion, and ca ee expec a ions. S uden s need oppo uni ies o examine
hei own esponses o AI in eg a ion, bo h exci emen abou new possibili ies
and conce ns abou p o essional au onomy, and conside how hese e ol ing
condi ions align wi h hei pe sonal alues and ca ee goals. The esul ing com-
men a y demons a es how s uden s de elop digi al e lexi i y by connec ing
heo e ical knowledge wi h p ac ical implemen a ion whils main aining p o es-
sional s anda ds.
T aining u u e sub i ling p o essionals equi es a comp ehensi e app oach
ha ecognises his mul i ace ed na u e o he p o ession and he impac o ech-
nological de elopmen s on es ablished p ac ices. Building on he e lec i e p ac-
ice ou lined abo e, encou aging s uden s o explo e echnical aspec s (synch o-
nisa ion, segmen a ion, o ma ing), linguis ic aspec s ( ansla ion, cul u al adap-
a ion, egis e ), and echnological aspec s (AI ool e alua ion, wo k low in eg a-
ion), while examining he po en ial in eg a ion o au oma ion in o wo k lows
should p o ide hem wi h oppo uni ies o de elop hei skills, sys ema ically
e alua e he pe o mance o LLMs, ASR, MT and o he AI sys ems, and e lec
on how hese ools a ec hei p o essional iden i y and wo k sa is ac ion. As
an example o his ype o e lec i e p ac ice, his sec ion ocuses on he use o
LLMs o he ansla ion o cul u al e e ences o show how sub i ling p ocesses
can be augmen ed h ough he c i ical use o au oma ion ools (See O’B ien 2024
on augmen a ion in ansla ion).
3.1 Con as ing and assessing ansla ions o sub i les
Con e sa ional LLMs and o he AI ools can be in eg a ed ea ly on in he ain-
ing p ocess. A common exe cise used in in oduc o y sub i ling classes is o sha e
examples o cul u al adap a ions aine s and s uden s ind in he sub i les hey
encoun e . A e selec ing some examples, s uden s e lec on he aspec s ele-
an o he ansla ion o cul u al e e ences in sub i ling (See Díaz-Cin as &
Remael 2021). These examples can be analysed using LLMs o allow s uden s o
assess sub i ling decisions and unde s and he adap a ion p ocess, con as ing
hei de eloping knowledge o sub i ling agains he decisions o o he sub i le s
and he explana ions p o ided by LLMs.
The example (1) below shows he ansc ip o he sub i les om Ana omy o
a Fall (di . T ie , 2023) shown on Amazon P ime in he UK. The o iginal e sion
o he ilm is mul ilingual, including dialogues in English, F ench and Ge man.
173
Da id O ego-Ca mona
(1) Ana omy o a Fall, 00:39:08:00-00:39:25:00
a. F ench:
T ois élémen s en pa iculie : un appo d’expe ise su
des p ojec ions de sang de M. Maleski e ou ées au domicile
du couple, e don la di ec ion semble indique qu’il a eçu
un coup iolen à la ê e alo s qu’il se ou ai su le
balcon du 2ème é age.
b. English:
Th ee elemen s in pa icula . Analyses o M Maleski’s blood
spa e ound a he couple’s domicile indica e he may ha e
ecei ed a blun o ce blow o he head when he was on he
hi d- loo balcony.
Ha ing access o he sou ce and a ge e sions o he ex , s uden s can use
LLMs as agen s o assess how he sys ems would espond o he asks and how
sub i ling, as a p ac ice, is depic ed in he ou pu . Example (2) shows he p omp
used and he esponses p o ided by h ee LLMs.
(2) P omp : You a e ac ing as an agen suppo ing he ask o ansla ing
sub i les. This ex is pa o he sc ip o he ilm Ana omy o a Fall.
T ansla e i and adap i o sub i les o he o iginal ilm in English: T ois
élémen s en pa iculie : un appo d’expe ise su des p ojec ions de
sang de M. Maleski e ou ées au domicile du couple, e don la di ec ion
semble indique qu’il a eçu un coup iolen à la ê e alo s qu’il se
ou ai su le balcon du 2ème é age.
a. Cha GPT (4o mini):
Sub i le Adap a ion:
"Th ee key elemen s: an expe epo on blood spa e
belonging o M . Maleski ound in he couple's home."
"The blood pa e n sugges s he was s uck iolen ly on he
head while on he second- loo balcony."
Le me know i his i s well o i any adjus men s a e
needed!
b. Copilo , GPT4:
Su e, he e’s he ansla ion and adap a ion o sub i les:
“Th ee key elemen s: an expe epo on M . Maleski’s blood
spla e ound a he couple’s home, indica ing he was
174
9 Teaching sub i ling in he imes o gene a i e AI
s uck iolen ly on he head while on he second- loo
balcony.”
Does his wo k o you needs?
c. Claude 3.5 Sonne :
Le me help you wi h he sub i le ansla ion and
adap a ion, ocusing on cla i y and p ope sub i le o ma .
Since his appea s o be pa o e idence being discussed in
he ilm, I'll o ma i app op ia ely o sub i ling:
Th ee elemen s in pa icula :
a o ensic epo on M . Maleski's blood spa e
ound in he couple's home,
whe e he di ec ion appea s o indica e
he ecei ed a iolen blow o he head
while on he second- loo balcony.
The ex is b oken down in o eadable chunks, each line
keeping unde he s anda d 42-cha ac e limi o sub i les.
This main ains eadabili y and ollows s anda d sub i le
iming p ac ices. Would you like me o adjus he line
b eaks o pacing di e en ly?
The esul s o his ini ial es indica e dispa i ies ac oss he sys ems. Apa
om p o iding a ansla ion, Claude (2c) p o ides con ex bo h abou he scene
and he na u e o sub i ling. The esul ing ansla ion also displays he limi a ion
o sub i les (segmen a ion, cha ac e limi a ion, iming and eadabili y), p o id-
ing s uden s wi h an oppo uni y o e lec on sub i ling guidelines and ins uc-
ions.
Ha ing access o he o icial ansla ion, s uden s can ask he sys ems o com-
pa e he wo e sions. When asked o con as he o iginal ansla ion agains
he new p oposed ansla ion, Claude again s ands ou by displaying con ex ual
awa eness (3). Fo ins ance, i s a es ha he o iginal e sion’s decision o ans-
la e domicile as ‘domicile’ ins ead o ‘home’ e lec s language ha is ‘mo e o -
mal/legal, i ing o a cou oom scene’.
(3) P omp : The ansla ion o he sub i les in English eads as below. Can
you explain he di e ences om you ansla ion? “Th ee elemen s in
pa icula . Analyses o M Maleski’s blood spa e ound a he couple’s
domicile indica e he may ha e ecei ed a blun o ce blow o he head
when he was on he hi d- loo balcony.”
175
Da id O ego-Ca mona
QC and documen a ion o e addi ional oppo uni ies o he in eg a ion o AI
ools in sel -assessmen . S uden s can use LLMs o iden i y inconsis encies and
e alua e he po en ial ecep ion o hei decisions. They can use he sys em o
ex ac eco ds o hei choices, e alua e consis ency, and gauge he a ionale
behind hei ansla ions as decoded by he sys em. This can p o e pa icula ly
use ul o eam p ojec s, whe e g oups need o coo dina e he ansla ion o
mul iple episodes. Wo king oge he , s uden s can lea n o de elop s a egies
ha ensu e consis ency ac oss episodes, p epa ing hem o eal-wo ld scena ios
whe e la ge eams collabo a e in localisa ion p ojec s.
Recen de elopmen s in media localisa ion and AI poin o u he changes in
how sub i les a e c ea ed and in eg a ed in o audio isual con en (Sla o 2024).
The eme gence o LLMs capable o p ocessing mul imodal inpu will likely con-
inue ans o ming sub i ling wo k lows. In he nea u u e, i is possible hese
sys ems will e icien ly iden i y speake s, analyse con ex based on isual ele-
men s and sugges iming based on scene composi ion. Some o hese ools a e
in de elopmen bu no ye deployed. E en i hese echnologies a e no cu -
en ly in eg a ed in o media localisa ion p ocesses a scale, p epa ing s uden s
o unde s and upcoming mul imodal de elopmen s is essen ial. The expansion
o accessibili y se ices, such as audio desc ip ion, and he de elopmen o au-
oma ic dubbing sugges p o essionals will s ill need o analyse in e ac ions be-
ween di e en meaning-making modes and make decisions conside ing hei
a ge audiences. Ac i i ies and cou ses ha suppo mul imodal analysis and in-
eg a e AI ools while main aining human agency in decision-making encou age
u u e p o essionals o hink s a egically abou sub i ling wo k lows and hei
ole in an e ol ing indus y.
4 Tailo ing wo k lows in sub i ling aining
The p oposed in eg a ion o AI ools in o sub i ling aining mo es beyond
eaching s uden s how o use speci ic sys ems o p omo e an unde s anding o
he conside a ions ele an o he p oduc s, p ocesses and p oduce s. I os e s
a p oblem-sol ing mindse ha enables s uden s o de elop lexible, con ex -
app op ia e and indus y-in o med wo k lows. By encou aging s uden s o
c i ically engage wi h di e en ools and app oaches, hey can de elop he
ounda ional knowledge and adap abili y needed in he cu en indus y land-
scape. Fu he , he con inuous e e ence o academic knowledge and indus y
expe ise empowe s s uden s o mo e swi ly be ween sec o s.
182
9 Teaching sub i ling in he imes o gene a i e AI
A key componen o his app oach is allowing s uden s o design hei own
sub i ling p ojec s and ake owne ship o all p ojec - ele an decisions: om ma-
e ial selec ion o he ailo ing o he ansla ion si ua ion. Adop ing a p ojec -
based app oach o ansla o aining (Ki aly 2006, Mi chell-Schui e oe de
2020) and a he han wo king wi h p e-de ined pa ame e s, s uden s selec
he audio isual p oduc s o be ansla ed, de ine he p oduc ion condi ions, se-
lec he mos sui able guidelines and e lec on he ecep ion equi emen s ha
ame hei ansla ions. This au onomous decision-making p ocess helps s u-
den s unde s and how di e en ac o s – audience needs, echnical cons ain s,
cul u al conside a ions, and echnological ools – in luence wo k low design.
Fo ins ance, a s uden migh choose o sub i le a documen a y aimed a a spe-
cialis audience, equi ing ca e ul conside a ion o e minology managemen
and po en ially bene i ing om AI ools o quali y assu ance. Ano he migh
wo k on a comedy clip whe e cul u al e e ences and wo dplay demand mo e
human in e en ion and suppo o c ea i e solu ions. Classes can also in eg a e
spaces o sha e and discuss he s uden s’ decisions along he way: sho p og ess
p esen a ions o online discussion o ums o p o ide p og ess upda es can help
s uden s lea n om each o he .
Th ough hese sel -di ec ed p ojec s, s uden s lea n o:
• E alua e which s ages o he wo k low can bene i om au oma ion
• Assess he sui abili y o di e en AI ools o speci ic asks
• Unde s and when human expe ise is mos c i ical and when au oma ion
p o ides he mos ad an ages
• T ansla e hei p oduc s, e lec ing on speci ic ansla ion needs
• Design QC p ocesses ha combine au oma ed and manual checks
• Documen hei decision-making p ocess o u u e e e ence
The lexibili y o explo e di e en wo k lows b idges academic expe ise wi h
indus y de elopmen s in meaning ul ways. While academic app oaches em-
phasise sys ema ic analysis and an in-dep h unde s anding o sub i ling p inci-
ples, indus y p ac ices ocus on e iciency, ma ke ele ance and echnological
inno a ion. By allowing s uden s o design hei wo k lows, his independen
p ojec -based app oach c ea es oppo uni ies o hem o unde s and how heo-
e ical p inciples suppo in o med decision-making in p ac ical scena ios. S u-
den s lea n o apply academic esea ch on eading speeds (K uge e al. 2022,
183
Da id O ego-Ca mona
Sza kowska & Bogucka 2019, Sza kowska e al. 2024) and segmen a ion (Ge be -
Mo ón e al. 2018) o e alua e he ou pu o au oma ed sys ems, o d aw on ans-
la ion heo y o assess MT sugges ions. This in eg a ion o pe spec i es helps s u-
den s de elop digi al e lexi i y (Tip on 2024: 86) o suppo indus y p ac ices.
As LSCs con inue o deploy new au oma ed solu ions ha challenge adi ional
wo k lows and companies de elop mo e ools, s uden s need o engage c i ically
wi h he p ocess, eso o solid ounda ional knowledge and be able o swi ly
in eg a e in o new echnological p oduc ion en i onmen s. Ha ing a sound he-
o e ical basis allows s uden s o c i ically assess hese de elopmen s and adap
hei p ac ice acco dingly a he han ollowing p esc ibed p ocedu es.
This p oposal p omo es collabo a i e lea ning en i onmen s ha bene i bo h
s uden s and aine s. When s uden s a e empowe ed o es di e en ools and
assess hei pe o mance, hey de elop he con idence o access new knowledge
au onomously. Ra he han posi ioning u o s as he sole sou ce o in o ma ion,
he esponsibili y o engaging wi h eme ging echnologies and indus y p ac ices
is sha ed be ween u o s and s uden s (Ki aly 2003, 2006). This app oach ecog-
nises ha aine s hemsel es a e na iga ing a apidly e ol ing landscape and
need o con inuously upda e hei skills. The esul ing dynamic suppo s aine s’
con inuing p o essional de elopmen as hey lea n alongside s uden s, os e ing
an en i onmen o mu ual g ow h ha be e e lec s he cons an echnological
changes in he indus y.
5 E hical conside a ions
The implemen a ion o au oma ed solu ions and, in pa icula , he popula isa ion
o LLMs has made discussions abou he e hical dimensions o hese p ac ices
essen ial (Moo kens 2020, 2024). Acco ding o he S a emen on Gene a i e A -
i icial In elligence published by AVTE (AudioVisual T ansla o s Eu ope, he Eu-
opean Fede a ion o Na ional Associa ions o Audio isual T ansla o s), he use
o AI in c ea i e ansla ion does no espond o c ea i e needs bu an in en ion
o inc ease p o i “while comp omising he li elihood o p o essional ansla o s
and o e ing audiences subpa p oduc s” (AVTE 2023b: 5).
T aining ac i i ies mus be designed and implemen ed wi h ca e ul conside a-
ion o e hical implica ions. Acknowledging ha engaging wi h AI ools is now
essen ial o employabili y in he indus y, bu i is also impo an o ecognise
and add ess he e hical and social issues o LLMs. Fi s ly, he use o copy igh -
p o ec ed ma e ials in LLM aining aises conce ns abou in ellec ual p ope y
igh s. C ea i e indus ies, such as audio isual ansla ion, ha e been pa ic-
184
9 Teaching sub i ling in he imes o gene a i e AI
ula ly ac i e in denouncing he deploymen o LLMs as an exploi a i e p ac-
ice whose impac is inc emen ed by he deploymen o AI-powe ed p oduc-
ion models (AVTE 2023b, ATRAE 2021, En chai e en Os 2024). Secondly, and
mo e impo an ly, aining should acknowledge how he implemen a ion o AI
ools migh a ec wo king condi ions in he indus y (AVTE 2023a, Ka akan a
e al. 2022, Koponen e al. 2020). P o essional associa ions ha e poin ed ou ha
he push o pos -edi ing wo k lows in ol ing AST, MT and LLMs comes wi h
a demand o educed a es and inc eased wo kload o ansla o s, changing
wo king condi ions and po en ially a ec ing job sa is ac ion (do Ca mo 2020,
Sakamo o e al. 2024).
Ensu ing he employabili y o g adua es demands aining ha is awa e and
esponsi e o he deploymen o AI (EMT 2022), bu educa ion equi es a c i i-
cal e alua ion o hese ools and p oduc ion models o equip s uden s wi h he
knowledge and skills o e alua e he impac o socie al, p o essional and ecolog-
ical sus ainabili y (Moo kens e al. 2024). These conce ns esona e wi h b oade
discussions abou he e hical implica ions o AI de elopmen , including issues
o bias in aining da a (Aka e al. 2021, Johnson e al. 2022), he en i onmen-
al impac o LLMs (Luccioni e al. 2023), and he isk o pe pe ua ing cul u al
hegemonies h ough au oma ion and s anda disa ion (Bende e al. 2021).
Ensu ing ha s uden s a e awa e o he capabili ies and limi a ions o he sys-
ems is also an e hical ask as aining p og ammes ha e a esponsibili y o p e-
pa e s uden s o he equi emen s o he ma ke . LSCs a e s eadily implemen ing
AI solu ions in o hei wo k lows (O ego-Ca mona 2024, Sla o 2024) and g ad-
ua es need hands-on expe ience wi h hese ools o success ully in eg a e in o
he p o ession. They need o be able o assess when he sys ems a e sui able o
no o a ask. This will only be achie ed h ough amilia isa ion. Fu he , when
he sys ems a e deemed sui able, i should also be u u e ained p o essionals
who decide how o deploy hem o ensu e hey a e augmen ing, no hinde ing,
he capabili ies o o he p o essionals (O’B ien 2024). Ac i i ies should os e
c i ical discussions abou he impac o AI on bo h he p o ession and socie y
a la ge, encou aging s uden s o conside how hese ools can suppo a he
han eplace human expe ise and de elop he p ac ical compe ences he ma ke
demands. This c i ical engagemen mus include he e lec i e p ac ice ha al-
lows s uden s o assess how di e en echnological implemen a ions a ec hei
wo k sa is ac ion and align wi h hei p o essional alues. The goal is o p epa e
s uden s o engage wi h AI ools e hically and s a egically, unde s anding bo h
hei po en ial and limi a ions, while de eloping he agency o ad oca e o ai
wo king condi ions and esponsible AI implemen a ion in an inc easingly au o-
ma ed indus y.
185
Da id O ego-Ca mona
6 Concluding ema ks
T aining sub i le s in he age o GenAI equi es balancing sub i ling-speci ic
skills wi h awa eness abou he ma ke and unde s anding media localisa ion
p ac ices. This ype o aining encompasses, i s ly, ensu ing g adua es de elop
he skills hey need o en e he indus y, which equi es a ho ough unde s and-
ing o sub i ling p inciples, wo k lows and ools. Secondly, he inc easing imple-
men a ion o au oma ed solu ions o educa e e lec i e p o essionals who can
adap o apidly changing p oduc ion se ings. And las ly, aining ha os e s
e hical awa eness o suppo esponsible engagemen wi h echnology and ad-
oca es o sus ainable p o essional p ac ices.
Fu u e-p oo ing sub i ling educa ion equi es in eg a ing eme ging echnolo-
gies while main aining co e p o essional compe ences. The p oblem-sol ing ap-
p oach p esen ed in his chap e esponds o hese needs by c ea ing spaces o
s uden s o explo e echnological possibili ies and de elop ounda ional knowl-
edge. Engaging wi h LLMs and o he GenAI ools allows s uden s o unde s and
he po en ial and limi a ions o au oma ion. By compa ing di e en solu ions and
e alua ing hei sui abili y o speci ic asks, s uden s lea n o make in o med de-
cisions abou implemen ing au oma ed solu ions. The abili y o iden i y speci ic
limi a ions h ough compa a i e analysis o di e ging ou pu s is a ans e able
skill ha will bene i s uden s in i ually any u u e ca ee in ol ing AI sys ems.
Fu he , his app oach encou ages hem o hink s a egically abou hei ole in
an e ol ing indus y.
The ac i i ies discussed demons a e how sub i ling aining can in eg a e
echnological de elopmen s main aining a ocus on quali y and p o essional s an-
da ds. These ac i i ies show s uden s how o ha ness GenAI ools o suppo hei
wo k a he han allowing au oma ion o dic a e hei p ac ice. The goal is o p e-
pa e adap able p o essionals who can na iga e echnological changes, main ain
high s anda ds and ad oca e o sus ainable wo king condi ions.
The con inuous de elopmen o GenAI ools sugges s he media localisa ion
indus y will keep e ol ing and g adua es need o be eady o assess, c i ique and
deploy GenAI-powe ed solu ions. This equi es ca e ul conside a ion o how au-
oma ion a ec s sub i ling p ocesses, os e ing c i ical engagemen wi h eme g-
ing echnologies and p omo ing e hical awa eness. Only by add essing hese
aspec s can aining p og ammes p epa e p o essionals who will con ibu e o
de eloping a sus ainable media localisa ion indus y.
186
9 Teaching sub i ling in he imes o gene a i e AI
Re e ences
Aka, Osman, Ken Bu ke, Alex Baue le, Ch is ina G ee & Ma ga e Mi chell. 2021.
Measu ing model biases in he absence o g ound u h. In P oceedings o he
2021 AAAI/ACM Con e ence on AI, E hics, and Socie y (AIES ’21), 327–335. Vi -
ual E en , USA: Associa ion o Compu ing Machine y. DOI: 10.1145/3461702.
3462557.
ATRAE. 2021. Comunicado sob e la posedición. Asociación de T aducción y
Adap ación Audio isual. h ps://a ae.o g/comunicado-sob e-la-posedicion/.
AVTE. 2023a. 2023 AVTE su ey on audio isual ansla o s’ wo king condi ions.
AudioVisual T ansla o s Eu ope. h ps:/ / p ezi . com/ i /a83j _ udsgd j / a e -
su ey-2022-23/.
AVTE. 2023b. AVTE s a emen on gene a i e a i icial in elligence. AudioVisual
T ansla o s Eu ope. h ps://a eu ope.o g/a e-s a emen - on- gene a i e-
a i icial-in elligence.
Bende , Emily M., Timni Geb u, Angelina McMillan-Majo & Shma ga e
Shmi chell. 2021. On he dange s o s ochas ic pa o s: Can language models
be oo big? In P oceedings o he 2021 ACM Con e ence on Fai ness, Accoun abil-
i y, and T anspa ency (FAccT ’21), 610–623. Vi ual E en , Canada: Associa ion
o Compu ing Machine y. DOI: 10.1145/3442188.3445922.
Bolaños Ga cía-Esc ibano, Alejand o. 2025. P ac ices, educa ion and echnology
in audio isual ansla ion. London, UK: Rou ledge.
Díaz-Cin as, Jo ge & Aline Remael. 2021. Sub i ling: Concep s and p ac ices. Lon-
don, UK: Rou ledge.
do Ca mo, Félix. 2020. “Time is money” and he alue o ansla ion. T ansla ion
Spaces 9(1). 35–57. DOI: 10.1075/ s.00020.ca .
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 .
En chai e en Os. 2024. Li é a u e, cinéma, p esse, jeux idéo: non à des aduc ions
sans âme. h ps://enchai e enos.o g/.
Es opace, Eden. 2017. Talen c unch hi s media localiza ion as Amazon, Ne lix
accele a e ma ke g ow h. h ps : / / sla o . com / alen - c unch - hi s - media -
localiza ion-as-amazon-ne lix-accele a e-ma ke -g ow h/.
Geo gakopoulou, Panayo a. 2019. Templa e iles: The holy g ail o sub i ling.
Jou nal o Audio isual T ansla ion 2(2). 137–160.
Geo gakopoulou, Panayo a. 2020. The aces o audio isual ansla ion. In The
many aces o ansla ion: F om ideo games o he Va ican (DG TRAD con e ence
187
Da id O ego-Ca mona
2019), 16–33. Eu opean Union. DOI: doi/10.2861/898773. h ps://op.eu opa.eu/
s/y4PS.
Ge be -Mo ón, Oli ia, Agnieszka Sza kowska & Bencie Woll. 2018. The impac
o ex segmen a ion on sub i le eading. Jou nal o Eye Mo emen Resea ch
11(4). 1–18. DOI: 10.16910/jem .11.4.2.
Iyuno SDI G oup. 2022. A simpli ied look a he sub i ling p oduc ion p ocess. h ps:
/ / iyuno . com / news / whi e - pape s / a - simpli ied - look - a - he - sub i ling -
p oduc ion-p ocess.
Johnson, Rebecca L, Giada Pis illi, Na alia Menédez-González, Leslye Denisse
Dias Du an, En ico Panai, Julija Kalpokiene & Donald Jay Be ul o. 2022. The
ghos in he machine has an Ame ican accen : Value con lic in GPT-3. a Xi .
DOI: 10.48550/ARXIV.2203.07785.
Ka akan a, Alina, Luisa Ben i ogli, Mau o Ce olo, Ma eo Neg i & Ma co Tu chi.
2022. Pos -edi ing in au oma ic sub i ling: A sub i le s’ pe spec i e. In P o-
ceedings o he 23 d Annual Con e ence o he Eu opean Associa ion o Ma-
chine T ansla ion, 261–270. Ghen , Belgium: Eu opean Associa ion o Machine
T ansla ion. h ps://aclan hology.o g/2022.eam -1.29.
Ki aly, Don. 2003. F om ins uc ion o collabo a i e cons uc ion: A passing ad
o he p omise o a pa adigm shi in ansla o educa ion? In B ian James
Bae & Geo ey S. Koby (eds.), Beyond he i o y owe . e hinking ansla ion
pedagogy, 3–27. Ams e dam, Ne he lands: John Benjamins. DOI: 10.1075/a a.
xii.03ki .
Ki aly, Don. 2006. P ojec -based lea ning: A case o si ua ed ansla ion. Me a
50(4). 1098–1111. DOI: 10.7202/012063a .
Koponen, Maa i , Umu Sulubacak, Kaisa Vi ikainen & Jö g Tiedemann. 2020.
MT o sub i ling: Use e alua ion o pos -edi ing p oduc i i y. In P oceedings
o he 22nd Annual Con e ence o he Eu opean Associa ion o Machine T ansla-
ion, 115–124. Lisboa, Po ugal: Eu opean Associa ion o Machine T ansla ion.
h ps://aclan hology.o g/2020.eam -1.13.
K uge , Jan-Louis, Na alia Wisniewska & Sixin Liao. 2022. Why sub i le speed
ma e s: E idence om wo d skipping and e eading. Applied Psycholinguis ics
43(1). 211–236. DOI: 10.1017/S0142716421000503.
Lommel, A le. 2020. Augmen ed ansla ion: a e we he e ye ? h ps://csa- esea ch.
com/Blogs-E en s/Blog/augmen ed- ansla ion-2020.
Luccioni, Sasha, Ch is ophe Akiki, Ma ga e Mi chell & Yacine Je ni e. 2023. S a-
ble bias: E alua ing socie al ep esen a ions in di usion models. In A. Oh, T.
Naumann, A. Globe son, K. Saenko, M. Ha d & S. Le ine (eds.), Ad ances
in neu al in o ma ion p ocessing sys ems, ol. 36, 56338–56351. Cu an Asso-
188
9 Teaching sub i ling in he imes o gene a i e AI
cia es. h ps : / / p oceedings . neu ips . cc / pape _ iles / pape / 2023 / ile /
b01153e7112b347d8ed54 317840d8a -Pape -Da ase s_and_Benchma ks.pd .
Ma king, Ma ion. 2022. Iyuno-SDI and Disney on o e coming he challenges o
localizing a global scale. h ps://sla o .com/iyuno-sdi-disney-on-o e coming-
he-challenges-o -localizing-a -global-scale/.
Mi chell-Schui e oe de , Rosema y. 2020. A p ojec -based app oach o ansla ion
echnology. London, UK: Rou ledge.
Moo kens, Joss. 2020. “A iny cog in a la ge machine”: Digi al Taylo ism in he
ansla ion indus y. T ansla ion Spaces 9(1). 12–34. DOI: 10.1075/ s.00019.moo.
Moo kens, Joss. 2024. ‘I am no a numbe ’: On quan i ica ion and algo i hmic
no ms in ansla ion. Pe spec i es 32(3). 477–492. DOI: 10.1080/0907676X.2023.
2278536.
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.
Ne lix. n.d. Timed ex s yle guide: sub i le empla es. h ps : / / pa ne help .
ne lixs udios.com/hc/en-us/a icles/219375728-Timed-Tex -S yle-Guide-
Sub i le-Templa es.
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.
O ego-Ca mona, Da id. 2024. Placing human agency in he AI-powe ed media
localisa ion indus y. Jou nal o Audio isual T ansla ion 7(2). 1–21.
Oziemblewska, Magdalena & Agnieszka Sza kowska. 2022. The quali y o em-
pla es in sub i ling. A su ey on cu en ma ke p ac ices and changing sub-
i le compe ences. Pe spec i es 30(3). 432–453. DOI: 10.1080/0907676X.2020.
1791919.
Pape cup. 2022. T ansla o alen c unch leads o in es men in machine lea n-
ing. h ps://www.pape cup.com/blog/ ansla o - alen - c unch-leads- o-
in es men -in-machine-lea ning.
Sakamo o, Akiko, Da en Van Laa , Joss Moo kens & Félix Do Ca mo. 2024. Mea-
su ing ansla o s’ quali y o wo king li e and hei ca ee mo i a ion: Concep-
ual and me hodological aspec s. T ansla ion Spaces 13(1). 54–77. DOI: 10.1075/
s.23026.sak.
Sla o . 2024. Sla o p o guide: Audio isual ansla ion. h ps://sla o .com/p o-
guide-audio isual- ansla ion/.
189
Da id O ego-Ca mona
Sza kowska, Agnieszka & Lidia Bogucka. 2019. Six-second ule e isi ed: An eye-
acking s udy on he impac o speech a e and language p o iciency on sub-
i le eading. T ansla ion, Cogni ion & Beha io 2(1). 101–124. DOI: 10.1075/ cb.
00022.sza.
Sza kowska, Agnieszka & Anna Jankowska. 2024. In oducing audio isual ans-
la ion. 1s edn. London, UK: Rou ledge.
Sza kowska, Agnieszka, Valen ina Ragni, Da id O ego-Ca mona, Sha on Black,
Sonia Szk iba, Jan-Louis K uge , K zysz o K ej z & B eno Sil a. 2024. Impac
o ideo and sub i le speed on sub i le eading: An eye- acking eplica ion
s udy. Jou nal o Audio isual T ansla ion 7(1). DOI: 10.47476/ja . 7i1.2024.283.
Tip on, Rebecca. 2024. The Rou ledge guide o eaching e hics in ansla ion and
in e p e ing educa ion. London, UK: Rou ledge.
Valdez, Susana, Hanna Pię a, Es e To es-Simón & Ri a Menezes. 2023. Sub i-
le s’ belie s abou pi o empla es. Ta ge 35(3). 426–454. DOI: 10.1075/ a ge .
00019. al.
190
