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
he ield ansla ion. The e is, indeed, ample anecdo al e idence o he use o
LLMs as MT ools.
(5) below is an example o a p omp ha could be used o ask Cha GPT o ans-
la e a piece o ex . Yamada (2023) explo ed how inco po a ing he ansla ion’s
pu pose and he a ge audience in o p omp s a ec s he quali y o ansla ions
gene a ed by Cha GPT. The e o e, in his speci ic example, he cha bo will be
gi en a conc e e pe sonali y, he language combina ion o be used will be English
in o F ench, he con en o be ansla ed is ela ed o enewable ene gy, and he
ansla ion is o be published in a go e nmen epo .
(5) Con en ansla ion p omp in Cha GPT. (C ea ed by Ramon Inglada)
I I we e o go ahead and en e his p omp in Cha GPT ollowed by a ex o
ansla ion, he LLM would p omp ly gene a e a ansla ed e sion o he con-
en in ques ion (in F ench in his case). As men ioned ea lie , he h ee ini ial
pa ag aphs om he English e sion o he Wikipedia a icle on he opic o e-
newable ene gy ha e been used as he sou ce ex . A his poin , I would ha e wo
op ions. Op ion 1 would simply en ail accep ing he gene a ed ansla ed con en
a ace alue, wi hou conduc ing any checks and wi hou e i ying he sui abil-
i y o he said con en (depending on a se ies o ac o s ha could include o e all
quali y, egis e , one, in ended audience, publica ion media and so on). Op ion 2
would in ol e ope a ionalising he concep o AI Li e acy. Using my co e skills, I
could pe o m all he checks men ioned be o ehand. I could selec o he po en ial
ansla ion al e na i es eques ed o he LLM o any passages ha I migh no
be sa is ied wi h ( o ins ance, asking Cha GPT o eph ase any gi en pa o he
ansla ed ex ). Finally, I could also assess he o e all sui abili y o he gene a ed
ou pu . The ele ance and use ulness o bo h he ini ial ansla ion-gene a ing
p omp and any addi ional p omp s en e ed in he cha bo ( o ins ance o check
o al e na i e o mula ions o o change he le el o o mali y in he ansla ed
ex ) would be g ea ly dependen on he use o e ec i e p omp ing echniques.
6.3 Example 3: Te minology ex ac ion
One o he po en ial uses o GenAI ools ha was quickly iden i ied as po en ially
highly in e es ing o ansla o s, in e p e e s, e minologis s, esea che s and
o he p o essionals was ha o a e minology ex ac ion ool. Use s can eques
Cha GPT o c ea e a lis o key e ms om a gi en passage o ex . Use s can ask
56
3 AI Li e acy: The concep o sui abili y and co e ansla ion skills
o a speci ic numbe o e ms o be iden i ied, and hey can e en eques he ool
o p esen he esul s in a able o ma , o acili a e u he wo k. An example o a
p omp ha could be used o his pu pose (using a ex in he ield o enewable
ene gy, including all he equi emen s men ioned abo e and also a pe sonali y)
is shown below:
(6) P omp : You a e a p o essional e minologis . Please ex ac a lis o 15
key e ms om he ollowing ex in he ield o enewable ene gy and
p esen hem in able o ma . Please also include a de ini ion.
(7) The i s ou e ms ela ed o enewable ene gy and hei de ini ions, as
p o ided by Cha GPT (C ea ed by Ramon Inglada)
As can be seen in (7), when using his p omp and p o iding Cha GPT 3.5 wi h
a ex o pe o m he e minology ex ac ion on, he LLM will e y quickly p o-
duce a con enien able con aining he speci ied numbe o e ms in he ield in
ques ion and also hei de ini ions. Once again, I could simply accep he au oma -
ically c ea ed selec ion o e ms and hei co esponding de ini ions as ele an
and sui able o ou needs (wha e e hey migh be). Howe e , I could also use
c i ical hinking and in his ins ance, again, apply he concep o AI Li e acy. The
speed and con enience wi h which Cha GPT has c ea ed his lis o e ms and
co esponding de ini ions is as ounding (and se e al o de s o magni ude as e
han pe o ming a simila p ocess manually), bu how can any use ensu e he
sui abili y o he gene a ed e ms and de ini ions i he co e skills a e lacking?
How can I selec he candida e e ms ha migh indeed be use ul o ou needs
in any gi en ask? How can he sui abili y (and e en accu acy) o he de ini ions
p o ided be assessed? And how can I speci y a conc e e numbe o e ms o be
57
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
