Chap e 8
Compu e -assis ed language media ion
in eaching human-cen ed augmen ed
ansla ion
Ma ia Zimina-Poi o
Uni e si é Pa is Ci é, F ance
In he e ol ing landscape o ansla ion educa ion, he in eg a ion o compu e -
assis ed language media ion has eme ged as a key s a egy o eaching human-
cen ed augmen ed ansla ion. This s udy explo es he implemen a ion o gene a-
i e AI (GenAI) models, in pa icula La ge Language Models (LLMs), in ansla ion
wo k lows o enhance ansla o compe ence while main aining e hical s anda ds.
Using cus om AI sys ems ha adap and e ol e wi h new ansla ions, s uden s can
be ained o c i ically e alua e AI-gene a ed con en and manage hyb id ansla-
ion wo k lows ha combine AI ou pu wi h human expe ise. The esea ch em-
phasises he impo ance o unde s anding he limi a ions and po en ial biases o
AI, and ad oca es a balanced app oach in which AI augmen s a he han eplaces
human judgemen . P ac ical examples and exe cises demons a e he s eng hs and
weaknesses o using AI in in o ma ion p ocessing and mul ilingual ex gene a ion,
wi h he ul ima e aim o aining a new gene a ion o ansla o s o use AI ech-
nologies esponsibly.
1 In oduc ion
1.1 GenAI in ansla ion educa ion
In ansla ion educa ion, in eg a ing gene a i e AI (GenAI) models in o ans-
la ion p ocesses p esen s challenges in main aining ansla o au onomy while
emb acing echnology (O’B ien 2024). E hical conce ns, including p i acy, al-
go i hmic bias, and esponsible use o machine ansla ion and linguis ic da a,
Ma ia Zimina-Poi o . 2026. Compu e -assis ed language media ion in eaching
human-cen ed augmen ed ansla ion. 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?,
147–165. Be lin: Language Science P ess. DOI: 10.5281/zenodo.17641078
Ma ia Zimina-Poi o
mus be ca e ully conside ed as ansla ion educa ion unde goes a pa adigm shi
(Ramí ez-Polo & Va gas-Sie a 2023, Slimi & Ca ballido 2023, Ta a es e al. 2023).
De eloping compu e -assis ed language media ion in ansla ion educa ion e-
qui es success ully in eg a ing da a analysis and AI assis an s in o pedagogical
wo k lows. In his espec , AI is one o he building blocks o ansla ion ech-
nologies, along wi h compu e -assis ed ansla ion en i onmen s, p ojec man-
agemen so wa e, on ologies, expe in ol emen and da a managemen acili-
ies (Ga cia 2023, He e al. 2023, Mi chell-Schui e oe de 2020).
1.2 Compu e -assis ed language media ion
Used as pa o a compu e -assis ed dynamic media ion amewo k (Hu e al.
2024), GenAI ools allow ansla ion s uden s no only o p oduce mul ilingual
con en , bu also o compa e and ine- une he gene a ed ou pu , os e ing a
deepe con ex ual unde s anding o he s eng hs and weaknesses o di e en
ansla ion models and enabling u u e ansla ion p o essionals o become mo e
p o icien e iewe s o au oma ically gene a ed con en (Zhuang e al. 2024).
This p ocess equi es new ansla ion skills and up- o-da e knowledge o AI as-
sis an s buil on unde lying AI models, such as La ge Language Models (LLMs)
(Douglas 2023).
As LLM echnology is in oduced o he ield o ansla ion, enabling ainees
o na iga e language a ia ions wi hin adap i e pa hways, he concep o a lin-
guis ic no m eme ges as an inc easingly c ucial analy ical ool, acili a ing e -
icien ansla ion ac oss di e se con ex s (Fab icius 2022, Schulz & Ollig 2023,
Sinne 2020). F om his pe spec i e, compu e -assis ed language media ion can
be augh as a se o new ansla ion compe ences, including he abili y o c e-
a e and use a language assis an , and po en ially o unde s and how o employ
LLM-enhanced linguis ic da a esponsibly (Raza e al. 2025).
F om a echnical s andpoin , building pe sonalised AI assis an s om sc a ch
equi es subs an ial esou ces, emphasising he e iciency o le e aging p e-
ained models (Shichkina & K inkin 2022). This p ocess can be pa o aining
p og ams speci ically ailo ed o p epa e da a scien is s specialised in ansla-
ion. Few-sho lea ning (FSL),1pa icula ly in-con ex lea ning (ICL),2empowe s
1Few-Sho Lea ning (FSL) is a machine lea ning app oach ha enables models o adap
o new asks wi h only a small amoun o labelled da a (Mosbach e al. 2023). Fo
mo e in o ma ion: h ps://medium.com/ubiai-nlp/s ep-by-s ep-guide- o-mas e ing- ew-sho -
lea ning-a673054167a0
2In-con ex lea ning, as popula ised in LLMs such as GPT-3 and GPT-4, in ol es a model’s
capaci y o comp ehend and execu e asks based on con ex ual in o ma ion embedded wi hin
he inpu sequence, wi hou modi ying he model’s pa ame e s (Dong e al. 2024).
148
8 Compu e -assis ed LM in eaching human-cen ed AT
ansla o s wi h limi ed na u al language p ocessing (NLP) expe ise. This ap-
p oach allows LLMs o apidly acqui e new capabili ies, he eby acili a ing
seamless in eg a ion o AI in o a ange o ansla ion wo k lows. In his ega d,
p omp enginee ing3 acili a es he de elopmen o a e sa ile app oach ha can
be applied o a ange o asks, including mul ilingual ansla ion, ine- uning o
ansla ion models (Doğ u & Moo kens 2024), in o ma ion ex ac ion, and ex
summa isa ion. T aining u u e ansla o s o e ec i ely p omp LLMs o hese
asks in digi al communica ion is essen ial. Ou analysis in Sec ion 3 will p o ide
speci ic examples ha illus a e hese app oaches in eaching.
1.3 In es iga ing biases and inconsis encies o GenAI
An impo an a ea o esea ch in eaching GenAI echnologies in ansla ion
cou ses is o make s uden s awa e o he po en ial biases and inconsis encies hey
may encoun e when using his new gene a ion o ools. The sho comings o AI-
gene a ed con en become appa en when he ansla ion ask is ully mas e ed.
Un o una ely, i s uden s a e no ully p o icien in he a ge language and lack
specialised knowledge and ansla ion p ac ice, hey may ind i di icul o e ec-
i ely assess he quali y o he ou pu (Ta a es e al. 2023). Unde s anding bo h
he s eng hs and limi a ions o GenAI h ough a se ies o p ac ical examples is
i al o e icien use o ansla ion ools (Fa ell 2023).
By exposing s uden s o eal-li e scena ios and p ac ical exe cises, i is possi-
ble o de elop a be e unde s anding o he quali ies and sho comings o au o-
ma ically gene a ed con en in he con ex o mul ilingual communica ion. This
hands-on app oach allows ainee ansla o s o iden i y po en ial biases, incon-
sis encies and e o s ha may a ise om elying solely on AI assis ance. In addi-
ion, by analysing and discussing hese p ac ical examples, s uden s can lea n o
c i ically e alua e he ou pu o AI assis an s and iden i y when human in e en-
ion o addi ional expe ise is equi ed o p oduce accu a e and con ex ually el-
e an con en . This knowledge is also essen ial o de eloping a new gene a ion
o ansla ion- e ision wo k lows ha success ully combine GenAI and human
expe ise.
3The ield o p omp enginee ing is conce ned wi h he design and op imisa ion o p omp s
o help a model pe o m be e on new asks wi h limi ed da a. Fo p omp s o be e ec i e,
hey mus clea ly de ine he ask a hand, include con ex o help he model use i s exis ing
knowledge, and g adually in oduce new in o ma ion o asks.
149
Ma ia Zimina-Poi o
2 Con ex ual unde s anding and au oma ion
A i icial in elligence is now capable o p ocessing a wide a ie y o in o ma ion,
including ex , images, speech, acial exp essions, and, o a ce ain ex en , body
mo emen s. These capabili ies can be used o de elop a la ge panel o AI assis-
an s (Yang e al. 2024). One o he mos in iguing aspec s o hese echnologies
is hei handling o con ex , which di e s signi ican ly om human unde s and-
ing (O’B ien 2024, Zhu e al. 2024).
Human unde s anding in ol esa combina ion o expe ience, emo ion, cul u al
backg ound and si ua ional awa eness. This allows people o na u ally in e p e
nuances, implici meanings and sub le ies in communica ion. Assessing human
unde s anding o con ex h ough w i en exp ession has long been a ocus o
esea ch, wi h me hodologies e ol ing o encompass cogni i e, linguis ic, and
echnological pe spec i es (O’B ien 2024). Resea che s in cogni i e psychology
and linguis ics ha e de o ed conside able ime and e o o s udying he mech-
anisms o language p ocessing, wi h a pa icula ocus on he ole o lexical ac-
cess, syn ac ic pa sing and discou se cohe ence in comp ehension. In addi ion,
discou se analysis has p o ided aluable insigh s in o how con ex ual cues shape
in e p e a ion, e ealing complex in e play be ween language and cogni ion in
human comp ehension (Van Dijk 2006, Pleye & Win e s 2015, Gledhill & Pec-
man 2018).
In con as , GenAI models p ocess con ex h ough he iden i ica ion o pa -
e ns wi hin he da a hey lea ned du ing he aining phase (Zhuang e al. 2024).
In con as o a ue unde s anding o con ex , hese models ely on as quan i-
ies o da a o iden i y and gene a e esponses based on s a is ical co ela ions. To
illus a e, when p esen ed wi h a sen ence, an AI assis an analyses he sequence
o wo ds and p edic s he mos p obable con inua ion based on he pa e ns i has
lea ned. Cu en ly, his app oach can e ec i ely imi a e human-like esponses,
bu i lacks genuine comp ehension and he capaci y o g asp mo e p o ound
meanings o unexpec ed cul u al nuances. Addi ionally, i ends o accen ua e
ecognised pa e ns in he gene a ed ou pu .
This di e ence in con ex ual unde s anding has signi ican implica ions o
ansla ion. While AI may be good a gene a ing cohe en and con ex ually ap-
p op ia e ex , i may s uggle wi h asks ha equi e cul u al sensi i i y o
complex p oblem-sol ing ha equi es nuanced human judgemen . Fu he mo e,
AI’s unde s anding o con ex is limi ed o he da a on which i has been ained.
I aining da a se s lack di e si y o a e biased, AI-gene a ed con en will e-
lec hese limi a ions (Ma ínez e al. 2023). Unde s anding hese di e ences is
150
8 Compu e -assis ed LM in eaching human-cen ed AT
c ucial when in eg a ing GenAI assis an s in o ansla ion wo k lows. While AI-
based echnologies can signi ican ly inc ease p oduc i i y and e iciency by pe -
o ming ou ine asks and p o iding quick in o ma ion, hey should complemen
a he han eplace human judgemen and expe ise in ansla ing ce ain ypes
o con en (Denning & A quilla 2022).
Balancing he s eng hs o AI in p ocessing and gene a ing ex wi h ad anced
human skills used o unde s and and in e p e complex con ex s can lead o
mo e e icien and nuanced applica ions o ansla ion echnology o ansla ion
p ojec managemen , known as human-cen ed augmen ed ansla ion (O’B ien
2024).
In p ac ical e ms, wha speci ic asks can help s uden s unde s and GenAI
weaknesses as e lec ed in he wa ning “Cha GPT can make mis akes. Check
impo an in o.” displayed by Cha GPT (h ps://cha gp .com)? In wha ypes o
si ua ions a e hese ools help ul, and wha a e he po en ial pi alls? The ollow-
ing sec ion explo es he poin o s udying biases and inconsis encies in GenAI
ope a ion h ough p ac ical examples o class oom ac i i ies.
3 Explo ing AI in p ac ice: del ing in o knowledge- ich
ex s
3.1 Tex summa isa ion and in e ac i e lea ning wi h AI assis an s
In his sec ion, we conside how p ac ical asks can help s uden s gain a deepe
unde s anding o GenAI. One o he easies ways o add ess hese complex
p ocesses in ansla ion educa ion is o use examples o knowledge- ich ex s
(Gledhill & Küble 2016). In his con ex , in o ma ion syn hesis is one o he key
capabili ies ha AI b ings o s eamline in o ma ion p ocessing, wi h ools such
as SciSpace and PDFgea being able o summa ise key in o ma ion almos in-
s an ly o e icien e ie al.4In his espec , linguis ic s anda ds, speci ica ions
and s yle guides p o ide an excellen basis o p ac ice (Gledhill & Zimina-
Poi o 2023). These well-s uc u ed documen s p ecisely ou line he s anda ds
ha should be applied in w i en communica ion. A he same ime, he leng h
and complexi y o hese documen s o en equi e de ailed human analysis and
ex ensi e p ac ice o ully mas e he ules.
To illus a e how AI can be used in eaching, le us conside a speci ic example.
In a Mas e ’s le el 2 cou se on echnical ansla ion and con olled languages, s u-
4SciSpace is an ad anced pla o m designed o esea che s ha uses na u al language p ocess-
ing and machine lea ning o acili a e e icien li e a u e e iew and analysis in scien i ic ields.
151
Ma ia Zimina-Poi o
den s a e expec ed o amilia ise hemsel es wi h ASD-STE100 (Simpli ied Tech-
nical English).5ASD-STE100 is an in e na ional speci ica ion o w i ing echni-
cal documen a ion in a con olled na u al language.6This comp ehensi e docu-
men o o e ou hund ed pages equi es conside able e o o mas e he p in-
ciples o STE o w i ing p ocedu es, desc ip i e echnical ex s, wa nings and
cau ions. In addi ion, he e is an ex ensi e “co e dic iona y” con aining a wide
ange o lexical i ems ha s uden s mus lea n o use in con ex . The dic iona y is
accompanied by illus a i e examples and con as ing cases (APPROVED: STE/
no app o ed: non-STE).
In eg a ing AI in o ASD-STE eaching can po en ially imp o e he lea ning
expe ience. The ollowing examples highligh se e al ways in which his can be
achie ed.
• Use case: Use ools such as PDFGea o SciSpace o quickly summa ise
leng hy sec ions o ASD-STE100.
• Bene i : By using a concise o ma in class oom se ings wi h limi ed each-
ing ime, s uden s can ocus on unde s anding and p ac ical applica ions
o he STE ules a he han being o e whelmed by he olume o ex .
In his con ex , AI assis an s can ac as in e ac i e u o s, answe ing s uden s’
ques ions abou speci ic language policies and p o iding immedia e eedback.
P omp cla i ica ion and ein o cemen can accele a e lea ning. In addi ion, AI
assis an s can quickly loca e speci ic sec ions o ules wi hin leng hy documen s
h ough enhanced esea ch and e e ence, sa ing ime and imp o ing e iciency.
The main bene i is ha s uden s can spend mo e ime applying he ules and
less ime sea ching o hem.
A he same ime, h ough his ype o exe cise, s uden s will ealise ha bo h
PDFGea and SciSpace copilo s a e qui e e ec i e a ex ac ing ele an knowl-
edge and p o iding a comp ehensi e lis o essen ial ules o be applied in w i ing.
Ne e heless, his knowledge ex ac ion is based on pa e ns in he ex and may
be supe icial when i comes o applying hese ules in w i ing and ew i ing
5See cu iculum o Mas e 2 ILTS, Uni e si é Pa is: h ps://od .u-pa is. / /o e-de-
o ma ion/mas e -XB/a s-le es-langues-ALL/ aduc ion-in e p e a ion-K6JMSAFS/
mas e - aduc ion-in e p e a ion-pa cou s-indus ie-de-la-langue-e - aduc ion-specialisee-
JRQNFV5Z.h ml
6ASD-STE100 (h ps://www.asd-s e100.o g) is a se o w i ing ules and a con olled ocabula y
o p oducing clea and simpli ied echnical documen a ion in English. I was de eloped in
he 1980s by he ae ospace indus y associa ions (AECMA/ASD and AIA) o make ai c a
main enance documen a ion mo e eadable o people wi h limi ed English p o iciency.
152
8 Compu e -assis ed LM in eaching human-cen ed AT
ex s. S uden s can apidly become awa e o hese limi a ions and gain insigh
in o he echnical aspec s o p oducing con en wi h GenAI. This dual awa eness
can help hem o unde s and bo h he po en ial and he limi a ions o using AI
assis an s in he con ex o echnical communica ion and ansla ion:
• Use case: Use PDFgea and SciSpace copilo s o gene a e examples and
coun e -examples based on he co e dic iona y and ASD-STE100 ules,
hen c i ically e alua e he AI’s ou pu .
• Bene i s: Fi s ly, his ype o exe cise encou ages ac i e lea ning and c i -
ical hinking as s uden s compa e AI-gene a ed con en wi h examples o
human ew i ing. Secondly, i p o ides insigh in o he limi a ions o au-
oma ic ex gene a ion in he p ac ical applica ion o w i ing ules.
The ollowing sec ion will demons a e how p e ious exe cises can be u he
de eloped in o a mo e conclusi e expe imen , using ools such as Cha GPT, Pe -
plexi y AI, o Claude.ai.7In his case, a p omp language model (inpu in Tech-
nical English and ou pu in Simpli ied Technical English) would be employed o
demons a e he ou comes o ew-sho lea ning on a speci ic da a se . In addi-
ion, p omp s can be used o highligh inconsis encies in he gene a ed ou pu ,
he eby di ec ing a en ion owa ds po en ial enhancemen s (such as he a oid-
ance o noun clus e s and he use o a icles encou aged by STE).
3.2 AI ou pu e inemen using p omp enginee ing and ew-sho
lea ning
F om a echnical pe spec i e, building LLMs om sc a ch equi es signi ican e-
sou ces, which highligh s he e iciency o using p e- ained models. This p ocess
can be in eg a ed in o eaching p og ams speci ically designed o ain da a scien-
is s specialising in ansla ion. Few-sho lea ning, especially in-con ex lea ning,
enables ansla o s wi h limi ed NLP expe ise o quickly acqui e new skills and
seamlessly in eg a e AI in o ansla ion wo k lows ac oss mul iple applica ions.
In he con ex o ew-sho lea ning, a model is p esen ed wi h a limi ed numbe
o labelled examples co esponding o a desi ed inpu -ou pu scena io. T ans e
lea ning p inciples a e embedded in he model o enable i o use hese examples
7Cha GPT is a con e sa ional AI model de eloped by OpenAI based on he GPT (Gene a i e
P e- ained T ans o me ) a chi ec u e. Based on OpenAI’s model, and a s andalone LLM wi h
NLP capabili ies, Pe plexi y AI is a con e sa ional sea ch engine ha answe s que ies using
na u al language p edic i e ex , using sou ces om he web. Claude.ai is ained by An h opic
using Cons i u ional AI (Bai e al. 2022).
153
Ma ia Zimina-Poi o
o adjus i s pa ame e s and ine- une i s ep esen a ions o enhance he p es-
ence o speci ic pa e ns. The main ad an age o ew-sho lea ning is i s abili y
o gene alise om limi ed examples o unseen da a, due o he ich abs ac ep-
esen a ions acqui ed du ing p e- aining. Howe e , e en in his scena io, he
esul s may no be en i ely sa is ac o y and may equi e human e inemen .
• Use case: C ea e a p omp language model wi h labelled examples o “Inpu
ex in Technical English” and “Ou pu ex in STE”, demons a ing he
esul s o ew-sho lea ning on a gi en da a sample.
• Bene i : This hands-on exe cise allows s uden s o in e ac di ec ly wi h
he echnology and gain p ac ical expe ience using p omp language mod-
els o ex gene a ion asks.
In addi ion, his app oach p o ides an oppo uni y o s uden s o obse e he
limi a ions o AI in ully cap u ing complex linguis ic nuances, highligh ing he
impo ance o human analysis and edi ing o e ine he ou pu o sa is ac o y
esul s.
3.3 Te minology ex ac ion wi h LLMs: unde s anding model
limi a ions
In pa allel, o he ini ia i es a e exploi ing he capabili ies o combining AI and
NLP o acili a e access o specialised knowledge. Teaching e minology ex ac-
ion om mul imodal sou ces using LLMs is one o he a eas which can ein o ce
lea ne s’ augmen ed easoning in ansla ion.
Be o e using LLMs wi h mul imodal capabili ies (such as Pe plexi y AI play-
g ound: h ps://labs.pe plexi y.ai o h ps://gemini.google.com) o he analysis
and ex ac ion o e ms om mul iple sou ces, i is bene icial o s uden s o
con as he ou comes gene a ed by LLMs wi h hose p oduced by con en ional
e m ex ac ion, ex -mining, and ex anno a ion ools. This compa ison enables
lea ne s o become awa e o po en ial biases and hallucina ions. Fo example,
he equencies o candida e e ms iden i ied by GenAI a e no always 100% ac-
cu a e compa ed o adi ional ex segmen a ion and e m ex ac ion me hods.
AI models can p opaga e e o s om misin e p e a ions in he ex , esul ing in
inaccu a e e m equencies. Mis eading sen ence s uc u e can lead o inco ec
coun s o ce ain e ms (Uchida 2024).
In addi ion, AI models ained on di e se and la ge da ase s may inapp op i-
a ely apply pa e ns lea ned in one con ex o ano he . Fo example, i a model
154
8 Compu e -assis ed LM in eaching human-cen ed AT
has equen ly seen “clima e change” discussed alongside e ms such as “policy”
and “ egula ion”, i migh o e es ima e he equency o hese e ms in any docu-
men men ioning “clima e change”, e en i he speci ic ex ocuses on scien i ic
da a a he han policy.
These e o s highligh a key limi a ion o AI models: hei eliance on pa e ns
and associa ions lea ned om as amoun s o da a, which can some imes lead o
misin e p e a ions i he speci ic con ex o a new ex di e s om hose pa e ns.
The gene a ed ou pu exagge a es he lea ned pa e ns. As a esul , he equency
o e ms iden i ied by GenAI may no always accu a ely e lec he ac ual ex
and i s con ex , especially in cases whe e a nuanced unde s anding o language
is equi ed. The e o e, compa ing AI-gene a ed esul s wi h hose o adi ional
ex segmen a ion and e m ex ac ion ools, which o en use s aigh o wa d
s a is ical me hods, can help iden i y and co ec such inaccu acies.
• Use case: Use LLMs o ex ac e ms om di e en ypes o documen s.
Check he ex ac ed e ms o accu acy and ele ance. Use ex -mining
and na iga ion ools (such as Voyan Tools: h ps:// oyan - ools.o g) o
c i ically e alua e he esul s p oduced by AI assis an s.
• Bene i : Lea n o use quan i a i e me hods o e alua e he quali y o AI-
gene a ed ex analysis.
No e ha eady-made p omp s o ex analysis can be ound on many websi es.8
Such models can be adap ed, o example, o e m ex ac ion (wo ks o Pe plex-
i y AI):
• Use case (sample p omp ): Ex ac he mos ele an e ms wi h con ex s
om he a ached documen . P esen he esul s in a ma kdown able wi h
he ollowing columns: e m, equency, con ex s, de ini ion.9
3.4 Towa ds “augmen ed easoning” in ansla ion: linguis ic da a and
LLMs
Many o he expe imen s wi h AI ools a e possible in he class oom, bo h o
demons a e he ole ha AI models can play in di e en applica ion se ings,
8Fo example, Keywo ds E e ywhe e Cha GPT P omp Templa es: h ps://
keywo dse e ywhe e.com/cha gp -p omp - empla es.h ml (also a ailable o Claude.ai).
9Mo e examples a e a ailable in #2024TEF -AI-powe ed e minology ex ac ion: A
hands-on guide o ansla o s by Josh Goldsmi h: h ps://you u.be/5Y5PhzyeMGI?lis =
PLLqIRaiVCGCR80 ysPOQ2AHJ30HEyI XX
155
Ma ia Zimina-Poi o
McKinnon, Ca ol Chen, Ca he ine Olsson, Ch is ophe Olah, Danny He nan-
dez, Dawn D ain, Deep Ganguli, Dus in Li, Eli T an-Johnson, E han Pe ez,
Jamie Ke , Ja ed Muelle , Je ey Ladish, Joshua Landau, Kamal Ndousse,
Kamile Lukosui e, Liane Lo i , Michael Selli o, Nelson Elhage, Nicholas
Schie e , Noemi Me cado, No a DasSa ma, Robe Lasenby, Robin La son, Sam
Ringe , Sco Johns on, Shauna K a ec, Shee El Showk, S anisla Fo , Tam-
e a Lanham, Timo hy Telleen-Law on, Tom Cone ly, Tom Henighan, T is an
Hume, Samuel R. Bowman, Zac Ha ield-Dodds, Ben Mann, Da io Amodei,
Nicholas Joseph, Sam McCandlish, Tom B own & Ja ed Kaplan. 2022. Cons i-
u ional AI: Ha mlessness om AI eedback. h ps://a xi .o g/abs/2212.08073.
Denning, Pe e J. & John A quilla. 2022. The con ex p oblem in a i icial in elli-
gence. Communica ions o he ACM 65(12). 18–21. DOI: 10.1145/3567605.
Doğ u, Gökhan & Joss Moo kens. 2024. Da a augmen a ion wi h ansla ion
memo ies o desk op machine ansla ion ine- uning in 3 language pai s.
Jou nal o Specialised T ansla ion (41). 149–178. DOI: 10.26034/cm.jos ans.
2024.4716.
Dong, Qingxiu, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia,
Jingjing Xu, Zhiyong Wu, Tianyu Liu, Baobao Chang, Xu Sun, Lei Li & Zhi ang
Sui. 2024. A su ey on in-con ex lea ning. h ps://a xi .o g/abs/2301.00234.
Douglas, Michael R. 2023. La ge language models. h ps://a xi .o g/abs/2307.
05782.
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 .
Fab icius, Anne. 2022. Wha ’s in a sociolinguis ic no m? The case o change in
p e ocalic / / in ecei ed p onuncia ion. In Janus Mo ensen & Kamilla K a
(eds.), No ms and he s udy o language in social li e, 157–184. Be lin, Ge many:
De G uy e . DOI: 10.1515/9781501511882-007.
Fa ell, Michael. 2023. P elimina y e alua ion o Cha GPT as a machine ansla-
ion engine and as an au oma ic pos -edi o o aw machine ansla ion ou -
pu om o he machine ansla ion engines. In P oceedings o he In e na ional
Con e ence on Human-in o med T ansla ion and In e p e ing Technology (HiT-
IT), 108–113. DOI: 10.26615/issn.2683-0078.2023_009.
Ga cia, Ignacio. 2023. Compu e -aided ansla ion sys ems. In Chan Sin-wai (ed.),
Rou ledge encyclopedia o ansla ion echnology, 2nd edn., 76–94. London, UK:
Rou ledge. DOI: 10.4324/9781003168348-4.
Gledhill, Ch is ophe & Na alie Küble . 2016. Wha can linguis ic app oaches
b ing o English o speci ic pu poses? ASp (69). 65–95. DOI: 10.4000/asp.4804.
162
8 Compu e -assis ed LM in eaching human-cen ed AT
Gledhill, Ch is ophe & Mojca Pecman. 2018. On al e na ing p e-modi ied and
pos -modi ied nominals such as aspi in syn hesis s. syn hesis o aspi in:
Rhe o ical and cogni i e packing in English science w i ing. Fachsp ache 40(1-
2). 24–46. DOI: 10.24989/ s. 40i1-2.1601.
Gledhill, Ch is ophe & Ma ia Zimina-Poi o . 2023. Human-machine in e ac ion:
How o in eg a e plain language ules in he e ision cycles o neu al machine
ansla ion ou pu . De Eu opa: Eu opean and Global S udies Jou nal Special
2022(1). Issue). 149. h ps://u-pa is.hal.science/hal-04050523.
Hao, Yu & An hony Pym. 2023. Whe e do ansla ion s uden s go? A s udy o he
employmen and mobili y o mas e g adua es. The In e p e e and T ansla o
T aine 17(2). 211–229. DOI: 10.1080/1750399X.2022.2084595.
He, Zhiwei, Tian Liang, Wenxiang Jiao, Zhuosheng Zhang, Yujiu Yang, Rui Wang,
Zhaopeng Tu, Shuming Shi & Xing Wang. 2023. Explo ing human-like ansla-
ion s a egy wi h la ge language models. h ps://a xi .o g/abs/2305.04118.
Hu, Bin, Chenyang Zhao, Pu Zhang, Zihao Zhou, Yuanhang Yang, Zenglin Xu &
Bin Liu. 2024. Enabling in elligen in e ac ions be ween an agen and an LLM: A
ein o cemen lea ning app oach. h ps://a xi .o g/abs/2306.03604.
Kim, Bokyeong, Soojin Hwang, Sanghoon Cha, Chang Hyun Pa k, Jongse Pa k
& Jaehyuk Huh. 2022. Suppo ing dynamic ansla ion g anula i y o hyb id
memo y sys ems. In 2022 IEEE 40 h In e na ional Con e ence on Compu e De-
sign (ICCD), 25–32. DOI: 10.1109/ICCD56317.2022.00015.
Ma ínez, Gonzalo, Lau en Wa son, Ped o Re i iego, José Albe o He nández,
Ma c Jua ez & Rik Sa ka . 2023. Towa ds unde s anding he in e play o gene -
a i e a i icial in elligence and he in e ne . h ps://a xi .o g/abs/2306.06130.
Mi chell-Schui e oe de , Rosema y. 2020. A p ojec -based app oach o ansla ion
echnology. London, UK: Rou ledge.
Mosbach, Ma ius, Tiago Pimen el, Shauli Ra ogel, Die ich Klakow & Yanai
Elaza . 2023. Few-sho ine- uning s. in-con ex lea ning: A ai compa ison and
e alua ion. h ps://a xi .o g/abs/2305.16938.
Moslem, Yasmin, Rejwanul Haque, John D. Kellehe & Andy Way. 2023. Adap-
i e machine ansla ion wi h la ge language models. In Ma y Nu minen,
Judi h B enne , Maa i Koponen, Si kku La omaa, Mikhail Mikhailo , F ed-
e ike Schie l, Tha indu Ranasinghe, E a Vanmassenho e, Se gi Al a ez Vi-
dal, No a A anbe i, Ma a Nunzia ini, Ca la Pa a Esca ín, Mikel Fo cada,
Maja Popo ic, Ca olina Sca on & Helena Moniz (eds.), P oceedings o he 24 h
Annual Con e ence o he Eu opean Associa ion o Machine T ansla ion, 227–
237. Tampe e, Finland: Eu opean Associa ion o Machine T ansla ion. h ps:
//aclan hology.o g/2023.eam -1.22/.
163
Ma ia Zimina-Poi o
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.
Pleye , Michael & James Win e s. 2015. In eg a ing cogni i e linguis ics and lan-
guage e olu ion esea ch. Theo ia e His o ia Scien ia um 11. 19. DOI: 10.12775/
hs-2014-002.
Ramí ez-Polo, Lau a & Chelo Va gas-Sie a. 2023. T ansla ion echnology and
e hical compe ence: An analysis and p oposal o ansla o s’ aining. Lan-
guages 8(2). 93. DOI: 10.3390/languages8020093.
Raus, Rachele, Alida Ma ia Sille i, Sil ia Domenica Zollo & John Humbley (eds.).
2023. Mul ilinguisme e a ié és linguis ique en eu ope à l’aune de l’in elligence
a i icielle. Milan, I aly: Ledizioni. h ps://www.collane.uni o.i /oa/i ems/
show/132.
Raza, Shaina, Oluwani emi Bamgbose, Sha dul Ghuge, Fa emeh Ta akol, Deepak
John Reji & Syed Raza Bashi . 2025. De eloping sa e and esponsible la ge lan-
guage model: can we balance bias educ ion and language unde s anding in la ge
language models? h ps://a xi .o g/abs/2404.01399.
Schulz, Wol gang & Ch is ian Ollig. 2023. Teaching no ms o la ge language mod-
els: The nex on ie o hyb id go e nance. Alexande on Humbold Ins i u e
o In e ne & Socie y. DOI: 10.5281/ZENODO.7957165.
Shichkina, Yulia & Ki ill K inkin. 2022. P inciples o building pe sonalized in elli-
gen human assis an s. In 2022 Fou h In e na ional Con e ence Neu o echnolo-
gies and Neu oin e aces (CNN), 148–151. DOI: 10.1109/CNN56452.2022.9912559.
Sin-wai, Chan. 2023. Rou ledge encyclopedia o ansla ion echnology. 2nd edn.
London, UK: Rou ledge. DOI: 10.4324/9781003168348.
Sinne , Ca s en. 2020. Linguis ic no m in sociolinguis ics. In F anz Lebsan &
Felix Tacke (eds.), Manual o s anda diza ion in he Romance languages, 145–
164. Be lin, Ge many: De G uy e . DOI: 10.1515/9783110458084-005.
Slimi, Zouhaie & Bea iz Villa ejo Ca ballido. 2023. Na iga ing he e hical chal-
lenges o a i icial in elligence in highe educa ion: An analysis o se en global
AI e hics policies. TEM Jou nal. 590–602. DOI: 10.18421/TEM122-02.
Ta a es, Célia, Lau a Tallone, Luciana Oli ei a & Sand a Ribei o. 2023. The chal-
lenges o eaching and assessing echnical ansla ion in an e a o neu al ma-
chine ansla ion. Educa ion Sciences 13(6). 541. DOI: 10.3390/educsci13060541.
Uchida, Sa o u. 2024. Using ea ly LLMs o co pus linguis ics: Examining Cha -
GPT’s po en ial and limi a ions. Applied Co pus Linguis ics 4(1). 100089. DOI:
10.1016/j.aco p.2024.100089.
Van Dijk, Teun A. 2006. Discou se, con ex and cogni ion. Discou se S udies 8(1).
159–177. DOI: 10.1177/1461445606059565.
164
8 Compu e -assis ed LM in eaching human-cen ed AT
Yang, Jian, Hongcheng Guo, Yuwei Yin, Jiaqi Bai, Bing Wang, Jiaheng Liu, Xin-
nian Liang, Linzheng Cahi, Liqun Yang & Zhoujun Li. 2024. M3p: owa ds mul-
imodal mul ilingual ansla ion wi h mul imodal p omp .h ps://a xi .o g/
abs/2403.17556.
Zhu, Yilun, Joel Ruben An ony Moniz, Sh u i Bha ga a, Jia ui Lu, Dhi ya Pi-
a ipe umal, Si e Li, Yuan Zhang, Hong Yu & Bo-Hsiang Tseng. 2024. Can
la ge language models unde s and con ex ? h ps://a xi .o g/abs/2402.00858.
Zhuang, Yan, Qi Liu, Yu ing Ning, Weizhe Huang, Zacha y A. Pa dos, Pa ick C.
Kyllonen, Jiyun Zu, Qingyang Mao, Rui L , Zhenya Huang, Guanhao Zhao,
Zheng Zhang, Shijin Wang & Enhong Chen. 2024. F om s a ic benchma ks o
adap i e es ing: psychome ics in AI e alua ion. h ps://a xi .o g/abs/2306.
10512.
Zimina, Ma ia. 2015. In e ac ions dans l’espace ex uel mul ilingue. Me a 60(2).
379. DOI: 10.7202/1032930a .
Zimina, Ma ia. 2019. Ve s une mémoi e de aduc ion dynamique e mul idi-
mensionnelle: T aduc ion e echnologie, ega ds c oisés su de nou elles p a-
iques. Des mo s aux ac es (8). 221–236. DOI: 10.15122/ISBN.978-2-406-09779-2.
Zimina, Ma ia & Se ge Fleu y. 2016. Pe spec i es de l’a chi ec u e ame/cad e
pou les alignemen s mul ilingues. Nou elles pe spec i es en sciences sociales
11(1). 325–353. DOI: 10.7202/1035940a .
Zimina-Poi o , Ma ia & Ch is ophe Gledhill. 2023. The e olu i e na u e o ans-
la ion and eaching wo k low: Tex ome ic analysis o mul iple e ision cycles
in a simula ed websi e ansla ion p ojec . mediAzioni 39. A116–A146. DOI:
10.6092/issn.1974-4382/18790.
165
