Drafting Definitions for Emerging Concepts and Terms Undergoing Semantic Shift Within the ARTES Knowledge Base: A Protocol for Integrating LLMS Into Terminological Analysis by Experimental Approach
Abstract
This paper presents a protocol for evaluating and integrating generative AI (GenAI) tools in the framework of the terminological analysis of emerging, semantically unstable terms, absent from established term bases. Implemented within the ARTES knowledge base, the protocol supports Master’s students in translation at Université Paris Cité, in their task consisting of conducting a terminological analysis required for their dissertation. The study focuses particularly on terms displaying semantic instability and variation, thereby giving rise to semantic neologisms, and on evaluating the effectiveness of GenAI in retrieving existing definitions and drafting new terminological definitions for such terms. A survey of students’ GenAI use and an experimental study on the concept of data pollution illustrate the approach. Findings show corpus-linguistic tools help structure conceptual knowledge and critically assess GenAI outputs, confirming the need for human oversight. The study proposes a model for prompt construction and evaluation, a systematic process for building a collection of effective prompts, and a methodology that combines LLMs and corpus linguistics’ techniques for terminology management.
Full text
Te minologija / Te minology
2025, ol. 32, pp. 1–43
Con en s link / Tu inio nuo oda
ISSN 2669-2198 (online / elek oninis)
Copy igh © 2025 Mojca Pecman. Published by he Ins i u e o he Li huanian Language. This is an Open
Access a icle dis ibu ed unde he e ms o he C ea i e Commons A ibu ion Licence, which pe mi s
un es ic ed use, dis ibu ion, and ep oduc ion in any medium, p o ided he o iginal au ho and sou ce a e
c edi ed. // Išleido Lie u ių kalbos ins i u as. Šis s aipsnis y a a i os p ieigos, pla inamas pagal „C ea i e
Commons‚ p isky imo licencijos sąlygas, leidžiančias ne ibo ai naudo i, pla in i i a ku i u inį be kokioje
laikmenoje, nu odan au o ių i šal inį.
Recei ed: / Gau a: 2025-09-10. Accep ed: / P iim a: 2025-09-22. 1
DRAFTING DEFINITIONS FOR EMERGING CONCEPTS AND
TERMS UNDERGOING SEMANTIC SHIFT WITHIN THE ARTES
KNOWLEDGE BASE: A PROTOCOL FOR INTEGRATING LLMS
INTO TERMINOLOGICAL ANALYSIS BY EXPERIMENTAL
APPROACH
Naujų są okų i seman iniu poslinkiu pasižyminčių e minų apib ėžčių engimas
ARTES žinių bazės kon eks e: didžiųjų kalbos modelių in eg a imo į e minologinę
analizę ekspe imen inis p o okolas
MOJCA PECMAN
Uni e si é Pa is Ci é
E-mail: mojca.pecman@u-pa is.
ORCID ID: h ps://o cid.o g/0000-0002-6753-1936
Fields o esea ch: ARTES knowledge da abase, co pus-based e minology, seman ic neologisms,
d a ing e minological de ini ions, gene a i e AI, p omp enginee ing o e minology.
h ps://doi.o g/10.35321/ e m32-02
------------------------------------------------------------------------------------------------------------------
ABSTRACT
This pape p esen s a p o ocol o e alua ing and in eg a ing gene a i e AI (GenAI) ools in he
amewo k o he e minological analysis o eme ging, seman ically uns able e ms, absen om es ablished
e m bases. Implemen ed wi hin he ARTES knowledge base, he p o ocol suppo s Mas e ’s s uden s in ans-
la ion a Uni e si é Pa is Ci é, in hei ask consis ing o conduc ing a e minological analysis equi ed o hei
disse a ion. The s udy ocuses pa icula ly on e ms displaying seman ic ins abili y and a ia ion, he eby gi -
ing ise o seman ic neologisms, and on e alua ing he e ec i eness o GenAI in e ie ing exis ing de ini ions
and d a ing new e minological de ini ions o such e ms. A su ey o s uden s’ GenAI use and an expe i-
men al s udy on he concep o da a pollu ion illus a e he app oach. Findings show co pus-linguis ic ools
help s uc u e concep ual knowledge and c i ically assess GenAI ou pu s, con i ming he need o human o e -
sigh . The s udy p oposes a model o p omp cons uc ion and e alua ion, a sys ema ic p ocess o building a
collec ion o e ec i e p omp s, and a me hodology ha combines LLMs and co pus linguis ics’ echniques o
e minology managemen .
2
KEYWORDS: ARTES knowledge da abase, co pus-based e minology, seman ic neologisms, d a ing
e minological de ini ions, gene a i e AI, p omp enginee ing o e minology.
ANOTACIJA
Šiame s aipsnyje p is a omas gene a y inio DI (GDI) į ankių e inimo e minologinės analizės
kon eks e i jų in eg a imo į e minologinės analizės sis emą, aikomą naujiems, seman iškai nes abiliems, į
p ipažin as e minų bazes neį auk iems e minams, p o okolas. Jis, įgy endinamas ARTES žinių bazėje,
paleng ina da bą Pa yžiaus uni e si e o e imo magis an ū os s uden ams, a liekan iems e minologinę
analizę, eikalingą baigiamiesiems da bams. Ty ime daugiausia dėmesio ski iama e minams, ku ie pasižymi
seman iniu nes abilumu i a ian iškumu, odėl ai lemia seman inius neologizmus, bei GDI eiksmingumo
a enkan esamas šių e minų apib ėž is i engian naujas šių e minų apib ėž is į e inimui. Me odą ilius uoja
a lik a s uden ų apklausa dėl GDI naudojimo i ekspe imen inis są okos da a pollu ion (klaidingų duomenų
į ašymas) y imas. Rezul a ai odo, kad eks ynų ling is ikos į ankiai padeda s uk ū izuo i koncep ualias žinias,
k i iškai į e in i GDI iš es is i pa i ina žmogaus a liekamos p iežiū os po eikį. Ty ime siūlomas užklausų
kū imo i e inimo modelis, sis emingas p ocesas eiksmingų užklausų inkiniui ku i bei didžiųjų kalbos
modelių i eks ynų ling is ikos echnikas sude inan i e minologijos aldymo me odologija.
ESMINIAI ŽODŽIAI: ARTES žinių bazė, eks ynais g indžiama e minologija, seman iniai neologizmai,
e minologinių apib ėžčių engimas, gene a y inis DI, užklausų kū imas e minologijai.
INTRODUCTION
In his s udy, we p esen he p o ocol o GenAI-assis ed e minological analysis,
implemen ed wi hin he ARTES p ojec
1
. The p o ocol is speci ically designed o add ess
eme ging concep s ha a e no ye documen ed in exis ing e minological da abases bu a e
inc easingly used in specialised li e a u e. Eme ging concep s mus i s and o emos be
de ined. The ARTES p ojec ocuses on c ea ing he necessa y linguis ic esou ces and
knowledge o add ess such concep s, which o en gi e ise o e minological a ia ion –
whe he o mal o seman ic – and a e gene ally absen om es ablished e m bases (c .
L’Homme 2024b). In his con ex , e ms a e analysed using a co pus-based me hodology
and eco ded in he ARTES knowledge base
2
, which unc ions as bo h a e m base and a
pedagogical ool o eaching e minology o u u e ansla o s a Uni e si é Pa is Ci é.
Mo eo e , ARTES is ac i ely used by Mas e ’s s uden s as pa o hei Mas e 's disse a ion
in e minology. E e y yea s uden s compile compa able co po a o conduc in-dep h
e minological analysis and d a e m eco ds o suppo specialised ansla ion.
This s udy aims o e alua e he con ibu ion o la ge language models (LLMs) o
e minological managemen in his speci ic educa ional con ex and highligh he
impo ance o conduc ing a co pus-based analysis p io o in e ac ing wi h gene a i e
a i icial in elligence (GenAI) ools. The s udy also illus a es he con ibu ion o
e minology o knowledge enginee ing and i s connec ion, in his con ex , wi h a i icial
in elligence (Condamines 2022). Using a pilo co pus composed o ex s ep esen ing
a ious con ex s o specialised and semi-specialised communica ion in English, we analyse
he concep o da a pollu ion by means o Knowledge-Rich Con ex s (KRCs) (Meye 2001).
1
A ailable a : h ps://al ae.u-pa isci e. /a es-aide-a-la- edac ion-de- ex es-scien i iques.
2
A ailable a : h ps://a es.app.uni -pa is-dide o . /a es-sym ony/web/app.php.
3
The meaning o he e m da a pollu ion began o shi in 2018, making i a aluable
candida e o s udying p ocesses o seman ic change and he challenges in ol ed in de ining
seman ic neologisms. While he e m is inc easingly used in English (alongside i s mo e
es ablished coun e pa , digi al pollu ion, which is well-documen ed in e m bases), i has ye
o be eco ded in majo e minological da abases wi h i s newly eme ging meaning.
This s udy ou lines he phases o an expe imen al p o ocol designed o explo e and
assess he con ibu ion o GenAI ools o e minological wo k, wi h pa icula ocus on
d a ing de ini ions. The goal is o unde ake an in o med adap a ion o he e minology
cu iculum in he Mas e ’s p og am o he 2025-2026 academic yea . The in eg a ion o
GenAI ools in o e m managemen is also mo i a ed by s uden s’ e ol ing needs iden i ied
h ough a su ey whose esul s a e p esen ed in his s udy. I is also d i en by he
equi emen s o he Eu opean Mas e ’s in T ansla ion (EMT), joined by Mas e ILTS
3
in
2009. The la es e sion o he EMT Compe ence amewo k (2022), co e ing he pe iod
2023-2028, places pa icula emphasis on he in eg a ion o echnologies and he need o
u u e ansla o s o con inuously upda e hei skills in ansla ion ools and echnologies. In
ligh o ecen ad ances in AI and neu al machine ansla ion (NMT), his aspec has
become cen al o cu en aining needs.
The i s sec ion o he s udy is de o ed o p esen ing backg ound con ex wi h
pa icula emphases on he in e connec edness be ween e minology, co pus linguis ics and
AI. We also p esen he ARTES amewo k o e minological analysis and he key s ep in
his p ocess: d a ing e minological de ini ions. We also in oduce he expe imen al
app oach o in eg a ing LLMs o co pus-based e m analysis. The second sec ion p esen s
he me hodology o de ising a p o ocol o GenAI in eg a ion and he su ey conduc ed
among Mas e ’s s uden s. The las sec ion p o ides he analysis o he p o ocol designed o
collec ing and d a ing e minological de ini ions by in eg a ing GenAI h ough c i ical
app oach. We also illus a e he inal s age o he analysis leading o d a ing a e m eco d
o he e m da a pollu ion in he ARTES knowledge base, and he model designed wi hin
his p o ocol o collec ing he mos e icien p omp s o e minology.
TERMINOLOGY, ARTIFICIAL INTELLIGENCE AND THE ARTES
KNOWLEDGE BASE
In his sec ion we e iew he la es e olu ions in e minology as a science and
p ac ice d i en by he p og ess in AI. We hen p esen he ARTES pla o m o d a ing
e m eco ds in line wi h ISO s anda ds and FAIR p inciples, and he possibili ies i o e s
o in e ac ing wi h AI. We also p esen he cen al ole o de ini ion in e minological
analysis, pa icula ly in he case o eme ging concep s and e ms.
3
Mas e Indus ie de la langue e aduc ion spécialisée (ILTS), UFR EILA, UPCi é.
4
Te minology and A i icial In elligence
Acco ding o Leona di (2025: 148), he landma king wo k by Eugen Wüs e , he
ounde o e minology science, is a ‚pa o a long adi ion o in e en ions in na u al
languages aiming a imp o ing hei ep esen a i e and communica i e e iciency.‛
Mo eo e , Leona di (2025) explains ha ‚ his adi ion con inues in con empo a y
o malised models de eloped in Na u al Language P ocessing (NLP) ha a e a he basis o
A i icial In elligence applica ions‛. Te minology and AI a e he e o e in insically linked.
Indeed, he con e gence be ween Te minology and AI can be aced back o 1993 when
Didie Bou igaul and Anne Condamines se a esea ch g oup ‚Te minology and A i icial
In elligence (TIA)‛. A he ime, e minology- ela ed AI was oo ed in knowledge
enginee ing and explo a ions o he in e ac ions be ween co po a and e minology by
implemen a ion o symbolic app oach using pa e ns o sea ching candida e e ms and
concep ual ela ionships (Meye e al. 1992; Bou igaul e al. 2001; Condamines, Rebey olle
2001; Condamines 2005). Consequen ly, Te minology and AI sha e he same goal:
knowledge modelling.
Fu he mo e, he in e sec ion be ween e minology and AI h ough hei sha ed
goal – knowledge modelling – is oo ed in he eme gence o co pus linguis ics in he la e
1980s, which p o ided access o digi al co po a and ools o hei explo a ion. O e ime,
co pus linguis ics became a leading app oach in e minology s udies (Pecman, Küble 2022).
I has also con ibu ed o he de elopmen o he ex ual app oach o e minology
(Bou igaul , Slodzian 1999), which is o en seen as a eac ion o he adi ional p esc ip i e
me hodology o he Wüs e ian school (Humbley 2022). This connec ion be ween
Te minology, Co pus Linguis ics and AI is also isible in he use o he e ms such as
compu a ional e minology and co pus-based e minology. In he iew o Condamines (2022),
his his o ical alignmen be ween e minology and AI pu suing common goals, and using
common ools such as co po a o la ge collec ions o language da a, has p og essi ely
e oded, e lec ing he shi ing pa adigms and accele a ed de elopmen wi hin he ield o AI
which culmina ed in 2020s wi h he in oduc ion o La ge Language Models (LLMs).
Ne e heless, language da a, which se es as a ese oi o knowledge modelling, is ano he
sha ed elemen be ween Te minology, Co pus Linguis ics and LLMs.
In app oxima ely hi y yea s, he join en u e be ween compu a ional linguis s,
NLP and IT specialis s culmina ed in he de elopmen o he esea ch p o o ype, Cha GPT.
Launched in No embe 2022, Cha GPT quickly e ol ed in o a widely used applica ion,
illus a ing he apid public up ake o a i icial in elligence inno a ions. The ollowing yea s
we e cha ac e ised by he p oli e a ion o di e en ypes o LLMs and GenAI ools:
Cha GPT, DeepSeek, Pe plexi y, Gemini, Claude, e c.
5
The pa adigm shi d i en by A i icial In elligence
The pa adigm shi b ough by AI is ongoing, b oadening he scope and possibili ies
o bo h esea ch and eaching, hence o h o ien ed owa d he e alua ion and in eg a ion o
AI. In ela ion o e minology, specialised ansla ion and co pus linguis ics, he ocus is on
c i ical assessmen o he e ec i eness and pi alls o AI-assis ance in linguis ic analysis,
ansla ion and e m managemen . While mos o he wo ks unde line he need o c i ical
app oach, pa icula ly in he con ex o specialised languages: e.g. Raus, Ma ioda 2024; San
Ma ín 2024; Da ies 2025; Küble , Pecman 2025; he a icles highly en husias ic owa ds
he possibili ies o e ed by AI ools can be ound oo, like Sch y e ’s (2023) s udy on he
e ec i es o Cha GPT in gene al language lexicog aphy. A he same ime, he esea ch
aiming o imp o e he pe o mance o LLMs models is speci ically in e es ed in domain-
speci ic e minology, in pa allel esou ces in pa icula , which se e as a high-quali y
aining da a: e.g. Ballie e al. 2021; Béna d e al. 2023; Zhu e al. 2023.
Mo eo e , in his con ex o apid p opaga ion o AI, he inc easing numbe o
publica ions and con e ences in i e scien is s o join he e lec ion on he ole o AI in
science and eaching, in p o essional and social p ac ices, on e hical and GDPR (Gene al
Da a P o ec ion Regula ion) issues, as well as on he issues ela ed o limi ed aining da a
o languages o lesse di usion: e.g. Ras ie 2021; Casal, Kessle 2023; Fina di 2023;
Lommel 2024. The ins i u ions con ibu e o he deba e by o ganising impo an e en s.
In e p e ing Eu ope Con e ence 2025
4
in i ed p o essionals, indus y expe s, academics and
s uden s o he discussion on he u u e o he in e p e a ion, as a p o ession, in ela ion o
ad ances in AI. A he 2025 A i icial In elligence Ac ion Summi (‚Somme pou l’ac ion
su l’in elligence a i icielle‛)
5
held in Pa is, o e 60 coun ies signed o he i s ime a
decla a ion aimed a p omo ing us wo hy, sus ainable, and inclusi e AI. A he same ime,
media and p ess ampli y he deba e by b inging he opic in o he public sphe e, spu ing
esea che s o each wide audience and conside ing a b oade ange o gen es and sou ces:
e.g. Falgas, Robe 2023; Fina di 2023; Da ies 2025.
I is he e o e no ewo hy ha p og ess in AI g ea ly impac s scien i ic, academic and
educa ional se ings. In he con ex o ansla ion and language s udies, he g owing
in luence o AI is signi ican ly eshaping s uden beha iou , pa icula ly in how hey sea ch
o and cons uc de ini ions. As no ed by P asznik and Lew (2025), Cha GPT has spa ked
deba e among lexicog aphe s. In hei ecen s udy on he impac o AI among Polish
s uden s o English, hey ound ha Cha GPT is widely used and ega ded as a aluable ool
o w i ing and ansla ion. The au ho s epo ha s uden s also u n o i o inspi a ion,
cu iosi y, and en e ainmen . They also s ess ha he u u e ole o AI in he ield emains
unce ain. As he echnological landscape e ol es, so do academic p ac ices, c ea ing a need
o e ise he educa ional and me hodological amewo ks o e lec hese changes.
4
A ailable a : h ps://knowledge-cen e- ansla ion-in e p e a ion.ec.eu opa.eu/en/e en s/in e p e ing-
eu ope-con e ence-2025.
5
A ailable a : h ps://www.elysee. /somme -pou -l-ac ion-su -l-ia.
6
Consequen ly, much o he cu en esea ch in e minology, language esou ce de elopmen
and dic iona y-making is now ocused on examining he pa adigm shi in oduced by AI
ools, along wi h hei c i ical assessmen and po en ial in eg a ion in o eaching and
esea ch. No able ecen ini ia i es include a special issue o he jou nal Te minology,
‚Te minology and AI‛, scheduled o 2027 and a 2025 wo kshop in Pa is explo ing he ole
o dic iona ies in he age o AI.
6
. As Al ameemi (2024: 429) s esses, ‚<…> he in eg a ion
o CL [Co pus Linguis ics] wi h Cha GPT holds g ea po en ial o he unde s anding o
language in he digi al age. <…> In o he wo ds, ins ead o being cau ious in applying
Cha GPT, linguis s should examine he impo ance o me ging CL and Cha GPT. E en
linguis ic academic p og ammes should conside he impo ance o applying echnology in
he s udy plan o hei deg ees. Mo eo e , i is no only linguis s who mus ake his poin
in o accoun , bu schola s in o he ields who should conside hese aspec s and hink abou
he e ec i e u ilisa ion o echnology in s udying ields o human knowledge.‛
In his con ex o p essing need o c i ical e lec ion on he ole o AI and he
de elopmen o app op ia e p ac ices, his s udy aims o explo e he possibili ies o
in eg a ion o GenAI ools in o he p ocess o e minological analysis in he con ex o he
ARTES knowledge base amewo k. The necessi y o such an e alua ion is all he mo e
c ucial as ‚wi h he ine i able in eg a ion o AI in o e minology wo k, he dis inc ion
be ween human-c ea ed and AI-c ea ed con en will become inc easingly blu ed‛, as
poin ed by San Ma ín (2024).
The ARTES knowledge base and AI
The ARTES knowledge base has been used since 2010 o eaching e minology o
Mas e ’s s uden s in T ansla ion a he EILA
7
depa men (c . Pecman, Küble 2011; Küble ,
Pecman 2012; Gledhill, Küble 2015; Pecman 2021). I p o ides aluable esou ces o
suppo ing he eaching and esea ch in e minology and specialised ansla ion. De eloped
by ALTAE
8
esea ch eam, ARTES consis s o wo pla o ms, one o collec ing
e minological and ph aseological esou ces by ansla ion s uden s and he o he o
que ying he da abase eely online. ARTES is a e m base; howe e , i s enhanced,
knowledge-o ien ed app oach o e minology and specialised discou ses also quali ies i as a
knowledge base (Pecman 2018).
The basic ene o he ARTES amewo k is ha e icien e m managemen elies on
he acquisi ion o specialis knowledge. This can be achie ed by conduc ing onomasiological
and semasiological co pus-d i en and -based e m analysis, o which, in he ARTES
amewo k, is added knowledge o concep ual ne wo ks analysis. I should be emphasised
6
Le dic ionnai e ace à l’esso de la aduc ion au oma ique e des in elligences a i icielles géné a i es, ISIT, Pa is,
12 h June 2025. A ailable a : h ps://www.isi -pa is. /le-dic ionnai e- ace-a-lesso -de-la- aduc ion-
au oma ique.
7
A ailable a : h ps://u-pa is. /eila.
8
A ailable a : h ps://al ae.u-pa isci e. .
7
ha bo h, co pus-based e minology (Pecman, Küble 2022) and co pus-based ansla ion
s udies (CBTS) (Küble e al. 2024), a e cha ac e ised by a s ong g ounding in au hen ic
da a d awn om co po a, used o add ess e minological and ansla ion challenges.
Fu he mo e, in e minology, co po a a e pa icula ly use ul o iden i ying and analysis he
newly coined e ms and seman ic neologisms.
Fu he mo e, he ARTES DB is de ised in acco dance wi h ISO s anda ds (i.e. ISO
1087:2019, 704:2022, 12620-1:2022, 5078:2025) and he FAIR da a p inciples ( indabili y,
accessibili y, in e ope abili y and eusabili y) ini ia ed by Wilkinson e al. (2016) and
adap ed o e minology by Vezzani e al. (2023). ARTES p o ides aluable esou ces o
NMT aining (c . SPECTRANS (Ballie e al. 2021; Zhu e al. 2023) and MaTOS (Béna d
e al. 2023) p ojec s). The aligned o pa allel da a is namely a ailable o i e ypes o i ems:
e ms, colloca ions, de ini ions ( e minological and encyclopaedic), ansla ion no es and
subjec o domain.
In he cu en 2025-2026 academic yea , he ARTES amewo k is being ex ended
o explo ing LLMs in eg a ion in o e minological analysis. The p o ocol is in ended o ake
in o accoun he a ious phases o e m analysis, ha is, o iden i ying e ms and
e minological a ian s, inding colloca ions, equi alen e ms, de ini ional con ex s, KRCs
and o d a ing de ini ions and no es. To ensu e a c i ical and expe imen al app oach o
e alua ing GenAI ou pu , emphasis will emain on acqui ing knowledge abou e ms
h ough a co pus-based app oach and de eloping he necessa y skills o p oducing high-
quali y de ini ions. In e minological analysis, de ini ions play cen al ole in he p ocess o
knowledge acquisi ion. Consequen ly, i is impo an o explo e he e ec i eness o GenAI
ools o iden i ying ele an exis ing de ini ions and d a ing e minological de ini ions.
D a ing e minological de ini ions wi h LLMs and co pus-based app oach
Wi h he eme gence o e minology as an independen discipline ollowing Wüs e ’s
(1968) p oposal o d a ing specialised dic iona ies, he de ini ion assumed a cen al ole as
a key elemen in e m analysis. In he in oduc ion o his English-F ench dic iona y o
Machine Tools, Wüs e (1968: 2.15) iden i ies he de ini ion as he i s and o emos
elemen essen ial o achie ing e minological p ecision. The de ini ion is also he backbone
o he onomasiological app oach ( om concep o linguis ic uni ) in e minological analysis.
De ining specialised concep s cons i u es a i s s ep in acqui ing bo h specialised knowledge
and e minological expe ise. In he Mas e ’s p og ams o e ed a he EILA depa men ,
u u e specialised ansla o s, in e p e e s and esea che s in Languages o Speci ic Pu poses
(LSP) de elop his skill h ough cou ses on e minology.
Collec ing and d a ing e minological de ini ions o he ARTES da abase is one o
he equi emen s o eco d design. I consis s in d a ing an o iginal de ini ion wi h he
pu pose o ha monise he da a s o ed in he e m base and o p o ide de ini ions o
p e iously unde ined concep s o he concep s o which only de ini ional con ex s a e
a ailable. The model o d a ing de ini ions ollows ISO s anda ds 1087:2019 and 704:2022,
8
widely used by e minologis s. S uden s d a de ini ions by selec ing he ele an
in o ma ion p o ided in de ini ional con ex s and KRC as well as by analysis o seman ically
ela ed e ms and seman ic ne wo ks. They also collec exis ing de ini ions when hey a e
a ailable in epu able e m bases (UNTERM, IATE, Te mium, e c., c . Figu e 7).
D a ing de ini ions o newly eme ging concep s o neologisms, speci ically in
ela ion o he e ms exhibi ing seman ic ins abili y, is highly challenging. In he ARTES
p ojec , pa icula a en ion is gi en o si ua ions whe e he use o a e m s a s changing and
showing a shi in meaning, he eby leading o he o ma ion o a new concep . E alua ing
he con ibu ion o LLMs o hei analysis is expec ed o be a complex ask, as discussed by
San Ma ín (2024) who obse es: ‚I c a ing adi ional de ini ions is labo -in ensi e, he
conside a ion o con ex ual and unc ional cons ain s makes he ask e en mo e ime-
consuming. This is an impo an ba ie o he c ea ion o lexible e minological de ini ions.
Gene a i e A i icial In elligence (GenAI) ools, especially hose powe ed by La ge
Language Models (LLMs) such as Cha GPT, can emo e hese ba ie s by educing he ime
and e o equi ed o c ea e hem. Howe e , he impac o GenAI can ex end well beyond
his, as i can p o oundly ans o m he me hods and pu poses unde lying he c ea ion and
consump ion o e minological de ini ions.‛
Fo ou s udy, we selec ed he e m da a pollu ion which we analysed h ough a
co pus-based app oach p io o es s p esen ed in his pape on AI-assis ed e minological
analysis. The e m da a pollu ion is a aluable candida e e m o he pu poses o he p esen
s udy because i exhibi s seman ic shi and is absen om cu en ly a ailable e minological
esou ces. Mo eo e , he case s udy o his e m enables a comp ehensi e assessmen o he
ARTES amewo k o e m managemen , he ad an ages o co pus-based e minological
analysis and he e icacy o LLMs in suppo ing his p ocess.
As is well-known, e ms unde going seman ic shi sub ly al e he unde lying
knowledge pa adigm, making i pa icula ly di icul o dis inguish he newly eme ging
meaning om he o iginal one. Ne e heless, such e ms e lec b oade concep ual ends
and a e o conside able e minological signi icance (c . Pecman 2012; 2014). Typically
e e ed o as seman ic neologisms (as opposed o o mal neologisms), and some imes as
neosemes (Renou 2020), his ype o e m esul s om a ange o linguis ic phenomena,
such as polysemy and mic osenses, which make hem pa icula ly challenging o s udy
(L’Homme 2024a; 2024b). As Lomba d e al. (2023) poin ou : ‚no el wo d senses a e
called ‘seman ic neologisms’ and ‚ hey esul om seman ic ex ension by means o
polysemy‛. Mo eo e , L’Homme (2024a: 216) explains ha ‚polysemy is a p e alen
phenomenon wi h which e minologis s a e o en con on ed‛, making he managemen o
polysemous e ms in e minological esou ces essen ial ‚in o de o a oid ambigui y in
communica ion‛.
The ollowing sec ion p esen s he gene al me hodology employed o de ising he
expe imen al p o ocol on he in eg a ion o GenAI ools o e m eco d d a ing, wi h
emphases on e minological de ini ions as co e elemen s p o ided in e m eco ds.
9
METHODOLOGY FOR INTEGRATING GenAI TOOLS INTO THE ARTES
CORPUS-BASED FRAMEWORK
In eg a ing LLMs in o he p ocess o inding and d a ing e m eco ds and
de ini ions in pa icula is expec ed o sa e ime and enhance he quali y o de ini ions (c .
San Ma ín 2024). Ne e heless, in ou app oach, co pus-based analysis is conduc ed be o e
in oducing LLMs o be able o e ec i ely in eg a e and assess hem.
Da ies (2025) explains, in his c i ical e alua ion o ‚how well he p edic ions o
La ge Language Models (o LLMs, like Cha GPT and Gemini) ma ched up wi h ac ual
co pus da a‛, ha ‚a co pus is much mo e imme si e and i ’s a much mo e connec ed
expe ience han wha a e o en jus he ba ebones displays in LLMs.‛ <…> ‚Full- ea u ed
co po a p o ide an imme si e lea ning en i onmen wi h ex ensi e links be ween wo ds.‛
Mo eo e , Da ies sugges : ‚bo h LLMs and co po a ha e hei ad an ages and he bes is
p obably o use LLMs in conjunc ion wi h co pus da a, since in many cases hese wo
‘sou ces o da a’ complemen each o he qui e well‛. Thus, o enable e ec i e in e ac ion
wi h LLMs and suppo a c i ical e alua ion o hei ou pu , a ho ough unde s anding o he
linguis ic da a unde analysis is essen ial.
Consequen ly, in ou expe imen al app oach, he in e ac ion wi h LLMs elies on
co pus-based e minology, combined wi h p omp enginee ing o e minology. Acco ding
o Boons a (2025: 7), ‚p omp enginee ing is he p ocess o designing high-quali y p omp s
ha guide LLMs o p oduce accu a e ou pu s. This p ocess in ol es inke ing o ind he
bes p omp , op imizing p omp leng h and e alua ing a p omp ’s w i ing s yle and s uc u e
in ela ion o he ask. In he con ex o na u al language p ocessing and LLMs, a p omp is
an inpu p o ided o he model o gene a e a esponse o p edic ion.‛
Fo p omp enginee ing, we used he empla es p oposed by Boons a (2025)
o iginally de eloped o Gemini-p o, and by Schulho e al. (2025), which we adap ed o
ca e o a c i ical app oach o LLMs and o se e ou objec i e o compiling a collec ion o
e ec i e p omp s o e m managemen . We hus excluded he pa ame e s ha a e designed
o in luence model beha iou , such as ‚ empe a u e‛ which con ols he deg ee o
andomness in oken selec ion, because hey a e no suppo ed by GenAI ools es ed he e.
9
They can howe e be adjus ed wi h p omp ph asing: e.g. ‚Be c ea i e and unexpec ed.‛
encou ages high- empe a u e beha iou while ‚Be concise and accu a e.‛ encou ages low-
empe a u e beha iou . In e m managemen , o inding ele an au hen ic da a, low-
empe a u e beha iou may be expec ed o yield be e esul s. Fo ou expe imen al
pu poses, we added a numbe o ields o Boonas a’s empla e, namely: p omp ing
echnique
10
, da e, commen and ele ance. We also adop ed he possibili y o de ine he
limi di e sely, by a numbe o cha ac e s, okens o i ems.
9
They a e suppo ed by Applica ion P og amming In e ace (API), which allows o cus omising he applica-
ions using LLMs.
10
The di e en ypes o p omp s acco ding o Boonas a (2025) include: ze o-sho (a p omp wi h no examples
p o ided), one-sho (wi h one example p o ided), ew-sho (wi h se e al examples p o ided), sys em
16
Figu e 6. The mos equen GenAI ools used by Mas e 1 and Mas e 2 s uden s o d a ing e m eco ds
In his s udy, we hus ocus on es ing Cha GPT by OpenAI, as well as DeepSeek
de eloped by he Chinese company o he same name, and Pe plexi y de eloped by
Pe plexi y AI which uses di e en LLMs, namely GPT de eloped by OpenAI and Claude by
An h opic. In he nex sec ions, we p esen a sys ema ic sequen ial p o ocol o hei
in eg a ion in o he ARTES e m managemen wo k low.
Using co pus-based e minology o es he e iciency o GenAI ools
As emphasised in he p e ious sec ion, in he expe imen al p o ocol we p opose, he
e minological analysis by co pus-based app oach is conduc ed p io o in e ac ing wi h
LLMs o ensu e he abili y o c i ically assess GenAI ou pu . We hus conduc ed he analysis
h ough all he s ages o he ARTES amewo k, ha is he s ages 1 o 7, lis ed in he
p e ious sec ion, namely using Ske chEngine
12
(Kilga i e al. 2014) o specialised co pus
design and explo a ion, alongside co po a in eg a ed in he Ske chEngine and wo mo e
ools wi h in eg a ed co po a, Google Books Ng am Viewe
13
(Michel e al. 2011) and
Ne speak
14
(Riehmann e al. 2012).
Figu e 7 illus a es he esul s o he phase 1, consis ing in e i ying i he e m is
eco ded in exis ing e m bases. The e m da a pollu ion is a ely eco ded in well-known
e m bases, as only one en y was ound in IATE, c ea ed in 2021, o he o iginal meaning
o he e m.
SEARCHED TERM: da a pollu ion
SEARCHED DATE: 25/05/2025
TERM BASE
RECORD
COMMENT
RELEVANCE
Te mium Plus
none
NA
NA
Vi ine linguis ique
none
NA
NA
12
A ailable a : h ps://www.ske chengine.eu.
13
A ailable a : h ps://books.google.com/ng ams.
14
A ailable a : h ps://ne speak.o g.
17
IATE
1 eco d
Reco d c ea ion da e: 2.3.2021,
Meaning: he 1s meaning o he e m.
P o ided ields and in o ma ion: de ini ion, con ex s,
sou ces ( om 2018 and 2019), equi alen s in 20 lan-
guages (amongs which F ench: pollu ion des données)
medium
WIPO
none
NA
NA
UNTERM
none
NA
NA
IGI Global Dic-
iona y
none
NA
NA
Figu e 7. Resul s o he sea ch o he e m da a pollu ion in exis ing e m bases
As he con en o he eco d is de o ed o he o iginal meaning o he e m, his
inding is conside ed o medium ele ance o he a ge ed eco d design. Figu e 8 shows
he eco d con en in IATE.
Figu e 8. Reco d o he e m da a pollu ion in IATE displaying in o ma ion on he o iginal meaning o he
e m
Ne e heless, he da a p o ided can be used o c ea e a e m eco d o da a pollu ion
wi h i s o iginal meaning (which could be glossed as ‘ he ac o pollu ing he da a’) in o de
o con as i wi h he eco d dedica ed o da a pollu ion used wi h he new meaning (which
could be glossed as ‘ he pollu ion caused by he da a’). The complexi y o de ining da a
pollu ion wi h i s newly eme ged meaning (some imes e o mula ed by pollu ion by da a), is
pa icula ly complex because no only i o e laps wi h he meaning o he e m digi al
pollu ion bu also because o he simul aneous exis ence o i s o iginal meaning (some imes
e o mula ed by pollu ion o da a, pollu ed da a). Ga he ing in o ma ion om exis ing
esou ces and eplica ing hem in e m eco ds d a ed in ARTES lends cohe ence o
analysis, by aking in o accoun he wide scope o in o ma ion on a e m (Figu e 9).
18
DEFINITION
injec ion o maliciously c a ed aining da a samples in o a aining se , causing he AI
sys em o lea n an inco ec model and subsequen ly misclassi y es ing samples
SOURCE
IATE, Council-EN, based on Yinzhi Cao e al. 2018: 1
ISONYME
da a poisoning
COMMENT
he i s meaning o he e m
RELEVANCE
medium
Figu e 9. In o ma ion selec ed om IATE o he ARTES DB
The absence o he e m in exis ing e m bases illus a es he use ulness o co pus-
based app oach. Figu e 10 shows he de ini ions e ie ed om co po a by using discou se
ma ke s o de ini ional con ex s (such as ‚is a‛, ‚is he‛, ‚ e e o‛, e c.):
1
DEF. KRC
Da a pollu ion is he se o ha ms gene a ed by economic ac i i ies ela ed o da a
collec ion, s o age and use, which ans e sally impac s indi iduals and hei li ing
en i onmen s.
SOURCE
Mo i e al. 2024: 155
COMMENT
he new meaning o he e m, ound in mo e ecen li e a u e
RELEVANCE
high
2
DEF. KRC
Da a pollu ion is he in e ela ed ad e se impac ha he gene a ion, s o ing,
handling, and p ocessing o digi al da a has on ou na u al en i onmen , social
en i onmen , and pe sonal en i onmen . I is he unsus ainable handling,
dis ibu ion, and gene a ion o da a esou ces.
SOURCE
Hasselbalch e al. 2022: 9–10
COMMENT
he new meaning o he e m, ound in ecen li e a u e
RELEVANCE
high
3
DEF. KRC
In e ne con en s (documen s, emails, cha s, images, ideos, e c.) ha a e pos ed on
he In e ne a e o en dissemina ed and eplica ed on di e en pee s o se e s,
gene a ing wha we e e o as Da a Pollu ion.
SOURCE
Cas elluccia & Kaa a 2009: 1
COMMENT
he i s meaning o he e m, ound in less ecen documen s
RELEVANCE
medium
4
DEF. KRC
In wha ollows he e m ‚da a pollu ion‛ is aken o e e o he accumula ion o all
‚con amina ions‛ o ‚dis o ions‛ which can esul om wo king wi h da a in he
in o ma ion echnology ield.
19
SOURCE
Zimme li 1986: 291
COMMENT
he i s meaning o he e m, ound in in less ecen documen s
RELEVANCE
medium
Figu e 10. De ini ional con ex s e ie ed om co pus
Al hough se e al de ini ional con ex s we e ound, only he i s wo e e o he
ecen meaning occu ed by seman ic shi . These de ini ional con ex s p o ided in expe -
o-expe communica ion a e o high ele ance o a ge ed e m eco d design. The i s
one is pa icula ly o ele ance as i comes om a pape published in p oceedings o an
in e na ional con e ence.
The second de ini ion ele an o d a ing he a ge ed e m eco d comes om a
whi e pape (Hasselbalch 2022), published subsequen ly o a monog aph (Hasselbalch 2021),
whe e he issue o da a pollu ion is add essed wi hin he amewo k o an independen
ini ia i e
15
.
The de ini ional con ex 3 and 4 a e conside ed o medium ele ance as hey poin
o he o iginal meaning o he e m.
In he nex s ep, he acquisi ion o knowledge on he concep o da a pollu ion
h ough co pus consis ed in analysing conco dances o he e m, and mo e speci ically in
iden i ying KRCs. Co pus analysis allowed us o e ie e mul iple con ex s om which we
selec ed he mos ele an ones o unde s anding he e m, and i s seman ic ela ion o
o he e ms cha ac e is ic o his discou se. These con ex s a e also use ul o p o iding he
examples on use o he e m. Figu e 11 shows a sample o collec ed KRCs (wi h he selec ed
pa s in bold o in e ac ing wi h LLMs, p esen ed in he las Sec ion):
1
KRC
Recognizing ha da a pollu ion is also a public p oblem deg ading an en i e
ecosys em and no me ely he indi idual sphe es o he da a gi e s, o e s a new
and ich pe spec i e on he exis ing solu ions–and in oduces new ones.
SOURCE
Ben-Shaha 2018: 133
COMMENT
he new meaning o he e m, ound in ecen li e a u e
RELEVANCE
high
2
KRC
Digi al in o ma ion is he uel o he new economy. Bu like he old economy’s ca bon
uel, i also pollu es. Ha m ul ‚da a emissions‛ a e leaked in o he digi al ecosys em,
dis up ing social ins i u ions and public in e es s. This a icle de elops a no el
amewo k–da a pollu ion– o e hink he ha ms he da a economy c ea es and he way
hey ha e o be egula ed.
SOURCE
Ben-Shaha 2018: 104
15
The Da a Pollu ion & Powe – Ini ia i e. A ailable a : h ps://www.da apollu ion.eu.
20
COMMENT
he new meaning o he e m, ound in ecen li e a u e
RELEVANCE
high
3
KRC
Two adi ional usages o he e m da a pollu ion can hus be combined. Fi s ly, da a
pollu ion can be unde s ood as he ad e se impac on pe sonal and social en i onmen s,
o ins ance on indi idual igh s, such as da a p o ec ion o he igh o p i a e li e, and
on democ a ic ins i u ions and balances o powe . Secondly, da a pollu ion can be un-
de s ood as he ma e ial ad e se e ec s on ou na u al en i onmen , e.g., he ca bon
oo p in o big da a.
SOURCE
Hasselbalch 2022: 22
COMMENT
he new meaning and he o iginal meaning o he e m, pu in con as
RELEVANCE
high
4
KRC
In he whi e pape , da a pollu ion is add essed simila ly as no only one ype o
en i onmen al impac , bu a he as he in e ela ed ad e se e ec s on delica e
balances in ou na u al, social and pe sonal ecosys ems and en i onmen s. As
desc ibed, he e m da a pollu ion is cu en ly used o emphasise he e y eal
and ma e ial ad e se en i onmen al impac o big da a on hese en i onmen s. As
ollows, he goal o a new ‘g een mo emen ’ o big da a is ‘da a sus ainabili y’, which
cu s ac oss he SDGs wi h sus ainabili y conside a ions connec ed o he a ious en i-
onmen al changes caused by he olume and di e si y o big da a, anging om i s e -
ec s on he na u al landscape o ou decisions and democ acy.
SOURCE
Hasselbalch 2022: 22–25
COMMENT
he example shows he en anglemen o he new meaning and he o iginal meaning o
he e m
RELEVANCE
high
Figu e 11. Re ie al o Knowledge Rich Con ex s (KRCs) in a specialised co pus
The KRCs 1 o 4 illus a e he e m used wi h a new meaning.
16
In he ollowing
s ep, on he bases o he collec ed de ini ional con ex s and KRCs (Figu es 10-11), we can
make a p oposal o a e minological de ini ion, p esen ed in Figu e 12:
DRAFTED
DEFINITION
speci ic ype o pollu ion ela ed o in o ma ion echnologies and digi al e a cha ac e ised
by he inc easing gene a ion o da a which is ans o ming p o essional, social and e e y-
day li e o indi iduals and hei en i onmen in he way ha appea s no o be in acco d-
ance wi h hei capabili ies o needs o e ie ing ele an in o ma ion (295 cha ac e s)
SOURCE
e m eco d au ho name, based on Mo i e al. 2024, Hasselbalch e al. 2022 and Ben-
16
We also collec ed he ele an KRCs illus a ing he i s meaning o he e m o d a ing he eco d o he
concu en e m in o de o link he wo eco ds and p o ide a no e on he seman ic link be ween he wo
e ms.
21
Shaha 2018
RELEVANCE
high
Figu e 12. D a ed e minological de ini ion based on co pus e ie al and KRCs
A his s age o ou expe imen al analysis, we obse ed a cogni i e bias: he
es ic ion o in e ac wi h LLMs be o e inalizing co pus-based analysis became an obs acle
as i u ned ou o equi e subs an ial sel -con ol. In he nex s ep, we hus in oduced
LLMs o inding KRCs and ele an sou ces, KRCs-in o med de ini ion d a ing, e ising
o (human) e m de ini ions, and o d a ing de ini ions h ough a ‚loop‛ in e ac ion wi h
LLMs.
In oducing GenAI ools o he p ocess o e m analysis
Finding Knowledge-Rich Con ex s (KRC) and ele an sou ces
We i s es ed LLMs capaci y o ind de ini ional con ex s. The a ge was o w i e a
p omp which would e u n he mos ele an de ini ional con ex s in quo es ollowed by a
e e ence. A e se e al a emp s, he p omp ing echnique illus a ed in Figu e 13 yielded
use ul esul s. Howe e , i did no e u n speci ically de ini ional con ex s, which in he case
unde scope is no su p ising. Co pus-based analysis showed ha he de ini ional con ex s
o da a pollu ion a e a e. The p omp e u ned use ul KRCs and ele an e e ences o
co pus design.
By using a combina ion o sys em p omp ing and s ep-back p omp ing, we ob ain
no only he esul s ele an o he new meaning o he e m, bu also in he desi ed
o ma . By es ic ing he sea ch o speci ic ype o li e a u e (scien i ic) and limi ing
numbe o de ini ions o look o ( h ee), we cons ained LLM o p oduce he mos ele an
ou pu (see P omp 2 in Figu e 13). Mo eo e , we i s asked one ques ion o allow he LLM
o ac i a e he gene al knowledge (P omp 1 in Figu e 13), be o e asking i o ind he mos
ele an examples. Figu e 13 p esen s he es conduc ed on Cha GPT:
GOAL
Finding KRCs and ele an e e ences
MODEL
Cha GPT-4- u bo
DATE
13/05/2025
TECH-
NIQUE
S ep-back p omp ing, sys em p omp ing
LIMIT
3 de ini ions
PROMPT
1
Wha is he di e ence be ween ‚da a pollu ion‛ de ined in hese e ms ‚speci ic ype o
pollu ion ela ed o in o ma ion echnologies and digi al e a cha ac e ised by he inc easing
gene a ion o da a which is ans o ming p o essional, social and e e yday li e o indi iduals
22
and hei en i onmen in he way ha appea s no o be in acco dance wi h hei capabili ies
o needs o e ie ing ele an in o ma ion‛ and ‚digi al pollu ion‛?
PROMPT
2
Find h ee mos ele an de ini ions o he e m ‚da a pollu ion‛ in scien i ic li e a u e and
p o ide he sou ces o each de ini ion ound. Re u n he de ini ions in he lis wi h he exac
wo ding ound in he ex s ollowed by he bibliog aphical e e ence o he sou ce and i s
URL.
OUTPUT
1
The e ms “da a pollu ion” and “digi al pollu ion” a e ela ed bu dis inc in ocus and
scope:
Da a Pollu ion
Focus: The ha m ul e ec s o excessi e o mismanaged da a–i s collec ion, s o age,
p ocessing, and use.
Key Conce ns:
O e load o i ele an , biased, o ha m ul da a
En i onmen al and social cos s o big da a in as uc u es
Impac s on human a en ion, p i acy, and decision-making
Dis up ion o social and in o ma ional ecosys ems
Analogy: Compa able o chemical pollu ion–con amina ion o he in o ma ion en i onmen
wi h ha m ul byp oduc s.
Digi al Pollu ion
Focus: The b oade en i onmen al and esou ce impac o digi al echnologies.
Key Conce ns:
Ene gy consump ion and ca bon oo p in o de ices, se e s, and ne wo ks
E-was e om ha dwa e and de ices
Li ecycle impac o p oducing and disposing o digi al ools
Analogy: Compa able o indus ial pollu ion–physical and ecological damage caused by
digi al in as uc u e.
Summa y o he Di e ence
Da a pollu ion is abou he con en and consequences o da a i sel and i s socio-
en i onmen al e ec s.
Digi al pollu ion is abou he en i onmen al cos o he digi al sys ems ha s o e, ansmi ,
and p ocess da a.
Would you like a side-by-side able compa ison o cla i y?
OUTPUT
2
‚Digi al in o ma ion is he uel o he new economy. Bu like he old economy’s ca bon uel,
i also pollu es. Ha m ul ‘da a emissions’ a e leaked in o he digi al ecosys em, dis up ing
social ins i u ions and public in e es s.‛
Ben-Shaha O. 2018: Da a Pollu ion 679, Uni e si y o Chicago Public Law & Legal Theo y
Pape Se ies. A ailable a :
h ps://chicagounbound.uchicago.edu/public_law_and_legal_ heo y/699/
‚Da a pollu ion is he in e ela ed ad e se impac ha he gene a ion, s o ing, handling and
p ocessing o digi al da a has on ou na u al en i onmen , social en i onmen and pe sonal
en i onmen . I is he unsus ainable handling, dis ibu ion and gene a ion o da a esou ces.‛
Hasselbalch G. e al. 2022: Da a Pollu ion & Powe – Whi e Pape o a Global Sus ainable
De elopmen Agenda on AI. A ailable a :
h ps://www. esea chga e.ne /publica ion/362174889_Da a_Pollu ion_Powe _-
_Whi e_Pape _ o _a_Global_Sus ainable_De elopmen _Agenda_on_AI_by_G y_Hasselbalch
_wi h_con ibu ions_ om_ he_Da a_Pollu ion_Powe _DPP_G oup_a _Bonn_Sus ainable_AI
_Lab
23
‚The concep o da a pollu ion includes in o ma ion pollu ion o ‘in ollu ion’ (Özdemi
2016), as coined by O am 1984. Howe e , in o ma ion pollu ion is di e en o some ex en ;
‘ he con amina ion o in o ma ion sou ces wi h i ele an , edundan , unsolici ed and low-
alue in o ma ion’ o as Cai and Zhang (1996) quo e: ‘… is a pile o widesp ead ye unwan ed
messages and ha one day, hese messages could deeply in luence he social li e bea ing
nega i e esul s.‛
Da a Pollu ion and Taxa ion 2022. A ailable a :
h ps://www.academia.edu/76896486/Da a_Pollu ion_and_Taxa ion
COM-
MENT
1s no a de ini ional con ex , a KRC, a quo e, sou ce p o ided and well-cons uc ed
2nd a de ini ional con ex , a quo e, sou ce p o ided and well-cons uc ed
3 d no a de ini ional con ex , a KRC, a quo e, misleading (consis s o wo de ini ions o
ano he e m), sou ce p o ided bu poo ly cons uc ed, no a scien i ic egis e (academic
egis e )
RELE-
VANCE
high
Figu e 13. P omp enginee ing o e m managemen : inding KRC and ele an e e ences wi h Cha GPT
Al hough only one o ou de ini ional con ex s ound by co pus-based app oach was
e u ned, alongside wo KRCs, he ou pu is ele an o he analysis o he concep o da a
pollu ion. Fi s , he ou pu shows ha co pus-based app oach canno be bypassed o
selec ing he bes de ini ional con ex s and KRCs, as well as o e ec i ely e alua ing he
LLM’s ou pu . Second, Cha GPT e u ned one sou ce ha was no iden i ied du ing he
phase o co pus design. Al hough i belongs o academic a he han scien i ic egis e
17
, i
can p o e use ul in he subsequen analysis on seman ically ela ed e ms, and be used o
augmen ing he co pus: he discu si e ma ke ‚includes‛ in e s ha in o ma ion pollu ion is a
me onym o da a pollu ion, and he ma ke s ‚o ‛ and simple quo es ha in ollu ion is a
synonym o in o ma ion pollu ion. Ne e heless, his hi d p oposal made by Cha GPT as a
de ini ional con ex is misleading because i con ains wo de ini ions o ano he concep
(in o ma ion pollu ion) and no de ini ion o he concep sea ched o .
Mo eo e , in a p e ious p omp , we used he same ins uc ion wi hou speci ying
ha he expec ed ou pu a e he quo es, which esul ed in e o mula ed quo es and
misleading da a (see Figu e 24 in Appendix).
KRC-in o med de ini ion d a ing
The nex es consis ed in asking LLMs o d a a de ini ion suppo ed by a selec ion
o KRCs. We selec ed 3 KRCs: he de ini ional con ex 1 (in Figu e 10) and he ex ac s
om KRC 1 and 4 (appea ing in bold in Figu e 11). Figu e 14 shows his es wi h
Cha GPT, DeepSeek and Pe plexi y.
17
In he ARTES amewo k, Mas e ’s disse a ions a e consul ed when specialis sou ces a e lacking. In gene al,
hey a e no used o cons uc ing specialised co po a. They a e conside ed as belonging o academic a he
han scien i ic egis e .
24
115b
bG115a
KRC-in o med de ini ion d a ing
DATE
13/05/2025
TECH-
NIQUE
Ze o sho
LIMIT
300 cha ac e s
PROMPT
D a he e minological de ini ion o he e m ‚da a pollu ion‛ in no mo e han 300
cha ac e s on he bases o hese knowledge ich con ex s:
Da a pollu ion is he se o ha ms gene a ed by economic ac i i ies ela ed o da a collec ion,
s o age and use, which ans e sally impac s indi iduals and hei li ing en i onmen s.
Da a pollu ion is also a public p oblem deg ading an en i e ecosys em and no me ely he
indi idual sphe es o he da a gi e s.
Da a pollu ion is add essed as no only one ype o en i onmen al impac , bu a he he
in e ela ed ad e se e ec s on delica e balances in ou na u al, social and pe sonal
ecosys ems and en i onmen s. The e m da a pollu ion is cu en ly used o emphasise he
e y eal and ma e ial ad e se en i onmen al impac o big da a.
MODEL
Cha GPT-4- u bo
DeepSeek-V3
Pe plexi y
OUTPUT
Da a pollu ion is he
ha m ul byp oduc o da a-
ela ed economic ac i i ies–
such as collec ion, s o age,
and use– ha dis up s
na u al, social, and pe sonal
ecosys ems, a ec ing bo h
indi iduals and he b oade
en i onmen in ways
compa able o adi ional
en i onmen al pollu ion.
“Da a pollu ion e e s o
he sys emic ha ms
caused by da a- ela ed
economic ac i i ies,
deg ading na u al, social,
and pe sonal ecosys ems
h ough ad e se
en i onmen al, collec i e,
and indi idual impac s
beyond me e p i acy
conce ns.”
(298 cha ac e s)
Key ea u es cap u ed:
1. Sys emic ha m – Em-
phasizes la ge-scale deg-
ada ion, no jus indi-
idual e ec s.
2. Economic oo – Links
o da a collec ion, s o -
age, and use.
3. T ans e sal impac –
A ec s na u al, social,
and pe sonal en i on-
men s.
4. Beyond p i acy – Posi-
Da a pollu ion e e s o he
ha ms caused by economic
ac i i ies in ol ing da a
collec ion, s o age, and use,
which deg ade na u al, so-
cial, and pe sonal ecosys-
ems. I is a public p oblem
wi h in e connec ed ad e se
e ec s ex ending beyond
indi iduals o en i e en i-
onmen s.
25
ions i as an ecologi-
cal/public issue.
This aligns wi h he aming
o da a pollu ion as an en i-
onmen al and socie al chal-
lenge. Le me know i you'd
like any adjus men s!
COM-
MENT
adequa ely o mula ed,
ela i ely clea , ai ly concise
adequa ely o mula ed,
ela i ely clea , ai ly
concise, wi h addi ional
in o ma ion
inadequa ely o mula ed
( wo u e ances), complex
(by he choice o hype nym
which is less clea ), wi h
high alignmen o KRC
examples, ai ly concise
RELE-
VANCE
high
high
low
Figu e 14. P omp enginee ing o e m managemen : KRC-in o med de ini ion d a ing wi h Cha GPT,
DeepSeek and Pe plexi y
The h ee ou pu s appea ele an o e m managemen , and his me hod can be
ega ded as sui able o d a ing e minological de ini ions wi h LLMs’ assis ance. Indeed,
p o iding KRCs guides he LLMs o p oduce a de ini ion o he newly eme ging meaning
o he e m. The ou pu s by Cha GPT and DeepSeek appea as mo e app op ia e, in
compa ison o Pe plexi y. In appendix, in Figu e 25, we p o ide an example o he isks o
using insu icien ly cons ained p omp s when managing e ms unde going seman ic shi
wi h LLMs.
Re ising and imp o ing bio-de ini ion by LLMs
In he nex es , we asked GenAI ools o imp o e he de ini ion we d a ed wi hou
he help o AI- ools, on he bases o collec ed in o ma ion, namely a ious KRC, and which
we p opose o call ‚bio-de ini ion‛ (Figu e 15):
GOAL
Imp o ing bio-de ini ion
DATE
13/05/2025
TECH-
NIQUE
Ze o sho
LIMIT
300 cha ac e s
PROMPT
Imp o e his de ini ion o he e m ‚da a pollu ion‛: speci ic ype o pollu ion ela ed o
in o ma ion echnologies and digi al e a cha ac e ised by he inc easing gene a ion o da a
which is ans o ming p o essional, social and e e yday li e o indi iduals and hei
en i onmen in he way ha appea s no o be in acco dance wi h hei capabili ies o needs
o e ie ing ele an in o ma ion
32
To highligh he e m’s dual meanings, ARTES p o ides wo sepa a e eco ds: one
o he o iginal meaning and ano he o he newly eme ging one (Figu e 21-22), and an
ex ensi e no e on he concu en e ms (Figu e 22).
Figu e 21. The ARTES DB in e ace showing he wo eco ds o da a pollu ion o dis inguish he o iginal
om he new meaning
Figu e 22. The ARTES DB in e ace showing he en y o da a pollu ion 2, he newly eme ged concep , along
wi h he in o ma ion on da a pollu ion 1 de ined as a concu en e m (and concep ) including
an explana o y no e
Finally, Figu e 23 shows he ollowing elemen s o he scheme de ised o c ea ing a
collec ion o e icien p omp s o LLMs-assis ed e m managemen wi hin ARTES
amewo k by expe imen al app oach: he homepage o he p omp collec ion in e ace, a
ex ield o eco ding he designed p omp , a ex ield o p o iding an example o he
ou pu p oduced, he equi ed ou pu size (in cha ac e s o wo ds), an assessmen o he
p omp ’s quali y (based on he ob ained ou pu ), and he ype o p omp echnique used.
33
Figu e 23. A scheme de ised o c ea ing a collec ion o e icien p omp s o LLMs-assis ed e m
managemen
CONCLUDING REMARKS
This s udy has explo ed he po en ial o in eg a ing GenAI ools in o e minological
analysis, speci ically o d a ing e m eco ds wi hin he ARTES knowledge base and
pedagogical amewo k used o eaching e minology o Mas e 's s uden s in ansla ion a
Uni e si é Pa is Ci é. We selec ed he e m da a pollu ion which exhibi s seman ic shi , as a
case s udy, o conduc an explo a o y analysis and cons uc a p o ocol o an e icien
in e ac ion wi h LLMs and GenAI ools. This p o ocol is now being in eg a ed in o he
cu iculum o e minology aining wi hin he Mas e ’s p og am. We also conduc ed an
inqui y among 50 Mas e ’s s uden s o ansla ion, which showed ha hal o he s uden s
a e using LLMs o e m eco d d a ing, which u he enhances he need o guide hem in
hei e icien use.
The expe imen al indings o ou s udy highligh he alue o co pus-linguis ic ools
in assis ing e minological analysis, enabling he cons uc ion and o ganisa ion o concep ual
knowledge, as well as assessing he quali y o he ou pu o GenAI ools. The p o ocol
designed o AI in eg a ion o e minological analysis showed ha GenAI ools can be use ul
o e ising and imp o ing bio-d a ed e minological de ini ions, bu ha he human
in e en ion is pa amoun . These indings align wi h ecen wo ks on in eg a ing AI in
eaching and esea ch (c . Küble e al. 2024; Raus, Ma ioda 2024; San Ma ín 2024;
Küble , Pecman 2025) and ein o ce he key ecommenda ions eme ging om hem,
34
namely he need o os e a c i ical app oach o AI echnologies and o de elop e icien
me hods o e alua ing machine-gene a ed ou pu .
The heo e ical indings o ou s udy highligh he c ucial ole o explo a o y
analyses in he cu en landscape o apidly e ol ing AI ools, pa icula ly in designing
e ec i e schemes o in eg a ing GenAI and LLMs. Cen al o his p ocess is he need o
p elimina y human analysis. Mas e y o domain-speci ic knowledge and linguis ic da a is
essen ial o assessing he quali y o GenAI ou pu and o c a ing e ec i e p omp s o
add ess inaccu a e esponses. In he con ex o e minology, his equi es expe ise in
specialised concep s and he applica ion o co pus-based app oaches o e m analysis. Co pus
linguis ics eme ges as a key me hod o in o med in e ac ion wi h GenAI ools and p omp
enginee ing.
A no ewo hy challenge encoun e ed du ing ou s udy was he cogni i e bias
in oduced by delibe a ely limi ing LLM in e ac ions, which demanded conside able sel -
discipline. In esponse, he p o ocol p oposed in his pape ad oca es a sequen ial
in eg a ion o GenAI ools ac oss a ious s ages o in o ma ion e ie al and e minological
analysis.
Fu u e esea ch will ocus on es ing and e ining his p o ocol wi h ansla ion
s uden s in e minology cou ses. We aim o expand he me hodology o co e all s ages o
e m eco d d a ing, wi h he objec i e o enhancing he e ie al and gene a ion o
e minologically ele an in o ma ion in in e ac ion wi h GenAI ools. One o he p o ocol’s
co e goals is o de elop a cu a ed collec ion o highly e ec i e p omp s o e m eco d
c ea ion. Ou s udy has laid he g oundwo k o his, by o e ing a model o p omp
cons uc ion and e alua ion, ailo ed o encompass di e en language models, ask ypes,
and linguis ic da ase s, and ha we will con inue o es , adap and e ine in u he s udies.
REFERENCES
Al ameemi Yase M. 2024: S a e-o - he-a Re iew o he Co pus Linguis ics Field om he
Beginning Un il he De elopmen o Cha GPT. – Theo y and P ac ice in Language
S udies 14(2), 423–431. A ailable a : h ps://doi.o g/10.17507/ pls.1402.13.
ARTES. A ailable a : h ps://a es.app.uni -pa is-dide o . /a es-sym ony/web/app.php.
Ballie Nicolas, Cho Dahn, Faye Bilal, Ke Zong-You, Ma ikainen Hanna, Pecman Mojca,
Wisniewski Guillaume, Yunès Jean-Bap is e, Zhu Lichao, Zimina-Poi o Ma ia 2021:
The SPECTRANS Sys em Desc ip ion o he WMT21 Te minology Task. – ACL
Rolling Re iew, a New Ini ia i e o he Associa ion o Compu a ional Linguis ics, 818–
825. A ailable a : h p://www.s a m .o g/wm 21/pd /2021.wm -1.80.pd .
Béna d Maud, Mes i ie Alexand a, Kuble Na alie, Zhu Lichao, Bawden Rachel, de la
Cle ge ie E ic, Roma y Lau en , Huguin Ma hilde, Nominé Jean-F ancois, Peng
Ziqian, Y on F ançois 2023: MaTOS: T aduc ion au oma ique pou la science
ou e e [MaTOS: Machine T ansla ion o Open Science]. – Ac es de l'a elie
35
“Analyse e Reche che de Tex es Scien i iques” (ARTS)@TALN 2023, 8–15. A ailable
a : h ps://aclan hology.o g/2023.jep aln eci al-a s.2.pd .
Boons a Lee 2025: P omp Enginee ing. Whi e pape . A ailable a :
h ps://www.kaggle.com/whi epape -p omp -enginee ing.
Bou igaul Didie , L’Homme Ma ie-Claude, Jacquemin Ch is ian 2001: Recen Ad ances in
Compu a ional Te minology, Ams e dam/Philadelphia: John Benjamins.
Bou igaul Didie , Slodzian Monique 1999: Pou une e minologie ex uelle [Towa ds a
Tex ual Te minology]. – Te minologies nou elles 19, 29–32.
Casal J. Ellio , Kessle Ma 2023: Can Linguis s Dis inguish Be ween Cha GPT/AI and
Human W i ing? A S udy o Resea ch E hics and Academic Publishing. – Resea ch
Me hods in Applied Linguis ics 2(3), 100068. A ailable a :
h ps://doi.o g/10.1016/j. mal.2023.100068.
Cha GPT, based on GPT-4- u bo, eleased on he 6 h No embe 2023, OpenAI. A ailable
a : h ps://cha gp .com.
Condamines Anne 2005: Linguis ique de co pus e e minologie [Co pus Linguis ics and
Te minology]. – Langages 157. La e minologie: na u e e enjeux, ed. L. Depecke , 36–
47.
Condamines Anne 2022: Te minologie, In elligence A i icielle e Psychologie cogni i e:
é lexions su les in e ac ions possibles dans l’é ude de la a ia ion en langues
spécialisées [Te minology, A i icial In elligence and Cogni i e Psychology:
Re lec ions on Possible In e ac ions in he S udy o Va ia ion in Specialised
Languages]. – De Eu opa. Eu opean and Global S udies Jou nal, Special Issue on
Mul ilingualism and Language Va ie ies in Eu ope in he Age o A i icial
In elligence, Uni e si à di To ino, 131–148.
Condamines Anne, Rebey olle Jose e 2001: Sea ching o and Iden i ying Concep ual
Rela ionships ia a Co pus-Based App oach o a Te minological Knowledge Base
(CTKB). – Recen Ad ances in Compu a ional Te minology, ed. D. Bou igaul , M.-
C. L’Homme, C. Jacquemin, Ams e dam/Philadelphia: John Benjamins, 127–148.
Da ies Ma k 2025: Co po a and AI/LLM: Compa ison o he P edic ions o LLMs (Cha GPT
and Gemini) o Ac ual Co pus Da a (mainly om English-Co po a.o g). Whi e pape .
A ailable a : h ps://www.english-co po a.o g/ai-llms/english-co po a-wi h-ai-
llms.pd , alongside a ideo pos ed on he 18 h o Ma ch 2025. A ailable a :
h ps://you u.be/WAzWoNzhZ9A?si=JT6TAhOMLkXjUoa4.
DeepSeek, based on DeepSeek-V3-0324, 深度求索 (DeepSeek). A ailable
a : h ps://cha .deepseek.com.
EMT Compe ence F amewo k 2022. A ailable a :
h ps://commission.eu opa.eu/sys em/ iles/2022-
11/em _compe ence_ wk_2022_en.pd .
36
Falgas Julien, Robe Pascal 2023: P ésen e l’IA comme une é idence, c’es empêche de
é léchi au numé ique ( ep is su le i e: Commen le discou s média ique su l’IA
empêche d’en isage d’au es possibles) [P esen ing AI as a Fac P e en s Re lec ion
on Digi al Technologies (based on he i le: How Media Discou se on AI P e en s
Conside ing O he Possibili ies)]. – The con e sa ion, 19 oc ob e 2023. A ailable a :
h ps:// hecon e sa ion.com/commen -le-discou s-media ique-su -lia-empeche-
den isage -dau es-possibles-211766.
Fina di Ky ia 2023: The Pa adox o ou Time and Role o Applied Linguis s in i : Con é ence
donnée à l’occasion du webinai e 2023 de l’AFLA. A ailable a : h p://www.a la-
asso.o g/webinai e-2023.
Gledhill Ch is ophe , Küble Na alie 2015: How T ainee T ansla o s Analyse Lexico-
G amma ical Pa e ns. – Jou nal o Social Sciences 11(3): Special issue on Ph aseology,
Ph aseodidac ics and Cons uc ion G amma (s), ed. M. I. González-Rey, 162–178.
Humbley John 2022: The Recep ion o Wüs e ’s Gene al Theo y o Te minology. –
Theo e ical Pe spec i es on Te minology: Explaining Te ms, Concep s and Specialized
Knowledge, ed. P. Fabe , M.-C. L’Homme, Book coll. ‚Te minology and
Lexicog aphy Resea ch and P ac ice‛, Ams e dam/Philadelphia: John Benjamins,
15–36.
IATE. A ailable a : h p://ia e.eu opa.eu.
IGI Global Dic iona y. A ailable a : h ps://www.igi-global.com/dic iona y.
ISO – TC 37 Language and Te minology – ISO S anda d 1087:2019(en) Te minology
Wo k and Te minology Science: Vocabula y. A ailable a :
h ps://www.iso.o g/obp/ui/#iso:s d:iso:1087:ed-2: 1:en.
ISO – TC 37 Language and Te minology – ISO S anda d 704:2022(en) Te minology Wo k
– P inciples and Me hods. A ailable a :
h ps://www.iso.o g/obp/ui/en/#iso:s d:iso:704:ed-4: 1:en.
ISO/TC 37/SC 3/WG 6 Te minology Managemen and A i icial In elligence – ISO
S anda d 5078:2025(en) Managemen o Te minology Resou ces – Te minology
Ex ac ion. A ailable a : h ps://www.iso.o g/obp/ui/en/#iso:s d:iso:5078:ed-
1: 1:en.
ISO/TC 37/SC 3/WG 6 Te minology Managemen and A i icial In elligence – ISO
S anda d 12620-1:2022 Managemen o Te minology Resou ces – Da a Ca ego ies –
Pa 1: Speci ica ions. A ailable a :
h ps://www.iso.o g/obp/ui/en/#iso:s d:iso:12620:-1:ed-1: 1:en.
Kilga i Adam, Baisa Ví , Buš a Jan, Jakubíček Miloš, Ko ář Voj ěch, Michel ei Jan,
Rychlý Pa el, Suchomel Ví 2014: The Ske ch Engine: Ten Yea s On. –
Lexicog aphy 1, 7–36.
37
Küble Na alie, Ma ikainen Hanna, Mes i ie Alexand a, Pecman Mojca 2024: Pos -edi ing
Neu al Machine T ansla ion in Specialised Languages: he Role o Co po a in he
T ansla ion o Ph aseological S uc u es. – Recen Ad ances in Mul iwo d Uni s in
Machine T ansla ion and T ansla ion Technology, ed. I. J. Mon i, G. Co pas Pas o ,
R. Mi ko , C. Manuel Hidalgo-Te ne o, Cu en Issues in Linguis ic Theo y 366,
Ams e dam/Philadelphia: John Benjamins, 57–78.
Küble Na alie, Pecman Mojca 2012: The ARTES Bilingual LSP Dic iona y: om
Colloca ion o Highe O de Ph aseology. – Elec onic Lexicog aphy, ed. S. G ange ,
M. Paquo , Ox o d: Ox o d Uni e si y P ess, 187–209.
Küble Na alie, Pecman Mojca 2025: Co pus Linguis ics: a Dinosau o an Elephan in he IA
Room? A Wo d om Te minology and T ansla ion S udies Pe spec i e: Phd Semina –
Win e school – Fo eign Li e a u es and Languages school, 5–6 Feb ua y 2025,
Uni e si à di Ve ona, I aly.
Leona di Na ascia 2025: Te minology Science, In e na ional Languages, and Knowledge
Communica ion. – Te minology Th oughou His o y. A discipline in he Making,
ed. K. Wa bu on, J. Humbley, Te minology and Lexicog aphy Resea ch and
P ac ice 24, Ams e dam/Philadelphia: John Benjamins Publishing Company, 148–
166.
L’Homme Ma ie-Claude 2024a: Managing Polysemy in Te minological Resou ces. –
Te minology 30(2), 216–249.
L’Homme Ma ie-Claude 2024b: Se en Good Reasons o a Be e Accoun o Fine-g ained
Polysemy in Te minological Resou ces. – Te minologija 31, 6–23. A ailable a :
h ps://jou nals.lki.l / e minologija/a icle/ iew/2407/2408.
Lomba d Alizée, Huyghe Richa d, Ba que Lucie, G as Do iane 2023: Regula Polysemy and
No el Wo d-sense Iden i ica ion. – The Men al Lexicon 18(1), 94–119.
Lommel A le 2024: The Rise o La ge Language Models In o med by no so La ge Co po a
o T aining Da a. – Digi al T ansla ion 11/1, 73–84. A ailable a :
h ps://doi.o g/10.1075/d .24006.lom.
Meye Ing id 2001: Ex ac ing Knowledge-Rich Con ex s o Te minog aphy: A Concep ual
and Me hodological F amewo k. – Recen Ad ances in Compu a ional Te minology, ed.
D. Bou igaul , Ch. Jacquemin, M.-C. L’Homme, Ams e dam/Philadelphia: John
Benjamins, 279–302.
Meye Ing id, Bowke Lynne, Eck Ka en 1992: Cogni e m: An Expe imen in Building a
Te minological Knowledge Base. – P oceedings o 5 h EURALEX In e na ional
Cong ess on Lexicog aphy, Tampe e, Finland. Tampe e: S udia T ansla ologica, 159–
172.
Michel Jean-Bap is e, Shen Yuan Kui, Aiden A i a P esse , Ve es Ad ian, G ay Ma hew K.,
B ockman William, The Google Books Team, Picke Joseph P., Hoibe g Dale,
38
Clancy Dan, No ig Pe e , O wan Jon, Pinke S e en, Nowak Ma in A., Aiden
E ez Liebe man 2011: Quan i a i e Analysis o Cul u e Using Millions o Digi ized
Books. – Science 331(6014), 176–82. A ailable a :
h ps://pubmed.ncbi.nlm.nih.go /21163965.
Pecman Mojca 2012: Ten a i eness in Te m Fo ma ion: a S udy o Neology as a Rhe o ical
De ice in Scien i ic Pape s. – Neology in Specialized Communica ion, ed. M. Te esa
Cab é Cas ell í, Rosa Es opá Bago , Ma ia C. Va gas Sie a. – Special issue o
Te minology 18(1), 27–58.
Pecman Mojca 2014: Va ia ion as a Cogni i e De ice: How Scien is s Cons uc Knowledge
Th ough Te m Fo ma ion. – Te minology 20(1), 1–24.
Pecman Mojca 2018: Langue e cons uc ion de connaisSENSes. Ene gie lexico-discu si e e
po en iel sémio ique des sciences [Language and he Cons uc ion o Meaning and
Knowledge: Lexico-Discu si e Ene gy and he Semio ic Po en ial o Science]:
monog aphie, p é ace de M.-C. L’Homme. Pa is: Edi ions L’Ha ma an.
Pecman Mojca 2021: 10 h Anni e sa y o he ARTES Te minological and Ph aseological
Da abase. – S ije od iječi: e minološki i e minog a ski ogledi (“The wo ld o wo ds:
Te minological and e minog aphic discussions), ed. I. B ač, A. Os oški Anić,
Izda alacko izdanje Ins i u a za h a ski jezik i jezikoslo lje (Publica ion o he
Ins i u o C oa ian language and linguis ics, Zag eb), 301–324.
Pecman Mojca, Küble Na alie 2011: ARTES: An Online Lexical Da abase o Resea ch and
Teaching in Specialized T ansla ion and Communica ion. – P oceedings om
In e na ional Wo kshop on Lexical Resou ces (WoLeR) 2011 a ESSLLI, 1–5 Augus
2011, Ljubljana, Slo enia, 87–93. A ailable a :
h p://alpage.in ia. /~sago /pub/WoLeR_2011_p oceedings.pd .
Pecman Mojca, Küble Na alie 2022: Tex Gen es and Te minology. – Theo e ical
Pe spec i es on Te minology: Explaining Te ms, Concep s and Specialized Knowledge,
ed. P. Fabe , M.-C. L’Homme, Book coll. Te minology and Lexicog aphy Resea ch
and P ac ice, Ams e dam/Philadelphia: John Benjamins, 263–289.
Pe plexi y, based on GPT-4o/Claude 4.0 Sonne , eleased on he 6 h o May 2025,
Pe plexi y AI. A ailable a : h ps://www.pe plexi y.ai.
P asznik Ba osz, Robe Lew 2025: Dic iona ies Ve sus AI Tools Th ough he Eyes o English
Majo s: In e na ional Jou nal o Lexicog aphy 38(2), 140–158. A ailable a :
h ps://doi.o g/10.1093/ijl/eca 005.
Ras ie F ançois 2021: Da a s co po a. – L’in elligence a i icielle des ex es. Des algo i hmes à
l’in e p é a ion [The A i icial In elligence o Tex s: F om Algo i hms o
In e p e a ion] 15, ed. D. Maya e, L. Vanni, coll. Le es numé iques, Pa is:
Édi ions Hono é Champion, 203–246.
39
Raus Rachele, Ma ioda Ma ia Ma ghe i a 2024: L’in elligence a i icielle en salle de classe:
la pe cep ion des é udian es e des é udian s [A i icial In elligence in he Class oom:
S uden s’ Pe cep ions]. – Mul ilinguisme eu opéen e IA: en e d oi , aduc ion e
didac ique des langues/Mul ilinguismo eu opeo e IA a di i o, aduzione e dida ica
delle lingue/Eu opean Mul ilingualism and A i icial In elligence: The Impac s on Law,
T ansla ion and Language Teaching, ed. R. Raus, F. Bisiani, M. M. Ma ioda,
M. Ton i, De Eu opa, Special Issue, 221–231.
Renou An oine e 2020: Seman ic Neology. Challenges in Ma ching Co pus-based
Seman ic Change o Real-wo ld Change. – Co po a and he Changing Socie y. S udies
in he E olu ion o English, ed. P. Rau ionaho, A. Nu mi, J. Klemola,
Ams e dam/Philadelphia: John Benjamins, 79–112.
Riehmann Pa ick, G uendl Henning, Po has Ma in, T enkmann Ma in, S ein Benno,
F öhlich Be nd 2012: WORDGRAPH: Keywo d-in-Con ex Visualiza ion o
NETSPEAK’s Wildca d Sea ch. – IEEE T ansac ions on Visualiza ion and Compu e
G aphics 18(9), 1411–1423. A ailable a :
h ps://downloads.webis.de/publica ions/pape s/ iehmann_2012.pd .
San Ma ín An onio P. 2024: Wha Gene a i e A i icial In elligence Means o
Te minological De ini ions. – P oceedings o he 3 d In e na ional Con e ence on
Mul ilingual Digi al Te minology Today (MDTT 2024), G anada: CEUR-WS.
A ailable a : h ps://ceu -ws.o g/Vol-3703/pape 1.pd .
de Sch y e Gilles-Mau ice 2023: Gene a i e AI and Lexicog aphy: The Cu en S a e o
he A Using Cha GPT. – In e na ional Jou nal o Lexicog aphy 36(4), 355–387.
Schulho Sande , Michael Ilie, Nishan Balepu , Kons an ine Kahadze, Amanda Liu,
Chenglei Si, Yinheng Li, Aayush Gup a, HyoJung Han, Se ien Schulho , P ana
Sandeep Dulepe , Sau a Vidyadha a, Dayeon Ki, Swe a Ag awal, Chau Pham,
Ge son K oiz, Feileen Li, Hudson Tao, Ashay S i as a a, He ande Da Cos a, Saloni
Gup a, Megan L. Roge s, Inna Goncea enco, Giuseppe Sa li, Igo Galynke , Denis
Pesko , Ma ine Ca pua , Jules Whi e, Shyamal Anadka , Alexande Hoyle, Philip
Resnik 2025: The P omp Repo : A Sys ema ic Su ey o P omp Enginee ing
Techniques. A ailable a : h ps://a xi .o g/pd /2406.06608.
Te mium Plus. A ailable a : h ps://www.b b. e miumplus.gc.ca.
UNTERM. A ailable a : h ps://un e m.un.o g/un e m2/en.
Vezzani Fede ica, Di Nunzio Gio gio, Ma ia Cos a Ru e 2023: ISO S anda ds o
Te minology Resou ces Managemen . A e They FAIR Enough? – Digi al T ansla ion
10/2, Ams e dam/Philadelphia: John Benjamins, 233–252. A ailable a :
h ps://doi.o g/10.1075/d .00009. ez.
Vi ine Linguis ique. A ailable a : h ps:// i inelinguis ique.oql .gou .qc.ca.
40
Wilkinson Ma k D., Dumon ie Michel, Aalbe sbe g IJsb and Jan, Apple on Gab ielle 2016:
The FAIR Guiding P inciples o Scien i ic Da a Managemen and S ewa dship. –
Scien i ic Da a 3, 160018. A ailable a : h ps://doi.o g/10.1038/sda a.2016.18.
WIPO. A ailable a : h ps://wipopea l.wipo.in / /linguis ic.
Wüs e Eugen 1968: The Machine Tool./La Machine-Ou il. An In e lingual Dic ona y o Basic
Concep s Comp ising an Alphabe cal Dic onna y and a Classi ied Vocabula y Wi h
De ini ons and Illus a ions: English-F ench Mas e Volume, Lond es, Technical
P ess. Aslib, 173 p.
Zhu Lichao, Zimina Ma ia, Béna d Maud, Namda Behnoosh, Ballie Nicolas, Wisniewski
Guillaume, Yunès Jean-Bap is e 2023: In es iga ing Techniques o a Deepe
Unde s anding o Neu al Machine T ansla ion (NMT) Sys ems Th ough Da a
Fil e ing and Fine- uning S a egies. – P oceedings o he Eigh h Con e ence on
Machine T ansla ion, Singapo e, Associa ion o Compu a ional Linguis ics, 275–281.
Re e ences o he ex s on da a pollu ion (used o he co pus and/o
appea ing in he ou pu s o GenAI ools)
Ben-Shaha Om i 2018: Da a Pollu ion. – Uni e si y o Chicago Public Law & Legal Theo y
Pape Se ies 679, 104–159.
Cao Yinzhi, Fangxiao Yu Alexande , Aday And ew, S ahl E ic, Me wine Jon, Yang Jun eng
2018: E icien Repai o Pollu ed Machine Lea ning Sys ems ia Causal
Unlea ning. – P oceedings o 2018 ACM Asia Con e ence on Compu e and
Communica ions Secu i y, Incheon, Republic o Ko ea, 4–8 June 2018 (ASIA CCS
’18).
Cas elluccia Claude, Kaa a Mohamed Ali 2009: Owne -Cen ic Ne wo king (OCN):
Towa d a Da a Pollu ion-F ee In e ne . – P oceedings o he 2009 Nin h Annual
In e na ional Symposium on Applica ions and he In e ne , 169–172.
Ge sen F ed 2022: Da a Pollu ion and Taxa ion: Mas e ’s Thesis, Execu i e Mas e Cybe
Secu i y, De ended on he 31s o Ma ch 2022, Uni e si ei Leiden, Ne he lands.
Hasselbalch G y 2021: Da a E hics o Powe : A Human App oach in he Big Da a and AI E a,
Glos (UK): Edwa d Elga .
Hasselbalch G y 2022: Da a Pollu ion & Powe , Whi e Pape o a Global Sus ainable Agenda
on AI, The Sus ainable AI Lab, Bonn Uni e si y.
Mo i Leona do, F ancey Alizée, Me le Tobias 2024: Da a Pollu ion: De ini ion and Policy
Responses. – P oceedings o he 16 h IFIP WG 8.5 In e na ional Con e ence, Elec onic
Pa icipa ion (ePa 2024), Ghen , Belgium, Sep embe 3–5 2024, 147–162.
Zimme li Wal he Ch. 1986: Who is o Blame o Da a Pollu ion? On Indi idual Mo al
Responsibili y Wi h In o ma ion Technology. – Philosophy and Technology II:
41
In o ma ion Technology and Compu e s in Theo y and P ac ice, ed. C. Mi cham,
A. Huning, Do d ech /Bos on: D. Reidel, 291–305.
APPENDIX
Figu e 24 shows a p omp wi hou he speci ica ion ‚wi h he exac wo ding ound in
he ex s‛, and he ou pu consis ing o wo e oneous, made-up con ex s, whe e he
de ini ion o in o ma ion pollu ion is c edi ed o da a pollu ion. Ne e heless, he p oposal by
Cha GPT ha ‚da a pollu ion encompasses he con amina ion o in o ma ion sou ces…‛,
can be ega ded as ue, i we ake in o accoun he seman ic ela ion be ween da a pollu ion
and in o ma ion pollu ion e ealed by he discu si e ma ke o me onymy ‚encompasses‛,
and acco ding o which in o ma ion pollu ion is a pa o da a pollu ion. The e o mula ion by
LLMs ‚The concep o da a pollu ion includes in o ma ion pollu ion‛ in o ‚da a pollu ion
encompasses [de ini ion o in o ma ion pollu ion] …‛ is espec ul o knowledge
ep esen a ion. Howe e , in e minological analysis, i is pa amoun o dis inguish
seman ically ela ed e ms, all he mo e so when hey a e nea -like synonyms.
GOAL
Finding de ini ional con ex s and hei e e ences
MODEL
Cha GPT-4- u bo
DATE
13/05/2025
TECH-
NIQUE
Sys em p omp ing
LIMIT
3 de ini ions
PROMPT
Find h ee mos ele an de ini ions o he e m ‚da a pollu ion‛ in scien i ic li e a u e and
p o ide he sou ces o each de ini ion ound. Re u n he de ini ions in he lis ollowed by
he bibliog aphical e e ence o he sou ce and i s URL.
OUTPUT
Da a pollu ion e e s o he ha m ul ‚da a emissions‛ leaked in o he digi al ecosys em,
dis up ing social ins i u ions and public in e es s. This concep e hinks he ha ms he da a
economy c ea es and he way hey ha e o be egula ed.
Ben-Shaha O. 2018: Da a Pollu ion 679, Uni e si y o Chicago Public Law & Legal Theo y
Pape Se ies. A ailable a :
h ps://chicagounbound.uchicago.edu/public_law_and_legal_ heo y/699
Da a pollu ion is he in e ela ed ad e se impac ha he gene a ion, s o ing, handling, and
p ocessing o digi al da a has on ou na u al en i onmen , social en i onmen , and pe sonal
en i onmen . I is he unsus ainable handling, dis ibu ion, and gene a ion o da a esou ces.
Hasselbalch G. e al. 2022: Da a Pollu ion & Powe - Whi e Pape o a Global Sus ainable
De elopmen Agenda on AI. A ailable a :
h ps://www. esea chga e.ne /publica ion/362174889_Da a_Pollu ion_Powe _-
_Whi e_Pape _ o _a_Global_Sus ainable_De elopmen _Agenda_on_AI_by_G y_Hasselbalc
h_wi h_con ibu ions_ om_ he_Da a_Pollu ion_Powe _DPP_G oup_a _Bonn_Sus ainable_
AI_Lab