GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
123
DOI: h ps://10.5281/10.5281/zenodo.17685835
GENDER BIAS IN ENGLISH–UZBEK MACHINE TRANSLATION: A
COMPARATIVE STUDY OF AI AND HUMAN OUTPUTS
Qud a ullaye a Muni axon Ab o jon qizi
Teache a Kokand Uni e si y
ANNOTATION
This a icle in es iga es gende bias in English-Uzbek machine ansla ion
sys ems by compa ing AI-gene a ed ansla ions wi h human ansla o ou pu s. The
s udy highligh s how AI models o en ep oduce o exagge a e gende s e eo ypes
embedded in English da ase s, while human ansla o s ely on con ex ual easoning
and cul u al adap a ion o mi iga e bias.
Keywo ds: Gende bias, machine ansla ion, Uzbek language, AI s human
ansla ion, neu al MT, s e eo ypes, bilingual communica ion.
АННОТАЦИЯ
В данной статье исследуется гендерная предвзятость в англо-узбекских
системах машинного перевода путем сравнения переводов, созданных
искусственным интеллектом, с переводами, выполненными человеком. В
исследовании подчеркивается, что ИИ-модели нередко воспроизводят или
усиливают гендерные стереотипы, присутствующие в английских датасетах,
тогда как человеческие переводчики опираются на контекстуальное мышление
и культурную адаптацию, чтобы смягчить предвзятость.
Ключевые слова: гендерная предвзятость; машинный перевод; узбекский
язык; перевод ИИ и человека; нейронные системы МП; стереотипы;
билингвальная коммуникация.
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
124
ANNOTATSIYA
Ushbu maqola ingliz-o‘zbek mashina a jimasi izimla idagi gende
a a kashlikni AI omonidan ya a ilgan a jimala a inson a jimonla i baja gan
a jimala o qali ahlil qiladi. AI modella ining ko‘pincha gende s e eo ipla ini
ak o lashi yoki kuchay i ishi, inson a jimonla ining esa kon eks o qali
a a kashlikni yumsha ishga in ilishi ko‘ ib chiqiladi.
Kali so‘zla : Gende a a kashlik, mashina a jimasi, o‘zbek ili, AI a inson
a jimasi, ney on MT, s e eo ipla , ikki illi kommunika siya.
In oduc ion.
Machine ansla ion (MT) is apidly eshaping how English-Uzbek bilingual
communica ion un olds in e e yday li e om s uden s w i ing essays o p o essionals
handling in e na ional co espondence. I s speed and con enience mean ha housands
o use s now ely on MT sys ems as an in isible linguis ic assis an . Howe e , benea h
his e iciency lies an eme ging sociolinguis ic issue: he sub le bu pe sis en
ep oduc .
1
One o he co e challenges s ems om he s uc u al di e ences be ween English
and Uzbek. English p onouns such as hey o occupa e a e inhe i ed “The nu se said
hey would co is o “The enginee explained he p oblem”.
This pa e n does no eme ge om malice bu om da a. MT sys ems lea n om
massi e ex co po a, and i hose ex s e lec adi ional gende oles, he sys ems
ep oduce hem. Ye , o eal people using hese ools, he consequences can eel
pe sonal. S uden s may eel hei w i ing is sub ly al e ed in ways ha mis ep esen
hei in en . Educa o s migh no ice ha ansla ions ein o ce ou da ed gende
expec a ions. In e e yday messaging, a gende -neu al s a emen can become
unin en ionally biased, c ea ing misunde s andings o disco.
2
1
Be dikulo , S. (2017). Issues o gende ep esen a ion in Uzbek linguis ic cul u e. Uzbek Linguis ic Jou nal, 4, 22–30.
2
Bolukbasi, T. e al. (2016). Man is o Compu e P og amme as Woman is o Homemake ?
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
125
Mo eo e , he issue becomes mo e impo an as MT is inc easingly in eg a ed
in o public se ices, educa ional pla o ms, and wo kplace communica ion. When an
au oma ed sys em consis en ly assigns men o posi ions o au ho i y and women o
ca egi ing oles, i shapes pe cep ions especially o younge use s who us digi al
ools. The ansla ion is no longe jus linguis ic; i becomes ideologic.
1
Bu he e is also a human side o his challenge. Many bilingual use s begin o
no ice hese biases and ac i ely co ec hem. Teache s discuss hem wi h s uden s,
ansla o s adap s a egies o coun e ac s e eo ypes, and o dina y use s de elop a
habi o checking MT ou pu s mo e c i ically. In his sense, MT bias becomes a sha ed
lea ning momen : a eminde ha echnology does no simply e lec language, bu
also he socie y om which ha language eme ges.
Ul ima ely, add essing gende bias in English–Uzbek MT is no only a echnical
ask i is a cul u al esponsibili y. De elope s, educa o s, and use s can wo k oge he
o c ea e ansla ions ha espec linguis ic neu ali y and human digni y. As machine
ansla ion con inues o e ol e, so oo mus ou awa eness o he alues we allow i o
encode.
Li e a u e e iew.
Gende bias in ansla ion is no an isola ed echnical law i is a e lec ion o he
deepe linguis ic and cul u al laye s ha shape how socie ies pe cei e gende . When
English-Uzbek ex s a e ansla ed, especially by machine ansla ion sys ems, h ee
in e wined mechanisms o en come in o play: lexical s e eo ypes, g amma ical
assump ions, and socio-cul u al expec a ions.
Lexical s e eo ypes eme ge when ce ain p o essions o oles a e habi ually
associa ed wi h one gende . Fo example, wo ds like nu se, eache , o sec e a y end
o be in e p e ed as eminine, while enginee , manage , o scien is a e o en ende ed
as masculine. These associa ions may no be explici ly encoded in English, bu Uzbek
ansla ions equen ly ac i a e hem. As a esul , a gende -neu al English sen ence
1
C ys al, D. (2011). In e ne Linguis ics.
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
126
becomes gende -ma ked in Uzbek simply because he sys em ills in wha i “expec s”
a he han wha he au ho in ended.
1
G amma ical assump ions also con ibu e o bias. Al hough Uzbek does no
g amma ically equi e gende ed p onouns, ansla o s bo h human and machine o en
eel compelled o choose one, especially in na a i e con ex s. English s uc u es ha
ely on neu al p onouns such as hey o omi gende al oge he can be challenging:
MT models end o esol e ambigui y by selec ing he s a is ically mos common
gende ound in hei aining da a.
2
Thus, ambigui y, which is a na u al pa o human
language, is ans o med in o ce ain y bu o en he w ong kind o ce ain y.
Socio-cul u al expec a ions add a deepe laye o his p ocess. Cul u es shape
language, and languages ein o ce cul u al no ms. I he aining co po a con ain ex s
whe e men domina e public, echnical, and leade ship domains, while women appea
in domes ic o ca egi ing oles, neu al MT models will encode hese pa e ns as
“de aul eali ies”. The esul is ansla ions ha sub ly ep oduce adi ional gende
hie a chies e en when he sou ce ex is neu al o in en ionally inclusi e.
Neu al MT sys ems do no in en hese biases; hey inhe i hem. Thei algo i hms
lea n by iden i ying s a is ical associa ions ac oss millions o sen ences. I he da a
skews owa d s e eo yped ep esen a ions, he model in e nalizes hese biases as
eliable pa e ns. In his way, gende bias is no me ely an e o bu a p edic able ou pu
o he sys em’s aining en i onmen .
3
Ye , ecognizing his p oblem has a human implica ion. Use s begin o no ice
when ansla ions do no e lec hei meaning. Educa o s and linguis s ad oca e o
mo e di e se, balanced co po a. De elope s expe imen wi h debiasing echniques.
Each o hese e o s acknowledges a simple u h: ansla ion is no jus he ans e
o wo ds i is he ans e o social alues.
1
E gashe a, M. (2022). Gende and cul u al nuance in Uzbek ansla ion p ac ices. In e na ional Jou nal o Linguis ics
and T ansla ion S udies, 5(2), 44–57.
2
Holmes, J., & Meye ho , M. (Eds.). (2020). The handbook o language and gende . Wiley.
3
Ho y, D. & Sp ui , S. (2016). Gende Bias in NLP.
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
127
By add essing gende bias in MT, we a e no only imp o ing linguis ic accu acy;
we a e pa icipa ing in a b oade cul u al shi owa d ai ness and ep esen a ion in
digi al communica ion.
Resea ch me hodology.
To unde s and how gende bias mani es s in eal English-Uzbek ansla ion
p ac ice, a da ase o 200 ca e ully selec ed English sen ences was es ed ac oss se e al
majo MT pla o ms Google T ansla e, DeepL, Cha GPT-based MT engines, and a
numbe o egional ansla ion ools commonly used in Cen al Asia. These sys ems
we e chosen because hey ep esen he ools ha o dina y use s, s uden s, educa o s,
and p o essionals mos equen ly ely on in hei daily communica ion.
The es se included a mix o gende -neu al sen ences, ambiguous s uc u es,
and con ex - ich examples in ol ing p o essions and pe sonal oles. By exposing each
sys em o he same linguis ic challenges, i became possible o obse e how
consis en ly (o inconsis en ly) di e en ools handled gende assignmen .
To p o ide a meaning ul benchma k, hese machine-gene a ed ansla ions we e
hen compa ed wi h ou pu s p oduced by expe ienced p o essional human ansla o s.
These ansla o s we e ins uc ed o p ese e neu ali y unless he con ex explici ly
equi ed speci ying gende a p ac ice aligned wi h mode n ansla ion e hics. Thei
wo k se ed as he “gold s anda d” agains which MT e o s, biases, and in e p e a i e
leaps could be e alua ed.
The compa ison e ealed no only he accu acy gaps bu also he sub le choices
each sys em made when con on ed wi h unce ain y. Some MT ools de aul ed o male
e e en s o oles associa ed wi h leade ship o echnical expe ise, while o he s
ended o eminize ca egi ing and se ice-o ien ed p o essions. Human ansla o s, in
con as , app oached hese sen ences wi h ca e ul conside a ion o p agma ic meaning,
cul u al nuance, and he au ho ’s likely in en ion.
This e alua ion p ocess unde sco es an impo an poin : machine ansla ion does
no simply con e ex i e eals he assump ions embedded in i s aining da a. By
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
128
compa ing MT ou pu wi h human judgmen , we begin o see whe e echnology aligns
wi h human easoning and whe e i d i s in o s e eo ype-d i en in e p e a ion.
1
Analysis and esul s.
The compa a i e analysis e ealed a clea pa e n: AI-based ansla ion sys ems
disp opo iona ely ely on masculine de aul s, especially when p ocessing sen ences
ela ed o p o essional o high-s a us occupa ions. Roles such as enginee , manage ,
esea che , lawye , o di ec o we e almos au oma ically ende ed wi h masculine
subjec s in Uzbek, e en when he o iginal English ex con ained no gende indica o s.
This endency e lec s he s a is ical imp in o aining co po a in which men appea
mo e equen ly in echnical, leade ship, o public- acing domains.
In con as , when he inpu sen ences in ol ed emo ional si ua ions, ca e- ela ed
esponsibili ies, o amily oles such as nu se, assis an , eache , ca egi e , o
pa en ing scena ios se e al MT sys ems displayed an opposi e bu equally
s e eo ypical pa e n: hey de aul ed o eminine in e p e a ions. In emo ional con ex s,
AI o en ampli ied adi ional assump ions, o ins ance ansla ing neu al s a emen s
abou com o ing, helping, o exp essing eelings wi h “u ayol” o o he ma ke s ha
implici ly signal eminini y. Such choices do no a ise om linguis ic necessi y bu
om he cul u al biases encoded in he da a ha ains hese models.
O e all, he indings illus a e ha AI does no me ely ansla e ex i
unin en ionally ep oduces and magni ies exis ing socie al biases, unless delibe a ely
co ec ed. Humans, meanwhile, ope a e wi h an awa eness o nuance, esponsibili y,
and con ex , making hei ansla ions mo e inclusi e and cul u ally balanced.
AI ends o exagge a e Wes e n gende s e eo ypes, s uggle wi h p onoun
ambigui y, and misin e p e Uzbek kinship no ms. Human ansla ions show s onge
con ex ual unde s anding and bias mi iga ion.
Conclusion.
1
Mi zaye a, D. (2020). T ansla ion ambigui y in Uzbek: A socio-linguis ic pe spec i e. Jou nal o Cen al Asian
Languages, 8(1), 15–34.
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
129
The s udy concludes ha add essing gende bias in English–Uzbek machine
ansla ion equi es mo e han inc emen al echnical adjus men s i calls o a deepe
e hinking o how AI sys ems a e ained and deployed. A cen al inding is ha cu en
MT models lack su icien Uzbek-speci ic aining da a, especially high-quali y
co po a ha ep esen mode n, inclusi e language use. Mos la ge-scale models ely
hea ily on mul ilingual da ase s domina ed by English and o he global languages,
lea ing Uzbek unde ep esen ed and cul u ally la ened. Wi hou iche , mo e di e se
Uzbek inpu , models con inue o ill gaps wi h biased assump ions lea ned om
un ela ed linguis ic con ex s.
Finally, he s udy highligh s he p omise o hyb id human-AI wo k lows. When
AI handles ini ial d a s and humans pe o m a ge ed edi ing especially in con ex -
sensi i e a eas like gende he esul is bo h e icien and e hically esponsible. Human
o e sigh can ca ch biases ha algo i hms o e look, while AI can accele a e ou ine
asks, allowing ansla o s o ocus on nuance. This syne gy ensu es ha echnological
speed does no come a he expense o cul u al sensi i i y o social equi y.
REFERENCES
1. Bake , M. (2018). T ansla ion and Con lic .
2. Be dikulo , S. (2017). Issues o gende ep esen a ion in Uzbek linguis ic
cul u e. Uzbek Linguis ic Jou nal, 4, 22–30.
3. Bolukbasi, T. e al. (2016). Man is o Compu e P og amme as Woman is o
Homemake ?
4. C ys al, D. (2011). In e ne Linguis ics.
5. Ecke , P., & McConnell-Gine , S. (2013). Language and gende . Camb idge
Uni e si y P ess.
6. E gashe a, M. (2022). Gende and cul u al nuance in Uzbek ansla ion
p ac ices. In e na ional Jou nal o Linguis ics and T ansla ion S udies, 5(2), 44–57.
7. Holmes, J., & Meye ho , M. (Eds.). (2020). The handbook o language and
gende . Wiley.
GOLDEN BRAIN ISSN: 2181-4120 VOLUME 3 | ISSUE 17 | 2025
Mul idisciplina y Scien i ic Jou nal No embe , 2025
130
8. Ho y, D. & Sp ui , S. (2016). Gende Bias in NLP.
9. Jukes, A. (2021). English in luence in Cen al Asian ansla ion.
10. Mi zaye a, D. (2020). T ansla ion ambigui y in Uzbek: A socio-linguis ic
pe spec i e. Jou nal o Cen al Asian Languages, 8(1), 15–34.
11. Sczesny, S., Fo manowicz, M., & Mose , F. (2016). Can gende - ai language
educe gende s e eo yping? F on ie s in Psychology, 7, 25–32.
