Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
114
ISSN:3030-3621
ОСОБЕННОСТИ ПЕРЕВОДА ИДИОМ ПРИ ИСПОЛЬЗОВАНИИ
ИСКУССТВЕННОГО ИНТЕЛЛЕКТА
Каримова Оминахон Махмудовна
студентка 3-го курса
Национальный университет Узбекистана
имени Мирзо Улугбека,
г. Ташкент, Узбекистан
Научный руководитель: д.ф.н.
(DSc)
Арустамян Я. Ю.
E-mail: omishka13@gmail.com
Аннотация
Несмотря на значительный прогресс в развитии машинного перевода
(МП), перевод идиоматических выражений остается сложной задачей. Идиомы
часто несут скрытый смысл, который нельзя передать дословно, а их правильный
перевод требует не только языковой, но и культурной адаптации. Современные
нейросетевые модели демонстрируют высокую степень беглости и
грамматической точности, но по-прежнему сталкиваются с проблемами в
передаче фигурального значения. Машинный перевод часто приводит к
буквальному переводу, потере смысловых оттенков и культурным
несоответствиям. Это подчеркивает важность человеческого участия в переводе,
особенно в контекстах, где необходимо учитывать культурные и стилистические
особенности. В работе рассматриваются основные трудности, возникающие при
передаче идиом в машинном переводе, и их влияние на межъязыковую
коммуникацию. Для повышения точности перевода необходимо дальнейшее
развитие технологий, способных учитывать контекст и смысловую
многозначность языка.
Keywords:
машинный
перевод,
идиоматические
выражения,
искусственный интеллект, культурная адаптация, семантическая точность,
стратегии перевода.
FEATURES OF IDIOM TRANSLATION USING ARTIFICIAL
INTELLIGENCE
Karimova Ominaxon Maxmudovna
3
rd
year student
National University of Uzbekistan named after Mirzo Ulugbek,
Tashkent, Uzbekistan
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
115
ISSN:3030-3621
Academic supervisor: DSc
Arustamyan Y. Y.
E-mail: omishka13@gmail.com
Abstract
Machine translation (MT) has significantly advanced with the development of
neural machine translation (NMT), improving fluency and grammatical accuracy.
However, translating idiomatic expressions remains a major challenge due to the
complexity of figurative language and cultural nuance. Idioms often carry meanings
that cannot be understood through direct translation, requiring a deep understanding of
both source and target languages. While AI-driven MT systems attempt to address
these challenges, they frequently produce errors such as literal translation, loss of
idiomatic meaning, and cultural misalignment. These limitations highlight the gap
between computational linguistic processing and human cognitive abilities in
translation. This study explores the difficulties MT systems face in idiomatic
translation and the implications for cross-linguistic communication. Despite
technological advancements, human expertise remains essential in contexts requiring
cultural adaptation. Future developments should focus on improving AI’s ability to
recognize figurative language, ensuring more accurate and contextually appropriate
translations.
Keywords:
machine translation, idiomatic expressions, artificial intelligence,
cultural adaptation, semantic accuracy, translation strategies.
Translation is more than a mechanical transfer of words from one language to
another; it is a complex process that requires a deep understanding of linguistic
structures, cultural context, and intended meaning. One of the most critical aspects of
translation is cultural adaptation, which involves modifying a text so that it aligns with
the cultural and linguistic norms of the target audience while preserving its original
intent. Without proper cultural adaptation, translations can become misleading,
unnatural, or even incomprehensible. This challenge is particularly evident in the
translation of idiomatic expressions, which often carry figurative meanings that cannot
be understood through direct, word-for-word translation. In recent years, machine
translation (MT) has seen remarkable advancements, with AI-powered tools such as
Google Translate, DeepL, and ChatGPT significantly improving linguistic accuracy
and fluency. However, despite these improvements, MT systems continue to face
substantial limitations in cultural adaptation. Unlike human translators, AI lacks real-
world experiences, cultural intuition, and the ability to interpret meaning beyond literal
definitions. Liu (2022) posits that the present AI technologies lack the requisite
advancement to entirely supplant human translators. Although AI has made notable
advancements in the domain of language translation, it still falls short of the nuanced
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
116
ISSN:3030-3621
comprehension of language and cultural context that human translators possess.
Moreover, human translators possess the capability to decipher idiomatic phrases and
colloquial language, which could pose a challenge for AI to precisely translate. [4; 935]
As a result, machine-generated translations often struggle with idiomatic expressions,
leading to errors such as literal translation, idiomatic loss, mistranslation, and semantic
shifts.
Translation has long been recognized as more than a mechanical process of
converting words from one language to another. At its core, it involves negotiating
meaning across cultural and linguistic boundaries. One of the central debates in
translation studies revolves around the degree to which a translated text should remain
faithful to the source language versus how much it should be adapted to fit the cultural
norms of the target audience. Lawrence Venuti’s (1995) theory of domestication and
foreignization offers a useful framework for understanding cultural adaptation in
translation. Domestication refers to a strategy where the translator modifies the text to
make it sound natural and familiar to the target audience, often replacing culturally
specific references with equivalents that are easily understood. On the other hand,
foreignization seeks to preserve the original cultural and linguistic characteristics of
the source text, even if they appear unusual or challenging to the target audience. While
domestication enhances readability and accessibility, foreignization maintains the
uniqueness and authenticity of the original expression. Idioms and culturally embedded
expressions present a unique challenge within this framework. A domesticated
translation might replace an idiom with a culturally appropriate equivalent in the target
language, while a foreignized approach might provide a literal translation, preserving
the original structure but potentially making the meaning less clear. The choice
between these approaches depends on the translator’s goals, the audience’s
expectations, and the context in which the translation is used.
With the rapid development of artificial intelligence, machine translation (MT)
has undergone significant transformations. Modern AI-based MT systems, such as
Google Translate, DeepL, and ChatGPT, rely on neural machine translation (NMT)—
a deep-learning approach that processes entire sentences rather than translating words
in isolation. These systems are trained on massive multilingual datasets, allowing them
to identify patterns and generate translations that are often more fluent and
grammatically correct than rule-based or statistical methods. However, AI translation
contains deficiencies and technical issues originating from natural language
processing. Neural networks work on fluency and coherence better yet they have some
errors that a human translator would not commit (Tomasello, 2019). Such problem
includes Homographs, Paronyms, and Ambigrammatical, which refer to words with
the same pronunciation but have different meanings and different syntactical functions
and the resultant effect is either grammatical inaccuracy or the production of a word
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
117
ISSN:3030-3621
that is alien in meaning to the subject in question. [1; 8] Moreover, AI-driven MT still
struggles with cultural adaptation and the translation of idiomatic expressions. Unlike
human translators, AI lacks an intuitive understanding of culture, context, and
pragmatics. Instead, it relies on statistical probabilities and pattern recognition, which
often lead to literal translations that fail to capture the intended meaning of idiomatic
phrases. Furthermore, since AI is trained on pre-existing translations, it can reinforce
biases, inconsistencies, and errors present in its training data. Another fundamental
limitation of MT is its inability to interpret and recreate metaphorical language in a
way that aligns with the cultural expectations of the target audience. While AI can
recognize frequently translated idioms, it often fails when faced with novel or low-
resource idiomatic expressions. This highlights a crucial gap between AI’s linguistic
processing capabilities and the deeper cognitive and cultural reasoning required for
effective translation.
Idioms are integral to every language, shaping the way people express ideas,
emotions, and cultural values. Idioms are fixed expressions whose meanings cannot be
deduced from the literal definitions of their individual words. Translating idioms has
always been considered a challenging decision-making process for translators, which
requires a lot of experience and creativity Even acknowledged and experienced
translators, who ideally have a well-founded knowledge of the target language and its
cultural aspects, cannot match the ability of native speakers in deciding when –
meaning in what text type or context – certain idioms would or would not be
appropriate A thorough knowledge of the source and target language is indispensable
in this process, which also requires creativity and the skill, willingness, and
perseverance to search for the best equivalent. [5; 86] From a translation perspective,
idioms require a nuanced approach, as their direct translations often lead to
misinterpretation or loss of meaning. Successful translation strategies may include:
Equivalence - finding an idiom with a similar meaning in the target language;
Paraphrasing - explaining the idiom’s meaning instead of providing a direct translation;
Literal Translation - a direct word-for-word translation, which often results in loss of
meaning; Cultural Substitution - replacing the idiom with a culturally relevant phrase
that conveys the same idea. Machine translation, however, frequently struggles with
idioms because it lacks the ability to contextualize figurative language. AI models often
default to literal translations, failing to recognize when an expression is being used
idiomatically. This results in translations that may be grammatically correct but
semantically incorrect or awkward.
To better understand the challenges AI faces in idiom translation, it is useful to
consider Mona Baker’s (1992) classification of idiom translation strategies. These
strategies are: A.
Using an Idiom of Similar Meaning and Form
- this approach entails
utilizing an idiom in the target language (TL) that has essentially the same meaning as
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
118
ISSN:3030-3621
the source language's idiom and also contains lexical elements that are similar. B.
Using an Idiom of Similar Meaning but Dissimilar Form
- finding a fixed phrase or
idiom in the target language that is composed of different lexical elements but has a
comparable meaning to the source language's expression or idiom may be achievable.
C.
Translation by Paraphrase -
due to variations in the stylistic preferences of the
source and target languages, this is now the most popular method
of translating idioms
when a correspondence cannot be found in the target language or when it seems
undesirable to
employ idiomatic language in the target translation.
D.
Translation by
Omission -
an idiom could occasionally be completely omitted in the TT, just like it
does with single words. It might not have a
close equivalent in the target language (TL),
its meaning may be difficult to interpret, or it might be for aesthetic
reasons. [2; 141]
While these strategies are widely used in human translation, AI systems often struggle
to apply them effectively, leading to literal translations or semantic distortions.
This study examines the ability of machine translation (MT) systems to process
idiomatic expressions by analyzing a random selection of Russian idioms. The idioms
were not chosen based on frequency or common usage but were selected arbitrarily to
assess how AI-driven translation tools handle non-literal and culturally embedded
expressions. This approach ensures that the evaluation reflects the AI systems’ ability
to recognize, interpret, and adapt figurative language, rather than relying on pre-learned
translations of well-known idioms. The translations were generated using three AI-
based MT systems: Google Translate, DeepL, and ChatGPT. These tools were selected
for their widespread use and advanced neural translation capabilities. The idioms were
translated from Russian into English using each system without human intervention to
ensure that the results reflect the raw output of AI processing.
Table 1: Translation
Russian Phrase Google
Translate
DeepL
ChatGPT
Божий
одуванчик
God’s dandelion God’s dandelion
/ Dandelion of
God
Sweet old dear/
Little old lady
Дубина
стоеросовая
Steros club
Stupid stooge /
Stool pigeon
Big oaf/
Thickheaded
Закля́дочный
друг
Bosom friend
A close friend/
A dear friend
Bosom friend /
Close friend
Заячья душа
Hare soul
Hare’s soul
Timid soul /
Cowardly soul
Казанская
сирота
Kazan orphan
A Kazan orphan
/ Kazan's orphan
False orphan /
Feigned orphan
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
119
ISSN:3030-3621
Жизнь бьёт
ключом
Life is in full
swing
Life is booming
Life is in full
swing / Life is
bustling
Кричит во всю
Ивановскую
Screams at the
top of Ivanovo
Shouting all over
Ivanovo
Shouting at the
top of one’s
lungs
Купаться в
роскоши
Bask in luxury
Bathe in luxury
To wallow in
luxury / To live
in the lap of
luxury
Не жизнь, а
малина
Not life, but
raspberries
Life is a
raspberry
Life is a bed of
roses
Делу время,
потехе час
Time for
business/ Time
for fun
It’s business as
usual
A time for work
and a time for
play
Table 2: Translation Strategy Definitions
Translation Strategy
Definition
Literal Translation
Word-for-word substitution without considering cultural
meaning.
Loan Translation
(Calque)
Directly borrowing structure but adapting it to the target
language.
Descriptive Translation Explaining the meaning instead of direct substitution.
Equivalence (Idiomatic
Translation)
Finding the closest equivalent phrase in the target
language.
Generalization
Using a broader term instead of a specific cultural
reference.
Adaptation (Cultural
Substitution)
Replacing a culturally specific term with one familiar to
the target audience.
Incorrect Translation
(Mistranslation)
Producing a wrong or misleading result.
Table 3: Comparative Analysis of Translations
Phrase
Google
Translat
e
Strategy
(Google)
DeepL
Strategy
(DeepL)
ChatGP
T
Strategy
(ChatGP
T)
Божий
одуванчик
God’s
dandelio
n
Literal
God’s
dandelio
n /
Dandelio
n of God
Literal /
Calque
Sweet
old dear
Adaptatio
n
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
120
ISSN:3030-3621
Дубина
стоеросова
я
Steros
club
Mistranslati
on
Stupid
stooge /
Stool
pigeon
Generalizati
on
Big oaf
Descriptiv
e
Закля́дочн
ый друг
Bosom
friend
Equivalence A close
friend, a
dear
friend
Generalizati
on
Bosom
friend /
Close
friend
Equivalen
ce
Заячья
душа
Hare
soul
Calque
Hare’s
soul
Calque
Timid
soul /
Cowardl
y soul
Equivalen
ce
Казанская
сирота
Kazan
orphan
Literal
A Kazan
orphan,
Kazan's
orphan
Literal
False
orphan /
Feigned
orphan
Descriptiv
e
Жизнь бьёт
ключом
Life is in
full
swing
Equivalence Life is
booming
Equivalence Life is in
full
swing /
Life is
bustling
Equivalen
ce
The analysis of AI-generated translations reveals significant differences in how
Google Translate, DeepL, and ChatGPT handle idiomatic expressions. The primary
challenges observed across all three systems include literal translation, idiomatic loss,
calque, mistranslation, and adaptation. While Google Translate demonstrated a strong
tendency for literal word-for-word translation, DeepL showed a more nuanced
approach but still struggled with idiomatic adaptation. ChatGPT, on the other hand,
exhibited the strongest ability to produce culturally adapted translations but
occasionally introduced over-interpretation, altering the original meaning.
Table 4: AI Translation Tools - Strengths & Weaknesses
Tool
Strengths
Weaknesses
Google Translate
Good for direct, basic
translations.
Too literal, often
produces unnatural
phrases.
Misunderstands
idioms.
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
121
ISSN:3030-3621
DeepL
More natural than
Google Translate.
Sometimes offers
multiple options.
Still struggles with
idioms. Lacks full
cultural adaptation.
ChatGPT
Best at adapting idioms
and cultural meanings.
Uses natural English
equivalents.
Sometimes over-
adapts, losing original
meaning. May
introduce subjective
interpretations.
The findings of this study indicate that machine translation (MT) systems
continue to face significant challenges in processing idiomatic expressions, with errors
stemming from the inability to account for cultural and figurative meaning. The most
prevalent issues observed in AI-generated translations include literal translation,
idiomatic loss, structural borrowing (calque), mistranslation, and over-adaptation.
Among the three systems analyzed, Google Translate demonstrated the highest
frequency of literal translations, producing outputs that closely followed the source
text’s structure but often failed to convey its intended meaning. DeepL, while more
contextually aware, frequently applied calque strategies, leading to syntactically
correct yet semantically unnatural translations. ChatGPT, by contrast, exhibited the
most advanced capacity for cultural adaptation, successfully recognizing idiomatic
meaning. However, it also displayed a tendency toward over-interpretation, sometimes
modifying the original phrase to enhance fluency at the expense of strict semantic
accuracy.
The inability of AI translation tools to accurately handle idiomatic expressions has
significant implications for cross-linguistic communication and translation quality.
While neural machine translation (NMT) systems have improved in fluency and
grammatical coherence, they remain limited in semantic depth and cultural awareness,
particularly in handling figurative language. The observed errors suggest that AI-
generated translations cannot yet replace human expertise in contexts where idiomatic
precision and cultural adaptation are essential. The limitations of AI in handling
idiomatic expressions and cultural nuances highlight the ongoing need for human
expertise in translation. However, the translation market is anticipated to be
significantly impacted by AIpowered translation. Although artificial intelligence (AI)
has novel prospects for the translation industry, the principal fallout from this
developing social phenomenon is a change in the qualifications of translators or even
the possibility of job displacement. [3; 18] As AI systems continue to evolve,
translators may need to adapt by focusing on post-editing, quality assurance, and
specialized translation tasks that demand cultural and contextual sensitivity. Future
Ta'lim innovatsiyasi va integratsiyasi
46-son_1-to’plam_Iyun -2025
122
ISSN:3030-3621
developments should focus on refining AI’s ability to process idiomatic meaning,
contextual variation, and cultural adaptation, ensuring that translated content maintains
both linguistic accuracy and communicative effectiveness.
The list of used literature:
1.
Charles-Kenechi, S. (2024). Artificial Intelligence in Translation Studies: Benefits
and
Challenges.
CJ,
2(1),
5-15.
https://cascadesjournal.com/index.php/cascades/article/view/31
2.
Faraj, B. R. A. (2024). Mona Baker’s Strategies Used for Translating the Arabic
HAND
Idioms.
Theory
and
Practice
in
Language
Studies.
file:///C:/Users/Admin/Downloads/16%20(1).pdf
3.
Hoda, A. (2024). Artificial Intelligence in Translation Studies: Benefits,
Challenges,
And
Future
Directions.
ةلجم
ةءارقلا
ةفرعملاو,
24.
https://mrk.journals.ekb.eg/article_381236_541d5e49e0c63391d7a6c327ca89a5a9
.pdf
4.
Khasawneh, M. A. S., & Al-Amrat, M. G. R. (2023). Evaluating the Role of
Artificial Intelligence in Advancing Translation Studies: Insights from Experts.
Migrat. Lett. https://migrationletters.com/index.php/ml/article/view/374
5.
Kovács, G. (2016). About the Definition, Classification, and Translation Strategies
of
Idioms.
Acta
Universitatis
Sapientiae,
Philologica,
8.
https://intapi.sciendo.com/pdf/10.1515/ausp-2016-0033
