Translation Studies Activity Worksheet Using Gen AI Tools for Translating Poems
Translation Studies Activity Worksheet Using Gen AI Tools for Translating Poems
This blog is assigned by Dr. Dilip Barad as part of the Thinking Activity on Translation Studies, based on the Activity Worksheet: Using Gen AI Tools for Translating Poems. The primary aim of this task is to analyse poem translations generated by Gen AI tools such as ChatGPT and Gemini through the application of theoretical frameworks of Translation Studies. The blog evaluates these translations in terms of sound, metre, syntax, tone, grammar, and cultural connotations, while also discussing the challenges involved in poetic translation and the possible strategies for resolving them.
In one voice the corpses spoke: ‘All is well, all is grand,’
O King, in your Ram Rajya, the Ganges bears the dead of the land.
O King, your crematoriums are full, the wood-piles have run dry,
O King, our pallbearers are exhausted, no more mourners left to cry,
From door to door, the Dance of Death performs its ghastly stride,
O King, in your Ram Rajya, the Ganges bears the dead on its tide.
O King, your chimneys smolder red, they ache for a moment’s rest,
O King, our bangles shatter, and hearts break within the chest,
Watching the fires, you play the fiddle—O ‘Billa-Ranga’ brave!
O King, in your Ram Rajya, the Ganges is a flowing grave.
O King, your robes are divine, your aura glows so bright,
O King, the entire city sees you now in the true light,
If there be a man among you, stand and shout it plain:
‘The King is naked!’—let the truth wash away the stain.
O King, in your Ram Rajya, the Ganges bears the dead again.
Hindi Translation:
शववाहिनी गंगा
एक साथ सब मुर्दे बोले ‘सब कुछ चंगा-चंगा’
राज, तुम्हारे रामराज्य में शववाहिनी गंगा।
राज, तुम्हारे मसान कम पड़े, कम पड़ गई लकड़ियाँ,
राज, हमारे कांधे कम पड़े, थक गईं रोती अखियाँ,
घर-घर जाकर यम की टोली करती नाच कुढंगा,
राज, तुम्हारे रामराज्य में शववाहिनी गंगा।
राज, तुम्हारी धधकती चिमनी अब तो राहत माँगे,
राज, हमारी चूड़ी टूटे, छाती फटे अभागे,
जलता देख फिडल तुम छेड़ो, वाह रे ‘बिल्ला-रंगा’!
राज, तुम्हारे रामराज्य में शववाहिनी गंगा।
राज, तुम्हारे दिव्य वस्त्र और दिव्य तुम्हारी ज्योति,
राज, तुम्हें अब असली रूप में सारी नगरी जोती (देखती),
हो हिम्मत तो आकर बोलो ‘राजा मेरा नंगा’,
राज, तुम्हारे रामराज्य में शववाहिनी गंगा।
Poem : 4
Choose a Hindi poem and translate it into Gujarati and English.
Here is Original poem
तुम तूफ़ान समझ पाओगे? by Harivanshrai Bachchan
गीले बादल, पीले रजकण,
सूखे पत्ते, रूखे तृण घन
लेकर चलता करता ‘हरहर’—इसका गान समझ पाओगे?
तुम तूफ़ान समझ पाओगे?
गंध-भरा यह मंद पवन था,
लहराता इससे मधुवन था,
सहसा इसका टूट गया जो स्वप्न महान, समझ पाओगे?
तुम तूफ़ान समझ पाओगे?
तोड़-मरोड़ विटप-लतिकाएँ
नोच-खसोट कुसुम-कलिकाएँ,
जाता है अज्ञात दिशा को! हटो विहंगम, उड़ जाओगे!
तुम तूफ़ान समझ पाओगे?
Poem translated by Chat GPT
Gujarati Translation
શું તમે તોફાનને સમજી શકશો?
ભીંજાયેલા વાદળ, પીળા ધૂળકણ,
સૂકા પાંદડા, કર્કશ તૃણઘન
લઈને ચાલે, ‘હર-હર’ ગાજે
શું તેનો આ ગીત તમે સમજી શકશો?
શું તમે તોફાનને સમજી શકશો?
સુગંધભર્યો આ મૃદુ પવન હતો,
એથી લહેરાતું મધુવન હતું,
અચાનક તૂટી ગયું તેનો જે મહાન સ્વપ્ન
શું તમે એ પીડા સમજી શકશો?
શું તમે તોફાનને સમજી શકશો?
વળગી-વળગી વેલોને મરડતો,
કળી-કુસુમને નોંચી ખસેડતો,
અજાણી દિશામાં દોડે છે!
હટો, વિહંગમ, ઊડી જશો!
શું તમે તોફાનને સમજી શકશો?
English Translation
Will you ever understand the storm?
Drenched clouds, yellow grains of dust,
Dry leaves, coarse clusters of grass,
Carrying all, it roars “har–har”
Will you grasp this song of the storm?
Will you ever understand the storm?
Once it was a fragrant, gentle breeze,
By it the honeyed groves would sway;
Suddenly its radiant dream lay shattered
Will you understand that breaking?
Will you ever understand the storm?
Twisting and snapping vines and branches,
Ripping apart buds and blossoms,
It rushes toward an unknown direction!
Away, O birds—fly, or be swept away!
Will you ever understand the storm?
Poem translated By Gemini
English Translation
Can You Understand the Storm?
Can you understand the storm?
Wet clouds, yellow dust-grains,
Dry leaves, thickets of brittle grass,
Carrying them all, roaring ‘Har-har’
Can you understand its song?
Can you understand the storm?
Once it was a breeze filled with scent,
The honeyed-forest swayed in its path,
Suddenly, its great dream was shattered
Can you understand that?
Can you understand the storm?
Twisting and mangling branches and vines,
Tearing and plucking flowers and buds,
It heads toward an unknown horizon!
Move, little bird, or you shall be swept away!
Can you understand the storm?
Gujarati Translation
તમે તોફાન સમજી શકશો?
તમે તોફાન સમજી શકશો?
ભીનાં વાદળ,
પીળાં રજકણ,
સૂકાં પાંદડાં, રૂખાં તૃણ-ઘન,
લઈને ચાલતું કરતું ‘હરહર’—
એનું ગાન સમજી શકશો?
તમે તોફાન સમજી શકશો?
સુગંધ-ભરેલો આ મંદ પવન હતો,
લહેરાતું આનાથી મધુવન હતું,
એકાએક એનું તૂટી ગયું જે સ્વપ્ન મહાનતમે સમજી શકશો?
તમે તોફાન સમજી શકશો?
મરોડી-તોડી ડાળીઓ અને લતિકાઓ,
ઝૂંટવી-તોડી કુસુમ-કલિકાઓ,
જાય છે અજ્ઞાત દિશામાં!
હટો વિહંગમ, તમે ઊડી જશો!
તમે તોફાન સમજી શકશો?
Comparative Analysis of GenAI Tools in Poetry Translation
A Translation Studies Perspective on ChatGPT and Gemini
Introduction
This study examines translations of four poems generated by ChatGPT and Google Gemini, evaluating their performance through the theoretical frameworks of translation studies. The corpus includes Faiz Ahmed Faiz's Urdu ghazal "Mujh Se Pehli Si Mohabbat," W.B. Yeats's "The Second Coming," Parul Khakhkhar's Gujarati poem "શબવાહિની ગંગા," and Harivansh Rai Bachchan's Hindi poem "तुम तूफ़ान समझ पाओगे?" This analysis applies Jakobson's, Catford's, Devy's, and Ramanujan's theoretical approaches to assess how AI tools navigate linguistic, cultural, and aesthetic dimensions of literary translation.
Poem 1: "Mujh Se Pehli Si Mohabbat" by Faiz Ahmed Faiz
Urdu → Hindi & English
Syntax, Sound, and Metre
The translation of Faiz's ghazal presents significant challenges in preserving the original's conversational intimacy while maintaining its lyrical quality.
ChatGPT's Hindi translation maintains the Urdu-Hindi linguistic proximity through choices such as "मुझसे पहली-सी मोहब्बत, मेरी महबूब, न माँग," using hyphens to retain the flowing quality characteristic of ghazal recitation. The refrain structure is preserved consistently throughout the poem, maintaining the cyclical nature essential to the ghazal form.
Gemini's Hindi translation adopts a more pedagogical approach, adding parenthetical glosses: "मैंने समझा था कि तू है तो दरख़्शाँ (रोशन) है हयात." While this provides clarity for readers unfamiliar with Persianate vocabulary, it disrupts the immersive quality of poetic reading and breaks the rhythmic flow.
In English translation, ChatGPT opts for "Do Not Ask of Me That Former Love, My Beloved," employing formal poetic syntax with measured line breaks. Gemini's title "Love, Do Not Ask Me for That Love Again" adopts a more conversational tone. Both approaches demonstrate different interpretive strategies—ChatGPT prioritizing formal distance, Gemini emphasizing colloquial intimacy.
Lexicon, Grammar, and Semiotics
The lexical choices reveal distinct translation philosophies. Examining the line "تیرا غم ہے تو غم دہر کا جھگڑا کیا ہے":
ChatGPT: "तेरा ग़म है तो ग़म-ए-दहर से क्या काम मुझे" (What business have I with the world's grief?)
Gemini: "तेरा ग़म है तो ग़म-ए-दहर का झगड़ा क्या है?" (What does the world's quarrel matter?)
ChatGPT's rendering emphasizes personal dismissal through "क्या काम मुझे," capturing the speaker's absorption in personal grief. Gemini's more literal translation of "झगड़ा" (quarrel) stays closer to the Urdu original but potentially diminishes the philosophical weight of "ग़म-ए-दहर" (world's sorrow/existential grief).
In the crucial sociopolitical imagery section, both translations handle the visceral description of bodies:
ChatGPT (English): "Bodies smeared with dust, bathed in blood"
Gemini (English): "Smeared in dust, bathed in blood's grim trace"
Gemini's addition of "grim trace" represents an interpretive enhancement not present in the original, while ChatGPT maintains directness that better serves the poem's stark social critique.
Cultural Connotations and Collocations
The poem's shift from romantic discourse to social commentary marks Faiz's characteristic fusion of personal and political. The translation of "अनगिनत सदियों के तारीक, वहशी से तिलिस्म" (countless centuries of dark, savage enchantment) demonstrates both tools' approaches to culturally loaded terminology:
ChatGPT: "Countless ages of dark, barbaric spell"
Gemini: "अनगिनत सदियों के तारीख़ (अंधेरे) बहीमाना (वहशी) तिलिस्म"
The concept of "तिलिस्म" carries specific connotations in Urdu-Hindi literary tradition, referring to both magical enchantment and, metaphorically, systemic oppression. ChatGPT's "spell" maintains this ambiguity, while Gemini's glossing approach, though educational, may direct readers toward singular interpretations.
The cultural weight of "वस्ल की राहत" (comfort of union) represents another translation challenge:
ChatGPT: "comfort of union"
Gemini: "joy of our embrace"
"Wasl" in Urdu poetic tradition encompasses spiritual union, physical intimacy, and mystical realization. ChatGPT's "union" preserves this multivalence, while Gemini's "embrace" concretizes the metaphor, potentially limiting its interpretive range.
Challenges and Resolutions
The primary challenge lies in maintaining the dual register of Faiz's poetry—the intimate address coexisting with public conscience. Both tools demonstrate competence in handling the semantic content but vary in their preservation of tonal complexity.
The refrain "और भी दुख हैं ज़माने में मोहब्बत के सिवा / राहतें और भी हैं वस्ल की राहत के सिवा" requires balancing philosophical abstraction with emotional resonance. Both translations successfully maintain the structural repetition, though ChatGPT's version sustains slightly better rhythmic consistency across Hindi and English versions.
Poem 2: "The Second Coming" by W.B. Yeats
English → Hindi & Gujarati
Syntax, Sound, and Metre
Yeats's apocalyptic vision presents unique challenges when translating into Indian languages, particularly in maintaining the poem's building sense of dread through syntactic accumulation.
ChatGPT's approach in Hindi preserves the opening's circular motion: "घूमता है, घूमता है, फैलते चक्र में समय," where the repetition mirrors Yeats's "Turning and turning." The Gujarati translation similarly employs "વળતો ને વળતો સમય," maintaining this rhythmic emphasis.
Gemini's Hindi translation renders this as "चौड़े होते चक्रवात में घूमते और घूमते," introducing "चक्रवात" (cyclone) which makes the image more meteorologically concrete. This represents a shift from Yeats's mystical "gyre" to a more naturalistic phenomenon, potentially altering the poem's metaphysical dimension.
Both tools maintain the structural parallelism of Yeats's verse, though achieving the original's iambic variations proves difficult in languages with different prosodic systems.
Lexicon, Grammar, and Semiotics
The translation of "Spiritus Mundi" reveals different strategies for handling culturally specific terminology:
ChatGPT: Retains "Spiritus Mundi" untranslated in both Hindi and Gujarati
Gemini: Hindi - "स्पिरिटस मुंडी," Gujarati - "સ્પિરિટસ મુંડી (જગત-આત્મા)"
Gemini's parenthetical gloss "(જગત-આત્મા)" (world-soul) provides cultural translation, though this pedagogical intervention may disrupt the poem's mystical atmosphere. ChatGPT's decision to leave the Latin phrase untranslated assumes reader familiarity with Western esoteric traditions.
The central sphinx imagery—"A shape with lion body and the head of a man"—receives distinct treatments:
ChatGPT Hindi: "सिंह-देह, मानव-मस्तक वाला एक आकार"
Gemini Hindi: "सिंह जैसा शरीर और मनुष्य जैसा सिर वाला एक आकार"
Gemini's use of "जैसा" (like) introduces a simile where Yeats employs direct identification. This shift from metaphorical certainty to comparative description subtly weakens the mythological authority of the vision.
Cultural Connotations and Collocations
Translating Yeats into Indian languages necessitates bridging Christian apocalyptic imagery with Hindu-Buddhist conceptual frameworks. Both tools appropriately retain "Bethlehem" (बेथलहम/બેથલેહેમ) without translation, recognizing its specific Biblical significance.
The phrase "The ceremony of innocence is drowned" carries civilizational weight:
ChatGPT: "मासूमियत के सारे अनुष्ठान डूब गए हैं"
Gemini: "निर्दोषता का उत्सव डूब गया है"
ChatGPT's "अनुष्ठान" (ritual/ceremony) better captures Yeats's sense of lost civilizational order than Gemini's "उत्सव" (celebration). The concept of "ceremony" implies structured social practices, while "celebration" suggests more spontaneous festivity.
Challenges and Resolutions
The fundamental challenge lies in conveying Christian eschatological imagery to readers grounded in cyclical rather than linear time concepts. The "Second Coming" itself—a specifically Christian doctrine—lacks direct equivalents in Hindu or Buddhist thought.
Both tools handle "slouches towards Bethlehem to be born" with different emphases:
ChatGPT: "घिसटता चला आ रहा है"
Gemini: "रेंग रहा है"
ChatGPT's "घिसटता" (dragging/slouching) emphasizes laborious movement, while Gemini's "रेंग" (crawling) creates more visceral, reptilian imagery. Both preserve the horror of the birth-as-threat paradox central to Yeats's conclusion.
Poem 3: "શબવાહિની ગંગા" by Parul Khakhkhar
Gujarati → Hindi & English
Syntax, Sound, and Metre
Khakhkhar's poem employs ballad-like repetition for political effect. The refrain "રાજ, તમારા રામરાજ્યમાં શબવાહિની ગંગા" (O King, in your Ram Rajya flows the corpse-bearing Ganga) structures each stanza's accusatory rhythm.
ChatGPT's Hindi translation maintains rhythmic accusation through "राज, तेरे रामराज्य में बहती है शव-वाहिनी गंगा," notably using the intimate/informal "तेरे" (your-informal) rather than formal "तुम्हारे." This register choice amplifies the sardonic tone—addressing power with deliberate disrespect.
Gemini's Hindi translation employs "तुम्हारे" (your-formal), which, while grammatically appropriate for addressing authority, may diminish the poem's deliberate irreverence. In protest poetry, register choice carries significant political meaning.
Both English translations attempt to preserve the refrain structure, though varying in their line arrangements and rhythmic patterns.
Lexicon, Grammar, and Semiotics
The opening line "એક અવાજે મડદા બોલ્યાં 'સબ કુછ ચંગા-ચંગા'" presents translation choices with ideological implications:
ChatGPT Hindi: "एक स्वर बोला— 'सब कुछ चंगा-चंगा!'" (one voice spoke)
Gemini Hindi: "एक साथ सब मुर्दे बोले 'सब कुछ चंगा-चंगा'" (all corpses spoke together)
Gemini's explicit rendering "मुर्दे बोले" (corpses spoke) makes the horror immediately visible, while ChatGPT's abstract "एक स्वर" (one voice) creates a moment of ambiguity before readers realize the voice belongs to the dead. Both approaches have merit Gemini's directness versus ChatGPT's delayed revelation.
The folk reference "વાહ રે બિલ્લા-રંગા" (a Gujarati folk character representing corrupt authority) poses cultural translation challenges:
ChatGPT English: "Bravo! Bravo! What colours here!"
Gemini English: "O 'Billa-Ranga' brave!"
ChatGPT's translation focuses on "રંગા" (colors) to create an ironic exclamation, losing the cultural specificity of the folk reference entirely. Gemini retains the transliterated name, preserving the cultural marker even if unexplained. Neither provides sufficient context for readers unfamiliar with Gujarati folk traditions.
Cultural Connotations and Collocations
The poem's central metaphor "રામરાજ્યમાં શબવાહિની ગંગા" (in Ram's kingdom, a corpse-bearing Ganga)—operates on multiple cultural levels. "Ram Rajya" represents the idealized just kingdom in Hindu political philosophy, while the Ganga carries profound sacred significance. The image of this sacred river bearing corpses invokes both pandemic reality and profound civilizational crisis.
Both tools appropriately retain "रामराज्य/Ram Rajya" untranslated in English, recognizing that any translation would lose the specific Indian political resonance, especially in contemporary discourse.
The lexical choice between "मसाण/श्मशान" (cremation grounds) demonstrates different approaches to register:
ChatGPT: "श्मशान" (formal Sanskrit term)
Gemini: "मसान" (colloquial, closer to Gujarati "મસાણ")
Gemini's choice maintains the folk register appropriate to the poem's ballad-like quality, while ChatGPT's Sanskritized term elevates the diction.
Challenges and Resolutions
The primary challenge lies in translating politically charged contemporary poetry where the cultural context (India's COVID-19 pandemic and governmental response) remains essential to interpretation. Both tools successfully transfer the poem's anger and irony, though readers unfamiliar with the specific political moment may miss the full resonance.
The final stanza's allusion to "The Emperor's New Clothes" receives different emphasis:
ChatGPT English: "If courage lives, then dare to say aloud: / 'My King stands bare before the crowd!'"
Gemini English: "If there be a man among you, stand and shout it plain: / 'The King is naked!'"
Gemini's "If there be a man among you" more directly challenges collective silence, while ChatGPT's "If courage lives" personalizes the moral imperative. Both preserve the Hans Christian Andersen allusion, though Gemini's formulation more explicitly invokes communal responsibility.
Poem 4: "तुम तूफ़ान समझ पाओगे?" by Harivansh Rai Bachchan
Hindi → Gujarati & English
Syntax, Sound, and Metre
Bachchan's poem employs anaphoric questioning—each stanza building toward the refrain "तुम तूफ़ान समझ पाओगे?" (Will you understand the storm?). This rhetorical structure requires consistent handling to maintain the poem's aggressive interrogation.
ChatGPT's translations preserve the questioning refrain effectively across both Gujarati ("શું તમે તોફાનને સમજી શકશો?") and English ("Will you ever understand the storm?"). The consistency of the refrain creates the hammering effect Bachchan intends.
Gemini's Gujarati translation varies in its refrain structure, sometimes dropping "શું" in repetitions, which creates rhythmic variation but may weaken the insistent questioning central to the poem's rhetorical strategy.
Lexicon, Grammar, and Semiotics
The poem's central metaphor transforms a gentle breeze ("गंध-भरा यह मंद पवन") into destructive storm, representing betrayed love or lost innocence becoming rage.
The line "गंध-भरा यह मंद पवन था" receives distinct treatments:
ChatGPT English: "Once it was a fragrant, gentle breeze"
Gemini English: "Once it was a breeze filled with scent"
ChatGPT's comma placement creates an elegiac rhythm emphasizing "gentle," while Gemini's construction focuses on the substantive quality "filled with scent." Both convey the meaning, but ChatGPT's version demonstrates greater attention to English poetic convention.
The critical line "सहसा इसका टूट गया जो स्वप्न महान" shows interpretive divergence:
ChatGPT: "Suddenly its radiant dream lay shattered"
Gemini: "Suddenly, its great dream was shattered"
ChatGPT adds "radiant," not present in the Hindi original, representing an interpretive enhancement. This addition amplifies the sense of loss not merely a "great" dream but a "radiant" one. Such additions raise questions about fidelity versus aesthetic enhancement in translation.
Cultural Connotations and Collocations
The exclamation "हर-हर" carries specific cultural weight in Indian context. Associated with Lord Shiva and chanted during cremations, it evokes destruction and transformation. Both tools retain the transliteration:
ChatGPT: "roars 'har–har'"
Gemini: "roaring 'Har-har'"
Neither provides explicit cultural explanation in the English translation, leaving international readers to either research or bypass this reference. This represents a foreignizing strategy that maintains cultural specificity at potential cost of accessibility.
The word "विहंगम" (bird) receives different treatment:
ChatGPT: "O birds" (plural)
Gemini: "little bird" (singular, diminutive)
Gemini's diminutive "little bird" creates a more intimate, protective warning, while ChatGPT's plural suggests a general admonition. This difference affects the poem's tone Gemini's version softens the violence with tenderness, ChatGPT's maintains impersonal force.
Challenges and Resolutions
The challenge in translating Bachchan's romantic register lies in navigating Hindi poetic conventions that may seem antiquated in English. Terms like "मधुवन" (honeyed forest/grove), "कुसुम-कलिकाएँ" (flowers and buds), and "विटप-लतिकाएँ" (trees and vines) belong to specific Hindi literary tradition.
"कुसुम-कलिकाएँ": ChatGPT "buds and blossoms," Gemini "flowers and buds"
These choices reveal translation philosophy whether to prioritize English poetic register or maintain closer structural parallels to the source language.
The poem's success in translation depends on preserving the emotional arc from gentleness to violence while maintaining the interrogative mood throughout. Both tools accomplish this with varying degrees of lyrical polish.
Theoretical Framework Analysis
Jakobson's Trichotomy of Translation
Roman Jakobson's influential essay "On Linguistic Aspects of Translation" (1959) distinguishes three types of translation: intralingual (rewording within the same language), interlingual (translation proper between languages), and intersemiotic (transmutation between sign systems).
This analysis primarily concerns interlingual translation, where Jakobson's concept of "equivalence in difference" becomes central. The AI translations demonstrate competence in achieving semantic equivalence but encounter difficulties with what Jakobson terms "poetry equivalence" where form and meaning are inseparable.
Jakobson argues that "poetry by definition is untranslatable," suggesting only "creative transposition" is possible. The AI translations validate this claim both tools successfully transpose semantic content but struggle with prosodic equivalence. For instance, Faiz's ghazal in Urdu carries specific metrical patterns (beher) that cannot be mechanically transferred to English or even to Hindi without the ghazal's traditional Persian-Arabic metrical system.
The tools demonstrate awareness of intersemiotic challenges when retaining culturally specific terms like "Spiritus Mundi," "Ram Rajya," or "Har-Har." These function as complex cultural signs whose meaning exceeds linguistic definition, requiring contextual knowledge for full comprehension.
Catford's Linguistic Theory of Translation
J.C. Catford's A Linguistic Theory of Translation (1965) provides taxonomies of translation shifts changes that occur in the process of moving from source to target language. Both AI tools demonstrate several types of Catford's shifts:
Level Shifts occur when source language grammar is translated by target language lexis, or vice versa. For example, Hindi's verbal aspects (continuous, perfective, habitual) often shift to English adjectives or adverbs. In Bachchan's poem, "धधकती चिमनी" (blazing-continuous chimney) becomes "chimneys that smolder" in English a shift from participial adjective to relative clause.
Category Shifts involve changes in grammatical categories. Gujarati's agglutinative structures frequently require prepositional phrases in English. Khakhkhar's "શબવાહિની" (corpse-bearing-adjective) becomes "that bears corpses" (relative clause) in English translation.
Class Shifts occur when translation equivalent belongs to different grammatical class than source item. Both tools navigate these necessarily, as linguistic structures differ fundamentally across language families.
Unit Shifts involve translating at different rank levels morpheme to word, word to phrase, etc. The AI tools demonstrate competence in recognizing when such shifts are necessary for natural expression in the target language.
Catford's framework helps explain why some translations feel more "literal" (fewer shifts) while others read more naturally (appropriate shifts). Gemini tends toward fewer category shifts, maintaining closer structural parallels to source texts, while ChatGPT allows more shifts for target-language fluency.
Ganesh Devy's Translation as Cultural Practice
Ganesh Devy's "Translation and Literary History: An Indian View" challenges Western translation theory's universalist assumptions, arguing that translation within the Indian multilingual context operates differently from translation between European languages or between European and non-European languages.
Devy emphasizes the shared cultural matrix of Indian languages, suggesting that translation between Indian languages represents a qualitatively different practice than translation across civilizational boundaries. The present analysis provides evidence supporting this thesis:
Gujarati → Hindi translations (Khakhkhar's poem) demonstrate smoother lexical and cultural transfer than English → Hindi/Gujarati (Yeats's poem)
Urdu → Hindi translation (Faiz's poem) shows both tools' awareness of the Hindustani continuum they retain Persianate vocabulary rather than over-Sanskritizing
The English → Indian languages direction encounters greater difficulties with cultural-symbolic content (Christian apocalyptic imagery in Yeats) than do translations within the South Asian linguistic sphere
Both AI tools demonstrate respect for what Devy calls the "Indian tradition of anuvad" they don't unnecessarily Sanskritize Hindi translations or eliminate Persianate vocabulary. For instance, both retain "ग़म-ए-दहर," "वस्ल," and "तिलिस्म" rather than substituting pure Sanskrit equivalents. This suggests their training data reflects actual linguistic practices within India's multilingual ecology.
Devy's argument about translation as cultural continuity rather than transfer between discrete systems finds validation in the AI tools' better performance with intra-Indian translation pairs.
A.K. Ramanujan's "Flowering of Context"
A.K. Ramanujan's essay "On Translating a Tamil Poem" argues that poetry translation must account for what he calls "context-sensitivity" the way meaning in Indian poetry emerges from complex intertextual and cultural contexts rather than from words alone.
Ramanujan identifies several types of context crucial to poetry translation: literary (allusions to other texts), cultural (shared assumptions and practices), linguistic (grammatical and lexical structures), and situational (historical and social circumstances).
Where the AI tools succeed in Ramanujan's terms:
Maintaining imagery clusters: Both tools recognize that metaphors don't translate in isolation but as part of larger symbolic systems
Preserving tonal consistency: Rather than word-by-word equivalence, both tools attempt to maintain overall emotional and aesthetic tone
Retaining culture-specific terms: Following Ramanujan's preference for annotation over domestication, both tools often retain transliterated terms rather than forcing Western equivalents
Where the AI tools fail Ramanujan's test:
"Billa-Ranga" in Khakhkhar's poem: Neither tool provides the folk-cultural context necessary for full comprehension
Historical situatedness: The tools lack awareness of composition circumstances Faiz's progressive politics, Khakhkhar's pandemic context, Yeats's post-WWI apocalypticism
Intertextual awareness: References and allusions go unrecognized and unexplained
Ramanujan emphasizes that "a poem does not mean in isolation" it carries its contexts with it. AI tools can handle linguistic transfer but struggle with cultural embedding. They can translate what is said but not always why it's said that way, or what it resonates with in the source culture's literary memory.
Synthesis: AI Translation and Theoretical Frameworks
Examining these translations through multiple theoretical frameworks reveals a consistent pattern: AI tools demonstrate impressive linguistic competence but limited cultural hermeneutics. They can:
Navigate complex grammatical structures (Catford)
Recognize when semantic equivalence requires creative transposition (Jakobson)
Respect linguistic ecology in multilingual contexts (Devy)
Preserve surface-level cultural markers (Ramanujan)
However, they cannot:
Make interpretive choices requiring deep historical knowledge
Explain translation decisions and their implications
Recognize and annotate intertextual references
Understand political subtexts beyond explicit content
Capture the full "context-sensitivity" Ramanujan describes
This aligns with what translation theorists have long argued: translation is not merely linguistic but profoundly cultural, historical, and political. AI's limitations paradoxically validate humanistic translation studies' core insights.
AI Translation in Literary Context
Linguistic Competence vs. Cultural Knowledge
The analysis reveals a fundamental distinction between linguistic transfer and cultural interpretation. Both ChatGPT and Gemini demonstrate sophisticated understanding of grammatical structures, semantic fields, and even stylistic registers. They can identify and preserve key metaphors, maintain structural parallelism, and adapt syntax appropriately for target languages.
However, cultural knowledge proves more elusive. When Khakhkhar references "Billa-Ranga," the tools can transliterate but cannot provide the folk-cultural context that gives the reference its full satirical power. When Yeats invokes "Spiritus Mundi," the tools can retain or gloss the term but cannot explain its significance within Yeats's occult philosophy or how it functions in the poem's apocalyptic vision.
This limitation becomes most apparent in politically charged texts. Khakhkhar's poem operates within specific discourse about India's pandemic response and governmental authority. Faiz's ghazal emerges from progressive literary movements in mid-20th century South Asia. These contexts shape every word choice, every image, every tonal shift yet they remain largely inaccessible to AI translation tools.
The Question of Domestication vs. Foreignization
Lawrence Venuti's influential work The Translator's Invisibility (1995) frames a central debate in translation studies: should translations make foreign texts accessible by conforming to target-culture norms (domestication), or should they preserve foreign elements to remind readers of the text's alterity (foreignization)?
The AI tools demonstrate interesting patterns in this regard. Both tend toward foreignization when handling culturally specific terms retaining "Ram Rajya," "Spiritus Mundi," "Har-Har" rather than substituting approximate Western equivalents. This represents a sound strategy for literary translation, where cultural specificity often carries aesthetic and thematic weight.
However, this foreignization occurs somewhat mechanically—the tools retain cultural markers without necessarily providing context for interpretation. A human translator might include footnotes, introductions, or subtle textual cues to guide readers. The AI tools simply preserve the foreign elements, assuming reader competence or accepting reader confusion.
Translation as Interpretation
Eugene Nida's distinction between formal equivalence (maintaining source-text structure) and dynamic equivalence (achieving equivalent effect on target readers) illuminates the different approaches visible in these translations.
Gemini tends toward formal equivalence, maintaining closer structural parallels to source texts and providing glosses for cultural terms. This approach privileges accuracy and educational value, making it potentially more useful for language learners or scholarly study.
ChatGPT leans toward dynamic equivalence, taking interpretive liberties to achieve smoother target-language expression. Adding "radiant" to "महान स्वप्न" (great dream) in Bachchan's poem represents such an interpretive choice—not strictly necessary for semantic accuracy but potentially enhancing aesthetic effect.
Neither approach is inherently superior; they serve different purposes and audiences. The significant point is that both involve interpretation—choices about what to prioritize, what to sacrifice, how to balance competing values. Translation theory has long recognized this interpretive dimension; AI translation demonstrates it through the variation between different tools' outputs.
Intertextuality and Literary Tradition
Both tools struggle with what Julia Kristeva and later theorists term intertextuality the way texts reference, echo, and respond to other texts. When Khakhkhar alludes to "The Emperor's New Clothes," when Yeats draws on Shelley and Blake, when Faiz engages with both Persian ghazal tradition and socialist realism—these intertextual dimensions shape meaning profoundly.
Human translators with deep literary knowledge recognize these echoes and can make choices that preserve intertextual resonances. AI tools, lacking such comprehensive literary memory, translate the surface text without awareness of its dialogue with literary history.
This limitation matters especially for poetry, where intertextual play often operates at subtle levels a phrase that echoes earlier usage, a metaphor that reverses traditional associations, a rhythm that invokes specific formal traditions.
Performance Patterns Across Language Pairs
The analysis reveals interesting patterns regarding translation directionality:
Within South Asian linguistic sphere (Urdu-Hindi, Gujarati-Hindi, Hindi-Gujarati): Both tools perform well, suggesting robust training data and genuine linguistic competence
From English to Indian languages: More difficulties emerge, particularly with culture-specific content (Yeats's Christian symbolism)
From Indian languages to English: Generally successful at semantic level, but cultural context often requires more annotation than the tools provide
This pattern supports Ganesh Devy's argument about qualitative differences in translation within vs. across civilizational contexts. It also suggests that AI translation tools have been trained more extensively on certain language pairs and cultural contexts than others.
The Question of Authorship and Agency
An interesting theoretical question emerges: who is the "translator" when AI tools produce these translations? The prompt-giver? The tool itself? The training data's original human translators?
This question matters because translation theory emphasizes translator agency the choices, judgments, and interpretive decisions that shape translation. Traditional translation criticism analyzes translator strategies, ideological commitments, and creative solutions. With AI translation, agency becomes distributed and obscured.
This has implications for translation ethics as well. If problematic translations emerge (culturally insensitive choices, political misrepresentations), who bears responsibility? The AI company? The user? The aggregated training data?
These questions extend beyond the scope of this analysis but merit serious consideration as AI translation becomes more prevalent.
Comparative Observations on ChatGPT and Gemini
Distinct Translation Philosophies
Through systematic comparison, two distinct approaches emerge:
ChatGPT demonstrates what might be termed an aesthetic-interpretive approach:
Prioritizes target-language fluency and natural expression
More willing to take interpretive liberties for poetic effect
Maintains consistent tone across entire poems
Produces translations that read smoothly as English or Hindi/Gujarati poems
Sometimes sacrifices literal accuracy for aesthetic coherence
Gemini adopts a pedagogical-literal approach:
Includes parenthetical glosses for culturally specific terms
Maintains closer structural parallels to source texts
Provides more literal lexical choices
Potentially more useful for language learning or comparative study
Sometimes disrupts poetic flow with educational interventions
Neither approach is definitively superior; they represent different valid translation philosophies serving different purposes and audiences.
Handling of Cultural Specificity
Both tools demonstrate awareness that certain cultural terms resist translation and should be retained. However, they differ in their support for reader comprehension:
Gemini's glossing strategy (e.g., "दरख़्शाँ (रोशन)") provides immediate clarity but at cost of immersive reading experience. This resembles academic editions with extensive footnotes—valuable for study but potentially intrusive for aesthetic appreciation.
ChatGPT's retention without glossing assumes reader cultural literacy or accepts interpretive gaps. This resembles literary editions that preserve foreign elements expecting readers to research independently or appreciate sound and context even without complete comprehension.
Register Sensitivity
The tools show varying awareness of linguistic register—the social dimension of language choice:
In Khakhkhar's poem, ChatGPT uses "तेरे" (informal/intimate) while Gemini uses "तुम्हारे" (formal) when addressing the King. This seemingly minor difference carries significant political weight—the informal address represents deliberate disrespect toward authority, amplifying the poem's satirical edge.
In Faiz's translation, both tools appropriately maintain the Hindustani register, preserving Persianate vocabulary rather than Sanskritizing. This demonstrates sensitivity to the poem's linguistic ecology and literary tradition.
Rhythmic and Prosodic Considerations
Poetry's musicality presents particular challenges for AI translation. While both tools attempt to preserve rhythmic patterns, neither fully succeeds in transferring prosodic qualities:
Faiz's ghazal metre (beher) cannot be mechanically transferred to English
Bachchan's Hindi rhythms don't map directly onto English stress patterns
Both tools focus primarily on semantic and structural equivalence, with prosodic considerations receiving secondary attention. This aligns with the practical difficulty of achieving metrical equivalence across linguistically different systems but represents a limitation for poetry translation where sound and sense are inseparable.
Implications for Translation Studies and Practice
AI as Translation Tool vs. Translator
The analysis suggests AI functions best as a translation tool rather than autonomous translator. Both ChatGPT and Gemini produce translations that are linguistically competent and semantically accurate, providing excellent starting points for human translators. However, they lack the cultural depth, historical awareness, and interpretive sophistication for final literary translation.
A productive model might position AI tools as draft generators in a multi-stage translation process:
AI produces initial translation capturing semantic content
Human translator refines for cultural nuance, political subtext, and aesthetic quality
Collaborative revision addresses intertextual dimensions and contextual embedding
Final editing ensures target-language literary quality
This hybrid approach leverages AI's linguistic processing capabilities while preserving human interpretive judgment.
Implications for Translation Pedagogy
For translation studies education, AI tools offer valuable pedagogical resources:
Comparative Analysis: Students can examine multiple AI translations alongside human translations, identifying strengths and limitations across different approaches
Error Analysis: AI mistakes and awkward renderings provide teaching moments about cultural context, linguistic nuance, and translation choices
Draft Generation: Students can use AI for initial drafts, then practice refinement and cultural enhancement learning translation through revision
Theoretical Application: As this analysis demonstrates, AI translations provide concrete texts for applying translation theory, examining how Jakobson's, Catford's, Devy's, and Ramanujan's insights illuminate actual translation challenges
The worksheet guiding this analysis represents precisely this pedagogical approach using AI translations as objects of critical study rather than final products.
Conclusion
This comparative analysis of ChatGPT and Gemini translations across four poems from multiple languages reveals both the impressive capabilities and significant limitations of current AI translation tools.
Capabilities demonstrated:
Sophisticated grammatical competence across multiple language pairs
Awareness of stylistic register and tonal consistency
Recognition of when cultural terms should be retained rather than translated
Ability to preserve structural patterns and refrain structures
Sensitivity to linguistic ecology in multilingual contexts (as per Devy)
Limitations identified:
Insufficient cultural hermeneutics and contextual knowledge
Inability to recognize and annotate intertextual references
Lack of awareness regarding historical and political contexts of composition
Mechanical rather than interpretive approach to cultural symbols
Difficulty capturing prosodic and musical dimensions of poetry
No capacity for explaining translation choices and rationale
The theoretical frameworks of Jakobson, Catford, Devy, and Ramanujan illuminate these findings. Jakobson's claim that poetry is "untranslatable" finds partial validation AI tools can transpose semantic content but struggle with formal equivalence. Catford's taxonomy of shifts helps explain variation between tools' outputs. Devy's emphasis on cultural context reveals AI's stronger performance within shared linguistic ecologies. Ramanujan's "context-sensitivity" identifies precisely what AI translation lacks the deep cultural embedding that shapes poetic meaning.
ChatGPT and Gemini demonstrate distinct translation philosophies aesthetic-interpretive versus pedagogical-literal neither inherently superior but serving different purposes. ChatGPT produces more fluent, readable translations at potential cost of literal accuracy. Gemini maintains closer fidelity to source texts while providing educational glosses that may disrupt poetic immersion.
Most significantly, this analysis validates the core insights of translation studies: translation is not merely linguistic transfer but cultural interpretation requiring historical knowledge, political awareness, and aesthetic judgment. AI's limitations in poetry translation paradoxically affirm what human translators have always known that translation is profoundly humanistic work demanding cultural expertise and interpretive sophistication.
The future of literary translation likely involves productive collaboration between AI tools and human translators machines for linguistic heavy lifting, humans for cultural wisdom and aesthetic refinement. As AI translation technology develops, maintaining critical awareness of both capabilities and limitations remains essential.
This study demonstrates that GenAI tools like ChatGPT and Gemini have achieved remarkable linguistic competence, making them valuable resources for translation practice and pedagogy. However, they remain tools requiring human guidance, judgment, and enhancement not autonomous translators replacing human expertise. For literary translation in particular, where cultural context, historical awareness, and aesthetic sensibility are paramount, human translators remain indispensable.
The integration of translation theory with practical AI translation analysis as modeled in this study provides a productive framework for understanding both the possibilities and limitations of AI in literary translation, contributing to ongoing discussions about technology's role in humanistic disciplines.
References
Bar Dilip. “Translation Studies Activity Worksheet Using Gen AI Tools for Translating Poems.” Research Gate, Jan. 2024, doi:10.13140/rg.2.2.32287.27041.
Barad, Dilip. Guidelines for Using Generative AI in Translation Studies: Theory and Practice. ResearchGate, Dec. 2024, doi:10.13140/RG.2.2.29351.25766.
Catford, J. C. A Linguistic Theory of Translation. Oxford UP, 1965.
Devy, Ganesh. “Translation and Literary History: An Indian View.” Postcolonial Translation: Theory and Practice, edited by Susan Bassnett and Harish Trivedi, Taylor & Francis, 1999, pp. 182–188.
Jakobson, Roman. “On Linguistic Aspects of Translation.” On Translation, edited by R. A. Brower, Harvard UP, 1959, pp. 232–239.
Nida, Eugene A. Toward a Science of Translating: With Special Reference to Principles and Procedures Involved in Bible Translating. Brill, 1964. Internet Archive, archive.org/details/towardscienceoft0000nida.
Ramanujan, A. K. “On Translating a Tamil Poem.” The Collected Essays of A. K. Ramanujan, edited by Vinay Dharwadker, Oxford UP, 2004, pp. 131–160.
Venuti, Lawrence. The Translator’s Invisibility: A History of Translation. Routledge, 2017.
Yeats, W. B. “The Second Coming.” The Collected Poems of W. B. Yeats, Scribner, 1996.
No comments:
Post a Comment