Translation

In short

Translation is the transfer of meaning from a source language (SL) into a target language (TL) so that a reader of the TL gets what the SL reader got — in sense, tone, register and effect. In professional-writing papers it is studied both as theory (equivalence, types, process, untranslatability) and as practice (literary, legal, technical, media translation, and now post-editing of machine output). It matters in India because the Constitution's Article 343 and the Official Languages Act, 1963 make Hindi-English bilingual output compulsory in Union government work, and because publishing, law, e-governance, OTT subtitling and IT localisation employ translators in 22 scheduled languages. It is examined in BA English professional-writing and translation-studies papers, in UGC-NET English (translation studies unit), and tested in placement rounds at publishers, LSPs (language service providers), subtitling houses and Bhashini-linked AI-language projects. Graduates work as translators, subtitlers, localisation QA testers, content localisers, terminologists and post-editors.

Explained

What translation is, and its basic vocabulary

• Source language (SL) / source text (ST): the language and document being translated from; target language (TL) / target text (TT): what you produce.
• Translation vs transliteration: translation carries *meaning* ("जल" → "water"); transliteration carries *script/sound only* ("जल" → "jal"). Names, brands and addresses are usually transliterated, not translated.
• Translation vs interpreting: translation is written and revisable; interpreting is oral and real-time (simultaneous, consecutive, whispered/chuchotage, liaison).
• Translation vs transcreation: transcreation recreates the message for a new market, allowed to change images, idioms and even the storyline of an advertisement, keeping only the intent.
• Translation unit: the smallest stretch you actually translate as a whole — word, phrase, clause, sentence or whole paragraph. Good practice: the sentence, checked against the paragraph.
• Directionality: translating *into* your dominant language (L1) is standard professional practice; translating out of it ("inverse translation") is common in India and needs a native-speaker reviser.
• Jakobson's three kinds: *intralingual* (rewording within a language — paraphrase, simplification), *interlingual* (true translation between languages), *intersemiotic* (language into another sign system — a text into a film, a chart, or Indian Sign Language).

Types and classifications

1. By degree of freedom (classic scale): word-for-word (interlinear) → literal → faithful → semantic → communicative → idiomatic → free → adaptation. (Newmark's semantic vs communicative, and Nida's formal vs dynamic equivalence, sit on this scale.)
2. Literal translation: keeps SL structure as far as the TL allows; safest for legal, statutory and technical text where wording carries liability.
3. Free (sense-for-sense) translation: keeps content, re-writes form; standard for journalism, fiction dialogue, marketing.
4. Equivalent-effect / dynamic translation: aims at the same reaction in the TL reader, even if the words differ — used in advertising, public-health messaging, scripture.
5. By subject: literary (poetry, fiction, drama), legal and official, technical and scientific, commercial/administrative, media (subtitling, dubbing, voice-over), audiovisual accessibility (SDH captions, audio description).
6. By agent: human translation (HT), machine translation (MT), machine translation + post-editing (MTPE), computer-aided (CAT) human translation.
7. Special Indian category: back translation — a second translator renders the TT back into the SL, blind, to verify accuracy. Mandatory in clinical-trial and ethics documents for Indian trial sites (ICMR National Ethical Guidelines require participant information sheets and consent forms in a language the participant understands, with certified translation and back-translation on record), and used for survey questionnaires.

Theory the syllabus asks for

• Nida — formal vs dynamic equivalence: formal equivalence matches form and content closely; dynamic (later "functional") equivalence matches reader response. Nida's three-stage model: *analysis → transfer → restructuring*.
• Catford — textual equivalence and shifts: *level shift* (SL grammar → TL lexis, e.g. a Sanskrit/Hindi aspect marker rendered by an English adverb) and *category shifts* (structure, class, unit, intra-system). Catford's "translation is substitution of TL material, not transference of meaning" is a standard quotation.
• Vinay and Darbelnet — seven procedures: *direct* — borrowing (SL word kept: *dharma*, *satyagraha*), calque (literal loan of a phrase: "Mother Tongue" ← *mātṛbhāṣā*), literal translation; *oblique* — transposition (change of word class), modulation (change of viewpoint), equivalence (idiom ↔ idiom), adaptation (cultural substitution).
• Newmark: semantic translation (author-centred, keeps SL flavour) vs communicative translation (reader-centred, smooth TL). Also his "translation loss" and the gain/compensation idea.
• Skopos theory (Vermeer, Reiss): the *purpose* of the TT decides the method — the same ST becomes a summary, a gist or a full translation depending on the brief. This is the theory a client brief actually follows.
• Polysystem / cultural turn: Even-Zohar, and Lefevere's "translation as rewriting"; Venuti's domestication vs foreignisation — whether you smooth a text into the TL culture or keep its foreignness visible.
• Indian thinking: translation as *anuvad* (speaking after), *rupantar* (change of form), *bhashantar* (change of language), *chhaya* (shadow/free rendering); A. K. Ramanujan's distinction between literal and "transposed" translation in his Tamil and Kannada renderings; Sujit Mukherjee's argument that Indian practice has always treated retelling as legitimate translation.
• Untranslatability (Catford): *linguistic* (puns, rhyme, grammatical gender, honorifics such as *aap/tum/tu*) and *cultural* (kinship terms like *mama/chacha/mausi*, caste and ritual terms, foods, festivals, *muhurat*, *jugaad*).

The professional process, step by step

1. Read the brief: language pair, purpose, audience, register, deadline, deliverable format, word count, reference material, whether MT is permitted.
2. Read the whole ST first and mark problems: terms, names, numbers, ambiguities, culture-bound items, tone.
3. Build a glossary / termbase before translating; freeze decisions on names, job titles, product names, statute names, units.
4. Draft at sentence level, keeping to the translation unit; leave flagged gaps rather than guessing.
5. Resolve queries with the client in one consolidated query sheet, not one message at a time.
6. Self-revise bilingually (TT against ST: accuracy, omissions, numbers, negation) then monolingually (does the TT read as if written in the TL?).
7. Revision by a second linguist (TEP = translation, editing, proofreading); then QA checks — tags, numbers, terminology, spelling, consistency.
8. DTP / layout check for text expansion: English → Hindi/Malayalam typically expands 15-30% in characters; Indic conjuncts need larger line height.
9. Deliver with a handover note: glossary, query log, assumptions, untranslated items.
10. Certification where required: a translator's declaration of accuracy; for court, visa and university use, affidavit/notarised translation, and for documents going abroad, apostille of the original under the Hague Apostille Convention (MEA handles apostille in India).
11. Quality metrics: error-typology scoring (accuracy, fluency, terminology, style, locale conventions) with severity weights — the MQM/DQF family; ISO 17100 is the international standard for translation services (requires a reviser other than the translator and defines competences). India's BIS does not publish a competing translation-service standard; where a tender quotes a standard it is normally ISO 17100.

Problem areas and how to handle them

• Culture-bound items (CBI/realia): options are borrowing with a gloss (*lakh* — "one hundred thousand"), cultural substitute, generalisation, description, omission as a last resort. Keep one strategy consistent through a text.
• Names and titles: transliterate personal names; translate descriptive titles (*Gram Panchayat* usually kept, *Tehsildar* kept with a gloss); never "translate" a registered company or scheme name.
• Numbers, units, money: convert the format, not the value — Indian digit grouping (1,25,000) vs international (125,000); crore/lakh → million/billion for export documents; keep SI units and add conversions in brackets if the ST used non-SI.
• Honorifics and pronouns: English "you" collapses *aap/tum/tu*; decide the relationship from context and stay consistent; Indic verb morphology forces a politeness choice every sentence.
• Idioms and proverbs: replace with a TL idiom of equal force, not a literal image ("अंधेर नगरी चौपट राजा" ≠ a literal "dark city").
• Poetry: decide in advance what you will keep — metre, rhyme, image, line count — because you cannot keep all; most Indian literary translation today keeps image and rhythm and drops rhyme.
• Legal text: keep sentence boundaries and defined terms exactly; translate a defined term the same way everywhere; never improve the drafting.
• Subtitling constraints: maximum two lines, roughly 35-42 characters per line, reading speed about 15-17 characters per second for adults, minimum duration about 1 second, maximum about 6-7 seconds; condense rather than truncate. Exact limits are set by the client's style guide (Netflix, Prime Video and Disney+ Hotstar each publish their own), so quote the brief, not a universal rule.

Machine translation, CAT tools and the Indian ecosystem

• CAT tool ≠ MT. A CAT tool (Trados Studio, memoQ, Phrase, free/open OmegaT, Matecat) splits the ST into segments and stores your own approved translations in a translation memory (TM); it does not generate text.
• TM metrics: exact match (100%), fuzzy match (usually counted in bands 95-99%, 85-94%, 75-84%), context/101% match, repetition. Agencies pay a discounted rate on matched words — the "CAT grid".
• MT generations: rule-based (RBMT) → statistical (SMT) → neural MT (NMT) → large language models. NMT and LLMs are fluent but *hallucinate*, drop negations, and invent terminology — which is why post-editing exists.
• Post-editing levels: *light* (make it accurate and understandable; ignore style) and *full* (publishable quality, matching human translation). ISO 18587 is the standard for full post-editing of MT output.
• Indian infrastructure: Bhashini (National Language Translation Mission, MeitY) offers translation, ASR and TTS APIs for Indian languages and powers the Anuvaad/Anuvadini-type tools; AI4Bharat (IIT Madras) released the IndicTrans and IndicTrans2 models and the Samanantar parallel corpus; the Supreme Court's SUVAS (Supreme Court Vidhik Anuvaad Software) translates judgments into Indian languages, and translated judgments are published on the eSCR portal. Coverage and accuracy vary by language pair — Hindi, Bengali, Tamil, Telugu, Malayalam, Marathi are far better served than, say, Bodo, Santali, Dogri or Manipuri.
• Institutions and recognition: Sahitya Akademi (translation prizes in each scheduled language, and its *Indian Literature* journal), National Translation Mission at CIIL Mysuru (textbook translation into scheduled languages), Central Translation Bureau and the Department of Official Language under MHA (Hindi-English official translation, Rajbhasha), National Book Trust, Kendriya Hindi Sansthan. The JCB Prize for Literature (which carried a separate translator award) was discontinued after 2024; the JLF-sponsored and international prizes continue, so check the current year before quoting a prize.
• Legal frame to know: a translation is a derivative work under the Copyright Act, 1957 — you need the rights holder's permission to translate and publish, and the translator holds copyright in the translation itself; Section 57 moral rights apply. Government documents in India follow the Official Languages Act, 1963 and the Official Languages Rules, 1976 (Regions A/B/C determine the language of correspondence); Article 348 requires authoritative texts of Acts, Bills and High Court/Supreme Court proceedings in English, and under Article 348(2) the Governor of a State may, with the previous consent of the President, authorise Hindi or the State's official language in that State's High Court proceedings (not in judgments/orders unless an English translation is issued). GST applies to translation services at the standard rate (18% in the current 5%/18%/40% structure effective 22 September 2025); freelancers cross the GST registration threshold for services at ₹20 lakh turnover (₹10 lakh in some special-category states).

Key points

• Translation - transfer of meaning from SL to TL preserving sense, tone and effect, not words.
• Transliteration - script/sound conversion only; used for names, brands, addresses.
• Interpreting - oral, real-time; sub-types simultaneous, consecutive, liaison, whispered.
• Transcreation - free recreation for a market; keeps intent, may change everything else.
• Jakobson's types - intralingual (rewording), interlingual (languages), intersemiotic (sign systems).
• Nida - formal vs dynamic equivalence; process = analysis, transfer, restructuring.
• Catford - textual equivalence, level and category shifts, linguistic vs cultural untranslatability.
• Vinay and Darbelnet - borrowing, calque, literal (direct); transposition, modulation, equivalence, adaptation (oblique).
• Newmark - semantic (author-centred) vs communicative (reader-centred) translation.
• Skopos - the purpose in the brief decides the method, not the source text's form.
• Domestication vs foreignisation - Venuti: smooth into TL culture vs keep the foreignness visible.
• Indian terms - anuvad, rupantar, bhashantar, chhaya; translation as legitimate retelling.
• Back translation - blind re-translation into the SL to verify accuracy; standard in clinical-trial and survey work.
• TEP - translation, editing, proofreading; ISO 17100 requires a second-person revision.
• TM and fuzzy match - reuse of your own approved segments, priced on a discount grid.
• MTPE - machine translation plus light or full post-editing; ISO 18587 covers full post-editing.
• Bhashini / AI4Bharat / SUVAS - India's public MT stack for scheduled languages and court judgments.
• Copyright - a translation is a derivative work under the Copyright Act, 1957; permission needed, and the translator owns the translation.
• Text expansion - English into Hindi or Malayalam usually grows 15-30%; plan layout for it.

Formulas and examples

Example 1 — Weighted word count and fee. A 10,000-word English manual is pre-analysed in a CAT tool against the client's TM: 4,000 new words, 1,500 in the 75-84% band, 2,000 in 85-94%, 1,000 in 95-99%, 1,000 exact (100%), 500 repetitions. Rate ₹2.00 per effective word, grid factors as above.

• New: 4,000 × 1.00 = 4,000
• 75-84%: 1,500 × 0.80 = 1,200
• 85-94%: 2,000 × 0.60 = 1,200
• 95-99%: 1,000 × 0.30 = 300
• 100%: 1,000 × 0.25 = 250
• Repetitions: 500 × 0.25 = 125
• Effective words = 4,000 + 1,200 + 1,200 + 300 + 250 + 125 = 7,075
• Fee = 7,075 × ₹2.00 = ₹14,150 (against ₹20,000 at raw count — a 29.25% saving for the client).
• GST at 18% (if registered) = ₹2,547 → invoice ₹16,697.

Example 2 — Schedule planning. The same job, human translation at 350 words/hour with 6 productive hours/day, and revision at 1,000 words/hour.

• Translation hours = 10,000 ÷ 350 ≈ 28.6 h → 28.6 ÷ 6 ≈ 4.8 days.
• Revision hours = 10,000 ÷ 1,000 = 10 h → 10 ÷ 6 ≈ 1.7 days.
• QA and DTP, say 0.5 day. Total ≈ 7 days, so a 5-day deadline needs two translators and one reviser, with a shared glossary frozen on day 1.

Example 3 — Subtitle condensation. A line of dialogue is on screen for 3.2 seconds. Its full translation runs to 71 characters.

• Allowed characters at 17 CPS = 17 × 3.2 = 54.4 → 54 characters.
• 71 > 54, so cut 17 characters (about 24%): drop the vocative and the filler, convert the passive to active, and use the shorter synonym. Two lines are permitted (2 × 42 = 84) but the limit here is the *reading speed*, not the line length — a common exam trap.
• If the scene allowed 4.5 s, capacity = 17 × 4.5 ≈ 76 characters, and the full 71-character line would fit on two lines.

Example 4 — Choosing the method from the brief (Skopos in practice). Same source sentence, three briefs: "Tea is life for the hills."

Example 5 — Procedure identification (Vinay and Darbelnet), the standard exam question.

Where it is used

• Publishing and literary translation: Penguin Random House India, HarperCollins India, Seagull, Oxford University Press, Sahitya Akademi and National Book Trust commission translations between English and scheduled languages; roles — translator, translation editor, rights assistant.
• Government and judiciary: Central Translation Bureau and every ministry's Rajbhasha/Hindi cell under the Official Languages Act, 1963; the Supreme Court's SUVAS-assisted judgment translation; Parliament and legislature translation branches; railway, bank and PSU Rajbhasha Adhikari posts (a standard recruitment line for BA English/Hindi graduates).
• Media and OTT: subtitling, dubbing scripts and SDH captions for Netflix, Prime Video, Disney+ Hotstar, ZEE5, SonyLIV through vendors such as Pixelogic, Deluxe, VistaIndia-type studios; roles — subtitler, QC checker, dubbing script adaptor.
• IT and localisation (LSPs and in-house): Lionbridge, RWS, TransPerfect, Crystal Hues, Braahmam, Mother Tongue, Lingual Consultancy, plus product-localisation teams at Google, Microsoft, Amazon and Flipkart for 10-22 Indian languages; roles — localisation specialist, linguistic QA tester, terminologist, MT post-editor.
• Pharma, CRO and health communication: translated participant information sheets and consent forms with certified back translation for Indian trial sites under the ICMR ethical guidelines and CDSCO/New Drugs and Clinical Trials Rules, 2019; patient-reported outcome instruments.
• Legal, financial and immigration documentation: certified and notarised translation of contracts, deeds, FIRs, degree certificates and annual reports; MEA apostille work for documents going abroad.
• AI and language-data industry: parallel-corpus creation, MT evaluation and annotation for Bhashini and AI4Bharat-linked projects — a fast-growing entry route for language graduates.

Interview questions

• Q: What is the difference between translation and transliteration? A: Translation transfers meaning into the target language; transliteration only converts script or sound, so "Ganga" stays "Ganga" rather than becoming a word meaning "river".
• Q: Formal versus dynamic equivalence — when would you choose each? A: Formal equivalence (close to source form) for statutes, contracts and scripture quoted as text; dynamic equivalence (same effect on the reader) for advertising, journalism and public-health messaging.
• Q: What does Skopos theory add that earlier theories did not? A: It makes the purpose stated in the client's brief the deciding factor, so the same source text can legitimately become a gist, a summary or a full translation.
• Q: Name the seven procedures of Vinay and Darbelnet with one example each. A: Borrowing (*dharma*), calque (mother tongue), literal, transposition (adjective to adverb), modulation (not difficult → easy), equivalence (idiom for idiom), adaptation (cricket replaced by a local sport).
• Q: How do you handle an untranslatable cultural term like *jugaad*? A: Keep the word and gloss it on first use, or use a descriptive phrase; choose one strategy and apply it consistently, and record it in the glossary.
• Q: What is back translation and why do clients insist on it? A: A second translator independently renders the translation back into the source language, blind to the original, so the client can detect meaning shifts — standard for clinical-trial consent forms and survey questionnaires.
• Q: Difference between a CAT tool and machine translation? A: A CAT tool stores and reuses your own approved translations as a translation memory and enforces terminology; machine translation generates new text itself. CAT is reuse, MT is generation.
• Q: What is light versus full post-editing? A: Light post-editing fixes only accuracy and comprehensibility; full post-editing brings the output to publishable, human-equivalent quality — the scope covered by ISO 18587.
• Q: A client sends a 5,000-word legal contract due in 24 hours, MT allowed. How do you proceed? A: Pre-translate with MT inside a CAT tool, lock defined terms with a termbase, do full post-editing on clauses and definitions first, run number and terminology QA, and flag anything I could not verify in a query sheet rather than guessing.
• Q: Who owns the copyright in a translation in India? A: The translation is a derivative work under the Copyright Act, 1957 — you need the rights holder's permission to translate and publish, and the translator holds copyright in their own translated text, with moral rights under Section 57.
• Q: Why is text expansion a practical problem? A: English into Hindi, Malayalam or Tamil usually runs 15-30% longer with taller conjunct glyphs, so buttons, tables, subtitles and brochures overflow unless layout is checked after translation.
• Q: Your subtitle translation is accurate but too long for the shot. What do you do? A: Condense — cut vocatives, fillers and repeated information, prefer shorter synonyms and active voice — keeping the meaning and the speaker's tone within the character and reading-speed limits of the client's style guide.
• Q: What is the official-language requirement that affects translation work in Union government offices? A: Article 343 makes Hindi in Devanagari the official language with English continuing in use, and the Official Languages Act, 1963 with the 1976 Rules requires bilingual documents and classifies states into Regions A, B and C for correspondence.

Common mistakes

• Translating word-for-word and producing TL sentences that are grammatical but unidiomatic — then defending it as "faithful".
• Translating what should be transliterated: personal names, brand names, scheme names and addresses.
• Dropping or mishandling negation, numbers and units — the single commonest real-world error, and the one QA always catches; also writing 1,25,000 as 1,250,000 or confusing crore with million.
• Not building a glossary first, so the same term appears three different ways in one document and the TM becomes unusable.
• Confusing theories in the exam: attributing formal/dynamic equivalence to Catford instead of Nida, or shifts to Nida instead of Catford.
• Treating MT output as a draft to tidy up rather than to verify against the source — fluent hallucinations and reversed negations survive a stylistic read.
• Ignoring register and honorifics, so a formal notice reads like a chat message, or *aap* and *tum* alternate for the same character.
• Skipping the second pair of eyes and self-certifying; ISO 17100 work requires a reviser other than the translator.

Related topics

Technical writing · Editing and proofreading · Content writing · Copywriting · Linguistics and semantics · Indian literature in English translation