Translation, Transliteration and Transcription: Meaning, Letters and Sounds

Three operations, three different targets

Each of the three works on a different layer of language.

Translation

Works on meaning. Output is in a different language. Example: Japanese 山 becomes English "mountain".

Transliteration

Works on writing. Each letter or character is mapped to another script, systematically. Example: Russian Москва becomes Moskva.

Linguistic transcription

Works on sound. Output represents pronunciation, either in phonetic notation or in the spelling habits of a target language.

Transliteration does not tell you what a word means; Moskva is still a Russian word, just written in Latin letters. Translation of the same word into English gives Moscow, which is an established English name rather than a letter-by-letter rendering.

A note on terms: in speech technology, transcription usually means turning recorded speech into text in the same language. That sense, and how it differs from translation, is covered in transcription vs translation. This page uses the linguistic sense, where transcription means representing sounds.

Strict transliteration versus practical transcription

In practice the line between transliteration and transcription blurs, and it helps to think of a spectrum.

At one end is strict transliteration: a fixed table mapping each source character to target characters, often with diacritics, designed so that a trained reader can reconstruct the original spelling. Library catalogs and academic work often use systems like this.

At the other end is practical transcription: spelling a name the way readers of the target language would naturally pronounce it. This is why one Russian name, Хрущёв, appears as Khrushchev in English, Chruschtschow in German and Khrouchtchev in French. Each language uses its own spelling habits to approximate the same sounds. Practical forms are easier to read but cannot be reversed reliably.

Most subtitles, news reports and everyday text sit near the practical end, using established spellings where they exist and simplified systems where they do not.

Romanization systems you will meet

Romanization means transliterating or transcribing into the Latin alphabet. Several languages have more than one system in use, and the choice affects how names look in English subtitles.

• Mandarin Chinese: Hanyu Pinyin is the standard system in mainland China and the most common internationally. Older texts and some names use Wade-Giles or other historical forms, which is why some place and personal names have two familiar spellings.
• Japanese: Hepburn romanization is the most widely used in English. Strict versions mark long vowels with macrons (Tōkyō), while everyday usage often drops them (Tokyo). Kunrei-shiki is an alternative system with different spellings for some sounds.
• Korean: South Korea's official Revised Romanization is used for place names and signage there; the older McCune-Reischauer system still appears in scholarship and in North Korean contexts. Many personal names follow neither, keeping a spelling the person chose.
• Russian and other Cyrillic languages: several systems coexist, from strict international standards to the practical forms used by newspapers and passports, so one name can have several valid spellings.
• Arabic: many systems exist, from scholarly ones with diacritics to simplified journalistic spellings. Informal online writing sometimes uses Latin letters with numerals for sounds English lacks.
• Hindi: scholarly systems use diacritics, while everyday and official use often relies on simpler spellings. Much informal Hindi is typed directly in Latin script.

Going the other way, names from Latin-script languages are transcribed into other scripts by sound. Japanese writes foreign names in katakana, Russian in Cyrillic, Hindi in Devanagari, Arabic and Hebrew in their own scripts, and Chinese with characters chosen for their sound and sometimes for a pleasant meaning.

Script conversion is not transliteration either

Some conversions change the form of writing without crossing a language boundary. Converting between Simplified and Traditional Chinese characters is the familiar example: same language, different character forms, and the mapping is not always one to one. Norwegian Bokmål and Nynorsk are two written standards of one language. None of these is translation or transliteration in the strict sense, though tools and menus sometimes label them that way.

When names are transliterated in translated video

In translated video, names follow the target language's script. That leads to a few predictable situations.

• Target language uses a different script from the source: names are rendered in the target script. English speakers' names appear in katakana in Japanese subtitles and in Cyrillic in Russian ones.
• Source in a non-Latin script, target in English or another Latin-script language: names are romanized, ideally using the established spelling where one exists (Tokyo, Beijing) rather than a strict system.
• Brand and product names: often kept in their original Latin form even inside non-Latin target text, depending on local convention and the brand's own practice.

Whether a name should be kept, transliterated, translated or explained is a policy decision, and that decision belongs to your style guide. The article on translating names and technical terms covers how to set that policy; this page covers the mechanics of scripts and spellings.

Transliteration also affects dubbing. A synthetic voice reads whatever spelling is in the translated text, so a name romanized one way may be pronounced differently from the same name romanized another way. The troubleshooting article on AI voices mispronouncing names shows how to work around that.

Common mistakes and pitfalls

• Mixing romanization systems in one file, so one person's name appears in two or three spellings.
• Ignoring established exonyms. Writing a strict transliteration of a city that has a familiar English name looks odd to viewers.
• Overriding a person's own preferred spelling of their name with a system's output.
• Assuming transliteration explains meaning. A romanized phrase is still unreadable to someone who does not know the language.
• Dropping diacritics in languages where they change meaning or pronunciation, or keeping them in subtitles where they will not render.
• Saving subtitle files in an encoding that cannot hold the target script. See subtitles showing strange characters if accents or characters turn into boxes or question marks.

How mydubly handles scripts and names

mydubly's video translator works with 21 languages written in several scripts, including Latin, Cyrillic, Arabic, Hebrew, Devanagari, Japanese, Korean Hangul and Chinese. Chinese output is Simplified Chinese, Arabic is Modern Standard Arabic and Norwegian is Bokmål. The spoken language is detected automatically, and the transcript is normally written in that language's usual script, though speech that mixes languages can produce surprises.

When you translate, names are rendered by the translation engine along with the rest of the text, so their spelling in the target script is the engine's choice. Expect occasional inconsistency for rare names and check them in the transcript and subtitle files. mydubly's SRT and VTT files are UTF-8 with a byte order mark, which keeps non-Latin scripts intact in most players and editors. If you dub, the voice reads the translated spelling, so correcting a name in your own script before re-recording or editing a line is the reliable fix. For a step-by-step overview of a translation project, see the guide on how to translate a video.

mydubly does not translate text shown on screen, and it does not offer a transliteration-only mode: it translates speech, it does not romanize it.

Next step

Before translating a video with names in another script, write a short list of the names, places and brands that appear, with the spelling you want in each target language. Use it to check the first draft, and add it to your style guide so the next video starts consistent.