Subtitles Are Not Dialogue: How to Spot What Translation Changes in Viewing

From spoken dialogue to read text: why they are not equivalent

Film dialogue reaches us through several channels at once. We hear words, but also hesitations, breaths, interruptions, emphasis, distance from the microphone and, often, overlapping voices. We also see who is speaking, who is silent, what reaction cuts a sentence short and what written information appears in the setting. Subtitles cannot reproduce that entire set in full: they turn part of the audiovisual event into a readable, timed sequence.

It is useful to begin with a basic distinction. Interlingual subtitling translates from one language into another and generally prioritises the viewer’s understanding of relevant verbal content. Subtitles for deaf and hard-of-hearing viewers, or SDH subtitles, add speaker identification, sound effects, music and other significant audio information according to the system and brief. Audio description is another operation: it adds verbal narration of visual elements in available gaps in the soundtrack. All three practices may coexist on a platform, but they do not meet the same need and should not be assessed by identical criteria.

For that reason, a track’s omission of an onomatopoeia, identification of a speaker or description of a song does not in itself prove poor translation. It may indicate an accessibility track rather than conventional interlingual translation; it may also follow different editorial rules. Before comparing lines, identify exactly the label offered by the player, the stated language and variant, and whether the track includes markers such as names in brackets, sound descriptions or music cues.

What subtitling makes necessary: space, speed, segmentation and readability

Every track operates under measurable constraints. A current Netflix guide for Spanish subtitles sets a reference point of up to 42 characters per line and a maximum speed of 17 characters per second for adult programmes. This is an example of an industry specification, not a universal law: other broadcasters, distributors, formats, languages, audience age groups and devices use their own parameters. But it helps explain why read text can rarely be a word-for-word copy of speech.

Speed is not experienced merely as a number. A short sentence can be tiring if it appears during a rapid cut, covers a decisive gesture or requires the eye to return to writing in the set. Equally, a relatively long sentence can be read without friction if it is held over a stable shot and remains on screen for long enough. Timing guides aim precisely to keep subtitles tied to the audio and responsive to editing, generally avoiding crossings over shot changes except where necessary and leaving reading time after an utterance ends.

How a verbal unit is divided also matters. Segmentation is not simply a line break: it determines which words are read together and where a sentence is interrupted. A break between a modifier and noun, or between a preposition and its complement, can hinder reading; a break at the end of a clause will usually preserve syntactic organisation better. When comparing two tracks, do not look only at what they say. Notice when they appear, how long they last, whether the sentence unfolds fluidly and whether the division forces you to mentally reconstruct a relationship that the audio delivered all at once.

Condensation arises from these pressures, but it does not automatically mean impoverishment. It may remove repeated forms of address, hesitations or repetitions of limited narrative value in order to retain decisive information. The critical question is different: what has been compressed, and what function did it serve? Repetition may be conversational filler, but it may also reveal nervousness, hierarchy, humour, an emotional relationship or a tactic of evasion.

Four frictions to watch for during a viewing

First friction: condensation. Pause after a particularly fast scene and note an utterance lasting several seconds. If you know the spoken language, compare the amount of information, not merely the word count. Ask whether connectives, forms of address, insults, repetitions or cultural references disappear. Then watch it again without pausing: the shorter version may let you look at the shot more closely; it may also erase an insistence that defined the character. Both observations may be true.

Second friction: register. Register encompasses choices such as politeness, distance, vulgarity, formality, dialect or intimacy. A loss should not be alleged simply because two expressions are not literally parallel across languages. It is, however, reasonable to describe a visible shift: for example, if a repeatedly ceremonial formula becomes neutral, or a harsh expression becomes softer. Frame it as an effect of the track: “the Spanish reduces the marker of deference maintained in the audio,” rather than as a verdict on the translator’s intention, unless a documented explanation exists.

Third friction: simultaneity. When two voices overlap, a film may convey conflict, intimacy, social noise or simply speed. A track may select one voice, alternate between them, identify them in turns or summarise. No solution can make two lengthy utterances readable with the same immediacy with which they can be heard. Look at whether the selection makes one voice more dominant than it was in the mix, and whether the subtitle preserves signs of interruption such as dashes, ellipses or speaker labels. Do not mistake the material impossibility of showing everything for indifference: first describe the loss or hierarchy that has been created.

Fourth friction: text within the image. A sign, text message, newspaper, diegetic credit and narrative graphic are not dialogue. Professional standards provide for forced narrative titles or superimposed translations when they are relevant to the plot; when they coincide with dialogue, a Netflix guide calls for prioritising the most relevant message and avoiding the accumulation of text to the point of illegibility. Here viewers may detect a genuine competition for their gaze: reading the original sign, the subtitle translating it and the spoken line may require three incompatible visual paths. Note which element goes unread and whether the track moves text so as not to cover important information.

How to compare two tracks without turning a preference into a universal verdict

Use a short, repeatable protocol. First, retain the source details: title of the work, edition or platform, date consulted, audio language, exact language and variant of each track, and device used. Tracks can vary across territories, reissues, dubs, restorations and provider updates; an isolated screenshot does not establish how a film is subtitled everywhere.

Second, select three to five identifiable moments and record their timecodes. Choose, where available, a fast conversation, an argument with overlaps, a scene marked by strong social register and a sequence with on-screen text. Third, transcribe what you can verify with confidence and separate three columns: what is audible, what is visible and what each subtitle says. If you do not command one of the languages, do not fill the gap with phonetic intuition: ask for a review from someone competent in that language pair.

Fourth, describe the operation before evaluating it. “The track removes two repetitions”; “it identifies the off-screen speaker”; “it translates the sign and postpones the end of the sentence”; “it replaces a form of address with a neutral pronoun.” Fifth, connect that operation to the shot, sound and act of reading. Only then offer a cautious interpretation: “the solution speeds reading but softens the insistence”; “the label guides comprehension of an invisible voice”; “the title forces a choice between two points of attention.”

This method makes precise disagreement possible. Two viewers may prefer different results because they have different reading speeds, language knowledge or accessibility needs. What remains open to debate does not disappear, but it stops being expressed as “these subtitles are bad” without specifying what text, what moment and what function are being judged.

Subtitles for translation and subtitles for accessibility: functions worth keeping separate

An interlingual subtitle track and an SDH track may show different text even when both are in the same target language. The latter may need to report a closing door, changing music, a call from off screen or the identity of a voice. These additions take up time and space, and can alter the balance between reading and image. They are not embellishments: they provide access to information that would otherwise arrive through hearing.

It is also important not to use “subtitles” as a synonym for audio description. Audio description turns visual material into voice, whereas subtitling makes audio information visible and, in interlingual cases, translates spoken language through text. When a platform labels its options ambiguously, viewers can document that ambiguity; they should not simply infer that all tracks pursue the same experience.

Quality therefore includes suitability for function. For an SDH track, systematically omitting narratively decisive sounds may be a problem distinct from condensing a filler word in a translation. For an interlingual track, an excess of labels may compete with the image if the brief does not call for them. A fair comparison begins by asking: whom is this track designed for, and what information is it required to make accessible?

Limits of the method: what only a reader competent in the relevant languages can assess

This protocol helps detect changes; it does not confer automatic linguistic competence. A viewer who does not know the source language can analyse duration, placement, density, overlaps, titles and the relationship to editing; they cannot independently establish that a word has a particular connotation, that a dialect has been neutralised or that an ambiguity is deliberate. Even a bilingual viewer must distinguish between understanding words and knowing the cultural context, regional variety or localisation instructions for a particular edition.

Responsible phrasing has degrees. “I could not follow the text message because the track coincided with dialogue” is a verifiable observation. “The translation simplifies a threat” requires linguistic comparison. “The translator intended to censor it” requires external evidence, such as a statement, an applicable editorial guide or documentation of the brief. Maintaining these distinctions does not make analysis less interesting: it makes it debatable, revisable and useful.

The next time a film seems harsher, clearer, funnier or flatter with a different track, do not immediately look for a culprit or a definitive version. Return to a specific minute. Listen, look, read and ask what information has been compressed, displaced or added. That is where a critique of subtitling begins that takes both translation and mise-en-scène seriously.

Sources and filmography consulted

1. dcmp.org. dcmp.org. Consulted 15 Sep 2026
2. dcmp.org. dcmp.org. Consulted 15 Sep 2026
3. www.esist.org. www.esist.org. Consulted 15 Sep 2026
4. www.ofcom.org.uk. www.ofcom.org.uk. Consulted 15 Sep 2026
5. dcmp.org. dcmp.org. Consulted 15 Sep 2026
6. dcmp.org. dcmp.org. Consulted 15 Sep 2026
7. www.ofcom.org.uk. www.ofcom.org.uk. Consulted 15 Sep 2026
8. partnerhelp.netflixstudios.com. partnerhelp.netflixstudios.com. Consulted 15 Sep 2026
9. dcmp.org. dcmp.org. Consulted 15 Sep 2026
10. partnerhelp.netflixstudios.com. partnerhelp.netflixstudios.com. Consulted 15 Sep 2026
11. www.ofcom.org.uk. www.ofcom.org.uk. Consulted 15 Sep 2026
12. www.ofcom.org.uk. www.ofcom.org.uk. Consulted 15 Sep 2026
13. partnerhelp.netflixstudios.com. partnerhelp.netflixstudios.com. Consulted 15 Sep 2026
14. www.ofcom.org.uk. www.ofcom.org.uk. Consulted 15 Sep 2026