Splitting Subtitle Text Into Cues and Lines That Read Naturally

Why segmentation changes how subtitles read

Viewers read subtitles in quick glances while watching the picture. Each glance takes in a chunk of text, and the brain tries to make sense of it before the eyes return to the image. When a cue or a line ends in the middle of a phrase, the reader holds an incomplete idea in memory, sometimes guesses wrongly where the sentence is going, and has to correct course when the next piece arrives. Good segmentation makes each piece a complete, or at least self-contained, unit of meaning.

Two levels are involved. Cue segmentation decides which words share a subtitle and when the subtitle changes. Line segmentation decides where, inside a cue, a second line begins. Both follow the same idea of keeping meaning together, but they have different constraints: cues are bound to timing, lines to screen width.

Dividing speech into cues

The ideal cue holds one sentence that fits comfortably in one or two lines and can be read in the time the speaker takes to say it. Real speech rarely cooperates, so subtitlers work through a short list of priorities:

1. End a cue at the end of a sentence where possible.
2. When a sentence is too long for one cue, split it at a clause boundary: before a conjunction such as "and", "but" or "because", before a relative clause, or after a comma.
3. Avoid starting a new sentence in the last few words of a cue; push those words into the next cue instead.
4. Keep a question and its tag together, and keep short replies with the line they answer when timing allows.
5. Follow the speech rhythm. A natural pause is usually a good place to change cue, because the timing and the sense agree.

Where a cue boundary lands also affects timing. A cue split must match when the second part is actually spoken, which ties segmentation to spotting; the timing side is covered in subtitle timing guidelines.

Keeping phrases together

Within a cue, some word groups should never be split across a line break, because separating them makes the first line look unfinished. Most style guides list the same groups:

• An article or determiner and its noun: "the / project" is a bad break.
• A preposition and its object: "in / the morning".
• An adjective and the noun it describes.
• A first name and a surname, or a title and a name.
• A number and its unit, and a date written in words.
• An auxiliary or modal verb and the main verb: "will / arrive", "has / been".
• A verb and its particle in phrasal verbs: "set / up", "turn / off".
• A negation and the word it negates.

Good break points fall between larger units: after punctuation, between a subject and a long predicate, before a conjunction, or before a prepositional phrase that starts a new idea.

One line or two

Many style guides prefer a single line whenever the text fits within the line-length limit, because one line is quicker to read and covers less of the picture. Two lines are used when the text needs them, and three are almost never allowed in professional work.

There are exceptions. Some guides prefer two short lines over one very long line, especially on wide screens where a single line can stretch from one side of the frame to the other. Dialogue between two speakers in a single cue is normally written as two lines, one per speaker, with the formatting conventions for that covered in subtitle formatting rules.

For Latin-script languages, commonly cited line limits sit around 37 to 42 characters, including spaces. The right number depends on the platform, the font and the viewing distance, so check the guide you are working to rather than borrowing a figure.

Line shape

When a cue has two lines, their relative lengths matter. A broadly balanced pair is easy to read; a long top line over a single word below looks awkward and makes the eye travel further than it needs to.

Guides differ on the preferred shape. Some favour a bottom-heavy pyramid, where the second line is slightly longer, because it keeps more of the picture visible near the top of the subtitle area. Others favour a top-heavy shape, or simply ask for lines of similar length. All agree that sense comes first: a well-placed break in an unbalanced pair is better than a perfectly balanced pair that splits a phrase.

Differences across languages and scripts

Segmentation rules travel only partly between languages. Each language has its own phrases that should not be split, and each script brings its own constraints.

German and Dutch

Long compound words can leave little room for good breaks; splitting a compound is not an option, so whole phrases move to the next line

French and Spanish

Clitic pronouns stay with their verbs; some styles also keep a space before certain punctuation in French

Japanese

Text has no spaces between words, so breaks fall between grammatical phrase units; full-width characters mean line limits are counted very differently from English

Chinese

Breaks fall between words, not inside multi-character words; character-based line limits are much shorter than for Latin scripts, and some styles replace commas with spaces

Arabic and Hebrew

Text runs right to left; punctuation and embedded numbers or Latin words can display in surprising positions if the file is handled carelessly

Hindi

Devanagari conjunct characters must not be split, and Hindi subtitles may mix in English words, which changes line width

Right-to-left subtitles bring player and punctuation issues of their own, which right-to-left video localization covers in detail. Translation also changes text length, so a segmentation that worked in the source language may need redoing after translation rather than copying the source line breaks.

Common segmentation mistakes

• Breaking lines purely by character count, so phrases are cut wherever the limit falls.
• Leaving one or two words of a new sentence at the end of a cue.
• Packing a whole long sentence into one cue with a long single line, then compensating with a very short reading time.
• Copying source-language line breaks into a translation where word order is different.
• Splitting names, numbers and units, or verbs and their particles.
• Treating the result as finished without watching it: segmentation that looks fine in a list can feel choppy over the picture.

Limits of rules and of automatic segmentation

No set of rules covers every sentence. Fast, run-on speech can force a choice between a sense-based break and a readable duration, and in those cases a short condensation of the text often serves viewers better than either compromise. Poetry, song lyrics and deliberately fragmented speech may also call for line breaks that follow the original rather than grammar.

Automatic tools segment by audio and length, not by meaning. A speech recognizer marks segment boundaries where it hears pauses and where its internal window ends, which often coincides with sentence ends but not always. A long, breathless sentence can arrive as one huge segment, and a hesitant speaker can produce several fragments for one sentence.

What mydubly's cues look like

mydubly's SRT and VTT files contain one cue per recognized speech segment, always on a single line, numbered in order. Those boundaries follow the recognizer's segments, so many cues are a sentence or clause, while others are fragments or long runs that need splitting. In a dubbed job, short fragments are merged into sentences before the voice is generated, and the target-language subtitles follow those dubbed lines, so a single cue can hold a whole sentence that is longer than one subtitle should be.

mydubly does not insert line breaks, rebalance lines or apply language-specific segmentation rules. Treat its files as accurately timed raw text: open them in a subtitle editor, add a second line where a cue is too long, split cues that hold more than one idea and merge fragments. The editing mechanics, including splitting a cue and renumbering, are covered in how to edit an SRT file, and reading speed checks are in subtitle timestamp alignment.

Next step: segment one real cue list

Generate subtitles for a short clip with the subtitle generator, or open a file you already have. Go through the first fifty cues and mark every line or cue break that splits a phrase, then fix them using the priorities above. If the file will be published as WebVTT or SRT, both formats store a two-line cue simply as two lines of text, so the result works in either.