Dialogue and pacing analyzer
Paste a scene or a chapter: the tool measures how much of it is dialogue, attributes each line to the characters the text names, picks out their verbal tics and highlights -ly adverbs and passive constructions. These are SagaScope's own measurements, computed in your browser: your text is neither uploaded nor stored.
Quotation marks or dialogue dashes (—). Up to 70,000 characters (about 12,000 words). Your text never leaves your browser.
When set, lines like "I said" or "dis-je" are attributed to the narrator.
Words per sentence, smoothed. Shaded zones are stretches without dialogue.
Character voices
11 of 11 lines are attributed to a character named in a tag; the rest stay unattributed rather than guessed.
| Character | Lines | Words per line | With "?" | With "!" | With "…" | Tics |
|---|---|---|---|---|---|---|
| Julie | 5 | 8.2 | 20% | 0% | 0% | — |
| Marc | 5 | 4.6 | 20% | 0% | 0% | — |
No "voices too alike" alert: with a few dozen lines per character, these figures vary as much within one character as between two.
- These are measurements, not judgements: a long stretch without dialogue or a very long sentence can be exactly what the scene needs.
- A line is attributed only when the text names the speaker next to it ("said Marc", "Julie asked"). Lines without a tag are left unattributed rather than guessed.
- Tics and punctuation rates are shaky below a few dozen lines per character: treat them as leads to check, not as a portrait.
- Dialogue without quotation marks or dashes cannot be detected.
The same measurements across your whole manuscript, chapter by chapter
In SagaScope these signals are computed for every chapter you analyse, using the characters of your story bible, with no credits and no AI for the measurements. You see the chapter's rhythm curve in its bulletin.
Join the waitlistEvery character's age at every event, even in an invented calendar.
How to read the results
Dialogue share is the proportion of spoken words: tags (‘said Marc’) and quoted thoughts don't count. There is no right percentage. An introspective novel may hold very little dialogue, a thriller a lot. As an example only, three published novels measured while building this tool sat between 31% and 44%. What matters is the gap between a scene and what you meant it to do.
The curve shows sentence length across the text, smoothed over five sentences. Shaded zones are stretches of at least 400 words with no line of dialogue at all. That isn't a flaw in itself: a description or an introspective passage may need it. The tool only tells you where it is.
The share of sentences over 35 words gives a sense of your prose's breath. A very long sentence, well built and well placed, can carry a whole scene; what tires the reader is a run of them. So check whether they come one after another. Average length, for its part, is best compared from scene to scene: an action scene is usually choppier than an introspective one, and that is as it should be.
-ly adverbs and passive voice are highlighted so you can judge each case: an adverb that doubles the verb (‘whispered softly’) often deserves a cut, one that changes the meaning doesn't. The passive it detects is limited to unambiguous forms (‘was taken by’, ‘had been found’).
The voices table counts, for each character, their lines, average length and the share of lines containing a question mark, an exclamation mark or an ellipsis. Tics are the words or phrases (one to three words) that this character uses far more than the other voices in the text, at least three times.
What stylometry cannot tell you
A line is attributed only when the text names its speaker next to it: ‘said Marc’, ‘Julie asked’, or ‘I said’ if you tell the tool who your narrator is. Lines without a tag stay unattributed. We would rather give you an incomplete table than a wrong one: when we tested alternation rules (‘A speaks, then B’), they held only about one time in two or three on real novels, so we kept them out of the voice statistics. If the table attributes only half of your lines, it may be a sign that a reader will also struggle to follow who is speaking: add a tag or a gesture that names the character wherever an exchange runs past three or four lines in a row.
The tool shows no ‘these two voices sound alike’ alert. We checked whether such figures tell the characters of a single novel apart: with a few dozen lines each, a character resembles themselves, from one half of the book to the other, as much as they resemble anyone else. A verdict of that kind would be noise presented as a conclusion. It would take far more text per character.
Finally, dialogue without quotation marks or dashes escapes the tool, and a very short text gives unstable punctuation rates. Treat the numbers as leads to check when you reread, never as a judgement of your style.
Three ways to use it
Talking heads. A scene that is 85% dialogue with no narration between the lines feels like characters talking in a void. Find the long lines and slip in a gesture, a place, a thought.
A tic that spreads. If one character's top tic is ‘you know’ or ‘anyway’, that may be a deliberate trait. If the same tic shows up for every character, it is yours: vary it from one voice to the next.
Sentences that stretch. When the curve climbs above 35 words over several consecutive sentences and -ly adverbs pile up in the same passage, the scene slows down without the reader knowing why. Cut, then read it aloud.
Common questions
- How does the tool know who is speaking?
- It reads the tags: a speech verb followed by a name (‘said Marc’, ‘Julie replied’) or preceded by one (‘Marc said:’). A pronoun (‘he said’) identifies nobody on its own. If your narrator speaks in the first person, enter their name so ‘I said’ lines can be attributed.
- Is my text sent anywhere?
- No. The analysis runs entirely in your browser, using the same measurements as SagaScope. Nothing is uploaded or saved; visit statistics only receive rounded numbers, never your text.
- What share of dialogue should I aim for?
- None: it depends on the genre, the scene and your intent. Compare your scenes with each other, or with what you meant to do. An exposition scene in pure dialogue deserves a second look, and so does an action scene without a single line.
- Why are some lines left unattributed?
- Because the text doesn't say who is speaking next to them. The tool doesn't guess: a wrong attribution would skew the character's statistics. The more your tags name the speaker, the fuller the table.
- Does the tool detect characters who sound too alike?
- No, on purpose. The figures we have do not reliably tell two characters apart on a few dozen lines; an alert would be a false signal.
- How is this different from the chapter pacing analyzer?
- That one compares chapter lengths and finds stretches where sentences are all the same length. The dialogue analyzer is about what is said and by whom: dialogue share, voices, tics, adverbs and passive voice.
And for the whole novel's consistency?
SagaScope builds your story bible from your text and alerts you when a detail contradicts itself. Join the waitlist.