Rhythm Reader●
An Interactive Model of Implicit Prosody in Silent Reading
A scansion lens for English text: it shows the beats, the phrases, and the feet that a silent reader may already be hearing — and lets you correct all of it.
Dictionary: loading…Text to analyze
Rhythm-enhanced text
Word-by-word analysis
Meter detail
Metrical lexicon
Reading the notation
Every syllable carries values on three tiers.
Word stress (lexical) is the fixed pattern a word has in the
dictionary, written with digits: 1 primary stress,
2 secondary stress, 0 unstressed —
so banana is 010.
Template is the word's rhythmic shape as a whole:
happy is a trochee (SW), above an iamb
(WS), understand an anapest (WWS),
wonderful a dactyl (SWW), banana an
amphibrach (WSW).
Beats (rhythmic) is how the word is realized in this
sentence: S strong, W weak — a word like
of has word stress 1 but usually no beat.
Feet group beats across words: iamb WS, trochee
SW, anapest WWS, dactyl SWW.
Feet are only drawn when the analysis is confident, and never across a
sentence-level phrase boundary. Some perfectly regular text supports two
scansions equally well — the app reports that honestly instead of picking one.
Editing
Click any syllable to toggle the selected tier (a beat on the Beats tier;
cycle 0 → 1 → 2 on the Word-stress tier). Double-click a word —
or use Inspect in the analysis table — to open the word inspector,
where you can choose template variants, split or merge syllables, and reset
the word to its dictionary analysis. Edited syllables are marked with a
dotted underline; every edit is recorded and exported.
The divergence measure
The analysis table reports a Prosodic Divergence Index
(PDI): how far the current annotation, after your edits, sits from the fully
automatic default, averaged over four components — word stress, template,
beats, and phrasing (each 0–1). The component breakdown matters more than
the composite: it says which representation you disagreed with.
In the Phrases display mode, boundaries themselves are editable: click
❘ to remove one, + to insert one.
Research mode
With research mode on (and an optional participant ID), the app records every edit as a process measure: what changed, on which tier, latency since the previous action, the confidence of the value being overridden, and whether the edit reverses an earlier one. A recording indicator is always shown while logging — the instrument never records covertly. Export the log via Research session log; it also rides along inside the JSON export. Note the resolution caveat: these are click-level latencies in a browser, informative about deliberation and ordering, not a substitute for eye-tracking-grade timing.
Sources and confidence
Each analysis is tagged with its source: the CMU Pronouncing Dictionary, a named fallback rule (for words not in the dictionary), or your own edit. Confidence scores are heuristic indications, not calibrated probabilities.
Running locally
This file is self-contained and runs offline. For full dictionary
coverage (126,052 words), place cmudict.js in the same folder;
without it, an embedded subset of common words is used and other words fall
back to transparent rules.