Implicit prosody.

The page is quieter than the reading

The other five pages in this section each describe something a speaker does out loud and print largely leaves out. Put them together and an obvious question follows. If so much of it is left unspecified, and no one is speaking, what happens when a reader reads silently?

The answer this site is built on is that the reader supplies much of what is missing. Syllables land more or less heavily, groups of words close and open, one word carries the main emphasis, a contour rises and falls toward a clause’s end — none of it audible, most of it unmarked on the page, and much of it never noticed. That is implicit prosody: prosodic structure the reader constructs rather than receives.

The term comes from Janet Dean Fodor, who proposed the implicit prosody hypothesis: in silent reading, a default prosodic contour is projected onto the text, and that projected contour can influence how the sentence is parsed (Fodor, 1998, 2002). The claim is more specific than it first sounds. It is not merely that readers can imagine a voice when asked to. The hypothesis proposes that readers ordinarily project some prosodic structure onto a text, and that this structure can affect comprehension.

Not a sixth component

It is tempting to file implicit prosody alongside stress, rhythm, phrasing, focus, and intonation, as one more item on the list. It is not one. It is a claim about all five at once — about where they come from when the source is a page rather than a voice.

• Stress
• Rhythm
• Phrasing
• Intonation

Words, syntax, and punctuation come from the page; context and prior knowledge from the reader. Focus is not a fifth form but a question the others answer: which word receives the prominence they carry.

In speech, listeners receive the words and the acoustic prosodic cues together. In print, readers receive the words along with partial cues from punctuation, syntax, spacing, and typography. Much of the prosodic structure still has to be inferred — and implicit prosody is the proposal that readers infer it routinely, and that the inference is not decoration but does work.

How do you study something no one can hear?

This is the central methodological problem, and it shapes everything the field can claim. The inner voice cannot be observed directly. Asking readers to describe it may change it. So the evidence is indirect: find a place where a prosodic difference would predict a measurable consequence, then look for the consequence.

Four approaches recur.

Ambiguity that prosody would resolve. If two readings of a sentence would be distinguished by a spoken boundary, and readers systematically prefer one reading with no boundary printed, the preference needs explaining. This is the original argument, and the next section takes it up.

Reading times where prosody would cost something. Eye-tracking makes it possible to ask whether a word that is prosodically awkward is also slower to read — even though nothing about its spelling or frequency is unusual.

Brain responses to printed boundary cues. Spoken phrase boundaries evoke a distinctive electrical signature, the closure positive shift. Steinhauer and Friederici (2001) found a comparable response in silent reading, at commas. A printed comma appears to evoke covert prosodic phrasing resembling the processing of an audible boundary.

Marking prosody explicitly and seeing what changes. If readers are constructing a prosodic representation, then printing part of it should interact with what they were building on their own. This is the approach the research behind this site has taken.

The evidence, in three pieces

Readers slow down for prosody that is not printed

Ashby and Clifton (2005) tracked eye movements while readers read words that differ in stress structure but not in spelling difficulty. Words carrying two stressed syllables took longer to read, and drew more fixations, than words carrying one.

Nothing in the orthography marks that difference. English does not print stress. If stress is nonetheless affecting how long the eye rests on a word, the reader must be supplying it.

Breen and Clifton (2011) sharpened the point using stress-alternating homographs — words like present, whose stress depends on what part of speech it is.

They embedded such words in limericks, where the metre sets up a strong expectation about where the beat should fall. When the word’s actual stress clashed with the metre the reader had been building, reading slowed. A rhythm no one printed and no one spoke was nonetheless there to be violated.

Readers group sentences that print leaves ungrouped

The oldest argument for implicit prosody comes from a cross-linguistic puzzle. Consider a sentence with two possible attachments for its relative clause:

English readers tend to attach the clause low, to the actress. Spanish readers tend to attach it high, to the servant (Cuetos & Mitchell, 1988). This challenged parsing theories that treated attachment preferences as universal, and became an important test case for competing explanations of sentence processing.

Fodor’s proposal was that the difference is prosodic rather than syntactic. Languages differ in how long a phrase has to be before it gets a boundary of its own. A relative clause that would sit inside the preceding phrase in one language becomes its own phrase in another — and a clause set off by a boundary attaches higher. On this account readers are not applying different parsing rules; they are hearing the same sentence differently in their heads, because their languages phrase it differently (Fodor, 1998, 2002).

A broader prediction follows: phrase length should affect where readers project a boundary. Hwang and Steinhauer (2011) tested that prediction in a different construction — ambiguous Korean garden-path sentences — by varying the length of sentence-initial subject noun phrases. Their ERP results were consistent with phrase length reshaping implicit phrasing and, in turn, syntactic processing.

For relative-clause length specifically, Hemforth and colleagues (2015) found more high attachments for long than for short relative clauses in English, Spanish, and German — while treating implicit prosody as one possible explanation rather than a settled cause. That hedge is doing real work, and the section below returns to it. The Phrasing page takes up what boundaries do; the question here is where a boundary comes from when nobody put it on the page.

Readers place emphasis where the discourse calls for it

Stress and phrasing can be studied without asking the reader anything. Emphasis is harder, because it depends on what the reader already knows. Gross, Millett, Bartek, Bredell, and Winegard (2014) approached it by borrowing a convention from dictionary pronunciation guides — capitals marking the emphasized word — and asking silent readers to choose between versions of a sentence after a short story had set up an expectation.

Readers preferred the version whose marked emphasis matched the newsworthy content of what they had just read, and a second experiment extended the finding to function words and added a reaction-time measure. The preference tracks discourse, which is what focus prosody does in speech. The pattern is consistent with readers representing discourse-appropriate focus; the capitals made that otherwise covert representation testable.

That study is the direct ancestor of the tools on this site. The Stress page describes what followed from it.

What this does not establish

Five limits are worth stating plainly, because implicit prosody is a hypothesis that invites overreach.

The inner voice is inferred, never observed. Every result above is a measurable consequence taken to imply an unmeasured cause. That is ordinary practice in cognitive science, but it means the evidence is always compatible with some other account of the same slowdown or the same preference. The case rests on convergence across methods rather than on any single decisive finding.

“A prosodic contour” is doing a lot of work. The hypothesis is often stated as though readers project one default structure. Real readers differ from one another, and the same reader differs across texts, reading speeds, and purposes. How detailed the projected representation is — whether it resembles a full spoken rendition or something much sparser — is not settled.

The attachment account remains contested. The story above runs: length creates a prosodic boundary, and the boundary shifts attachment. The middle step is the one least well supported. Jun and Kim (2004) tested it directly in Korean and found that although Korean readers preferred high attachment, speakers did not produce the larger break between the relative clause and the adjacent noun phrase that the hypothesis predicts — their default was to phrase each word separately. Hwang and Steinhauer (2011) further note that earlier length manipulations sometimes changed referential or syntactic complexity along with prosodic length, leaving possible confounds involving discourse representation, memory load, and semantic plausibility. Length effects on attachment are robust; that implicit prosody is what produces them is a further claim, and one still under debate.

Influence is not the same as necessity. Showing that implicit prosody can shift an interpretation does not show that reading depends on it. Skilled silent reading is fast, and much of it may proceed with very little prosodic elaboration; the effects are clearest in exactly the places researchers construct to make them visible, which are not typical sentences.

The link to reading skill is largely correlational. Prosodic sensitivity and comprehension travel together: a meta-analysis reports a moderate correlation, r = .31, while emphasizing variation across readers and tasks and pathways that run indirectly through vocabulary, word reading, and metalinguistic skill (Tong, Lentejas, Deng, An, & Cui, 2023). Which way the influence runs is a separate question, and one that observational designs cannot settle. A recent training study from this project found that marking stress and rhythm improved adult readers’ prosodic sensitivity (Gross, Roldan, Kiessel, Esselink, & Dunlap, 2026) — a change in the prosodic measure itself, which is not the same as a change in reading.

Why the tools exist

If print fully represented the prosody readers supply, this site would have nothing to do. The reason for the tools is the gap: readers construct a great deal that the page never shows them, and they have little direct way to inspect what they built.

Marking it changes that in two ways. It gives a reader something to compare against, which is one of the few ways to notice a process that normally runs unattended. It also turns an implicit structure into an explicit one, allowing this project to test how readers respond when printed prosodic markings align — or conflict — with the structure they may be constructing.

That is not a new thought here. Gross, Winegard, and Plotkowski (2018) ended their study of stress marking by asking whether a free application could turn ordinary text into typographically enhanced text — and posed the question in the convention they had been studying:

Try it

This page has no tool of its own, because implicit prosody is not a separate thing to display — every tool here is already a display of it.

The comparison is the exercise. Read a sentence silently first and notice what your inner voice did with it. Then paste the same sentence into Rhythm Reader, which proposes one pattern of stress and rhythm with every syllable tappable, or Rhythm Reader Pro, which shows phrase structure and marks each phrase’s main accent and lets you move the boundaries yourself.

Agreement gives you a useful point of comparison, not proof: the tool offers a plausible analysis, not an independent standard against which an inner voice could be checked. Disagreement may be the more informative case. Something that normally passes unnoticed has become visible enough to examine — and to argue with. That is the whole idea.

References and further reading

This completes the Learn section. Start at Stress to read it in order.