Reading Research Quarterly

The science of reading has persuasively shown that a to‐be‐recognized word in print is influenced by detailed knowledge of phonology (Van Orden & Kloos, 2005). In contrast, the extent to which silent reading represents prosody, the rhythm and melody of language, is less‐well understood. The impetus for our investigations into the role of prosody in skilled, adult reading stems from the central role that prosody plays in speech. Prosody is a universal feature of all languages (Endress & Hauser, 2010). Prosodic speech acoustically varies in duration, frequency, amplitude, and tempo (Selkirk, 1986). The singsong quality of infant‐directed speech is an exaggerated example (Bryant & Barrett, 2007). In American Sign Language, the physical correlates of prosody involve variation in displacement, velocity, jerk, and facial features (Wilbur & Martínez, 2002).

Prosody serves diverse functions in speech. Prosodic variations reveal features of the speaker (e.g., emotional state, intentions) as well as the form of the utterance (e.g., statement, request) that may not be captured by word selection, sentence construction, or punctuation. For example, the expression “Brian bought a book” would bear different prosodic qualities to signal a statement, question, exclamation, or sarcasm (Nespor & Vogel, 1986). Additionally, prosody resolves polysemy (Schafer, Speer, Warren, & White, 2000), reduces ambiguity (e.g., DeDe, 2010; Kjelgaard & Speer, 1999; Snedeker & Trueswell, 2003), flags irony (Nakassis & Snedeker, 2002), influences online parsing decisions (DeDe, 2010), and signals turn taking in a conversation (Oliveira & Freitas, 2008). Prosodic cues are redundantly present when other cues offer disambiguation (Schafer et al., 2000; Snedeker & Trueswell, 2003). As a testament to its central role in speech comprehension, prosody is inextricably part of remembered speech, such that “old” words with new prosody are rarely falsely recognized (Speer, Crowder, & Thomas, 1993). The absence of prosody can extract a toll on the listener, as spoken utterances lacking rich prosody are harder to understand (Cutler, Dahan, & van Donselaar, 1997).

Like speech, expressive reading is rich in prosody. Reading aloud “with appropriate speed, accuracy, and proper expression” is a hallmark of fluency among nascent readers according to the National Reading Panel (National Institute of Child Health and Human Development, 2000, p. 11). Expressive reading necessitates deciphering the correct pronunciations of the individual words on the printed page (segmental phonology) as well as rendering the appropriate pattern of undulating stress and pulsating beat across connected text (suprasegmental prosody).

Written English is based on an alphabetic system that maps graphemes to phonemes. Becoming a fluent reader is linked to fast and efficient decoding of the letter–sound correspondences in English (Schwanenflugel, Hamilton, Kuhn, Wisenbaker, & Stahl, 2004). Decoding speed also forecasts prosodic reading and better reading comprehension (Miller & Schwanenflugel, 2008; Schwanenflugel et al., 2004). Miller and Schwanenflugel (2008) suggest that young readers apply their understanding of spoken prosody to oral reading first and then internalize these prosodic codes later as an expressive inner voice. Prosodic reading has many communicative benefits. When reading aloud, natural prosody facilitates sentence organization in memory and increases recall (Koriat, Greenberg, & Kreiner, 2002). By comparison, nonprosodic reading compromises comprehension (National Research Council, 1999). Early versions of synthetic speech (text‐to‐speech assistive technologies) yielded unnatural, monotonous renderings of spoken text, causing cognitive processing costs, such as fatigue (Paris, Thomas, Gilson, & Kincaid, 2000).

Although prosodic reading is a hallmark of fluency, the exact role of suprasegmental prosody in the translation of print to speech is a nascent field of investigation. The term prosody appears five times in The Science of Reading: A Handbook (Snowling & Hume, 2005), and most of these entries mention its omission from the field of reading research. As Treiman and Kessler (2005) noted, the omission of prosody is unsurprising given that distinctions of length, tone, pitch, stress, and intonation are largely ignored by writing systems. Punctuation and syntax assist but underspecify a prosodic rendering of connected text. The lack of prosodic transparency in written English means few cues for expressive reading. Struggling readers produce oral readings that are prosodically ill formed (Levy, di Persio, & Hollingshead, 1992). Repeated readings foster prosody development in good readers but not necessarily in poor readers (Levy, Nicholls, & Kohen, 1993).

In two studies, we sought to understand whether the inner voice of skilled, adult readers preserves the prosodic richness of speech and expressive reading. Adult readers typically report the phenomenological experience of an expressive “voice in the head” that seemingly captures the nuances of lively speech (Huey, 1908/1968). In support of an expressive inner voice in adult readers, reading research has persuasively shown that printed word recognition is influenced by detailed knowledge of segmental phonology. Just as tongue twisters are notoriously difficult to articulate properly, visual tongue twisters trick the silent reader. Readers are significantly slower in judging the semantic acceptability of visual tongue twisters (e.g., “The detective discovered the danger and decided to dig for details”) compared with control sentences (McCutchen, Bell, France, & Perfetti, 1991).

The tongue‐twister effect has generalized to deaf readers (Hanson, Goodell, & Perfetti, 1991) as well as to tongue twisters comprised of legal nonwords when spoken, typed, recalled, or recognized (Acheson & MacDonald, 2009). In a critical test of the phonological nature of the tongue‐twister effect in working memory, McCutchen and colleagues (1991) found that the phonetic content of digit names held in working memory interacted with the phonetic content of the sentences being read, suggesting interference at the level of sound‐based codes used in working memory. Neuroimaging research revealed that the tongue‐twister effect implicates cortical areas involved in articulatory‐phonological processing as well as speech programming (Keller, Carpenter, & Just, 2003), a finding that is consistent with the idea that the voice in the head exploits the broader language system (Mattingly, 1972).

Phonological features of the inner voice emerge in adult readers when performing a range of visual word recognition tasks. In lexical decision tasks, target words with phonetically longer vowels (e.g., plead) or consonant clusters take longer to respond to than targets with respectively shorter vowels (e.g., pleat) or consonant clusters, when orthographic lengths are controlled (Abramson & Goldinger, 1997; Lukatela, Eaton, Sabadini, & Turvey, 2004). Observing eye movements, Ashby and Clifton (2005) found that polysyllabic words with two stressed syllables (e.g., fundamental) were read 36 ms more slowly and received more fixations than did polysyllabic words with one stressed syllable (e.g., significant), a finding that is consistent with the longer pronunciation times for stressed compared with unstressed vowels (Selkirk, 1986).

Recent brain activation research reveals that skilled, adult readers activate phonological features during the initial moments of visual word recognition (i.e., by 80 ms), suggesting that phonological codes may be guiding (rather than a by‐product of) lexical access (Ashby, Sanders, & Kingston, 2009). To establish the time course of phonological processing of vowels in word recognition, Ashby, Treiman, Kessler, and Rayner (2006) observed the eye movements of skilled readers who were presented with parafoveal previews of the vowel for a to‐be‐read target word embedded in a sentence. Shorter reading times were observed when readers received congruous, compared with incongruous, vowel previews, suggesting that phonological processing of vowels begins early (before foveal fixation; Ashby et al., 2006).

Even though written English is based on an alphabetic system, skilled readers extract phonological units larger than single phonemes when computing the relations between print and speech. Onsets (the initial consonant or consonant cluster) and rimes (the vowel and any following consonants) play a special role in mapping letters to sounds (Treiman, 1994). Orthographic rimes, in particular, are phonologically reliable and guide the pronunciation of written words (Treiman, Mullennix, Bijeljac‐Babic, & Richmond‐Welty, 1995). Orthographic rimes are further divisible into two phonological units—a vocalic nucleus and syllable‐final coda—as demonstrated by their cohesiveness in a Reicher (1969) and Wheeler (1970) letter detection task (Gross, Treiman, & Inman, 2000). Suggesting that phonological units as large as syllables may be guiding lexical access, neurophysiological evidence reveals that skilled, adult readers prelexically activate syllable‐level information during the initial moments of visual word recognition (Ashby, 2010). When syllables are previewed parafoveally, compared with incongruous previews (with one letter more or less), the reading of low‐frequency words is benefited in particular (Ashby, 2006; Ashby & Rayner, 2004). Thus, the English orthography has a lexical level (represented by the spaces between words), a graphemic level where a letter or a group of letters represents a single sound, and an intermediate level such that readers use large phonological units (e.g., syllables, rimes) when recognizing words in print.

When translating the relations between print and speech, emerging evidence suggests that adult readers extract suprasegmental prosodic features, including lexical stress, metrical stress, prosodic phrasing, punctuation, and acoustic features of the implied author. By observing readers' eye movements when reading stress‐alternating homographs, Breen and Clifton (2011) evaluated the influences of lexical and metrical stress. In homographs, unique meanings correspond to the different pronunciations (e.g., ABstract [noun] vs. abSTRACT [verb]; caps have been used to signal stress assignment in these examples, although no such aid was available to participants in the study). In the experiment, readers experienced a “cost” (e.g., longer reading times) when homographs were syntactically biased to have a noun interpretation (e.g., ABstract) yet had to be prosodically disambiguated as a verb (e.g., “The brilliant abSTRACT the…”), a garden path phenomenon. To explore metrical stress across phrases, Breen and Clifton embedded stress‐alternating homographs (e.g., PRESent vs. preSENT) in limericks, noted for their catchy, predictable rhymes (e.g., “There once was a clever, young gent who had a nice talk to present”). When silent readers encountered a mismatch between the predicted meter and actual stress pattern of a homograph, they experienced difficulty (i.e., lower probability of skipping the critical word, longer fixations times).

Implicit prosodic phrasing plays a role in silent reading (Bader, 1998; Hwang & Schafer, 2009; Swets, Desmet, Hambrick, & Ferreira, 2007). According to the implicit prosody hypothesis (Fodor, 2002), silent readers project prosody onto written sentences to aid syntactic parsing decisions. The ambiguous sentence “The old man the boat” may cause the silent reader to stumble if “old man” is mistakenly parsed as the subject of the sentence. Proper phrasal parsing of this garden path sentence requires “man” to be parsed as the verb. Garden path sentences seemingly require the reader to reanalyze both the syntactic structure and the prosodic structure of the sentence (Bader, 1998). The ambiguous phrase “the maid of the princess who scratched herself in public” has two plausible interpretations concerning who did the scratching: the maid or the princess. Swets et al. showed that individual differences in working memory capacity among adults affect syntactic ambiguity resolution. Participants with low working memory tended to insert a prosodic break between “maid of the princess” and “who scratched herself,” rendering the conclusion that the maid did the scratching—a high attachment preference. In contrast, participants with high working memory were more likely to interpret the princess as the self‐scratcher—a low attachment preference. The prosodic breaks created on the fly influence attachment preferences (Bader, 1998), and working memory seems to play a role in prosodic chunking strategies (Swets et al., 2007).

Punctuation guides emphatic oral reading (e.g., Stop!) as well as silent reading. Punctuation in text and altered prosody in speech were found to affect word recognition and comprehension in a similar fashion (Cohen, Douaire, & Elsabbagh, 2001). Moreover, commas and speech boundaries were found to reliably elicit a similar online brain response (event‐related brain potential), suggesting a correspondence between punctuation and an inner prosodic voice (Steinhauer, 2003). The “voices” of the story characters influence silent reading times (e.g., Alexander & Nygaard, 2008; Kurby, Magliano, & Rapp, 2009). After being familiarized to the authors' voices, reading rates were slower for texts “written” by slower talking speakers than faster talking ones (Alexander & Nygaard, 2008), suggesting that silent readers impose an author's pronunciation rate onto their voice in the head.

Building on recent research, we explored whether the inner voice of skilled, adult readers represents focus prosody when translating the relations between print and speech. Fodor (2002) notes that the role of implicit prosody in silent reading is easily overlooked and difficult to substantiate and that a manipulation of implicit prosody often requires a manipulation of another linguistic feature. To subtly manipulate implicit prosody without compromising interpretative clarity, we exploited stylistic emphasis of print (e.g., caps) to elicit prosodic prominence in the silent reader's inner voice. We borrowed the usage of caps as a marker for prosodic emphasis from the pronunciation guides of dictionaries (e.g., for aioli, /ahy‐OH‐lee/), comic strips (e.g., POW!, OUCH!), and the phenomenological impression when reading an e‐mail message typed in all caps (e.g., CAN YOU HEAR ME NOW?), which seems as though the sender has raised his or her voice. E‐mail etiquette forums caution writers to restrict the use of caps because “typing in all caps is considered yelling, screaming or at the very least adding emphasis to the word you type” (Kallos, n.d., para. 3).

Experiments 1 and 2 investigated whether silent readers represent focus prosody in their inner voice. If silent readers internalize the prosody of their spoken language, then linguistic theories of focus prosody in spoken English should apply to the voice in the head. We extrapolated from Selkirk's (1986, 1996) theory of focus marking in spoken English, which postulates that speakers prosodically emphasize new or important content in an ongoing discourse. In theory, silent readers should give higher helpfulness ratings for the final sentences of a paragraph when new or important content is cap‐emphatic and previously given information is not stylistically emphatic, compared with incongruously matched stimuli. For example, the cap‐emphatic text in “JAMES stole the bracelet” should be preferred by silent readers when the paragraph builds up to the question, “Who stole the bracelet?” In contrast, the cap‐emphatic text in “James stole the BRACELET” should be preferred when the just‐read story leaves the reader wondering, What did James steal?

Experiment 2 used a reaction time task to augment the preference judgments used in experiment 1 and additionally explored whether typically elusive function words could bear prosodic prominence when contextually new. For example, “He CAN” should be judged more quickly and accurately following “Can he swim?” “HE can” should be judged more quickly and accurately following “Who can swim?”

Experiment 1

Focus prosody plays an important role in speech production, speech perception, and expressive reading. Experiment 1 investigated whether silent readers represent focus prosody when translating print to a speech‐based code.

Across a range of speech perception studies, acoustically salient (longer, louder) information is understood as new or important content in the ongoing discourse, whereas de‐accented constituents are understood as given information (Bolinger, 1978; Rooth, 1992; Speer et al., 1993; Warren, 1996). New content is not derivable from the story or the implied context (Halliday, 1967). For example, the question, “Did John read Don Quixote?” implies the noun book. Because book is implied, Ladd (1980) contends that the fitting response emphasizes the new information, “No, John doesn't READ books.” (John may read journal articles.)

Consistent with linguistic theories, English speakers responded to pragmatic needs in an ongoing discourse by placing the main stress on the most newsworthy content (Nava & Zubizarreta, 2010; Rooth, 1992). Speakers were engaged in a scripted question‐and‐answer dialogue. When asked, “What's been happening?” 81% of the speakers placed the main stress on the unexpected verb (“A dog is SINGING”), rendering the verb as highly noteworthy (Nava & Zubizarreta, 2010). When answering the question, “Why are you buying that old stamp?” the majority (82%) of speakers responded by narrowly placing emphasis on the newsworthy verb (“because I COLLECT stamps”). When asked, “What was that crashing sound,” which implies something broke, 97% of the speakers emphasized the new subject (“A GLASS broke”).

Complementary research reveals that listeners of English appreciate when contrastively focused content is prosodically marked by speakers. In question‐and‐answer dialogues, listeners rate focus marked by pitch‐accenting and the de‐accenting of the nonfocused contents of the sentence as more appropriate than other intonational patterns (Welby, 2003). Focal pitch accenting is used by listeners to both encode and remember contextually prominent words in spoken discourse, and better memory for pitch‐accented words persists at least one day later (Fraundorf, Watson, & Benjamin, 2010).

Focus prosody is present in the expressive reading of children and adults. Burgeoning readers in the third grade are sensitive to linguistic focus. When reading aloud, these early readers marked contrastive focus with higher pitch and greater intensity compared with the same words appearing in noncontrastively focused contexts (Schwanenflugel, Westmoreland, & Benjamin, 2013). Skilled, adult readers dynamically responded to the context that established given information and produced the same sentence with different intonational patterns that emphasized the contrastive information (Cooper, Eady, & Mueller, 1985; Eady & Cooper, 1986).

In Cooper et al.'s (1985) study, readers first heard one of four prerecorded questions (e.g., “Did William or Chuck like the present that Shirley sent to her sister?”). Afterward, readers read aloud a declarative sentence: “Chuck liked the present that Shirley sent to her sister.” Across a range of speakers and stimuli, the acoustical analysis consistently showed that the focal words (i.e., Chuck, present, Shirley, sister) in stressed compared with nonstressed contexts was much longer in duration and perceptually distinctive because declination occurred on either side of the focused content (Cooper et al., 1985; Eady & Cooper, 1986). According to the implicit prosody theory, the prosody of sentences read aloud (e.g., those in Cooper and colleagues' studies) perfectly matches the prosody projected onto a silently read sentence in a similar context (Fodor, 2002).

Just as new information is pitch accented and old information is de‐accented in an ongoing discourse and in sentences read aloud, we investigated whether new information receives prominence in the reader's inner voice in experiment 1. Selkirk's (1986, 1996, 2002) theory of focus marking in spoken English stipulated a precise, testable framework for how contextual newness might implicitly influence our silent readers. In the following examples from Selkirk (1986, 1996), focus is marked by caps, and the boundaries of focus marking (which necessarily contain the most prominent word in that phrase) are marked by [_F].

When answering the question, “What did Mary buy a book about?” the appropriate response is narrowly focused on the new content that receives heavy stress: “Mary bought a book about [_F BATS].” If the question was, “What did Mary buy?” the appropriate response, “Mary bought [_F a book about BATS],” pitch‐marks the new content, and focus propagates up the syntactic tree to the phrasal level. Prosodic prominence can be contrastive, and focus is narrowly restricted to the contrasting information (e.g., “I don't think she [_F SNIFFLED], she [_F SNEEZED].” Similarly, a fitting response to the question, “Why don't you eat French [_F TOAST]?” is “I've forgotten how to [_F MAKE] French toast,” where heavy stress is narrowly focused on the contrasting information (Ladd, 1980). Focus prosody presupposes that the speaker and hearer agree on shared information, and the speaker prosodically highlights new or important content (Halliday, 1967).

In experiment 1, participants silently read short stories in which the final sentence of each featured new content in light of the just‐furnished details of the story. To avoid reliance on our subjective sense of newness when writing short stories, we morphed examples from Selkirk (1986, 1996, 2000, 2002) and drew on the degree of givenness theory (Baumann & Grice, 2006; Baumann, Grice, & Steindamm, 2006). The degree of givenness theory postulates that givenness systematically varies along a continuum from most accessible to least accessible. In theory, immediate repetition of a given referent yields low newsworthiness (e.g., “Sam stood up in the canoe. Sam fell out”). Textually displaced repetition yields increased newsworthiness (e.g., “Sam and Sally went canoeing. The river was turbulent. The canoe rocked back and forth and someone fell out. Sam fell out”). Therefore, when writing our short stories, we exploited novelty and textual displacement to yield new or important content, and immediate repetition to yield old information.

If the inner voice captures the prosodic liveliness of focus prosody according to Selkirk's theory (1986, 1996, 2002) of spoken English, we reasoned that our silent readers should give higher helpfulness ratings when new or important content is cap‐emphatic and previously given information is not stylistically emphatic (focus congruous condition), compared with matched, incongruous stimuli. For example, the reader first learned about the family's two dogs, Rover and Fido. Next, the reader learned a hole was dug under the fence, allowing one dog to squirm away. “[_FROVER] escaped” as the story‐final sentence should receive higher preference ratings compared with “[_F Rover] ESCAPED” because cap‐emphasis congruently maps onto the prosodic focus in the former sentence but not in the latter one. If the silent reader instead learned about the family's (only) pet dog, Rover, and learned that Rover just dug a hole under the fence, then “Rover [_F ESCAPED]” should be rated more favorably than the incongruous “ROVER [_F escaped].”

Methods

Experiment 2

Experiment 2 implemented two experimental modifications to offer converging evidence for an inner voice during silent reading that flags new information. Instead of asking participants to make off‐line preference judgments using a Likert scale, experiment 2 asked participants to make speedy, dichotomous judgments of helpful/unhelpful when reading the target sentence, while their reaction times and accuracy were recorded. Additionally, experiment 2 broadened our exploration of prosodic prominence by featuring function words (e.g., auxiliary verbs, prepositions, pronouns, conjunctions) in both new and given roles.

Function words are known for their elusiveness in both spoken and written English. In discourse, functions words typically refer to established content and, thus, are prosodically weaker than the content words in the same context (Selkirk, 1986, 1996). For example, Nava and Zubizarreta (2010) found that pronouns (e.g., them) that referred to previous content (e.g., tomatoes) were de‐emphasized 100% of the time by English speakers (e.g., “Do we have tomatoes? No, I didn't BUY them”). Characteristically unstressed function words correspondingly have reduced vowels when pronounced (Selkirk, 1995), rendering them more difficult to isolate in the acoustic stream (Gleitman & Wanner, 1982).

Although pronouns are typically prosodically weaker than the content words in the same context (Selkirk, 1986, 1996), there are contexts in which pronouns convey new or important information. Ladd (1980) offers the following example. Consider the question, “Why don't you have some French toast?” The narrowly focused response, “There's nothing to make French toast OUT of,” de‐emphasizes what is known (French toast) and prosodically emphasizes the contextually important pronoun. Using similar sentences, German, Pierrehumbert, and Kaufmann (2006) asked participants to play the role of speaker B in a conversation by naturally reading a scripted response. After hearing speaker A's prerecorded sentence, which established the context of “Are the children playing their game?” participants read, “Paul took down the tent that they play their game in.” Linguistic theories of focus marking (Schwarzschild, 1999; Selkirk, 1986, 1995) predict that the preposition in should be pitch‐accented. Yet, even in this narrowly focused condition in which the preposition in does not have an antecedent in the discourse that is salient or implied, the preposition received pitch‐accenting only 32% of the time. Participants preferred (64% of the time) to place focal stress on the entire noun phrase, “their game in.” Across all the conditions, speakers preferred to accent nouns and verbs rather than prepositions, suggesting that these novel prepositions embedded in larger noun phrases are poor candidates for focal stress (German et al., 2006).

To evaluate more fully the contextual conditions that may lead to focus on function words, experiment 2 featured a diverse array of function words (e.g., he, can, in) embedded in a wider range of sentence types than used by German and colleagues (2006). We asked silent readers in experiment 2 to respond quickly to passages in which function words were noteworthy and bore stylistic prominence (e.g., “While shaking their heads no, the ladies pointed down….We read THAT book”), as well as passages in which function words were nonsalient both contextually and stylistically (congruous trials). Congruous function stimuli were pitted against perfectly matched, incongruous function stimuli. Seeking converging evidence for an inner voice that flags new information, a subset of congruous and incongruous content stimuli from experiment 1 was included in experiment 2.

Based on Selkirk's (1986, 1996) theory of focus marking, we predicted that participants would more quickly read, and accurately judge, congruous function and content stimuli as helpful compared with matched, incongruous stimuli. We also reasoned that participants would have an advantage when swiftly judging content stimuli compared with function stimuli. As mentioned, content words are more commonly stress‐marked in English compared with function words. Moreover, among function words, there are auxiliary words (e.g., in, a) that rarely receive contextual emphasis (Weber, 2006), and such uncharacteristically noteworthy auxiliaries were included in experiment 2 (e.g., “I only wanted A cookie”). In light of these differences, we hypothesized that participants' judgments about function stimuli would not be as easy as their judgments about content stimuli. Differential effort could be manifested as slower response times and less accuracy.

Methods

General Discussion

Our writing system encodes our spoken language. English readers engage phonology when they read (Perfetti & Bell, 1991; Van Orden & Kloos, 2005). Our results suggest that readers also engage prosody when they read silently. Inspired by the central role that prosody plays in speech and oral reading, we investigated the extent to which skilled, silent reading represents prosodic focus. We exploited stylistic emphasis in print to elicit prosodic emphasis in the silent reader's inner voice. The results of experiments 1 and 2 suggest that our skilled, silent readers “heard” the prominence of new or important information, a natural form of highlighting. Consistent with linguistic theories of focus marking, silent readers of short stories in experiment 1 gave higher preference ratings when cap‐emphasized words congruously mapped onto new or important content and when background information was stylistically de‐emphasized, compared with matched, incongruous trials. Experiment 2 generalized the benefits of congruous stylistic emphasis and contextual importance to function words (in uncharacteristically newsworthy roles) and to a new behavioral measure (reaction time). The participants in experiment 2 tended to judge focus congruence more accurately in content than function stimuli, although this effect did not generalize across all experimental stimuli. In short, the inner speech of reading seemingly contains some of the prosodic richness typical of lively speech.

We found converging evidence that our skilled, silent readers seemingly “heard” prosodic focus. Our findings supplement the compelling evidence for the role of segmental phonology in skilled reading (Abramson & Goldinger, 1997; Acheson & MacDonald, 2009; Ashby & Clifton, 2005; Ashby et al., 2006; Gross et al., 2000; Hanson et al., 1991; Lukatela et al., 2004; McCutchen et al., 1991; Treiman, 1994; Treiman et al., 1995). Our findings complement recent evidence that silent readers extract suprasegmental prosodic features, such as lexical stress and metrical stress, when computing the relations between print and speech (Breen & Clifton, 2011), for when these readers encountered an inconsistency between the predicted meter and the required stress of a homograph, they suffered longer fixation times (Breen & Clifton, 2011). Silent reading has been shown to be influenced by implicit prosodic phrasing (e.g., Bader, 1998; Hwang & Schafer, 2009; Swets et al., 2007), punctuation (Cohen et al., 2001), and the voices of the story characters (Alexander & Nygaard, 2008; Kurby et al., 2009). Our experiments 1 and 2 go beyond previous research by demonstrating readers' sensitivity to focus prosody, in accordance with Selkirk's theory. Thus, there is growing evidence that suprasegmental prosodic sensitivity plays a role in skilled, silent reading.

Although our focus was on prosodic awareness in skilled, adult reading, other research reflects an increasing awareness that young children's reading proficiency includes prosodic awareness in addition to phonological awareness. Fast and efficient phonemic decoding of words is essential for learning English. Yet, the relation between letters and sounds is not one to one. Rather, a single sound can have several spellings (e.g., /k/ as in cat, bake, and back), and the same grapheme can be used to spell different sounds (e.g., thin, then). Prosodic units, like rimes, can be a more stable unit of analysis (Treiman et al., 1995).

In a longitudinal study, sensitivity to speech rhythm among 5–8‐year‐old children predicted their reading attainment and phrasing (a measure of fluency) one year later, after controlling for phonological processing skills, vocabulary, and age (Holliman, Wood, & Sheehy, 2010). In fourth graders, prosodic skills at the word and phrase levels (assessed by means of a reiterative speech task) predicted unique variance in the students' reading accuracy and reading comprehension, while controlling for general rhythmic sensitivity and phonological awareness (Whalley & Hansen, 2006). Contending that speech rhythm sensitivity may be a universal requirement when learning to read and write, Goswami and colleagues (2011) compared children with developmental dyslexia against matched samples of normal readers learning one of three very different spoken and written languages (English, Spanish, Chinese). The researchers found that rhythmic sensitivity uniquely predicted phonological awareness and reading in three different languages.

Extrapolating from Goswami et al (2011) findings on the importance of speech rhythm sensitivity in learning to read and write, we wonder if beginning or struggling readers of English might benefit from reading lessons that draw a more explicit link between the rhythm in speech and the rhythm in the writing system. Prosthetically enhanced poetry seems to be the perfect reading material for such a lesson. Poetry is an artistic example of regular stress patterning across connected text, and three types of poetic meter (trochaic, iambic, anapestic) accentuate the different, undulating rhythms of spoken English (Selkirk, 1986). In poetry, an iambic meter consists of an unstressed syllable followed by a stressed syllable (e.g., The man is small), a trochaic meter consists of a stressed syllable followed by an unstressed syllable (e.g., City lights are near), and an anapestic meter consists of two unstressed syllables followed by one stressed syllable (e.g., And the house is the place). Beginning readers might benefit from “seeing” the stress pulses of the trochaic rhythm that is common in children's poetry (e.g., “Twinkle, twinkle, little star”; Taylor & Taylor, 1849, p. 30). Reading tutorials could advance to less familiar rhythms (e.g., iambic: “did gyre and gimble in the wabe”; Carroll, 1917, p. 26) and advance further to obscure rhythms (e.g., anapestic: “The immortal desire of Immortals we saw in their faces, and sighed”; Yeats, 2010, p. 374).

Perhaps beginning readers' syllable segmentation abilities would profit from prosodically enhanced reading materials (e.g., but, butter, butterfly) while playing a modified tapping game (Liberman, Shankweiler, Fischer, & Carter, 1974). In the original game, children tapped out the number of segments (e.g., phonemes, syllables) in spoken utterances. After hearing the words, but, butter, and butterfly, the correct answers when counting syllables were one, two, and three taps, respectively. In the proposed game, young readers would be given reading materials that cap‐emphasize syllable beats (e.g., “Itsy Bitsy Spider”; North, 1910, p. 346). Then, the readers would play the game by tapping out the number, and type (stressed vs. unstressed), of syllables in words. The readers would be instructed to tap out beat sequences using stronger taps for stressed syllables and quieter taps for unstressed syllables. Thus, the correct answer when reproducing the meter in “Itsy Bitsy Spider” is “TAP‐tap, TAP‐tap, TAP‐tap.”

A diverse array of prosthetic‐prosody aids may uniquely benefit struggling readers, nonnative speakers, or readers challenged by difficult text. For example, prosthetic cap‐emphasis might aid the assignment of stress when silently reading (a) ambiguous heteronyms (e.g., PERmit vs. perMIT), (b) esoteric, multisyllabic words (e.g., esoTERic), (c) words that are spelled similarly but have different pronunciations and meanings (e.g., deFER vs. DIFfer), and (d) root words with pronunciation changes across derivations (e.g., DIFfer vs. differENtial). For example, we wonder if late speakers of English might have faster reading times and fewer pronunciation errors when provided with visual aids marking stress in ambiguous heteronyms, as in the following examples:

• The farmer's market usually has a wide variety of PROduce.
• Rapidly boiling water in the large pot will proDUCE hot steam.

Could college students prepare more efficiently for standardized vocabulary tests by studying from prosthetically enhanced word lists (e.g., aCERbic, abSCOND, aMALgamate, buCOLic, caCOPHony, caNONical, DESiccate, disPARage, ePHEMeral, GARrulous)?

Could prosthetic‐prosody aids facilitate the acquisition of function words among early readers? In reading research, function words are noted for their particular difficulty, including their irregular spellings, subtle meanings, weak stress assignment, and vowel pronunciation differences between citation form and connected text (Weber, 2006). Function words are read less accurately than content words in lists as well as connected text by normal and impaired readers (Blank, 1985), even though some function words are among the most frequent words appearing in print (Morgan, Shi, & Allopenna, 1996). In the proposed intervention, young readers might benefit from reading materials like the function stimuli in our experiment 2. For example, these readers might benefit from seeing newsworthy function words emphasized in print (e.g., “Sam asked, ‘Did you read THAT book?' Sally replied, ‘No, I read THIS book’”). Our future investigations will focus on some of these potential uses of prosthetically enhanced text.

Do skilled, adult readers represent metrical prosody in their inner voice? Our findings in experiments 1 and 2 are limited by our concentration on the role of focus marking in silent reading. Our future studies will swap prose for poetry to investigate the role of prosodic meter in the voice in the head. Extrapolating from linguistic theories of metrical alternation biases in English (Selkirk, 1986), skilled readers' ratings of the helpfulness of the stylistic enhancements of text should be significantly more favorable when stylistic emphasis and syllable stress are congruous (e.g., “Will be a totter'd weed of small worth held”; Shakespeare, 1997, p. 115) compared with incongruously matched stimuli (e.g., “Will be a totter'd weed of small worth held”). Additional evidence for a prosodic inner voice during silent reading could be gathered by using complementary methodologies. If the inner voice is melodious, then would silent reading times benefit by first listening to a song with a congruous beat (compared with an incongruous beat)? For example, could rap music (with a strong, predictable meter) prime the reading times of poetry or prose with an identical, implicit meter? Could silent reading times benefit by asking readers to first tap out a meter that is congruous (compared with incongruous) with the beat inherent in the text?

Future investigations could investigate whether contextual newness varies on a continuum from low to high newsworthiness, according to the degree of givenness theory (Baumann & Grice, 2006; Baumann et al., 2006). Such a study could explore the nature of the relationship between a referent and its anaphor (e.g., synonyms, hyponyms, hypernyms, converseness, part–whole relationships). For example, the immediate mention of a synonym should have low newsworthiness. Would the mention of a whole (e.g., hand) automatically activate its parts (e.g., fingers), rendering the parts as “old” information?

The integrity of our findings assumes a direct link between stylistic emphasis in print and implicit emphasis in the inner voice. We can probably never know for certain whether an inner voice guides silent reading. Our silent readers may have been responding to linguistic features that systematically covaried with our experimental manipulations of focus. Our claims for an expressive inner voice are strengthened by having used linguistic theory to make specific, testable predictions. By exploiting different stimuli (e.g., newsworthy content words, atypically newsworthy function words) and different measures of behavior (preference ratings, reaction times), we found converging evidence that our skilled, silent readers seemingly “heard” contextual focus, which is a feature of prosodic speech.

To conclude, prosodic speech is pervasive (e.g., infant‐directed speech, gifted college lectures, stage acting, voice of an irate mother). Our findings suggest that silent reading also captures aspects of this prosodic vigor.