The role of syllables in word recognition among beginning Finnish readers: Evidence from eye movements during reading

This article was downloaded by: [Turku University]On: 26 January 2015, At: 23:50Publisher: RoutledgeInforma Ltd Registered in England and Wales Registered Number: 1072954 Registered office: MortimerHouse, 37-41 Mortimer Street, London W1T 3JH, UK

Click for updates

Journal of Cognitive PsychologyPublication details, including instructions for authors and subscription information:http://www.tandfonline.com/loi/pecp21

The role of syllables in word recognition amongbeginning Finnish readers: Evidence from eyemovements during readingTuomo Häikiöa, Jukka Hyönäa & Raymond Bertrama

a Department of Psychology, University of Turku, FIN-20014 Turku, FinlandPublished online: 27 Nov 2014.

To cite this article: Tuomo Häikiö, Jukka Hyönä & Raymond Bertram (2014): The role of syllables in word recognitionamong beginning Finnish readers: Evidence from eye movements during reading, Journal of Cognitive Psychology,DOI: 10.1080/20445911.2014.982126

To link to this article: http://dx.doi.org/10.1080/20445911.2014.982126

PLEASE SCROLL DOWN FOR ARTICLE

Taylor & Francis makes every effort to ensure the accuracy of all the information (the “Content”)contained in the publications on our platform. However, Taylor & Francis, our agents, and our licensorsmake no representations or warranties whatsoever as to the accuracy, completeness, or suitabilityfor any purpose of the Content. Any opinions and views expressed in this publication are the opinionsand views of the authors, and are not the views of or endorsed by Taylor & Francis. The accuracy ofthe Content should not be relied upon and should be independently verified with primary sources ofinformation. Taylor and Francis shall not be liable for any losses, actions, claims, proceedings, demands,costs, expenses, damages, and other liabilities whatsoever or howsoever caused arising directly orindirectly in connection with, in relation to or arising out of the use of the Content.

This article may be used for research, teaching, and private study purposes. Any substantial orsystematic reproduction, redistribution, reselling, loan, sub-licensing, systematic supply, or distributionin any form to anyone is expressly forbidden. Terms & Conditions of access and use can be found athttp://www.tandfonline.com/page/terms-and-conditions

The role of syllables in word recognition amongbeginning Finnish readers: Evidence from eye

movements during reading

Tuomo Häikiö, Jukka Hyönä, and Raymond Bertram

Department of Psychology, University of Turku, FIN-20014 Turku, Finland

The eye movements of Finnish first and second graders were monitored as they read sentences wherepolysyllabic words were either hyphenated at syllable boundaries, alternatingly coloured (every secondsyllable black, every second red) or had no explicit syllable boundary cues (e.g., ta-lo vs. talo vs. talo =“house”). The results showed that hyphenation at syllable boundaries slows down reading of first andsecond graders even though syllabification by hyphens is very common in Finnish reading instruction, asall first-grade textbooks include hyphens at syllable boundaries. When hyphens were positioned within asyllable (t-alo vs. ta-lo), beginning readers were even more disrupted. Alternate colouring did not affectreading speed, no matter whether colours signalled syllable structure or not. The results show thatbeginning Finnish readers prefer to process polysyllabic words via syllables rather than letter by letter.At the same time they imply that hyphenation encourages sequential syllable processing, which slowsdown the reading of children, who are already capable of parallel syllable processing or recognisingwords directly via the whole-word route.

Keywords: Beginning reading; Eye movements; Reading development; Syllables; Word recognition.

To be able to read, readers need to understand howgraphemes and phonemes correspond to eachother. This is a central principle in developmentaltheories of reading (e.g., Ehri, 1995; Ehri & McCor-mick, 1998). For instance, according to Ehri’s (1995)model, after completing the pre-alphabetic phasereaders start to utilise some letter–sound pairings inthe partial-alphabetic phase before moving to thefull-alphabetic phase, in which they can fully utilisethese correspondences to read words letter by letter.When reading skill improves, readers start to utiliselarger reoccurring units such as syllables in a phasethat is called the consolidated-alphabetic phase.

The multiple-route model of orthographic pro-cessing (Grainger & Ziegler, 2011; see also

Grainger, Lété, Bertand, Dufau, & Ziegler, 2012)incorporates a similar view of reading develop-ment. According to this model, increasing readingskill offers more possibilities for word recognition.Beginning readers usually read words letter byletter, but as they become more skilled they donot have to rely solely on serial letter identificationand can also process letters in a word in parallel. Atthis point, they can use the fine-grained route whichutilises letter chunks with sensitivity to the preciseletter order. In other words, this route presumesthat letter chunks like syllables can be used asfunctional processing units. Parallel letter proces-sing also allows readers to make use of the so-calledcoarse-grained route, by which they can recognise

Correspondence should be addressed to Tuomo Häikiö, Department of Psychology, University of Turku, FIN-20014 Turku,Finland. E-mail: [email protected]

This work was supported by Suomen Kulttuurirahasto (Finnish Culture Foundation); Suomen Kulttuurirahaston Varsinais-Suo-men Rahasto (Finnish Culture Foundation, Varsinais-Suomi Regional Fund); Oskar Öflunds Stiftelse (Oskar Öflund Foundation) tothe first author.

© 2014 Taylor & Francis

Journal of Cognitive Psychology, 2014http://dx.doi.org/10.1080/20445911.2014.982126

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

words on the basis of letter clusters that are notnecessarily contingent. Thus, via this route moredirect access to whole-word representations can beaccomplished.

Several studies have demonstrated that syllablesare used by skilled readers in lexical access. Forinstance, Macizo and Van Petten (2007) showedthat syllable frequency affects word reading timesin English. Furthermore, Ashby (2010) showed inan event-related potential (ERP) study that phono-logical representations of syllables become acti-vated early on during visual word recognition.Similarly, Ashby andMartin (2008) found a syllablecongruency effect in amodified lexical decision taskwhere words were preceded by parafoveal pre-views that either were or were not congruent withthe initial syllable of the target word. Furthermore,Ferrand, Segui, and Humphreys (1997) foundsyllable priming effects in English. In addition,studies in French and English show reliable num-ber-of-syllable effects in lexical decision (Chetail,2014; Yap & Balota, 2009). Finally, the syllablestructure may—at least for relatively short words—be already extracted from the parafovea, as demon-strated by a larger skipping rate for five-lettermonosyllabic than five-letter disyllabic words (Fitz-simmons & Drieghe, 2011).

However, the evidence for the role of syllables inword processing is not unequivocal. For instance,Schiller (2000) failed to replicate the syllable prim-ing effect of Ferrand et al. (1997). Furthermore,even though there is evidence that increasing thenumber of syllables increases word recognitiontimes, this effect either disappears (e.g., Ferrand &New, 2003) or reverses (Yap & Balota, 2009) forhigh-frequency words. There is further evidenceshowing that word frequency modifies syllableeffects. Colé, Magnan, and Grainger (1999) founda syllable compatibility effect for low frequency(e.g., format) but not for high-frequency words(e.g., balcon) in an experiment where participantsresponded whether a carrier word (e.g., balcon)started with a previously presented target syllable(e.g., bal) or not. Ashby (2006) found that readingtimes for low frequency (e.g., detest) but not forhigh-frequency words were faster when the paraf-oveally presented preview was identical to theinitial syllable of the target word (de) than when itcontained an extra letter (det). These findingsimply that for adults the syllable is a functionalprocessing unit but more in low than in highfrequency words.

Several studies have found syllable effects forchildren as well (e.g., Colé et al., 1999; González &

Valle, 2000; Hautala, Aro, Eklund, Lerkkanen, &Lyytinen, 2012; Maïonchi-Pino, Magnan, & Écalle,2010). For instance, syllable compatibility effectswith high-frequency syllables were found forFrench first-grade readers; for third- and fifth-grade readers the same effects were found inde-pendent of syllable frequency (Colé et al., 1999;Maïonchi-Pino et al., 2010). Similarly, Gonzálezand Valle (2000) demonstrated that the presenceof frequent initial syllables facilitated beginningreaders’ word recognition. For Finnish, Hautalaet al. (2012) demonstrated a number-of-syllableeffects for dysfluent second-grade readers readingeasy words but found no reliable number-of-syllable effects for more proficient second graders;that is, they responded equally fast to six-letterbisyllabic words as to length-matched trisyllabicwords. All in all, the findings suggest that syllablesare used in word processing but that with increas-ing proficiency the role of the syllable may change.It is possible though that the role of syllables inpolysyllabic word processing is language-depend-ent, as syllable structure is more complex andsyllables boundaries more obscure in some lan-guages than others (for a comparison between 16languages, see Seymour, Aro, & Erskine, 2003).

The present study was conducted in Finnish.Finnish is a syllable-stressed language with straight-forward syllabification rules.1 In fact, Finnish is theleast complex European language with regard to thenumber of existing syllables (ca. 3,000 syllables),syllable structure (no heavy consonant clusters forinstance) and the unambiguity of syllable boundar-ies (Seymour et al., 2003). Furthermore, in Finnishthe primary stress is always on the first syllable of aword. Finally, Finnish has a transparent ortho-graphy with an almost perfect grapheme–phonemecorrespondence (Karlsson, 1999). Due to thesecharacteristics, it is not unlikely that Finnish readerswould make more frequent use of smaller readingunits in word recognition than is the case in otherlanguages. This would be in line with the psycho-linguistic grain size theory (Ziegler & Goswami,2005), which posits that readers of transparentorthographies make more frequent use of smallerunits in word recognition than readers of moreopaque orthographies. One could hypothesise thatfor Finnish syllables are likely to be involved inword recognition due to words being typically

1 In Finnish, a syllable boundary is always (1) before thelast letter of a consonant cluster or (2) between twodifferent vowels that do not constitute a diphthong.

2 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

polysyllabic and having simple syllable structureand clear syllable boundaries.

The importance of syllables in word recognitionis emphasised in Finnish reading instruction. Oneexample of this is that in all Finnish first-grade ABCbooks (be it mother tongue, religion or naturalsciences) the syllable structure of words is signalledby hyphens at syllable boundaries. From the sec-ond-grade onwards, hyphens do not appear inbisyllabic words anymore, but they are preservedin difficult and long words until the end of thesecond grade. The use of hyphenation reflects theidea that making children aware of syllables andsyllable structure aids the development of readingand writing.

However, despite the use of hyphenation inearly reading instruction, recent evidence showsthat hyphens within multi-unit words may disruptword recognition. For instance, Häikiö, Bertram,and Hyönä (2011) had second-grade readers readsentences containing compound words which wereeither hyphenated at the morpheme boundary(as prescribed by Finnish spelling regulations, e.g.,ulko-ovi = “front door”) or concatenated (sivuovi =“side door”). Hyphenation was beneficial forthe less proficient readers, but the more profici-ent second graders actually read concatenatedcompounds faster than hyphenated compounds.2

Häikiö et al. argue that the hyphen directs initialattention mainly to the first morpheme, and thatconsequently morphemes are processed more se-quentially than when they are presented without amorpheme boundary hyphen. They further arguethat sequential morpheme-based processing is dis-ruptive for proficient second graders who might beable to recognise these words as holistic units ormay gather information from both morphemessimultaneously. In contrast, the hyphen would bebeneficial for less proficient readers who processpolymorphemic words in a morpheme-by-morph-eme fashion anyway. In addition, it has been shownthat hyphenation at syllable boundaries slowsdown second graders despite its prominent use inreading instruction (Häikiö, Bertram, & Hyönä,2014). In this study, second-grade children readsentences with target words which were eitherconcatenated or hyphenated at the syllable level.

Häikiö et al. also manipulated word length andshowed that hyphenation was even more disruptivefor longer words with more than two syllables.They argue that hyphenation directs initial atten-tion to the first syllable and thus interferes withparallel access to multiple syllables or with directwhole-word access. In contrast, first graders in theearly stages of reading instruction are likely toprocess syllables sequentially during polysyllabicword recognition. If so, hyphens at syllable bound-aries would be facilitative, as the syllables areeasier to detect in hyphenated words. The currentstudy tested this hypothesis.

As mentioned earlier, hyphenation at syllablelevel disrupts reading—at least for second graders.Yet, there are other ways to visually signal syllableboundaries. One such cue is alternate colouring(every second syllable black, every second red). Infact, it has been established that both adults andchildren can use alternate colouring as a syllableboundary cue (e.g., Carreiras, Vergara, & Bar-ber, 2005; Chetail & Mathey, 2009; Prinzmetal,Hoffman, & Vest, 1991; Rouibah & Taft, 2001).Prinzmetal et al. (1991) demonstrated that whenparticipants were required to report the brieflypresented colour of a letter, they made more errorswhen the colour was incongruent with the syllablestructure than when it was congruent. Doignon andZagar (2006) replicated the findings of Prinzmetalet al. (1991) for first- to fifth-grade children.Carreiras et al. (2005) showed in an ERP study aP200 effect for incongruent syllable colouring.Finally, Chetail and Mathey (2009) showed thatincongruent syllable colouring slowed downweaker second-grade children in a lexical decisiontask. Interestingly, for more proficient secondgraders this effect was reversed. Chetail andMathey interpreted these findings as a demonstra-tion of syllable colouring activating syllableinformation. When readers do not have access tomany lexical candidates, as is the case for weakersecond graders, syllabic activation helps them toidentify the word swiftly. For proficient secondgraders syllabic activation leads to lexical competi-tion since the coloured syllables activate morewords than for weaker second graders, leading toslower word recognition. It may also be the casethat the more proficient readers were slowed downdue to being encouraged to access words viasyllables, whereas normal recognition would go byvia the whole-word orthographic route. In sum, thelexical decision and ERP studies indicate thatalternate colouring may facilitate word recognition.The current study investigated whether alternate

2 It can be argued that faster reading does not necessar-ily correlate with more proficient reading. However, in theHäikiö et al. (2011) study as well as—to foreshadow thepresent results—in the present study, reading skill (meas-ured by a standardised reading test) was connected to fasterreading speed.

THE ROLE OF SYLLABLES IN READING 3

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

colouring of syllables would facilitate word recog-nition during reading.

In sum, the current study investigates the role ofsyllables and syllable cues (hyphenation and altern-ate syllable colouring) in polysyllabic word recogni-tion. In Experiment 1, we investigated these issuesin early reading development by testing first graderswho had received only 2 months of formal readinginstruction. In Experiment 2, we investigated thedevelopment in polysyllabic word recognition bycomparing the reading performance of first andsecond graders.

EXPERIMENT 1

The evidence reviewed earlier suggests thathyphenation is disruptive for proficient Finnishsecond-grade readers but may be beneficial for lessproficient readers. In Experiment 1, we investigatedwhether first-grade readers benefit from hyphena-tion. In addition, we tested whether a visually lesssalient cue, alternate colouring of syllables (usingred and black, e.g., talo, with italics denoting the redcolour) would speed up reading among first graders.Participants read normal non-cued sentences andsentences where words contained explicit syllableboundary cues (hyphenation or alternate colours).We hypothesised that if the first graders havereached the consolidated-alphabetic phase (i.e.,they can make use of letter chunks like syllablesduring polysyllabic word recognition; see Ehri,1995), they should benefit from explicit syllableboundary cues.

However, it is also possible that first graders arestill in the full-alphabetic phase (Ehri, 1995) andprocess polysyllabic words on a letter-by-letter basis.In order to investigate this issue, we introducedconditions where hyphens or alternate colouringwere inserted within a syllable (t-alo, talo, with italicsdenoting the red colour); we called these conditions“illegal”, since the hyphens do not inform readersabout syllable boundaries as they usually do in first-grade textbooks. If syllables areutilised in polysyllabicword processing, inserting a hyphen or alternatingcolours at a non-syllable boundary should slow downreading in comparison to syllable boundary hyphena-tion. On the other hand, if first graders read words ona letter-by-letter basis, “illegal” hyphenation/alternatecolouring should not lead to a different pattern ofresults when compared to “legal” hyphenation/altern-ate colouring.

Finally, it is possible that first graders already uselarger units than syllables or access multiple

syllables in parallel, in which case syllable boundarycues would be disruptive. If so, hyphenation—beinga more salient visual cue—is expected to be moredisruptive than alternate colouring.

Method

Participants. Seventeen monolingual first graders(on average 7:5 years, range 7:0–7:10) were recruitedfrom a Finnish elementary school. At the time oftesting (October/November) they had receivedapprox. 2 months of formal reading instruction buthad learned to read to some extent already beforethis. All participants had normal or corrected-to-normal vision. None of them had previously partici-pated in an eye movement experiment. Permissionfrom children’s parents was acquired prior to theexperiment. The participants received candy orstickers as reward for participation.

Apparatus. Eye movements were recorded mono-cularly with a table-mounted model of Eyelink1000 (SR Research, Canada). A sampling rate of1,000 Hz was used. The eye-tracker is an infraredvideo-based tracking system with hyperacuityimage processing and spatial resolution of 0.5degrees. A chin rest was used to minimise headmovements. The texts were presented on a 20-inchViewSonic G225f computer screen (refresh rate of100 Hz, resolution 1,024 × 768).

Materials. The target word selection started with alist of words for which the age of acquisition (AoA)and familiarity ratings had been acquired earlier.The list was then complemented with additionalwords. For these words, the AoA and familiarityratingswere acquired from 12members of the TurkuUniversity community. For all the words, AoA wasrated on a 7-point scale (1= 0–2 years; 2 = 3–4 years; 3= 5–6 years; 4 = 7–8 years; 5 = 9–10 years; 6 = 11–12years; 7 > 12 years). Familiarity was rated on a 5-point scale (“How often do you encounter or use theword?” 1 = very seldom; 2 = seldom; 3 = sometimes; 4= often; 5 = very often). On the basis of these ratings,we selected 90 words for the experiment. Withrespect to AoA, all words were rated by at least80%of the raters with 4 or lower and had an averagerating below 4 (i.e., acquired before 7–8 years).

The 90 words were divided into three groupssince there were three experimental conditions(hyphenated, coloured and control). The groupswerematchedwith the other groups on word length,word frequency, average bigram frequency, initial

4 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

trigram frequency, family size (i.e., the number ofderivations and compounds beginning with thestem of the target word), AoA, and familiarity, eachp > .3. All values, apart from AoA and familiarity,were extracted from a newspaper corpus containing22.7 million word forms (Laine & Virtanen, 1999).3

The lexical statistics of the target words are pre-sented in Table 1.

The target words were embedded in sentences.Since there were three matched groups of 30target words, 30 sentence triplets were created. Asentence frame identical up to word N + 1 wasconstructed for each word triplet. The target wordnever appeared in the beginning or end of thesentence. The sentences of each triplet were ratedby nine Turku University employees to be equallynatural. In addition to the control conditionsentences, two versions of the sentences werecreated, one with a cue at the syllable boundariesin the target words, so-called legal syllable bound-ary cues, and one with cues within a syllable, so-called illegal syllable boundary cues. The otherwords in the sentences with cued target wordswere “legally” cued in both conditions, that is, allcues were at syllable boundaries. In the illegal cuecondition, for half of the target words the cue wasmoved one character towards to the beginning ofthe word, and for the other half one charactertowards the end of the word. The experimentalsentences were preceded by six practice sentences,followed by a block of 45 experimental sentences,containing 15 control condition sentences, 15sentences with legal target word cues (7 or 8coloured sentences and 8 or 7 hyphenated sen-tences) and 15 sentences with illegal target wordcues (8 or 7 coloured sentences and 7 or 8hyphenated sentences). The order of sentenceswithin each block was randomised for each parti-cipant. Each participant saw each target word onlyonce, presented either within the hyphenated,coloured or control sentences and either in thelegal or illegal condition. An example with the fiveconditions is presented in Table 2.

The sentences were presented in Courier fontso that each character position was of equal width.With a viewing distance of 60 cm, one characterspace subtended approximately 0.4 degrees of

visual angle. Sentences were always presentedhalfway between the top and the middle of thescreen. Sentences were all single-line sentenceswith a maximum of 67 characters per line.

After the eye movement experiment, the chil-dren’s reading skill was assessed with a classroomsubtest (word recognition) of Ala-asteen Lukutesti(ALLU) (Lindeman, 1998), a standardised Finnishreading test for elementary-school children. In thistask, children see a picture (e.g., of a pencil =kynä) and have to choose which of the four words,all phonologically similar to the correct word (e.g.,kylä = village; kyllä = yes; kynä = pencil; andkylmä = cold), corresponds to the picture. Thesubtest comprises 80 picture–word pairings. Thereis a time limit of 5 min, and one point is awardedfor each correctly chosen word. The descriptivestatistics of the ALLU test included in the analysesare presented in Table 3.

Procedure. The participants were instructed toread sentences for comprehension at their ownpace and were encouraged to read silently. Theywere further told that after varying intervals theywould get an oral question (after approx. every sixsentences) about a sentence to which they had togive a yes/no answer. The participants answeredthe questions with a minimum of 75% accuracy.Furthermore, the participants were told that textpresentation could be somewhat unusual but thatthey should nevertheless read the sentences totheir best ability. The eye-tracker was calibratedusing a 3-point calibration grid extending hori-zontally over the computer screen. Before pre-senting a target sentence, the participant fixatedon a calibration point at the left side of the screen,after which the sentence appeared.

3We acknowledge that a newspaper corpus is not idealfor indexing word frequencies for children. However, as wehad no access to a children’s corpus, we used it as anapproximation. Moreover, we collected AoA-ratings tomake sure that all the target words were familiar to theparticipants.

TABLE 1Lexical statistics of the target words and sentences

Measure Mean SD Min. Max.

Frequencya 83 148 0 929Length in characters 6.31 1.47 4 10Length in syllables 2.60 0.61 2 4Bigram frequencyb 7.81 3.18 0.84 14.00Initial trigram frequencyb 0.69 0.74 0.01 6.15Family size 105 124 3 649Age of acquisitionc 2.47 0.57 1.30 3.95Familiarityd 3.11 0.77 1.76 4.76Sentence length in characters(excluding hyphens)

38.98 5.73 29 54

Sentence length in words 5.74 0.70 5 7

aPer million; bPer thousand; cOn a scale of 1–7; dOn a scaleof 1–5.

THE ROLE OF SYLLABLES IN READING 5

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

Dependent variables and predictors. We used twostandard, word-based eye movement measures asthe dependent variables, gaze duration (i.e.,summed fixation durations on the target wordbefore exiting it for the first time) and go-pasttime (i.e., summed durations of fixations madebefore exiting the word to the right for the firsttime). Furthermore, we analysed one global meas-ure, namely sentence reading time. We analysedthe data twice. For the first analyses, we includedthe critical predictor Syllable Boundary Cue withthree levels, Control, Colour and Hyphen. Herelegality of the cue was not considered. In the secondanalyses, we excluded the control condition andentered next to Syllable Boundary Cue (which nowhad two levels, Colour and Hyphen) the factorLegality with two levels (Legal and Illegal).

Furthermore, for the two target word measures,we considered a number of variables that havebeen established as reliable predictors of wordprocessing: word length (Len), word frequency(Freq), AoA, word familiarity (Fmlrty) and bigramtrough ratio4 (Trough). We were also interested inhow the number of syllables affected the targetword reading. As expected, there was a highcorrelation between number of syllables andword length, so we computed the residuals ofnumber of syllables over word length and includedthe resulting variable as a predictor (Syl.res). Forboth the word and sentence measures, we includedone participant measure, reading level as assessedby the reading skill test (Allu). Despite the factthat there were many predictors, we had noconvergence problems within the models. The

kappa value measuring collinearity of the continu-ous predictors was below 8.5 in each model.

Statistical considerations. The measures were log-transformed to normalise the data. Furthermore,values 2.5 SDs smaller or larger than the grandmean were excluded from the duration measures.Finally, all the continuous predictor variables werecentred.

We used multiple regression mixed-effects mod-elling with participants and items as crossed ran-dom effects. Other variables did not improve therandom effect structure.Wewill only report modelswith the effects retaining statistical significance inthe stepwise backward elimination procedure. Inthis procedure, we first included all the predictors(including all relevant interactions) in the model.We then removed the least predictive predictor ineach round until we ended up with amodel in whichall the predictors were significant, |t| > 1.96. We alsomade sure by model comparison that each pre-dictor significantly improved the explanatory valueof the model. The analyses were conducted usingthe languageR library of R statistical software(R Development Core Team, 2007). The modelsreported in Appendix 1 present the output of thepvals.fnc() function of the library.

Results

Trials in which the target word was initiallyskipped were excluded from the analyses (3.2%

4A bigram trough (Seidenberg, 1987; Seidenberg &McClelland, 1989) is a letter pair of lower frequency thanthe preceding and following bigram. We assessed the effectof bigram trough and its interaction with explicit syllableboundary cues. To this end, we constructed a trough ratiousing the formula {[(preceding bigram frequency + follow-ing bigram frequency)/2]/bigram frequency at first syllableboundary}. Thus, words without a trough had a ratio below1 and words with a trough a ratio above 1.

TABLE 2An example sentence triplet

Condition Finnish sentence Translation

Control Äiti kertoi, että ikkuna oli likainen Mother said that the window was dirtyColour-legal Äiti kertoi, että aurinko oli kirkas Mother said that the sun was brightColour-illegal Äiti kertoi, että aurinko oli kirkas Mother said that the sun was brightHyphenated-legal Äi-ti ker-toi, et-tä naa-pu-ri o-li vi-hai-nen Mother said that the neighbour was angryHyphenated-illegal Äi-ti ker-toi, et-tä na-apu-ri o-li vi-hai-nen Mother said that the neighbour was angry

Target words are presented here in italics for illustrative purposes. Bolding denotes the red colour.

TABLE 3Descriptive statistics of ALLU reading performance in

Experiment 1

Mean (SD) Min. Max.

Raw scorea 35.6 (12.3) 19 58Reading levelb 4.8 (1.4) 3 7

aMaximum of 80; bOn scale 1–9.

6 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

of trials). Furthermore, the analyses wererestricted to first-pass sentence reading, whichended when the reader reached the last wordand when possible regressions were positionedbefore the second to last word in the sentence.No further data trimming was done. The non-transformed means for each measure are pre-sented in Table 4. The full models are presentedin Appendix 1. One participant was excluded dueto not completing the whole experiment, yielding16 participants to be included in the analyses.

Gaze duration. There was a significant differencebetween Control and Hyphen, t > 2; hyphenationdisrupted target word reading. The differencebetween Control and Colour was not significant,t < 1, whereas the contrast between Colour andHyphen was significant, t > 2 (not shown in themodel). There were also significant main effects ofLen, Fmlrty, Trough andAllu, all ts > 2. Shorter andmore familiar words shortened gaze duration, asdid bigram trough at the syllable boundary. Goodreaders had shorter gaze durations than weakerreaders. None of these variables interacted withHyphen or Colour.

Additional analyses did not reveal a main effectof Legality, but there was a tendency for a SyllableBoundary Cue × Legality interaction, t = 1.63. Thistendency reflected that words including hyphenswithin syllables elicited longer gaze durations thanwords with hyphens at syllables boundaries (t =2.30, 293 ms), whereas the difference betweencolours alternating within syllables or at syllableboundaries was negligible (t < 1, 20 ms).

Go-past time. The differences between Control andHyphen as well as Colour and Hyphen weresignificant, ts > 2. Hyphenation slowed down go-past times in comparison to the other two condi-tions, which did not differ from each other, t < 1.Moreover, the main effects of Syl.res,AoA, Troughand Allu were significant as well as the interaction

between Len and Hyphen, ts > 2. Go-past timeswere longer when the words had more syllables,were acquired later or did not have a bigram troughat the syllable boundary. Weak readers had longergo-past times than good readers. In addition, thedisruptive effect for hyphenation was enlarged inlong words in comparison to short words.

Additional analyses showed a significantHyphen×Legality interaction. Illegality disrupted reading ofhyphenated words, t = 3.57, but not of colouredwords, t =−.21. For hyphenated words the within-syllable manipulation elicited go-past times thatwere on average 484 ms longer than the syllableboundary manipulation, whereas the differencebetween within-syllable and syllable boundary col-our alternation was only 22 ms.

Sentence reading time. For these analyses, onlysentences with legally cued target words wereincluded. There was a significant differencebetween Control and Hyphen, t > 2, as well asbetween Colour and Hyphen, t > 2, indicating thatsyllable boundary hyphenation lengthened sentencereading times in comparison to the other twoconditions. There was no significant differencebetween Control and Colour, t < 1.96. Furthermore,there was a significant main effect of Allu, t > 2:Good readers read sentences faster than weakreaders. None of the variables interacted withHyphen or Colour.

Discussion

In Experiment 1, it was shown that hyphenationdisrupted reading both at word and sentence level.Moreover, colouring did elicit shorter reading timesthan hyphenation, but no disruption or facilitationappeared in comparison to the control condition.With regard to the legality of the syllable boundarycues, we found a Hyphenation × Legality interac-tion in go-past time as well as a trend for an

TABLE 4Means for each eye movement measure (in milliseconds) in Experiment 1, as a function of word presentation style and syllable

cue legality

Colour Hyphenated

Control Legal Illegal Legal Illegal

Gaze duration 1,284 (1,005) 1,248 (883) 1,268 (943) 1,313 (1,087) 1,606 (1,213)Go-past time 1,433 (1,036) 1,442 (1,032) 1,464 (1,099) 1,616 (1,262) 2,100 (1,581)Sentence reading time 6,630 (3,570) 6,948 (3,768) – 7,637 (3,961) –

For sentence reading time only legally cued sentences were included. Standard deviations are in parentheses.

THE ROLE OF SYLLABLES IN READING 7

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

interaction in gaze duration. Illegal hyphenationdisrupted reading more than legal hyphenation.

The results imply that already during the first-grade proficient beginning readers have movedpast the stage of accessing words on a letter-by-letter basis. If they still did so, the hyphen shouldhave affected word processing in the same wayindependent of the position where it is inserted.Since the hyphen at the syllable boundary clearlyfacilitates reading in comparison to hyphensinserted at non-boundary position, we can concludethat first graders can make use of syllables duringpolysyllabic word processing. Other effects point tothe same direction. Go-past time became longerwith increasing number of syllables, independent ofword length. In addition, another cue indicatingsyllable structure, namely the bigram trough at thefirst syllable boundary, speeded up word reading.Thus, in line with earlier research in other lan-guages (e.g., Ashby, 2010; Colé et al., 1999; Gon-zález & Valle, 2000; Maïonchi-Pino et al., 2010), itseems that syllables are indeed functional proces-sing units during lexical access among beginningFinnish readers.

However, the results also show that even thoughhyphenation at the syllable boundary might bebeneficial in comparison to hyphenation withinthe syllable, it is certainly not beneficial in compar-ison to the non-cued, “normal” presentation ofwords. This is best seen in the sentence readingtimes, where syllable boundary hyphenation eli-cited a 1-s delay in comparison to normal presenta-tion. How can this disruptive hyphenation effect beexplained?

One possibility is that visual acuity plays a role;the further the letters are away from the fixationpoint, the more the final letters of the word arevisually degraded. Since hyphens extend the lengthof a word, it brings the final letters further awayfrom the fixation point. Visual acuity constraintsmay indeed contribute to the 29 ms effect in gazeduration, even though it is unlikely that one or twohyphens more would generate an effect of this size.However, it can only account a small part of the 183ms effect in go-past time, as this measure includesregressions unlikely to be significantly influencedby visual acuity.

Häikiö et al. (2011) argued that in compoundword processing a hyphen at the constituent bound-ary of two-constituent compound words (e.g., ulko-ovi = front door) directs attention to the firstconstituent resulting in the processing of the secondconstituent being delayed. As a result, hyphenatedcompound words are processed constituent-by-

constituent, whereas short compound words with-out hyphens can be processed via whole-wordforms or via parallel processing of constituents.Following this line of reasoning, it may be arguedthat hyphenated polysyllabic words are processedsyllable-by-syllable, whereas their normal counter-parts can be processed holistically or via simultan-eous syllable processing.

Given the bigram trough and the number-of-syllable effects, onemay be tempted to interpret theresults to support the latter option. However, theresults also show whole-word related effects. Famil-iar words elicited shorter gaze durations than lessfamiliar words, and early acquired words elicitedshorter go-past times than late-acquired words (seereviews of Blythe & Joseph, 2011; Rayner, 1998,2009, for similar evidence). These effects indicatethat words may also be recognised as holistic unitswithout the involvement of syllables. The mostlikely scenario is that both syllable-based andholistic processing make a contribution to therecognition of polysyllabic words, in line with theparallel route architecture of the Grainger andZiegler’s (2011) model.

Alternate colouring of syllables did not disturbreading in comparison to normal presentation andgenerated faster reading times than syllable bound-ary hyphenation. This implies that alternate colour-ing does not direct attention to the first syllable in asimilar way as hyphenation does. Chetail andMathey (2009) showed that at least second graderswere affected by syllable colouring, be it positive(non-proficient second graders) or negative (profi-cient second graders). Our data show trends in bothdirections. First, there was a non-significant numer-ical trend of colouring slowing down reading.Second, there was a trend for a Colour × TrialNumber interaction (t = 1.85) in total fixationduration (a measure not reported here since ityielded very similar results to go-past time). Thetrend suggests that while alternate colouring dis-rupted word reading in the beginning of the experi-ment, it started to facilitate word reading towardsthe end of the experiment. To explore whetherthese trends could turn into significant effects, weincreased the statistical power in Experiment 2.

EXPERIMENT 2

A possible reason for beginning readers not relyingon syllable boundary cues in Experiment 1 may bethat these cues were partly unreliable, as illegalcues were also included. Therefore, in Experiment

8 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

2 we only included polysyllabic words with legalsyllable boundary cues. To increase statisticalpower, we doubled the number of trials per parti-cipant. As a result, we expected to obtain largereffects of both hyphenation and alternate colour-ing. Furthermore, since Chetail and Mathey (2009)demonstrated that the effect of alternate colouringwas facilitatory for weaker second graders andinhibitory for more proficient second graders, weexpected to find an interaction with reading skill.To this end, and to directly assess the developmentof polysyllabic word processing from first to secondgrade, children of both grades were included inExperiment 2.

Since early reading development proceeds at arapid pace, we tested the same children twice atdifferent stages. In this way we could examinepossible developmental changes within the sameparticipant groups. We also examined whetherhyphenation is more beneficial in oral than silentreading. Since syllables are emphasised in Finnishreading and writing instruction, signalling syllablestructure may facilitate oral reading more thansilent reading where words are not mouthed. Oralreading was included in the study also for the reasonthat beginning readers are more used to oral thansilent reading in early reading instruction.

Method

Participants. Twenty monolingual first graders (onaverage 7:3 years during the first testing, range 6:10–7:9) and 21 monolingual second graders (on average8:6 years during the first testing, range 7:10–8:9) wererecruited from a Finnish elementary school. At thetime of the fall testing (October/November) the firstgraders had received approx. 2 months and thesecond graders approx. 1 year and 2 months offormal reading instruction. Practically all first gradershad learned to read before entering school. Allparticipants had normal or corrected-to-normal vi-sion and participated in an eye movement experi-ment for the first time. Permission from thechildren’s parents was acquired prior to the experi-ment. All the participants were also tested duringspring (February/March). The participants receivedcandy or stickers as reward for participation.

Apparatus. The eye-tracker was identical to theone used in Experiment 1.

Materials. For the fall testing, the sentences fromExperiment 1 were used with the difference that the

participants read all 90 sentences instead of 45sentences. For the spring testing, half of the wordswere embedded in new sentences, while half of theold sentences were used to make sure that theobserved effects were not context-dependent.The target words were matched between the oldand new sentences on the same variables as inExperiment 1, each p > .2, apart from familiarity,which was slightly higher for the old than the newmaterial (3.30 vs. 2.93, respectively). The newsentences were constructed following the samecriteria as in Experiment 1. Each new sentencewas rated to be equally natural with the othersentences of the triplet by eight Turku Universityemployees.

The experiment proper was preceded by sixpractice sentences and included 90 sentences forfirst graders and 120 sentences for second graders,divided into two blocks of equal size. First gradersread only the experimental sentences, whereas thesecond graders also read 30 filler sentences. Thesentences were assigned to three matched listscounterbalanced across participants. Each particip-ant saw 30 hyphenated, 30 coloured and 30 non-cued sentences. The block order was counterba-lanced across participants. The sentence order wasrandomised for each participant within each block.For the fall testing, each participant saw each targetword only once, either in the hyphenated, colouredor control condition. In the spring testing, eachparticipant saw the target words in the same formatas in the fall testing, with the only difference beingthat half of the sentences were new.

After the fall experiment, the children’s readingskill was assessed with a classroom subtest ofALLU (Lindeman, 1998). For first graders, theword recognition task was used (see Experiment 1).Second graders completed a sentence comprehen-sion task where they saw a picture (e.g., peoplesailing) and had to choose which of the foursentences (e.g., “They swim”, “They dive”, “Theysail” and “They dance”) corresponds to the picture.The subtest comprises 20 picture–sentence pair-ings. There is a time limit of 120 s, and one point isawarded for each correctly chosen sentence. Thedescriptive statistics of the ALLU test are pre-sented in Table 5.

Procedure. The eye-tracker calibration and sen-tence presentation were performed identically toExperiment 1. The instructions about reading andcomprehension questions were almost the same asin Experiment 1 the main difference being thatchildren got instructions about the oral reading

THE ROLE OF SYLLABLES IN READING 9

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

session as well. The participants answered thequestions with a minimum of 75% accuracy. Forthe fall experiment, half of the participants readthe first block orally and the second block silently,while the other half did the opposite. For thespring experiment, each participant did the oppos-ite; those who had previously first read orally nowread the first block silently and the second blockorally, and vice versa. For both testing sessions,there was a short optional break between blocks.

Design and data analyses.There were a few changesin the analyses of Experiment 2 in comparison tothose of Experiment 1. First, the legality of thesyllable boundary cue was not manipulated andtherefore not included as a factor. Second, anumber of new predictors were included in theanalyses:Grade (1 vs. 2), reading mode (Mode: oralvs. silent), sentence novelty (Novelty: new vs. old)and experiment (Xp: Experiment 1, fall vs. Experi-ment 2, spring) as predictors. Third, because firstand second graders completed different Allu subt-ests, standardised scores were used in the analyses.The kappa value measuring collinearity of thecontinuous predictors was below 5 in each model.The models were constructed in a similar way as inExperiment 1.

Results

Trials in which the target word was initiallyskipped were excluded from the analyses (4.9%

of trials). Furthermore, the analyses wererestricted to the first-pass sentence reading as inExperiment 1. No further data trimming was done.The non-transformed means for each measure arepresented in Table 6. The full models and t valuesof the predictors can be found in Appendix 2.Seven 1st graders and four 2nd graders wereexcluded due to excessive track loss or not com-pleting the whole experiment after getting tootired. In the end, 13 first graders and 17 secondgraders were included in the analyses. In thefollowing, we will not discuss results for Mode,Novelty and Xp since, they did not interactsignificantly with the Syllable Boundary Cue,even though the effects of Mode and Xp weresignificant in each measure, all ts < −5.5 (forNovelty, all ts < 1.36). Silent reading was overallfaster than oral reading and reading was generallyfaster at the second than at the first testing pointfor each measure. Because of the lack of interac-tions involving Xp, Mode and Novelty, we col-lapsed all of the data in the same analyses toincrease statistical power. In case Syllable Bound-ary Cue interacted with Grade, we followed up theinteraction by testing the effect of Syllable Bound-ary Cue individually for both grades.

Gaze duration. The effect of Hyphen was qualifiedby significant interactions with Allu and Grade, ts> 2. The interaction with Allu indicated that themore proficient readers were more disrupted byhyphenation than the less proficient ones. With

TABLE 5Descriptive statistics of ALLU reading performance in Experiment 2

First grade Second grade

Mean (SD) Min. Max. Mean (SD) Min. Max.

Raw scorea 36.2 (13.0) 19 66 7.8 (2.5) 4 13Reading levelb 4.7 (1.6) 3 8 2.8 (1.6) 1 6

aMaximum of 20 for first grade, 80 for second grade; bOn scale 1–9.

TABLE 6Means for each eye movement measure (in milliseconds) in Experiment 2, as a function of word presentation style and grade

Control Colour Hyphenated

First grade Second grade First grade Second grade First grade Second grade

Gaze duration 1,305 (1,229) 939 (769) 1,326 (1,200) 931 (742) 1,337 (1,170) 1,012 (786)Go-past time 1,461 (1,324) 1,033 (880) 1,496 (1,335) 1,043 (860) 1,579 (1,388) 1,177 (929)Sentence reading time 6,647 (4,086) 4,698 (2,551) 6,778 (4,238) 4,732 (2,479) 7,440 (4,439) 5,394 (2,763)

Standard deviations are in parentheses.

10 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

regard to Grade, hyphenation disrupted secondgraders while the effect was non-significant forfirst graders (Control vs. Hyphen, t = 1.72 for firstgraders and t = 3.03 for second graders; Colour vs.Hyphen, t = .64 for first graders and t = 3.85 forsecond graders). There was neither a differencebetween Control and Colour nor an interactioninvolving Colour, ts < 1. The main effects of Len,Syl.res, AoA and Freq were significant. Wordswere read faster when they were frequent or earlyacquired. Furthermore, gaze durations increasedas a function of word length and number ofsyllables.

Go-past time. Hyphen interacted significantly withLen, Allu and Grade, ts > 2. The disruptive effectof hyphenation was larger for long than for shortwords. With regard to reading level (Allu),hyphenation disrupted more proficient readersmore than less proficient ones. Finally, hyphena-tion disrupted second graders to a larger degreethan first graders (Control vs. Hyphen, t = 3.57,Cohen’s d = .110 for first graders and t = 7.91 forsecond graders, Cohen’s d = .220; Colour vs.Hyphen, t = 2.24, Cohen’s d = .169 for first gradersand t = 7.27, Cohen’s d = .257 for second graders).For Colour, there was neither a significant differ-ence with the Control condition nor interactions, ts< 1. Finally, there were significant main effects ofSyl.res, Fmlrty and AoA. Go-past time was longerwhen the word was late acquired, less familiar orcontained several syllables.

Sentence reading time. The effect of Hyphen wasqualified by significant interactions with Allu andGrade, ts > 2. Hyphenation disrupted more profi-cient readers to a larger extent than less proficientones and second graders more than first graders(Control vs. Hyphen t = 6.66, Cohen’s d = .201 forfirst graders and t = 11.08, Cohen’s d = .288 forsecond graders; Colour vs. Hyphen t = 5.58,Cohen’s d = .167 for first graders and t = 9.83,Cohen’s d = .257 for second graders). The differ-ence between Colour and Control was not signi-ficant, t < 1.

Discussion

For the most part, the main results of Experiment1 were replicated in Experiment 2. The colouralternation condition yielded similar reading timesas normal presentation in every measure, andinteractions involving colour did not come close

to significance. We conclude that at least innormal sentence reading colour alternation doesnot provide a sufficiently strong cue to affectpolysyllabic word recognition.

In contrast, and similarly to Experiment 1,hyphenation disrupted polysyllabic word proces-sing in all measures. In addition, we found thatsecond graders were disrupted by hyphenation toa larger degree than first graders. In gaze durationthe hyphenation effect was present only for secondgraders; in go-past and sentence reading time itwas larger for second than first graders. If habitu-ation to hyphenation would have been an issue,second graders should have been less disturbed byhyphens than first graders, as they have beenexposed to hyphenated textbooks at school for alonger time. A similar pattern as for grade wasfound for reading proficiency, with stronger dis-ruptive effects of hyphenation obtained for moreproficient readers.

Regarding the other predictor variables,increase in word length was associated with longergaze durations, the number of syllables with bothincreased gaze durations and go-past times, lateacquired words with longer gaze durations and go-past times than early acquired words, and familiarwords with shorter go-past times than less familiarwords. These effects are similar to the onesobtained in Experiment 1 and in line with earlierresearch (see reviews of Blythe & Joseph, 2011;Rayner, 1998, 2009). Unlike in Experiment 1,there was no effect of bigram trough. Finally, asexpected, readers were overall faster during thesecond testing which took place approx. 4 monthslater than the first testing. The improvement inreading skill was considerable: First graders readon average 60 words per minute (wpm) in the falland 85 wpm in the spring term; second graderswent up from 85 wpm in the fall to 107 wpm in thespring.

With regard to oral vs. silent reading, wewitnessed the usual finding (for a review, seeRayner, 1998) that oral reading is slower than silentreading. However, reading mode did not modify theimpact of syllable boundary cues. Against ourexpectations, syllable boundary cues did not facilit-ate oral reading, even though it can be hypothesisedthat syllables—by virtue of having strong phonolo-gical representations—play a more prominent rolein oral than in silent reading. However, we notedthat among beginning readers oral and silent read-ing may not qualitatively differ from each other.Even when asked to read silently, they oftenmouthed words and were even whispering what

THE ROLE OF SYLLABLES IN READING 11

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

they were reading. This was quite pronounced forfirst graders but also apparent for second graders.Thus, even though syllables are used as wordrecognition units, reading mode does not modifytheir role and the hyphen is equally disruptive inboth oral and silent reading.

GENERAL DISCUSSION

The present study demonstrated that hyphenationdisrupts first graders more when hyphens appearwithin a syllable (t-alo) than when they appear at asyllable boundary (ta-lo). We took this as evidencethat Finnish first graders do not recognise wordssolely on a letter-by-letter basis anymore, but thatthey make use of syllabic units in written wordrecognition and hence have moved past thealphabetic phase (see Ehri, 1995). However, theresults indicate that this is only part of the story, ashyphens at syllable boundaries clearly disruptpolysyllabic word processing of beginning Finnishreaders. Most strikingly, the disruptive hyphena-tion effect is already found in the very early stagesof reading instruction, in the beginning of the firstgrade.

As we argued in Häikiö et al. (2011), the hyphenis a visually very salient cue that clearly divides theword into separate parts. It is likely that inhyphenated polysyllabic words visual attention isinitially mainly allocated to the first syllable demar-cated from the rest of the word by a salient visualcue and attentional allocation to the subsequentsyllables is therefore delayed. In addition, as thevisual acuity decreases the further the letters areaway from the fixation point, the more the finalletters of the word are visually degraded. Thisproblem is magnified the longer the word is andthe more hyphens the word includes, as is the casewith the trisyllabic and quadrosyllabic words usedin this study. The interactions between hyphen andword length in go-past time in Experiments 1 and2 of the present study and in Häikiö et al. (2014)indeed reflect that hyphenation becomes moreproblematic with increased word length. However,the visual acuity account does not explain thesizeable effect in go-past times. Moreover, it doesnot explain why the disruptive effect of hyphena-tion is larger for second graders than first graders,even though the perceptual span gradually devel-ops during the elementary school years (Häikiö,Bertram, Hyönä, & Niemi, 2009). It thus seems thathyphens encourage children to process syllables

within words serially even when they are capable ofprocessing more than one syllable at a time.

As argued in the Introduction, recognition ofpolysyllabic words in Finnish may take differentshapes. First, it is possible that children recognisewords via syllables by simultaneously accessingmore than one syllable. The number-of-syllableeffects obtained in Experiments 1 and 2 supportthis interpretation. However, we also found anAoA and familiarity effect, which point to holisticprocessing of polysyllabic words.

In Figure 1 we adapted the model of Graingerand Ziegler (2011) to describe how the develop-ment of reading a polysyllabic word like talo(=house) may take place. It is clear that at thebeginning of reading development children startout using Route 1 and read words via serial letteridentification and phonological recoding (seeGrainger et al., 2012, for experimental evidence),a stage called the alphabetic stage by Ehri (1995).As reading skill develops, words may start to berecognised via the so-called fine-grained route(Route 2) whereby readers make use of frequentlyco-occurring letter clusters like syllables. We sus-pect that initially this route may be phonologicallymediated, so the syllables ta- and -lo may addressthe phonological representation/talo/, before theorthographic representation and the meaning ofthe word are retrieved. However, eventuallywords may be processed entirely via the ortho-graphic routes. Also, as argued earlier, syllablesmay be addressed sequentially or simultaneously,with the latter option related to a more advancedstage in reading development. In this moreadvanced stage, hyphens at constituent boundarieswill become disruptive.

At the most advanced stage of reading devel-opment, orthographic representations may beaccessed via the holistic or coarse-grained routewithout phonological involvement. At this stageletters would be recognised in parallel and directlymapped onto an orthographic representation(Route 3b) or mediated—as Grainger and Ziegler(2011) suggest—via open bigrams (combinationsof letters that are not necessarily contingent).5

Routes may operate in parallel and the balancebetween the routes may depend on factors such asword frequency. We argue that hyphenation isdisruptive for either route, as it narrows down the

5Open bigram coding is not the only possible mechan-ism for extracting the relative position of letters in words(for an alternative coding scheme, see Gomez, Ratcliff, &Perea, 2008).

12 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

initial focus on the first syllable, and therebyeliminates a number of open bigrams and/or thepossible direct mapping of the whole stimulus ontothe whole-word orthographic representation.

The present results suggest that Finnish firstand second graders recognise words mainly via thefine-grained route and the more direct routes(Routes 3a and 3b). Less proficient readers makemore use of the fine-grained route than moreproficient readers, and—given that they are lessdisturbed by hyphenation—they may sometimesaccess syllables serially. It is nevertheless slightlysurprising that the balance is already so muchshifted towards the more direct routes for ourearly first graders, as they have been exposed toformal reading instruction for a few months only.However, it should be noted that the majority ofthe children were able to read to some extentalready when entering the elementary school.Moreover, the shallow orthography and the clearsyllabic system in Finnish allows for quick readingdevelopment (see Seymour et al., 2003). Pro-longed use of hyphens at syllable boundaries inFinnish ABC books may therefore well disturb ordelay reading development rather than facilitate it.

REFERENCES

Ashby, J. (2006). Prosody in skilled silent reading: Evid-ence from eye movements. Journal of Research in

Reading, 29, 318–333. doi:10.1111/j.1467-9817.2006.00311.x

Ashby, J. (2010). Phonology is fundamental in skilledreading: Evidence from ERPs. Psychonomic Bulletin& Review, 17(1), 95–100. doi:10.3758/PBR.17.1.95

Ashby, J., & Martin, A. E. (2008). Prosodic phonolo-gical representations early in visual word recognition.Journal of Experimental Psychology: Human Percep-tion and Performance, 34, 224–236. doi:10.1037/0096-1523.34.1.224

Blythe, H. L., & Joseph, H. S. S. L. (2011). Children’s eyemovements during reading. In S. P. Liversedge, I. D.Gilchrist, & S. Everling (Eds.), Oxford handbook oneye movements (pp. 643–662). Oxford: Oxford Univer-sity Press.

Carreiras, M., Vergara, M., & Barber, H. (2005). Earlyevent-related potential effects of syllabic processingduring visual word recognition. Journal of CognitiveNeuroscience, 17, 1803–1817. doi:10.1093/brain/122.5.839

Chetail, F. (2014). Effect of number of syllables in visualword recognition: New insights from the lexicaldecision task. Language, Cognition and Neu-roscience, 29, 1249–1256. doi:10.1080/23273798.2013.876504

Chetail, F., & Mathey, S. (2009). The syllable frequencyeffect in visual recognition of French words: A studyin skilled and beginning readers. Reading and Writ-ing, 22, 955–973. doi:10.1007/s11145-008-9135-9

Colé, P., Magnan, A., & Grainger, J. (1999). Syllable-sized units in visual word recognition: Evidence fromskilled and beginning readers of French. AppliedPsycholinguistics, 20, 507–532. doi:10.1017/S0142716499004038

Doignon, N., & Zagar, D. (2006). Les enfants en coursd’apprentissage de la lecture perçoivent-ils la syllableà l’écrit? [Can children perceive the syllable in

Figure 1. Model of Grainger and Ziegler (2011) adapted to elucidate the reading development of Finnish children when readingpolysyllabic words like talo (=house). Adapted with permission from Figure 6 of Grainger and Ziegler (2011, p. 8). © 2011 byGrainger and Ziegler.

THE ROLE OF SYLLABLES IN READING 13

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

written words during reading acquisition?]. CanadianJournal of Experimental Psychology, 60, 258–274.

Ehri, L. C. (1995). Phases of development in learning toread words by sight. Journal of Research in Reading,18, 116–125. doi:10.1111/j.1467-9817.1995.tb00077.x

Ehri, L. C., & McCormick, S. (1998). Phases of wordlearning: Implications for instruction with delayedand disabled readers. Reading and Writing Quarterly,14, 135–163. doi:10.1080/1057356980140202

Ferrand, L., & New, B. (2003). Syllabic length effects invisual word recognition and naming. Acta Psycholo-gica, 113, 167–183. doi:10.1016/S0001-6918(03)00031-3

Ferrand, L., Segui, J., & Humphreys, G. W. (1997). Thesyllable’s role in word naming. Memory & Cognition,25, 458–470. doi:10.3758/BF03201122

Fitzsimmons, G., & Drieghe, D. (2011). The influence ofnumber of syllables on word skipping during reading.Psychonomic Bulletin & Review, 18, 736–741.doi:10.3758/s13423-011-0105-x

Gomez, P., Ratcliff, R., & Perea M. (2008). The overlapmodel: A model of letter position coding. Psycholo-gical Review, 115, 577–600. doi:10.1037/a0012667

González, J., & Valle, I. (2000). Word identification andreading disorders in the Spanish language. Journal ofLearning Disabilities, 33, 44–60. doi:10.1177/002221940003300108

Grainger, J., Lété, B., Bertand, D., Dufau, S., & Ziegler,J. C. (2012). Evidence for multiple routes in learning toread. Cognition, 123, 280–292. doi:10.1016/j.cognition.2012.01.003

Grainger, J., & Ziegler, J. C. (2011). A dual-routeapproach to orthographic processing. Frontiers inPsychology, 2, 54. doi:10.3389/fpsyg.2011.00054

Häikiö, T., Bertram, R., & Hyönä, J. (2011). The devel-opment of whole-word representations in compoundword processing: Evidence from eye fixation patternsof elementary school children.Applied Psycholinguist-ics, 32, 533–551. doi:10.1017/S0142716411000208

Häikiö, T., Bertram, R., & Hyönä, J. (2014). The role ofsyllables in word recognition among 2nd grade chil-dren: An eye movement study. Manuscript submittedfor publication.

Häikiö, T., Bertram, R., Hyönä, J., & Niemi, P. (2009).Development of the letter identity span in reading:Evidence from the eye movement moving windowparadigm. Journal of Experimental Child Psychology,102, 167–181.

Hautala, J., Aro, M., Eklund, K., Lerkkanen, M.-K., &Lyytinen, H. (2012). The role of letters and syllables intypical and dysfluent reading in a transparent ortho-graphy. Reading and Writing: An InterdisciplinaryJournal, 26, 845–864. doi:10.1007/s11145-012-9394-3

Karlsson, F. (1999). Finnish: An essential grammar. Lon-don and New York, NY: Routledge.

Laine, M., & Virtanen, P. (1999). WordMill lexicalsearch program. Turku: Center for Cognitive Neu-roscience, University of Turku.

Lindeman, J. (1998). ALLU: Ala-asteen Lukutesti[ALLU: Reading test for elementary school]. Turku:Center for Learning Research, University of Turku.

Macizo, P., & Van Petten, C. (2007). Syllable frequencyin lexical decision and naming of English words.Reading and Writing, 20, 295–331. doi:10.1007/s11145-006-9032-z

Maïonchi-Pino, N., Magnan, A., & Écalle, J. (2010).Syllable frequency effects in visual word recognition:Developmental approach in French children. Journalof Applied Developmental Psychology, 31, 70–82.doi:10.1016/j.appdev.2009.08.003

Prinzmetal, W., Hoffman, H., & Vest, K. (1991). Auto-matic processes in word perception: An analysis fromillusory conjunctions. Journal of Experimental Psycho-logy: Human Perception and Performance, 17,902–923. doi:10.1037/0096-1523.17.4.902

Rayner, K. (1998). Eye movements in reading andinformation processing: 20 years of research. Psycho-logical Bulletin, 124, 372–422. doi:10.1037/0033-2909.124.3.372

Rayner, K. (2009). Eye movements and attention inreading, scene perception, and visual search. TheQuarterly Journal of Experimental Psychology, 62,1457–1506. doi:10.1080/17470210902816461

R Development Core Team. (2007). R: A languageand environment for statistical computing. Vienna.Retrieved from http://www.R-project.org

Rouibah, A. & Taft, M. (2001). The role of syllabicstructure in French visual word recognition. Memory& Cognition, 29, 373–381. doi:10.3758/BF03194932

Schiller, N. O. (2000). Single word production inEnglish: The role of subsyllabic units during phono-logical encoding. Journal of Experimental Psycho-logy: Learning, Memory, and Cognition, 26, 512–528.doi:10.1037/0278-7393.26.2.512

Seidenberg, M. (1987). Sublexical structures in visualword recognition: Access units or orthographicredundancy? In M. Coltheart (Ed.), Attention andperformance XII: The psychology of reading (pp. 245–263). Hillsdale, NJ: Lawrence Erlbaum.

Seidenberg, M. S., & McClelland, J. L. (1989). Adistributed, developmental model of visual wordrecognition and naming. Psychological Review, 96,523–568. doi:10.1037/0033-295X.96.4.523

Seymour, P. H. K., Aro, M., & Erskine, J. M. (2003).Foundation literacy acquisition in European ortho-graphies. British Journal of Psychology, 94, 143–174.doi:10.1348/000712603321661859

Yap, M. J., & Balota, D. A. (2009). Visual word recogni-tion of multisyllabic words. Journal of Memory andLanguage, 60, 502–529. doi:10.1016/j.jml.2009.02.001

Ziegler, J. C., & Goswami, U. (2005). Reading acquisi-tion, developmental dyslexia, and skilled readingacross languages: A psycholinguistic grain sizetheory. Psychological Bulletin, 131(1), 3–29.doi:10.1037/0033-2909.131.1.3

14 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

APPENDIX 1

Final models for each measure in Experiment 1. t valuesbelow −1.96 and above 1.96 correspond to p values below.05. If an interaction was significant, its main effects arereported as well. Please note that these effects are notindependently interpretable in the lmer() output. Weseparately present the model with Legal in go-past timesince it was the only measure where we got a significantinteraction.

APPENDIX 2

Final models for each measure in Experiment 2. t valuesbelow −1.96 and above 1.96 correspond to p values below.05. If an interaction was significant, its main effects arereported as well. Please note that these effects are notindependently interpretable in the lmer() output.

TABLE A1Experiment 1, gaze duration: Fixed effects

Estimate SE t value

(Intercept) 7.805974 0.290164 26.902Colour 0.001357 0.053361 0.025Hyphen 0.110735 0.05384 2.057Trough −0.023036 0.011 −2.094Fmlrty −0.073988 0.033092 −2.236Len 0.121075 0.017961 6.741Allu −0.025279 0.007571 −3.339

Gaze duration values have been log-transformed.

TABLE A2Experiment 1, go-past time: Fixed effects

Estimate SE t value

(Intercept) 7.874534 0.289333 27.216Colour 0.027308 0.042896 0.637Hyphen 0.248257 0.043273 5.737Len 0.080213 0.023696 3.385Allu −0.023364 0.007501 −3.115AoA 0.134561 0.041154 3.270Syl.res 0.192788 0.074844 2.576Trough −0.023552 0.010120 −2.327Colour: Len 0.031835 0.029758 1.070Hyphen: Len 0.077072 0.029951 2.573

Go-past time values have been log-transformed.

TABLE A3Experiment 1, go-past time with legality included: Fixed effects

Estimate SE t value

(Intercept) 7.862555 0.298296 26.358Hyphen −0.123957 0.135828 −0.913Legal −0.004445 0.060218 −0.074Syl.res 0.179259 0.088654 2.022Len 0.134976 0.019319 6.987AoA 0.117217 0.048151 2.434Trough −0.025479 0.011954 −2.131Allu −0.021953 0.007382 −2.974Hyphen: Legal 0.231410 0.085697 2.700

Go-past time values have been log-transformed.

TABLE A4Experiment 1, sentence reading time: Fixed effects

Estimate SE t value

(Intercept) 9.666621 0.285504 33.86Colour 0.049952 0.026349 1.90Hyphen 0.167758 0.026552 6.32Allu −0.028639 0.007588 −3.77

Sentence reading time values have been log-transformed.

TABLE A5Experiment 2, gaze duration: Fixed effects

Estimate SE t value

(Intercept) 8.464647 0.218885 38.67Colour −0.036877 0.062597 −0.59Hyphen −0.159727 0.062633 −2.55Allu −0.285896 0.039770 −7.19Grade −0.769127 0.139736 −5.50Syl.res 0.120882 0.050117 2.41Mode −0.132324 0.013760 −9.62Freq −0.025361 0.010505 −2.41Xp −0.282228 0.013763 −20.51Len 0.106148 0.010851 9.78AoA 0.139794 0.027706 5.05Colour: Allu 0.012766 0.011374 1.12Hyphen: Allu 0.041675 0.011395 3.66Colour: Grade −0.003973 0.039809 −0.10Hyphen: Grade 0.126237 0.039814 3.17

Gaze duration values have been log-transformed.

THE ROLE OF SYLLABLES IN READING 15

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15

TABLE A6Experiment 2, go-past time: Fixed effects

Estimate SE t value

(Intercept) 8.709415 0.232893 37.40Colour −0.006386 0.055403 −0.12Hyphen −0.061819 0.055481 −1.11Len 0.116766 0.012954 9.01Allu −0.305500 0.042324 −7.22Grade −0.854613 0.148718 −5.75Mode −0.162171 0.012188 −13.31Syl.res 0.186051 0.052251 3.56AoA 0.112354 0.030413 3.69Xp −0.293758 0.012190 −24.10Fmlrty −0.045520 0.022392 −2.03Colour: Len 0.005947 0.010305 0.58Hyphen: Len 0.027136 0.010327 2.63Colour: Allu 0.007378 0.010070 0.73Hyphen: Allu 0.030909 0.010095 3.06Colour: Grade −0.000362 0.035252 −0.01Hyphen: Grade 0.130117 0.035266 3.69

Go-past time values have been log-transformed.

TABLE A7Experiment 2, sentence reading time: Fixed effects

Estimate SE t value

(Intercept) 8.46032 0.22276 37.98Colour −0.0645 0.06585 −0.98Hyphen −0.15204 0.06596 −2.3Allu −0.28519 0.04027 −7.08Grade −0.76567 0.14147 −5.41Xp −0.28653 0.01444 −19.85Mode −0.1285 0.01443 −8.9Colour: Allu 0.01482 0.01198 1.24Hyphen: Allu 0.03992 0.012 3.33Colour: Grade 0.02344 0.04182 0.56Hyphen: Grade 0.12345 0.04183 2.95

Sentence reading time values have been log-transformed.

16 HÄIKIÖ, HYÖNÄ, BERTRAM

Dow

nloa

ded

by [

Tur

ku U

nive

rsity

] at

23:

50 2

6 Ja

nuar

y 20

15