Introduction

Competent reading requires fast and automatic recognition of individual words. Research on adults shows that efficient visual word recognition involves rapid decomposition of morphemic constituents (Amenta & Crepaldi, 2012; Rastle, Davis, & New, 2004). For instance, in a seminal study, Taft and Forster (1975) showed that in a lexical decision paradigm, rejecting nonwords consisting of existing prefixes and stems (e.g., dejuvenate) is harder than rejecting nonwords consisting of an existing prefix and a nonexisting stem (e.g., depertoire). This interference effect indicates that morphological decomposition takes place prior to lexical access as the non-existing stem in depertoire renders an additional check of the legitimacy of prefix-stem combination unnecessary.

While morphological processing during visual word recognition has been extensively investigated in adults, developmental mechanisms are as yet largely unclear. Although we know that from about 2 years on children apply inflectional as well as derivational morphology productively in their spoken language (Berko, 1958; Penke, 2012) and can successfully perform morphological awareness tasks at the age of seven (e.g., Kirby et al., 2012), much less is known about how automatic morphological decomposition develops during reading acquisition.

There is evidence that children as young as 7 years read morphologically derived words (e.g., hilly) faster and more accurately than matched monomorphemic, “pseudoderived” words (e.g., silly, Carlisle & Stone, 2005), indicating that they make implicit use of the morphological structure during word recognition. Similarly, across grades 1 to 8, students are more likely to complete a word fragment (e.g., H M) with a primed word when the prime is morphologically related (e.g., HARMED, HARMFUL), then when the prime is orthographically, but not semantically related (e.g., HARMONY). Morphological effects were equally large for inflected (HARMED) and derived (HARMFUL) words (Deacon, Campbell, Tamminga, & Kirby, 2010; Feldman, Rueckl, DiLiberto, Pastizzo, & Vellutino, 2002; Rabin & Deacon, 2008). Recently, Dawson, Rastle, and Ricketts (2018) replicated the morphological interference effect (Taft & Forster, 1975) in lexical decision among children and adolescents. However, while all age groups (7 years to adults) were less accurate in rejecting pseudomorphemic nonwords (e.g., earist) than control nonwords (e.g., earilt), only adults and older adolescents (16 to 17 years) also showed higher response times for pseudomorphemic than control nonwords, indicating developmental change.

A central paradigm used in many studies on morphological processing during reading is masked priming. The crucial evidence is that response times to targets (e.g., CAR) are shorter for morphologically related (e.g., cars) than for unrelated primes that are similar to targets on either semantic (e.g., bus) or orthographic (e.g., cap) dimensions (e.g., Beyersmann, Casalis, Ziegler, & Grainger, 2015; Beyersmann, Duñabeitia, Carreiras, Coltheart, & Castles, 2013; Beyersmann, Iakimova, Ziegler, & Cole, 2014; Crepaldi, Rastle, Coltheart, & Nickels, 2010; Diependaele, Sandra, & Grainger, 2005, 2009; Duñabeitia, Perea, & Carreiras, 2008; Longtin & Meunier, 2005). Based on this paradigm, there is a current debate whether the priming effect is form-based or meaning-based. Proponents of the form-based, morpho-orthographic segmentation view claim that initially the morphological decomposition process in visual word recognition is based solely on orthographic information, prior access to meaning. Semantic information is activated only in a later phase of visual word processing (Rastle & Davis, 2008). This hypothesis is based on the finding that against an unrelated baseline, similar priming effects were observed for pseudo-morphologically suffixed words (e.g., corner—CORN) and for true morphologically suffixed words, (e.g., darkness—DARK). A number of studies found support for this theory, suggesting that early morphological processing is semantically blind (Beyersmann, Castles, & Coltheart, 2011; Lavric, Clapp, & Rastle, 2007; Lavric, Elchlepp, & Rastle, 2012; Lázaro, Illera, & Sainz, 2016; Longtin & Meunier, 2005; Longtin, Segui, & Hallé, 2003; McCormick et al., 2008, 2009; Rastle, Davis, & New, 2004).

Contrarily, the second hypothesis suggests that both morpho-orthographic and morpho-semantic information affect the early processing of morphologically complex words. This view mostly rests on studies reporting a larger magnitude of true morphological priming (e.g., darkness—DARK) than of pseudo-morphological priming (e.g., corner-CORN), indicating semantic contributions during the initial decomposition of words into their morphemic constituents (e.g., Diependaele, Dunabeitia, & Keuleers, 2011; Feldman, Kostić, Gvozdenović, O’Connor, & del Prado Martín, 2012; Feldman, O’Connor, & Moscoso del Prado Martín, 2009; Feldman, Soltano, Pastizzo, & Francis, 2004).

Interestingly, there is evidence for developmental change of the basis of morphological decomposition, but findings are mixed: English children aged 7 to 10 years showed priming for true, but not for pseudosuffixed words, while older children and adults showed similar facilitation in both conditions (Beyersmann et al., 2012). Similarly, German children in Grades 2 to 5 showed target facilitation from real suffixed words (kleidchen—KLEID), suffixed nonwords (kleidtum—KLEID), and nonsuffixed nonwords (kleidekt—KLEID), relative to unrelated control primes (träumerei—KLEID). However, in contrast to adults, they did not show differences between suffixed and nonsuffixed conditions (Hasenäcker, Beyersmann, & Schröder, 2016, 2020). These findings suggest that English and German elementary school children may not (yet) make use of morpho-orthographic segmentation. On the other hand, evidence for early morpho-semantic as well as morpho-orthographic priming was reported for French (Quémart, Casalis, & Colé, 2011) and Hebrew (Schiff, Raveh, & Fighel, 2012).

Obviously, this inconsistency may be related to differences between the languages and orthographies investigated. Most studies on morphological processing in children have been obtained for English and French, deep orthographies with complex and often opaque correspondences between graphemes and phonemes. In deep orthographies readers employ larger grain size units (Prior, 2012) and knowledge of morphemes is important for reading acquisition (Verhoeven & Perfetti, 2003), therefore morphological effects are not unexpected. Another language feature that is likely to be relevant is morphological structure. Particularly English morphology is not extensively productive compared to other languages (Meunier & Segui, 2002).

Effects of morphological processing during visual word recognition have also been demonstrated for a range of other languages including German (Hasenäcker et al., 2016), Dutch (Perdijk, Schreuder, Baayen, & Verhoeven, 2012); Brasilian Portuguese (De Oliveira & Reis Justi, 2017), Spanish (Lázaro, Camacho, & Burani, 2013), and Italian (Burani et al., 2002). While these languages vary largely in grapheme-phoneme consistency as well as morphological complexity, each of them has homophonic word spellings which sound the same and can only be differentiated based on morpho-semantic knowledge (e.g., Engl.: to—two—too; German: viel—fiel [engl.: many—fell], Italian: lago—l´ago [engl: lake—nail]. This is not the case for Bosnian, which is one of the few orthographies with a one-to-one relationship between letters and sounds. The high transparency of Bosnian orthography (Duranovic, 2017) may lead to the assumption that morphology does not have a strong influence on reading because reliance on grapheme-phoneme correspondence rules is sufficient (Defior, Martos, & Herrera, 2008). However, Bosnian is a morphologically rich language, particularly with respect to inflectional morphology. Changes in word forms are typically generated by adding monosyllabic suffixes (consisting of a vowel only or a vowel-consonant combination (Lukatela, Mandić, Gligorijević, Kostić, & Savić, 1978). Almost 800 different suffixes are available for word formation (Cedic, 2001). Lukatela et al. (1978) postulated that nouns are not represented in the lexicon with their grammatical cases. Instead, root morphemes for all nouns, as well as the small set of inflectional morphemes, are stored in the lexicon. Appropriate combinations of a root and its inflections are determined by separately stored syntactic rules. We argue that parsing a word into morphemic units would be of great value for languages with such a rich morphology. In a language with a complex morphological system, where every word can have numerous inflectional forms, the representation of each word as a whole unit would be cognitively extremely costly and recognition of words would be impossible if a particular inflectional form is encountered for the first time. Therefore, decomposing a word into its constituent morphemes would be useful (Lehtonen & Bryant, 2005). Thus, the present study aims to investigate whether morphological effects on word reading may be observed in Bosnian language with its highly transparent orthography and rich morphology, and if so, how such effects might change across different levels of age and reading experience.

Another characteristic of reading acquisition in Bosnian language is also particularly interesting in the context of morphological priming effects. Conflicting evidence regarding morpho-orthographic and morpho-semantic decomposition in the early stages of visual word processing may partly be due to the high orthographic similarity between prime and target words. Although target words are usually presented in upper case letters in order to make them less similar to the lower case prime, there is still considerable overlap for many letters (see, for example the first two letters in corner—CORN). Bosnian language is an interesting test case here, as two different alphabetic writing systems are used in Bosnia and Herzegovina, which have only minimal visual overlap: Latin is the more frequent writing system, which is taught in Grade 1 and is mostly used in everyday contexts. However, from Grade 3 on all children are taught the Cyrillic alphabet and practice reading and spelling throughout their school career. By presenting the very same set of primes in Latin as well as Cyrillic, we aimed to reduce orthographic overlap between (Cyrillic) primes and (Latin) targets to a minimum.

In our masked priming experiment, we presented each target with Latin as well as Cyrillic primes. Target words presented in Latin (the more familiar orthography) were preceded by either (a) a suffixed word prime in Latin (e.g., žena-žene), (b) the same suffixed word prime in Cyrillic (e.g., žena- жeнe), (c) an unrelated prime in Latin (e.g., žena–igra), or (d) the same unrelated prime in Cyrillic (e.g., žena– игpa). If morphological units are accessed during word reading in the highly transparent Bosnian orthography, we expected to see significant facilitation of morphological compared to unrelated primes. If early morphological processing during visual word recognition is semantically blind (e.g., Longtin & Meunier, 2005; Rastle et al., 2004), we would expect no significant effect of morphological priming for suffixed word primes in Cyrillic, as there is no orthographic overlap between prime and target. If, however, semantic information has an influence on the initial decomposition of words into morphemic constituents (e.g., Diependaele et al., 2011; Feldman et al., 2004, 2009, 2012), morphological priming for suffixed word primes should be obtained in Cyrillic as well as Latin.

In order to examine developmental mechanisms of morphological decomposition, we compared priming effects in fourth-grade children who have only just mastered the grapho-phonemic code of Cyrillic, sixth- and eighth-grade children and adults with different levels of accumulated orthographic lexicon in the two scripts (Latin and Cyrillic). The priming paradigm also allows to investigate whether morphological effects occur early or late during visual word processing (Casalis et al., 2009). In order to identify the timeline of priming effects for morphologically related primes, short (50 ms) and longer (250 ms) prime durations have been used. Morphological priming effects in adults have been found even for very short Stimulus Onset Asynchronies (SOAs) of 30 to 60 ms (Frost, Kugler, Deutsch, & Forster, 2005; Rastle et al., 2004). In the present study we applied a short prime duration of 50 ms, which should be sufficient to ensure that the youngest group of 4th graders could process the prime. Especially for the primes written in the less familiar writing system of Cyrillic, it seemed important to introduce a longer prime duration of 250 ms SOA as well, which allowed us to contrast early vs. late effects.

Methods

Participants

A total of 320 participants, all native speakers of Bosnian, took part in a priming experiment. Four age groups with 40 participants each were recruited: Students of Grades 4 (27 male and 13 female), 6 (20 male and 20 female), and 8 (20 male and 20 female), from two elementary schools located in Tuzla, in the northeastern part of Bosnia and Herzegovina, and a group of students from the University of Tuzla (4 boys and 36 girls). All participants had normal or corrected-to-normal vision. The Ministry of Education of Tuzla Canton approved all study procedures. Parents gave written permission for participation of their children and university students volunteered for the study. From each age group, half of the participants performed the experiment with the 50 ms prime time, and the other half performed the experiment with the 250 ms prime time. Participant characteristics are presented in Table 1.

Table 1 Number of words read correctly in 1 min for each grade group
Full size table

Participants were typically developing readers with average reading skills assessed with a 1-min word reading test (Duranovic, 2013), presenting a list of 120 familiar Bosnian words (increasing in length and difficulty) in Latin script. The raw score was the number of words read correctly in 1 min. Since we were interested in typical reading development, poor readers who performed below the 25th percentile on the Latin 1-min reading test, were not admitted to the study. In order to get an impression of participants' reading abilities in Cyrillic, the items from the Latin test were converted to Cyrillic and participants were again asked to read as many words as possible within 1 min. No norms are available for the Cyrillic test version. Still, Table 1 shows that participants had reasonable reading abilities in Cyrillic script, although the number of words read correctly was almost twice as high in Latin in the younger groups and still about 40% higher in the older age groups.

Design and stimuli

Sixty-one Bosnian nouns in singular form were selected as targets (see “Appendix”). Their mean surface frequency (the number of times that a given word appears, Lau, Rozanova, & Phillips, 2007) was 303 occurrences per 1.5 million, based on the Oslo Corpus of Bosnian Texts (1997, http://www.Tekstlab.uio.no/Bosnian/Corpus.html#cont). Thirty-one of the targets were free stems (which are existing words themselves; e.g., “sat”) and 30 targets were bound stems (the base morpheme of the noun is not identical to any existing word, i.e. „sob”, and the singular noun is created by adding an inflectional suffix “sob” +  “a” =  “soba”). Target words were always presented in lower case Latin letters and paired with three different primes in Latin (morphological, orthographic, and unrelated) and two different primes in Cyrillic (morphological and unrelated). In the morphological condition, primes constituted suffixed forms of the targets. For the 31 free-stem targets, plural noun primes were created by adding a suffix (e.g., “sat + i” =  “sati”) and for the 30 bound-stem targets plural noun primes were created by replacing the singular suffix (e.g., “sob - a” +  “e” =  “sobe”). In the orthographic condition, primes comprised words that were morphologically unrelated but orthographically similar to the targets (e.g., “glava” and “glavni”), where the target and prime shared letters within the word or they shared either the initial or the final letters (see Colombo, 1986, for similar procedures). About 70% of the orthographic primes shared three or more letters with the target. In the unrelated condition, primes comprised words that were completely unrelated to the targets (e.g., “glava” and “ništa”). However, note that in many cases, the morphological primes had a larger visual overlap with the target word than the orthographic primes due to restrictions of the lexicon. To ensure that orthographic primes provided a sufficient level of control for orthographic similarity Levenshtein distance was used for assessing the orthographic similarity of prime and target. The Levenshtein distance between two words is the minimum number of single-character edits (insertions, deletions or substitutions) required to change one word into the other (Levenshtein, 1966). As evident from Table 2, Levensthein distance was indeed somewhat higher for the orthographic than the morphological condition, and this difference was significant (p < .001). Not surprisingly, Levenshtein distance was also clearly higher for unrelated primes than for the two other conditions (both ps < .001). The three sets of Latin primes were matched pairwise for length, stem, suffix, and bigram frequency (see Table 2).

Table 2 Item characteristics per condition (Latin)
Full size table

The Cyrillic prime conditions used the words of the morphological and unrelated prime conditions in Cyrillic script. As orthographic overlap for Cyrillic items was minimal, no orthographic condition was created for Cyrillic. All words had regular grapheme-phoneme correspondences.

Each participant was assigned to a prime duration condition (50 or 250 ms) and was administered all conditions (three with Latin primes and two with Cyrillic primes) in two different sessions, so each participant was presented with all priming conditions. The experiment manipulated the five priming conditions (Latin morphological, orthographic and unrelated as well as Cyrillic morphological and unrelated) as repeated measures, and the prime duration (50 ms vs. 250 ms) and age (Grades 4, 6, 8 and university students) as group factors.

Procedure

Participants were tested individually, seated approximately 40 cm in front of a monitor in a quiet room. Stimulus presentationwas controlled by PsychoPy software (Peirce, 2007; 2009). Vocal responses were recorded by a microphone fitted to each participant by means of a headset. SV-1 Voice Key was used as a device designed specifically for experiments requiring a vocal response.

Participants were told that they would see a series of hash tags (####) followed by words presented in lowercase letters. Their task was to read aloud the words as quickly and as accurately as possible. The presence of primes was not mentioned to the participants. Stimuli were presented to each participant in a different random order, following ten practice trials. Each trial started with the presentation of a forward mask (####) that remained on the screen for 500 ms. The prime in Latin or Cyrillic was then presented in lowercase letters for 50 ms or 250 ms, followed by the target (always in Latin lowercase letters) which acted as a backward mask to the prime. The target words appeared in black on a white background (12-point Arial font) and remained on the screen for 2000 ms or until participants responded, which ever happened first. Response times were measured from appearance of the target word on the screen until the voice key was triggered.

Results

Due to the high transparency of Bosnian orthography, reading accuracy was close to ceiling across age groups, with only 0.06% errors on average. Thus, reading accuracy will not be further analysed.

Response latencies were analyzed for correct responses only. During preliminary data analysis, it turned out that due to an error in pseudorandomization, participants saw some items more than once while others were not presented. We thus checked carefully to what extent this error affected data quality. It turned out that between 1773 and 2365 items (instead of the originally planned 61 items × 40 participants per age group = 2440 items) were presented per condition and age group. The minimum number of items that any participant saw in one condition was 28 (out of 61) and the minimum number of participants per age group that saw a particular item ranged from 3 to 11 (out of 20). As a sufficient number of data points was available in each condition, we proceeded with the analysis. For items that were presented more than once, only the first presentation was included.

Data inspection also indicated, that a considerable number of responses (17.5% in the 50 ms prime condition and 22.4% in the 250 ms prime condition) were shorter than 300 ms. We assume that these RTs resulted from faulty voice key responses and excluded them from analysis. We also excluded responses above 3 SDs of the grade level mean (0.8% of the responses for 50 ms priming time and 0.9% for 250 ms priming time). Table 3 presents the number of items per condition and age group for which RTs were available after data trimming.

Table 3 Number of items per condition after data timing
Full size table

Reaction times (RTs) of accurate responses were analyzed by generalized linear mixed-effect models (GLMMs) in R (R Core Team, 2017) using the inverse Gaussian distribution and identity link function (Lo & Andrews, 2015). GLMMs were fitted by the mixed function of the afex package (version 0.18-0, Singmann, Bolker, Westfall, & Aust, 2016), which is based on the package lme4 (version 1.1-14, Bates, Maechler, Bolker, & Walker, 2015). In a first model we wanted to test whether priming effects in our sample are in line with earlier studies and focused on the Latin script prime conditions only. For this analysis, we entered grade (fourth, sixth, eighth, university), prime time (50 ms, 250 ms) and condition (morphological, orthographic, unrelated) as fixed effects including interaction terms. In the second model, we directly compared priming effects for the two different scripts (Latin vs. Cyrillic) for those conditions that were presented in both scripts (morphological vs. unrelated). More specifically, grade (fourth, sixth, eighth, university), prime time (50 ms, 250 ms), script (Latin, Cyrillic) and condition (morphological, unrelated) were entered as fixed effects including interaction terms. In each model we entered subjects and items as crossed random effects for which we specified random slopes for each within-subjects and within-items factor, respectively (see Barr, Levy, Scheepers, & Tily, 2013). Note that this type of analysis controls for effects of over additivity between age groups. To achieve better convergence, we removed random correlations (Barr et al., 2013). Likelihood ratio tests were used to compute p values for all fixed effects as implemented in the mixed function. To decompose significant main and interaction effects post hoc contrasts were performed using the lsmeans package (version 2.27-61, Lenth, 2016) and the Benjamini–Hochberg method for adjusting p values (Benjamini & Hochberg, 1995).

Prime effects for Latin morphological, orthographic and unrelated prime conditions

First we fit a GLMM for Latin script prime conditions only: Grade × Prime Time × Condition (Latin: morphological, orthographic, unrelated) (see Table 4). Figure 1 presents boxplots for each grade level and priming condition. We observed significant main effects for Grade, Prime time, and Condition, and a significant interaction between Prime time and Condition. All contrasts between grade levels were significant (4th–6th: p = .0078; all other contrasts p ≤ .0001) except the contrast between 8th grade and and University students (p = .09). Contrasts were also significant between all Latin conditions (morphological—orthographic: p < .0001; morphological—unrelated: p < .0001; orthographic—unrelated: p = .0001), faster in the orthographic than in the unrelated condition, and again faster in the morphological than the orthographic as well as unrelated condition. Post hoc contrasts for the Prime time × Condition interaction showed highly significant differences between morphological and orthographic as well as unrelated conditions for both prime durations (p < .0001). The contrast between the orthographic and the unrelated condition was significant at p = .008 in the 50 ms prime duration and marginally significant in the 250 ms prime duration (p = .057)..

Table 4 Results of the GLMM for Latin script prime conditions only
Full size table
Fig. 1
figure 1

Box plot of reaction times (RTs) as a function of grade, prime time and condition in Latin only. The median is represented as band inside the box. Individual points indicate outliers and stars indicate extreme outliers

Full size image

While the differences between conditions were suggestive of morphological priming, it should be noted that priming condition was systematically confounded with visual-orthographic overlap between prime and target in our item set. This is particularly crucial for the comparison of morphological and orthographic conditions. In order to check whether the difference between these two conditions was indeed due to morphological segmentation, we ran a second GLMM model for these two conditions only, in which we introduced Levenshtein Distance as an additional factor (see Table 5). From this analysis, we excluded items with a Levenshtein Distance of 5 or 6 (comprising only 438 data points altogether), which only occurred in the orthographic condition. In this Grade × Prime Time × Condition × Levenshtein Distance model, Levenshtein Distance did neither show a significant main effect (p = .31), nor any interactions, while the condition effect was still highly significant (p < .0001).Thus, even when the visual overlap between prime and target is matched across the two conditions, participants profited more strongly from a prime that had a morphemic overlap with the target compared to a prime that was orthographically similar, but morphologically unrelated. The differences between Grade levels were again significant (all ps < .0001, except 4th–6th grade: p = .031, and 8th grade—university students: p = .019)..

Table 5 Results of the GLMM for Latin script, morphological and orthographic conditions only including Levenshtein distance
Full size table

In summary, the analysis of Latin conditions clearly confirmed earlier findings on morphological priming effects. Across all age groups investigated, RTs were always lower for target words that were preceded by a morphologically related prime compared to an unrelated prime. Importantly, while orthographic primes also induced lower RTs than unrelated primes, this effect was significantly smaller than the priming induced by morphological primes. The interaction with grade level was not significant (p = .61), suggesting that in the highly consistent Bosnian orthography morphological and orthographic priming effects are established by Grade 4. Even though the effect of morphological relatedness was still evident for items that were identical in orthographic overlap (as quantified by Levenshtein Distance), it is important to note, that the difference between Latin prime conditions was most likely influenced by differences in orthographic overlap. Thus, it will be interesting to see whether similar effects can be found for primes in Cyrillic script, which show the same morphological relatedness with target words, but no orthographic overlap.

Prime effects for morphological versus unrelated primes in Latin and Cyrillic

In another GLMM, we directly compared priming effects for the two scripts: Grade × Prime Time × Condition (morphological, unrelated) × Script (Latin, Cyrillic). Mean scores are presented in Fig. 2. The analysis revealed significant effects of Grade, Condition, and Script (see Table 6). Furthermore, the interactions Prime Time × Condition, Condition × Script, and Prime Time × Condition × Script were significant. All grade contrasts were significant (4th–6th: p = .0069; 8th—University: p = 0029, all other contrasts: p < .0001). Overall, response times in the morphological prime condition were significantly lower than in the unrelated prime condition (p < .0001) and response times were higher in Cyrillic compared to Latin (p < .0001).

Fig. 2
figure 2

Box plot of reaction times (RTs) as a function of grade, prime time, condition and script. The median is represented as band inside the box. Individual points indicate outliers and stars indicate extreme outliers

Full size image
Table 6 Results of the GLMM directly comparing priming effects for the two scripts
Full size table

The interactions with script were of particular relevance in this analysis: Although the condition effect was clearly smaller for primes in Cyrillic than in Latin (see Fig. 2), the contrast between morphological and unrelated condition was overall still significant for Cyrillic primes (p = .0087). However, when we ran contrasts separately per prime duration, the condition effect was only marginally significant in Cyrillic (50 ms: p = .08; 250 ms: p = .06) but still highly reliable in Latin (both ps < .0001).

Discussion

The main aim of the current study was to investigate morphological decomposition during visual word recognition in the highly transparent Bosnian orthography. This is important, as it might be assumed that in an orthography with highly transparent and reliable grapheme-phoneme correspondences, fast and automatic processing of larger junks like morphemes might not be needed. Nevertheless, in Latin, the script Bosnian readers are most familiar with, we found significantly lower response times for morphologically related primes than for unrelated as well as for orthographically similar primes. Thus, just like in more complex orthographies like English (e.g., Deacon, et al., 2010; Feldman, et al., 2002; Rastle, Davis, Marlsen-Wilson, & Tyler, 2000) or French (Casalis et al., 2009; Quémart, Casalis, & Colé, 2011), fluent word recognition in Bosnian entails fast and automatic decomposition of morphologically complex words.

In our study, priming effects were similarly large for short primes of only 50 ms duration and longer primes of 250 ms duration, indicating that morphological decomposition happens very early during the word recognition process (Beyersmann et al., 2011). With respect to the important issue of developmental mechanisms we observed that response times for word recognition decreased across the investigated age groups (Grades 4, 6, 8 and adults), while priming effects were mostly similar across age groups. Thus, our evidence indicates that in Bosnian orthography, morphological decomposition is well established after 4 years of formal reading instruction, while we did not find strong evidence for major developmental changes later on. It is, however, possible that our cross-sectional design was not sufficiently sensitive to identify minor developmental changes. It is also possible that a constant priming effect in relation to systematically decreasing reading times actually reflects changes in the developmental mechanisms underlying the priming effects. It will be important to further investigate developmental mechanisms of morphological decomposition longitudinally, starting with younger age groups. The main reason why we did not include younger age groups in the current study is that we wanted to exploit the fact that Bosnian language is written in two scripts, Latin and Cyrillic. While Latin is the dominant script, all children learn and practice reading in Cyrillic from Grade 3 on. Thus, we assumed that after more than 1 year of practice, fourth grade children should have sufficient reading skills to show priming effects in the less familiar writing system of Cyrillic as well. Investigating priming effects in Cyrillic is highly interesting, as it can help to differentiate between priming effects that are induced by visual-orthographic overlap and those that have a truly morpho-semantic basis. Indeed, we found a significant difference between morphological and unrelated priming conditions in both orthographies in both scripts, but this effect was clearly smaller in Cyrillic than in Latin script. When the condition effect was analysed separately per prime condition, it was actually no longer significant in Cyrillic, perhaps due to power issues.

It is quite impressive that a morphological priming effect is evident for Cyrillic primes, which do not have any visual overlap with the target, providing strong evidence that the effect is morphology based. In an earlier study with Serbian adults, Feldman et al. (2012) had actually reported similarly large facilitation effects for Latin and Cyrillic scripts. The fact that we found stronger effects for Latin than Cyrillic may be related to the larger visual overlap in the Latin compared to Cyrillic primes. We assume that it is also due to the fact that Cyrillic writing is more frequently used in Serbia than in the Federation of Bosnia and Herzegovina where the study was conducted. In order to exploit the bi-scriptal situation of Bosnian, future studies could reverse the script of primes and targets and present primes always in the more familiar Latin script, while targets could be presented in Latin as well as Cyrillic. This would ensure that participants have sufficient time to process words in the less familiar Cyrillic script. Another critical issue may be that most studies on morphological priming, including the one by Feldman et al. (2012) used a lexical decision paradigm, while we thought it more natural to let children and adults simply read targets aloud.

A number of methodological limitations should be considered when interpreting our findings. While in masked priming paradigms primes are typically presented in lower case and targets in upper case letters in order to reduce orthographic overlap, we chose to use the same (lower case) letter set in the Latin conditions. We could, however, show that morphological effects were evident when we controlled for orthographic overlap. Furthermore, visual similarity was minimal in the conditions with Cyrillic primes, and we still found evidence for morphological priming. As a matter of fact, earlier research showed that priming effects are similar in conditions with upper- and lowercase presentation of the target (Bowers, Vigiliocco, & Haan, 1998).

Second, in studies on English, pseudosuffixed items like corner—CORN, which are semantically unrelated, turned out to be of particular interest from a developmental perspective, as only older children and adults showed a priming effect for these items (Beyersmann et al., 2012), suggesting that with increasing reading experience similarity in form becomes more important. The rich morphological structure of Bosnian did not allow us to include such a pseudosuffixed condition as “pseudo-suffixed” words are very rare and of low frequency. They possibly exist in combination with derivation suffixes, but not with inflectional suffixes which were used in this study. And finally, as already mentioned, our decision to use a reading aloud paradigm, which seemed most natural, especially for the younger readers, may have induced unwanted noise based on articulatory processes. Lexical or semantic decision paradigms may be more adequate.

Thus, in summary, this first study on development of morphological decomposition in the biscriptal, phonologically highly transparent and morphologically rich Bosnian language provides tentative evidence that morphological priming effects are established by Grade 4 and may not undergo major developmental change later on. We attribute the minor facilitation effects of primes in Cyrillic mostly to low familiarity with this letter set in our Bosnian sample. The fact that we found significant priming in both scripts suggest that the observed effects are caused by morpho-semantic rather than morpho-orthographic overlap. Further investigation of reading acquistion in the bi-scriptal language of Bosnian is certainly worthwhile and can reveal important insights for the general theory of reading.