One key theoretical milestone in visual word recognition is a widely replicated finding from masked priming, showing that briefly (50 ms) presented affixed and pseudo-affixed primes facilitate the lexical decision to their embedded stems (e.g., farmer–FARM; corner–CORN), while nonaffixed words do not (e.g., cashew–CASH), suggesting that skilled readers are experts at rapidly extracting morphological information from print (e.g., see Diependaele et al., 2005, for evidence in Dutch; Longtin et al., 2003, in French). Morphological first language (L1) priming effects have been observed across various languages (see Coughlin & Tremblay, 2015, for evidence in French; Kazanina et al., 2008 in Russian; Kırkıcı & Clahsen, 2013 in Turkish; Norman et al., 2017, in Hebrew) and stimuli types, including truly-affixed words (e.g., Beyersmann et al., 2016; Kahraman, 2022; Taft, 2004; Taft & Forster, 1976), pseudo-affixed words (e.g., Devlin et al., 2004; Longtin et al., 2003; Marslen-Wilson et al., 2008; Rastle et al., 2004), and affixed nonwords such as farmity (e.g., Beyersmann et al., 2013; Longtin & Meunier, 2005).

In contrast to the large body of evidence on morphological processing in L1, morphological processing in second language (L2) speakers, and specifically the mechanisms of cross-language transfer between the embedded morphemes of a bilingual’s two languages, are less well understood (for a review, see Kahraman & Beyersmann, 2023) and motivated the current investigation. While several studies have shown that bilinguals are indeed proficient at extracting morphological information from their second language, as evidenced in L2 learners of English (e.g., Feldman et al., 2010), French (e.g., Coughlin & Tremblay, 2015), Turkish (e.g., Kırkıcı & Clahsen, 2013), and Russian (e.g., Gor & Jackson, 2013), comparatively little work has investigated the mechanisms of cross-language morphological transfer between the embedded morphemic constituents of the two active languages (Kahraman & Beyersmann, 2023).

Of central importance in the investigation of cross-language morphological processing has been the masked translation priming paradigm (Chung et al., 2019; Duñabeitia et al., 2013; Kim et al., 2011; Kim & Wang, 2014; Ko & Wang, 2015; Voga, 2020; Voga & Anastassiadis-Symeonidis, 2018; Voga et al., 2020; Voga & Grainger, 2007; Wang et al., 2021; Wen & van Heuven, 2018). For example, using semantically transparent complex primes, Voga and Grainger (2007) showed that late bilinguals responded faster and more accurately to L2 French monomorphemic cognate words (e.g., CANON [CANNON]) preceded by L1 Greek affixed words (e.g., κανονιά /kanonia/ [cannon–shot]) relative to a phonological control condition. This initial evidence for early cross-script morphological translation priming was later replicated in Greek–French bilinguals (Voga et al., 2020) and using noncognate stimuli in Chinese–English bilinguals (Wang et al., 2021; Wen & van Heuven, 2018), Korean–English bilinguals (Chung et al., 2019; Kim & Wang, 2014; Ko & Wang, 2015).Footnote 1 For example, Wen and van Heuven (2018) showed that lexical decisions to L2 English monomorphemic target words (e.g., FACT) were facilitated by prior presentation of complex Chinese translations in L1 (e.g., 事实), suggesting that the morphemic units in the two languages are activated rapidly and simultaneously in bilinguals (e.g., Duñabeitia et al., 2013; Ko & Wang, 2015; Voga & Grainger, 2007). However, at least a proportion of the previously observed priming effects can be explained by the shared semantic relationship between the prime and the target in addition to the shared morphological relationship per se (Kahraman & Beyersmann, 2023).

The current study

To shed light on the role of orthography, morphology, and semantics in cross-language priming, the current study examined morphologically complex nonword primes constructed from illegal combinations of the L1 translations of two real embedded morphemic constituents in L2 relative to a control condition (Experiment 1), as well as the illegal combinations of L1 translations of embedded words that were semantically related to the target (Experiment 2). Since nonwords do not hold any whole-word lexical representations, the reading system relies on the identification of the embedded morphemic constituents instead. As such, the use of complex nonwords has the potential to inform distributional and decompositional theories of morphological processing. Distributional accounts view morphological structure as a by-product of orthography and semantic overlap and reframe it as a result of the statistical regularities in associations between form and meaning (e.g., Stevens & Plaut, 2022). This approach is therefore fundamentally based on the idea that morphology does not have a distinct level of representation in the mental lexicon. Decompositional accounts, on the other hand, propose morphology as an explicit level of representation where words are organized based on their morphological structure in the mental lexicon (see Amenta & Crepaldi, 2012; Marelli et al., 2020, for reviews).

Masked cross-script morphological translation priming for noncognate pairs seems robust (e.g., Schoonbaert et al., 2009; Voga & Grainger, 2007; see Wen & van Heuven, 2017, for a meta-analysis), however, it is not clear if priming for noncognate translation equivalents can be evidenced in similar-script languages. Only one prior study investigated masked morphological translation priming effects using noncognate words in similar-script languages (Duñabeitia et al., 2013 Spanish–English in Experiment 1 and Basque–Spanish in Experiment 2) using affixed word primes with a stimulus-onset asynchrony (SOA) of 50 ms (e.g., doloroso [painful]–PAIN) but found no significant priming effects. It might thus be the case that the competing orthographies of two similar-script translation equivalents (e.g., doloroso and pain) may cancel out facilitation at the level of morphology (Kim & Davis, 2003; Voga & Grainger, 2007). Since cross-language priming might require more processing depth due to the translational processes involved, the current study used 200 ms SOAs. The current study is a very first attempt at advancing our understanding of cross-language morphological processing using a carefully-matched set of complex nonword primes.

Experiment 1

Experiment 1 sought to examine cross-language morphological processing using a masked morphological translation priming paradigm in Turkish–EnglishFootnote 2 bilinguals. A group of English L1 monolinguals was assessed as a control group. The goal was to clearly tease apart morphological and orthographic influences on cross-language processing using morphologically complex nonword primes. L2 English stem targets (ICE) were preceded by three different types of complex L1 Turkish noncognate nonword primes: affixed L1 nonwords (buzca [iceish], nonaffixed L1 nonwords (buznak [iceald]), and unrelated L1 nonwords (tuşku [keyment]). This experiment used morphologically complex nonword primes to prevent lexical inhibition between the prime (iceish) and the embedded stem (ice) which may have been one of the reasons for the absence of facilitation from complex real word primes (icy) in Duñabeitia et al.’s (2013) study. Prime–target pairs were selected such that there was no orthographic overlap. We hypothesized that if morphological information is transferred across languages, we would expect to see significant embedded stem priming effects in both the affixed and nonaffixed nonword conditions relative to the unrelated control condition. Moreover, if the presence of the affix additionally facilitates cross-language priming, we would expect to see significantly more priming in the affixed compared with the nonaffixed nonword condition. If not, we would not expect any differences between affixed and nonaffixed cross-language priming. These hypotheses were pre-registered, along with an analysis plan (https://aspredicted.org/uu5vt.pdf).

Method

Participants

The required sample size for linear mixed-effects models with slightly over 80% power was determined through the use of simulation to estimate study design power within the R environment (Version 2021.09.2; R Development Core Team, 2016) using ‘simr’ package. The ‘lme4’ package was used for modelling, and the model was fitted using a competitive model testing approach and stored as the model for power analysis on inverse response times (RTs). Based on this analysis, 72 advanced L2 speakers of English (50 females, age: M = 30.72 years, SD = 4.79, range: 21–43) and 70 L1 speakers of English (57 females, age: M = 22.8 years, SD = 7.35, range: 17–47) participated in the experiment in exchange for monetary reimbursement or course credit. Bilinguals were all late learners of English with a high level of English proficiency (see Table 1 for participant demographics).

Table 1 Participant demographics
Full size table

Five L1 English participants who reported having language or learning disabilities or having acquired English at a late age, and 10 L2 participants who were identified as nonnative speakers of Turkish or who acquired English early in life were excluded prior to the analyses. Participant demographics were collected via The Language Experience and Proficiency Questionnaire (LEAP-Q; Marian et al., 2007). Only some components of LEAP-Q were used since this shorter version was more feasible in an online data collection setting where participants were more likely to be affected by environmental distractors with lengthy tasks. This short questionnaire asked participants to list all languages they know in order of dominance and in order of acquisition. They were also asked to list the percentage of the time they were exposed to each language and to name the cultures with which they identify. The number of years participants spent in an English-speaking family and a school and/or working environment was additionally asked. The rest of questions as well as participant responses are presented in Table 1.Footnote 3

Materials

For the lexical decision task, 90 L2 English stem targets (e.g., ICE) were selected as targets. The list of word trials is reported in Appendix 1 (see also https://osf.io/snxm5/ for a full set of materials). It was ensured words were frequent (M = 4.67, SD = 0.58), concrete noncognate items with no pseudo-suffixes. Word frequencies were extracted from the SUBTLEX-UK database (van Heuven et al., 2014) using the Zipf values. Concreteness ratings were obtained from Brysbaert et al. (2014) (M = 4.54, SD = 04.69, range: 3.16–5). Of 90 targets, 70 were nouns, 11 verbs, and nine adjectives. English target words were translated into Turkish. The translated Turkish translations acted as stems embedded in affixed and nonaffixed prime stimuli. The English words and their Turkish translations were matched on number of phonemes (M = 3.3 for English targets; M = 3.5 for Turkish translations, p = .26) and, though less robustly, on length (M = 4 for English targets; M = 3.7 for Turkish translations, p = .08). However, the number of syllables of English words was lower than their Turkish translations (M = 1.04 for English targets; M = 1.37 for Turkish translations, p < .0001), something that is caused by structural differences between the Turkish and English languages (e.g., Turkish words are easier to break into syllables; Durgunoğlu & Öney, 1999).

Affixes and nonaffixes (–CAFootnote 4 vs. –KI) were matched on number of letters and positional specific bigram frequency (P23). It was ensured that nonaffixes would not stand as a stem and would always be part of a real word (e.g., cash–ew). Like nearly all other suffixesFootnote 5 in Turkish, six suffixes used in current study had more than one form on basis of the vowels that precede it, which exhibits one form of vowel harmony (see Table 2 for suffix details). In addition to these vowel alternations, consonants in suffixes are also subject to alternations on basis of voicing assimilation rule, which requires the devoicing of an initial voiced consonant of a suffix (i.e., c, d, g turns into ç, t, k, respectively) when the base ends in a voiceless consonant (f, s, ş, h, p, ç, t, k). Therefore, six different suffixes were repeated 15 times in different forms (e.g., –CA; –ca, –ce, –ça, –çeFootnote 6) in the affixed nonword condition and 5 times in each list.

Table 2 Turkish suffixes, their English translation equivalents, and their meanings
Full size table

Once the stems were combined with potential endings, the affixed and nonaffixed Turkish nonwords were matched on positional specific bigram frequency (suffixed nonwords: M = 4,709, SD = 2,933; nonsuffixed nonwords: M = 4,022, SD = 1,966).

Stems in the unrelated condition (e.g., tuş) were selected such that they closely matched the stem in the related condition (e.g., buz) on length, word frequency, bigram frequency, number of phonemes, number of syllables, orthographic neighbourhood, and OLD20. Half of the unrelated stems were combined with affixes, the other half with nonaffixes, thus mirroring the experimental design used for related nonwords.

For the purposes of the lexical decision task, 90 nonword filler stimuli were created by changing one to two letters in real English words (e.g., end–enp). Real word and nonword targets were matched on length and bigram token frequencies. They were not matched on the orthographic neighbourhood size or positional specific bigram frequencies, however, due to the need to control for the other experimental factors. Their Turkish primes were constructed by combining a new set of Turkish nonwords that were constructed by changing one to two letters in real Turkish words (e.g., yut [swallow]–kuf) with the same affixes and nonaffixes as for the experimental words. Three experimental lists were created using a Latin square design, each list containing only one of the prime-target pairs to ensure each participant responded to each target only once.

English L2 target words were preceded by three different types of complex L1 Turkish nonword primes (see Table 3 for characteristics): affixed L1 nonwords (e.g., buzca [iceish], nonaffixed L1 nonwords (e.g., buznak [iceald]), and unrelated L1 nonwords (e.g., tuşku [keyment]).

Table 3 Mean item characteristics (SDs)
Full size table

Procedure

Participants were tested online using the Gorilla platform (https://gorilla.sc). Each trial consisted of a 500-ms fixation cross, 500-ms forward mask of hash keys, followed by a 200-ms prime in lowercase, then the uppercase target. The target remained on screen until the response was made or until 3 seconds elapsed. Participants were instructed to respond as quickly and accurately as possible. After the lexical decision task, participants also responded to a short version of LEAP-Q (Marian et al., 2007) to ensure that participants are late bilinguals who speak Turkish as their L1 and English as their L2, and that they are proficient in their L2.

Analysis

All data and analyses script are available (https://osf.io/snxm5/?view_only=b3115a5d31f0431d89ce955a526407e5). Response time (RT) analysis was conducted on correct trials only. As a criterion for removal of extreme outliers, RT distributions as well as error rates for items and subjects were visually inspected separately for L1 English and L2 English groups. Three participants from each group were removed since their nonword response times or accuracies were two standard deviations above the mean. Individual data points below 250 ms or above 2,000 ms were also removed (L1: 0.4 %; L2: 0.5 % of the total data). Data points whose standardized residuals were greater than 2.5 standard deviations (Baayen, 2008) were removed (L1: 2.8 %; L2: 3.2 %). Table 4 shows mean RTs and error rates (ERs) across conditions for the L1 and L2 speakers of English. The results focused on response times only because the ERs were very low (see Table 4). The significance level is set at 0.05.

Table 4 Mean response times (RTs) and error rates (ERs) across conditions, in ms (SDs)
Full size table

The L1 and L2 datasets were then merged, and a Box-Cox power transformation indicated RTs be inverse-transformed for the merged analysis of data to normalize RT distributions and reduce the effect of outliers. Priming effects were tested in a mixed-effect model which included the two fixed effects of group (L1, L2) and of prime type (affixed nonword, nonaffixed nonword, unrelated) using ANOVA Type III (Version 3.0-12; Fox & Weisberg, 2019). Trial number was normalized and added as an additional predictor to the model to discard the effects of fatigue or habituation. The data were modelled in a model with reduced random slope structureFootnote 7 (Barr et al., 2013).

Results and discussion

The main effects of group and prime type were significant, F(1, 137.95) = 7.32, p < .001; F(2, 136.41) = 8.14, p < .0001, respectively, with a robust two-way interaction between group and prime type, F(2, 132.14) = 10.81, p < .0001, suggesting that priming effects differed across L1 and L2 groups. To further unpack the interaction between prime type and group, we analyzed the effect of prime type separately for each participant group.

Statistical analyses were conducted on inverse-transformed RTs for the L2 group, while no transformation was required for the L1 group on basis of Box-Cox power transformations. Linear mixed-effect modelling was used to fit RTs and response accuracies for both groups. The model for each group contained prime type (affixed nonword, nonaffixed nonword, unrelated) as a fixed effect and subjects and items as random effects factors (random intercepts). Trial number was normalized and also added as an additional predictor to the model to discard the effects of item presentation. The data was modelled in a maximal random slope structure (Barr et al., 2013), and pairwise comparisons between conditions were computed using the emmeans package (Lenth, 2019).

The main effect of prime type was non-significant in L1 English speakers, χ2(2) = 0.80, p = .66. However, the model for L2 English speakers yielded a robust main effect of prime type, χ2(2) = 36.91, p < .0001. Trial order was a significant covariate in L1 speakers, χ2(1) = 4.15, p = .041, and L2 speakers, χ2(1) = 37.37, p < .0001. The pairwise comparisons of factor prime type for L2 speakers are presented in Table 5. Lexical decisions to words preceded by affixed and nonaffixed related nonword primes (e.g., buzca–ICE; buznak–ICE) were significantly faster than in the unrelated control condition (z = −6.06, p < .0001; z = −3.77, p < .001), demonstrating cross-language embedded stem priming effects. Facilitation of embedded stems occurred independently of the morphemic structure (i.e., for both affixed and nonaffixed nonwords) due to the lexical activation of the translated stem (i.e., buzca [iceish]–ice) in both conditions. In addition, there was a significant difference between the affixed and nonaffixed nonword prime conditions (z = −2.47, p = .01), showing that the presence of the affix additionally facilitated cross-language priming. This result suggests that cross-language morphological priming occurred in absence of whole-word access and orthographic overlap.Footnote 8

Table 5 Pairwise contrasts between prime types in the L2 group
Full size table

AoA and L2 priming

A more exploratory, non-preregistered analysis of the L2 data was performed in order to assess the modulation of cross-language morphological priming by age of English acquisition (EngAoA), which previous studies have shown is an important predictor for cross-language priming (e.g., Sabourin et al., 2014). Even though all bilinguals of this study were characterized as late bilinguals (see Table 1 for participant characteristics), we could still examine the effect of AoA by comparing bilinguals who acquired the language at an earlier age versus later age. This variable was added to the modelFootnote 9 to calculate the interaction with factor prime type. The interaction between prime type and EngAoA was computed using the emmeans package. Results demonstrated a two-way interaction, χ2(2) = 11.10, p = .003, showing that prime type was modulated by age of L2 acquisition. Further post hoc analysis comparing earlier and later acquirers of English in the form of 0.05 and 0.95 quantiles (i.e., 12 and 25 years of age; Table 6) revealed that the interaction was due to more robust priming effects in participants who acquired English at an earlier age (Fig. 1). Critically, the difference between the affixed and nonaffixed conditions increased with increasing EngAoA (the red line, Fig. 1), suggesting that late acquirers were more reliant on morphological structure.

Table 6 Priming as a function of age of acquisition (EngAoA) comparing earlier (E) and later (L) acquirers of English with estimates for each contrast, standard errors (SE), degrees of freedom (df), z ratio, and p values
Full size table
Fig. 1
figure 1

Priming effects (in milliseconds) as a function of age of acquisition (EngAoA in years). Note. All three priming effects interacted significantly with AoA. Green and blue lines indicate affixed priming (e.g., buzca [iceish]) and nonaffix priming (e.g., buznak [iceald]), whereas the red line shows participants’ sensitivity to morphological structure in the stimuli. (Color figure online)

Full size image

In conclusion, the results of Experiment 1 provide evidence for significant morphological translation priming with both L1 affixed and nonaffixed nonword primes towards the visual recognition of L2 stem targets. Affixed nonword primes yielded significantly more facilitation than nonaffixed primes. However, the size of the affixed and nonaffixed priming varied as a function of the age of L2 acquisition. Late acquirers of L2 English were more reliant on morphological structure, while early acquirers were primarily guided by the identification of embedded words, a point we return to in the General Discussion.

Experiment 2

The purpose of this experiment was to investigate the extent to which cross-language embedded word translation priming effects are affected by semantics at 200 ms SOA. L2 English stem targets (e.g., ICE) were preceded by three different types of L1 Turkish primes. Critically, the stimuli were composed of nonaffixed stimuli to simulate a priming manipulation that is less biased towards morphemic parsing.Footnote 10 In the translation condition, targets were preceded by L1 nonwords with embedded words that were noncognate translation equivalents of the target (e.g., buznak [iceald]). Naturally, the embedded stem and its translation equivalent also shared a semantic relationship. In the semantic condition, targets were preceded by L1 nonwords with embedded words that were semantically related to the target (e.g., suzur [waterew]), but did not represent direct translations. In the control condition, targets were preceded by L1 unrelated nonwords (e.g., tuşzur [keyew]). Note that the nonaffixed L1 translated condition was identical to Experiment 1. If cross-language priming is purely due to the shared semantic relationship between the embedded word of the prime and the target, we would expect significantly greater priming in the semantic relative to the unrelated condition, and greater priming in the translated than the semantic condition. If, however, cross-language priming is specifically due to the lexico-semantic relationship that is shared between the Turkish stems and their English target translation equivalents, we would still predict greater priming in the translated compared with the semantic condition, but crucially, priming in the semantic versus unrelated condition should be absent.

Method

Participants

Seventy-six advanced Turkish–English bilinguals (34 females, age: M = 30.76 years, SD = 4.95, range: 19–46) participated in the experiment in exchange for monetary reimbursement or course credit. Using participant responses to LEAP-Q, seven participants who reported having acquired Turkish at a late age and had a low proficiency in Turkish, or who acquired English early in life were excluded prior to the analyses. Participant responses are presented in Table 7.

Table 7 Participant demographics
Full size table

Materials

L2 English targets (e.g., ICE) were preceded by nonaffixed L1 translated nonwords (e.g., buznak [iceald]), nonaffixed L1 semantically related nonwords (e.g., suzur [waterew]), and nonaffixed L1 unrelated nonwords (e.g., tuşzur [keyew]). The translated nonword condition was identical to Experiment 1. The semantic primes were selected by retrieving (and then translating) semantic associations of the English target words from the Small World of Words (smallworldofwords.org; De Deyne et al., 2019). All items were concrete words (nouns, verbs, or adjectives) that contained no pseudo-suffixes. These were then combined with the same nonaffixes as in the translated nonword condition. The Turkish translations and semantic associations were matched on length (M = 3.07; M = 3.08, p = 0.3, respectively), frequencyFootnote 11 (M = 392.6; M = 200.3, p = .2, respectively), number of syllables (M = 1.4; M = 1.5, p = .1, respectively), number of phonemes (M = 3.5; M = 3.6, p = 0.4, respectively), OLD20 (M = 1.4 M = 1.4, p = 0.3, respectively), orthographic neighbourhood (M = 11.1; M = 10.5, p = .6, respectively), and positional specific bigram frequency (M = 90.9; M = 98.3, p = .7, respectively). The full list of stems is reported in Appendix 2.

Once the stems were combined with 30 potential endings, nonword stimuli in the three conditions were matched on length (M = 5.9, M = 6.0, M = 5.9, respectively). For the purposes of the lexical decision task, 90 nonword filler stimuli from Experiment 1 were combined with the same nonaffixes as the experimental words. Three experimental lists were created using a Latin square design, each list containing only one of the prime–target pairs to ensure each participant responded to each target only once.

Procedure

The procedure and analysis were the same as in Experiment 1. The planned analyses were preregistered prior to data collection (https://aspredicted.org/sc5ky.pdf). A Box-Cox power transformation of the RTs indicated RTs not be transformed (please refer to the following OSF link for all data and analyses script: https://osf.io/sy4d8/?view_only=66a20e7007c946b88147f9d7f16f3416).

Results and discussion

The main effect of prime type was significant, χ2(2) = 13.70, p < .001. Trial order was also a significant covariate, χ2(1) = 7.70, p < .001. Table 8 shows mean RTs and error rates across conditions. Similar to Experiment 1, the results in Experiment 2 focused on RTs only since error rates (ERs) were very low.

Table 8 Mean response times (RTs) and error rates (ERs) across conditions, in ms (SDs)
Full size table

The pairwise comparisons of factor prime type for L2 speakers are presented in Table 9. Lexical decisions to words preceded by nonaffixed translated nonword primes (e.g., buznak–ICE) were significantly faster than in the unrelated control condition (z = −3.69, p < .001), demonstrating cross-language embedded priming effects. In addition, there was a significant difference between the nonaffixed translated and nonaffixed semantically related nonword conditions (z = 2.46, p = .02). The difference between the semantic and the unrelated conditions was not significant (z = −1.15, p = .24), demonstrating that semantic prime–target overlap alone was not sufficient to trigger priming.

Table 9 Pairwise contrasts between prime types in Experiment 2
Full size table

Although priming in the semantic condition did not differ statistically from the unrelated condition, it is possible that priming in the translated condition might have been at least partially due to the shared semantic relationship between the stem of the prime (buz) and the target (ice), which was proportionately larger in the translated than in the semantic control condition. However, as we discuss in more detail below, what really seems to be driving morphological translation priming effects is the direct lexico-semantic relationship between the embedded word and its translation equivalent. This second experiment thus sheds further light on the nature of cross-language morphological priming by providing important insights into the shared lexical and semantic contributions in this process.

General discussion

Experimental evidence reported in the current study showcased for the first time the potential of morphologically complex nonwords and nonmorphological embedded words as effective noncognate translation primes in similar-script bilinguals. The results of Experiments 1 and 2 showed that robust noncognate cross-language embedded word and morphological priming effects were present in similar-script bilinguals. We focused our cross-language investigation on morphologically complex nonword stimuli, including affixed and nonaffixed nonwords in Experiment 1 as well as nonaffixed semantically related nonwords in Experiment 2, to disentangle the separate effects of stems and affixes.

Morphological influences on cross-language processing

Several key conclusions can be derived from the current findings. First, priming effects observed in Experiment 1 in the affixed (buzca–ICE) and nonaffixed (buznak–ICE) nonword conditions relative to the unrelated condition (tuşku–ICE) suggest that bilinguals were able to rapidly extract the stems embedded in the Turkish L1 nonword primes and map them onto their corresponding L2 (English) translations, which in turn facilitated their L2 lexical decision response. Although primes (which were all nonwords) in this study were visible (200 ms SOA), it is clear that the identification of affixes in this task was due to a prelexical embedded stem activation given that the nonword primes did not correspond to existing whole-word lexical representations.

Second, cross-language activation was further facilitated by the activation of the L1 Turkish affix in Experiment 1, as evidenced by the significantly larger priming effects in the affixed compared with the nonaffixed condition. The here reported cross-language morphological priming effects show that bilinguals are sensitive to morphological structure of words and that they are experts at rapidly extracting the stem (buz) embedded in a nonword prime, independently of whether it was accompanied by a real affix (buzca) or a nonaffix (buznak), which then in turn leads to the activation of its stem’s translation equivalent (ice), thus producing target word facilitation. Third, given that none of the prime–target pairs shared any orthographic relationship in either experiment, it can be ruled out that the observed priming effects were simply due to lower-level form overlap between the two languages.

Fourth, the results of Experiment 2 revealed significant priming in the translated (buznak [iceald]–ICE) but not in the semantic condition (suzur [waterew]–ICE) relative to the unrelated condition, with a statistically significant difference between the translated and semantic conditions. These findings indicate that semantics alone is not sufficient to trigger cross-language morphological effects (for similar arguments, see Voga, 2020), and instead suggest that bilinguals were able to map the stems embedded in the L1 primes onto the lexical representations of the translated L2 stem targets. It is possible that semantic relatedness partially contributed to cross-language embedded word priming effects in the translation condition at 200 ms SOA, as has been previously shown to be the case in cross-language priming studies with monomorphemic words (e.g., Schoonbaert et al., 2009).

Duñabeitia et al.’s (2013) study was the first to examine cross-language priming in Spanish–English (similar-script) bilinguals; they, however, failed to report significant priming effects in their noncognate stimuli using 50 ms SOA. The absence of facilitation in their study may perhaps be explained by some level of competition between the lexical representations of the Spanish primes (doloroso) and the English target words (pain), thus counteracting any potential facilitation that may have come from the presence of the affix (–oso). In the two experiments, these potentially competing mechanisms were avoided since none of the primes formed existing real words. Also, Duñabeitia and colleagues used a relatively short prime duration, which may have been too brief to trigger cross-language morphological priming effects. In the current study, via exposing primes for a longer duration, participants were allowed more time to thoroughly process the prime thereby increasing the magnitude of morphological facilitation (Feldman, 2000).

Cross-language priming as a function of age of L2 acquisition

The results further showed that cross-language morphological co-activation is more pronounced in bilinguals with earlier age of L2 acquisition. One possible explanation for the AoA influences on L2 priming is that morphological organization of lexical representations differs in different bilingual groups. Bilinguals who acquired English at earlier ages could effectively map sublexical orthography in L1 Turkish onto full-form orthographic representations in L2 English, and therefore show less sensitivity to the morphological information in the L1 stimuli, resulting in similar magnitudes of priming from affixed and nonaffixed nonwords. Bilinguals with less experience, on the other hand, employ visual word recognition processes that are guided more by the processing of L1 affixes, potentially because these individuals may be applying morphological parsing strategies from their dominant agglutinative Turkish L1. This finding adds to the growing number of studies showing that individual variability in AoA determines the strength of lexical connections between morphemic constituents in the bilingual lexicon (e.g., Dimitropoulou et al., 2011; Sabourin et al., 2014).

Theoretical implications, limitations, and conclusion

The current findings do not allow us to clearly tease apart the assumptions of distributional theories of morphological processing (e.g., Amenta et al., 2020; Plaut & Gonnerman, 2000; Stevens & Plaut, 2022) whereby semantics and morphology are inevitably intertwined, and decompositional theories of morphological processing (e.g., Beyersmann & Grainger, 2023; Crepaldi et al., 2010; Grainger & Beyersmann, 2017; Taft, 2004) that view morphemes as being independently represented in the reading system. However, given the study’s focus on cross-language morphological transfer in bilinguals, we will centre the below discussion primarily around existing interactive activation models of bilingual word recognition. The observed cross-language priming effects from affixed and nonaffixed nonword primes are broadly consistent with current bilingual models of lexical access such as the Bilingual Interactive Activation + (BIA+ model; Dijkstra & van Heuven, 2002) and Multilink models (Dijkstra et al., 2019) that propose an integrated mental lexicon enabling the rapid simultaneous activation of bilinguals’ two languages. However, although the BIA+ and Multilink models provide a sophisticated approach to addressing cross-language mechanisms of monomorphemic word identification, based on the parallel, simultaneous activation of the two languages within a single integrated lexicon, they are less able to capture the processing of morphologically complex words. Given the underspecification of existing cross-language processing theories with regard to the identification of morphologically complex words, Kahraman and Beyersmann (2023) proposed a tentative theoretical framework integrating the basic principles of the word and affix model (Beyersmann & Grainger, 2023; Grainger & Beyersmann, 2017) and Multilink model (see Fig. 2).

Fig. 2
figure 2

Hypothetical flow of activation in bilingual complex word identification system for affixed nonwords (Panel 1), nonaffixed nonwords (Panel 2), and nonaffixed semantically related nonwords (Panel 3). Note. Activation between layers flows via bidirectional connections. Stronger connections are indicated via thicker lines, and stronger activations are indicated via thicker rings. This mimics Turkish–English bilinguals who speak similar-script languages. The visual recognition of complex words in bilinguals occurs in a three-layered network of orthographic input, orthographic lexicon, and semantic representations. In visual input ‘buzca’ in the top panel (Engl. ‘iceish’), its morphemic constituents (buz, –ca) as well as translation equivalents (ice, –ish) are activated simultaneously within an integrated orthographic lexicon. The nonaffixed nonword ‘buznak’ in the middle panel (Engl. ‘iceald’) simultaneously activates the embedded word (buz) as well as its translation equivalent (ice) within an integrated orthographic lexicon. The additional facilitation for affixed (buzca) compared with nonaffixed nonwords (buznak) is explained by Turkish–English bilingual’s sensitivity to morphological structure in the processing of complex nonwords in L1 Turkish (indicated via “L1 parsing strategy transfer” from their agglutinative native language to English in Panel 1). Whenever a letter string is exhaustively decomposable into morphemes (e.g., buz + ca), the reading system rises the activation levels of the corresponding morphemes, which explains the comparatively larger priming effects in the affixed compared with the nonaffixed nonword conditions. As shown in Panel 3, embedded words (e.g., su in suzur [Engl. ‘water’ in ‘waterew’]) will strongly activate their corresponding semantic representations that is shared between languages (‘su’/’water’) which in turn will be mapped onto the lexical representation of the translation equivalents in L2. In addition, the embedded word (su) will weakly activate other semantic neighbours (e.g., the semantic representation of ‘buz’/’ice’; indicated by a node in opaque grey colour). The latter point provides a hypothetical explanation for the absence of ‘suzur–ICE’ priming in our data

Full size image

Within this bilingual network, it is proposed that there are two distinct mechanisms running in parallel towards the processing of a complex word. While stem activation is handled by the non-morphological embedded word activation mechanism (e.g., buz in buzca and buznak), the morphological affix activation mechanism triggers the activation of affixes (e.g., –ca; see top panel in Fig. 2). Importantly, both mechanisms operate bottom-up simply by mapping letter strings onto existing lexical and affix representations in the orthographic lexicon. The activation of stems tolerates minor orthographic alterations. A given stimulus input activates words with slight orthographic deviations in addition to those that provide an exact match with the input. Importantly, the proposed mechanisms operate within a single integrated bilingual lexicon where the Turkish morphemes (buz, –ca,) as well as their English translation equivalents (ice, –ish) are simultaneously activated, thereby facilitating the recognition of target words as a real word in the other language than the prime. Activated units that share relevant meaning representations in both languages then send and receive activation from the semantic level. Critically, activation flows between layers via bidirectional excitatory connections. As can be seen in Fig. 2, the here observed cross-language embedded word priming effects are captured in the bidirectional links between the semantic representation and the lexical representation of the translated stem. Given that relationships between translation equivalents in this model are mediated via semantics, it is further predicted that the magnitude of cross-language morphological priming depends on the degree of semantic overlap between L1 prime and the L2 target at 200 ms SOA (see Panel 3 in Fig. 2 for the processing of nonaffixed semantically close nonwords that failed to activate the target ICEFootnote 12).

An important question for future research is the question of automaticity of cross-language morphological transfer effects. The exploration of cross-language priming effects with shorter SOAs or the simultaneous recording of event-related brain potentials (ERPs) and behavioural measures within a cross-language morphological priming study may be able to shed light on the time-course of cross-language activation processes in bilinguals. Another prospect for future work lies in the investigation of cross-language transfer effects using translation equivalent affixes, rather than translation equivalent morphemic stems. Although the current data clearly provide evidence for cross-language priming of stem targets, it is less clear if these results generalize to cross-language transfer between the affix representations of L1 and L2, which would provide further evidence for a highly specialized morphemic transfer system that bilinguals can draw on in their reading. In addition, a replication of our Experiment 2 with semantically related words embedded in affixed nonword primes (e.g., suca [waterish]–ICE) using a longer prime duration might yield further insights into the independent contribution of lexical, morphological, and semantic influences on cross-language morphological translation priming effects. Finally, it remains to be seen if the here observed priming effects generalize to other languages with a less productive morphology than Turkish. It has been reported that morphological processing in L1 changes as a function of morphological productivity of a language (e.g., Beyersmann et al., 2020, 2021; Mousikou et al., 2020). Examining the influence of language-specific differences in morphological complexity therefore represents an important avenue for the investigation of cross-language morphological transfer effects in bilinguals.

In sum, the two experiments reported in the current study provided evidence for the dynamic interplay between morphological representations across languages in similar-script bilinguals. A general trend towards more robust cross-language morphological priming was observed in early compared with late bilinguals, thus pointing to the important role of individual variability in bilingual research (e.g., Andrews & Lo, 2012; Kahraman & Kırkıcı, 2021). The observed difference between affixed and nonaffixed nonword priming demonstrates that bilinguals with a morphologically complex L1 employ a word recognition strategy that is sensitive to morphological information in L1 Turkish and highlights the need for further specifications of current bilingual models of visual word recognition to integrate morphological information. We here propose a tentative theoretical bilingual framework to accommodate cross-language morphological transfer effects. Further work is required to test the generalizability of the proposed theoretical constraints to other languages.