Prefix priming within and across languages in early and late bilinguals

Abstract

In contrast to ample evidence for cross-linguistic priming of monomorphemic words, cross-linguistic representation of affixes is not well understood. The current study examines cross-linguistic prefix priming among early and late English-Spanish bilinguals, focusing on prefixes that have the same form and meaning in the two languages. We first confirm robust prefix priming among English monolingual speakers (Experiment 1). We also observe prefix priming from first-language English to second-language Spanish but only for early bilinguals (Experiment 2). On the other hand, both early and late bilinguals do not show reliable prefix priming effects that are dissociated from orthographic or semantic priming from Spanish to English (Experiment 3) or from Spanish to Spanish (Experiment 4). The results suggest that for early bilinguals, the tested prefixes in their L1 and L2 have shared representations. Less reliable results for late bilinguals may reflect their weaker sensitivity to morphological structure in a second language.

Highlights

• • Prefix priming in English and Spanish and between those languages was tested with early and late Spanish-English bilinguals.

• • Robust prefix priming in one’s dominant language (English) is replicated, but not in one’s less dominant language (Spanish).

• • Cross-language prefix priming from one’s dominant language to a less dominant language was found among early bilinguals, but not late bilinguals.

1. Prefix priming within and across languages in early and late bilinguals

It is widely accepted that words in two languages that share their meaning are connected to some extent in the lexicon of bilingual language users, and more so for cognates than noncognates (Carramazza and Brones, 1979; De Groot and Nas, 1991; Dijkstra et al., 1998, 2010). However, it remains uncertain how morphologically complex words and their constituents, i.e., roots and affixes, are represented in the bilingual lexicon. In the present study, we examine cross-language affix representation throughout four experiments that test prefix priming in Spanish and English both within a first language (L1; Experiment 1), a second language (L2; Experiment 4) and between L1 and L2 (Experiment 2 and Experiment 3) using a masked priming paradigm. Based on the results, we propose an integrated representation of cognate prefixes within existing frameworks for bilingual word recognition.

2. Bilingual lexicon and morphological representation

A prominent topic in bilingualism has been the organization of the bilingual lexicon with a specific focus on how words in two languages with the same meaning are represented. Priming is a popular method to probe this matter. In this paradigm, a pair of words is presented in a sequence, and the time to recognize the latter word is measured. It has been repeatedly reported that recognition of a word is facilitated when preceded by its translation equivalent in the other language among bilinguals. This pattern is asymmetric such that L1–L2 priming is more robust than L2–L1 priming (e.g., Jiang, 1999; Smith et al., 2019; Wang, 2013; see Wen and van Heuven, 2017 for meta-analysis).

This priming pattern is captured in various models of bilingual word recognition. For example, the Revised Hierarchical Model (RHM; e.g., Kroll and Stewart, 1994; Kroll and Tokowicz, 2001, 2005) proposes that L1 and L2 words are stored in separate lexicons but are linked at the conceptual level. Asymmetric priming arises because the links between L2 words and their corresponding concepts are weaker than those between L1 words and the concepts. On the other hand, the Bilingual Interactive Activation models (BIA and BIA+; Dijkstra and Van Heuven, 2002; Van Heuven et al., 1998) assume a unified lexicon of L1 and L2 words which are represented across feature, letter, word, meaning and language levels. L2 words, due to their low frequency for unbalanced bilinguals who generally use their L1 more frequently than their L2, have lower resting levels and hence are more costly to recognize, leading to smaller L2–L1 priming effects.

These models also predict greater overlap or association for cognates as compared to noncognates, leading to differences in processing. Such differences include faster recognition of cognates than noncognates, known as the cognate facilitation effect (Carramazza and Brones, 1979; Dijkstra et al., 2010). According to the RHM, this occurs because cognates have greater form overlap and thus stronger links at the lexical level. Therefore, retrieving cognates in L2 is faster because they take advantage of their L1 counterparts. According to the BIA+, this facilitation effect arises due to shared representations of cognates at both the semantic and orthographic levels.

These models are centered around monomorphemic words, and it is less understood how morphologically complex words are represented in bilinguals. For example, are morphemic units smaller than words, i.e., roots and affixes, also connected across languages? It is widely agreed that roots and prefixes show robust priming effects within the same language at both short and long-stimulus onset asynchronies (SOAs), supporting the notion that these morphological pieces have independent representations (c.f., Amenta and Crepaldi, 2012; Cho et al., 2024). Do these same units show priming effects cross-linguistically as well? Results thus far are mixed with respect to cross-linguistic root priming. Studies by Kim and colleagues (Kim et al., 2011; Kim and Wang, 2014) show statistically reliable priming effects for Korean–English bilinguals from Korean-derived words (e.g., 매력적 maylyek-cek ‘attract-ive’) to English words (e.g., attract) that share the same root at short SOAs (36 ms, 48 ms and 72 ms) as well as long SOA (150 ms). These results extend the literature on monomorphemic words to non-cognate morphologically complex words in two languages and demonstrate the shared representation of roots at the morphological level. On the other hand, Voga and Grainger (2007) who used cognate roots with form overlap (e.g., κανονιά “cannon-shot”-canon “cannon”) failed to find cross-language root priming effects that are statistically larger than form-based priming (e.g., κανόνας “rule”-canon).

How affixes are represented in bilinguals’ lexicon, specifically for those that are cognates, has not been addressed in the literature, yet prefix priming effects reported in monolingual studies as reviewed in the following section provide a useful basis to probe the representation of prefixes across L1 and L2 and within L2, which is the primary focus of the current study.

3. Monolingual prefix priming

Five studies to date have found robust prefix priming effects for monolingual speakers in French (Giraudo and Grainger, 2003), Setswana (Ciaccio et al., 2020), Spanish (Domínguez et al., 2006, 2010) and English (Chateau et al., 2002) (See Cho et al., 2024 that reports robust prefix priming across these studies in a meta-analysis). Giraudo and Grainger (2003) examined affix priming in French at three different SOAs (34 ms, 57 ms and 115 ms). Primes and targets shared either the same prefix (e.g., enjeu-envol) or suffix (e.g., fumet-muret). Across the three SOAs, prefixes yielded robust priming effects ranging from 28 ms to 40 ms, whereas suffixes did not show such priming effects. Similarly, Ciaccio et al. (2020) report statistically reliable prefix priming effects for inflection as well as derivation in Setswana, but not for suffix priming (but see Crepaldi et al., 2015 who report robust English suffix priming). Domínguez et al. (2006, 2010) report statistically reliable prefix priming effects in Spanish across different SOAs (33 ms, 132 ms and 200 ms) in both reaction times and event-related potentials, where prefix overlap between primes and targets yielded early positivity at target words (150–250 ms time window) in contrast to orthographic overlap that yielded a larger N400 effect, reflecting lexical inhibition.

Finally, in Chateau et al. (2002), prefix priming effects were tested in English in a masked priming paradigm (SOA = 45 ms). Prefixes were further categorized depending on their form-meaning consistency, high consistency if they are more frequently used as prefixes and low consistency if they are frequently used in pseudo-prefixed words (e.g., de in desert) as well. Both types of prefixes yielded faster recognition of target words that have the same prefix although the magnitude of facilitation was larger for high-consistency prefixes (36 ms) than low-consistency prefixes (24 ms). Orthographic or semantic overlap, on the other hand, did not show statistically reliable facilitation.

Based on these findings, it can be postulated that within the same language, prefixes function as independent access points from the early stage of visual word recognition, just as roots do (e.g., Rastle et al., 2000). It is this status as lexical access points that yields facilitatory priming effects even at short SOAs. Whether prefixes yield similar priming effects from one language to another has not yet been examined. We investigate this question with English-Spanish bilinguals in the current study. If prefixes show reliable cross-language priming effects, we take this as evidence for the shared representation of prefixes between languages, such that their activation in one language facilitates the subsequent access in another language.

4. L2 morphology and the effect of age of acquisition

For morphological priming to occur, individuals must be sensitive to the morphological structure of words. Previous research suggests that the degree of sensitivity to L2 morphological structure may vary among bilinguals depending on their age of L2 acquisition (AoA). For example, a series of studies report that late Turkish–German bilinguals show smaller priming effects from L2 inflected words (e.g., geprüft ‘checked’) to their root (e.g., prüft ‘check’) compared to early bilinguals (Babcock et al., 2012; Basnight-Brown and Altarriba, 2007; Veríssimo et al., 2018). In particular, Veríssimo et al. (2018) report a discontinuity of priming where priming effects were similar to monolingual speakers for bilinguals with AoA of less than 5 and decreased with higher AoA. In addition, Heyer and Clahsen (2015) examined root priming effects from derived words with bilinguals with relatively high AoA (average AoA = 9.78, SD = 2.12) and report that morphological priming did not dissociate from orthographic priming effects, indicating that late bilinguals might rely more on orthographic information than morphological structure when processing those words.

Such AoA effects in morphological processing can be considered a part of the vast literature on the ‘Critical Period Hypothesis’ of language learning. While this hypothesis was originally proposed for first language acquisition, it has been suggested that there may also be a critical period for acquiring a second language as well (e.g., Birdsong, 2009, 2018; Johnson and Newport, 1989). According to this hypothesis, language acquisition is more uniform among individuals at an earlier age (before puberty), whereas it is more variable after this period due to the maturation of the neural systems (Lenneberg, 1967; Long, 1990). Applying this hypothesis to second language acquisition, the Shallow Structure Hypothesis (SSH; Clahsen and Felser, 2006) claims that late L2 learners depend on fundamentally different mechanisms than early learners. Specifically, an extended version of SSH to morphological processing maintains that late L2 learners tend to rely on storing the whole word form rather than computing its internal morphological structure when processing morphologically complex words (Clahsen et al., 2010).

Given this background, the present study investigates differential prefix priming effects for early versus late bilinguals. Based on the findings in Heyer and Clahsen (2015) and Veríssimo et al. (2018), we define early bilinguals as those who learned L2 before the age of 5 and late bilinguals as those who learned L2 after the age of 10.

5. The present study

The present study examines priming effects of prefixes during visual word recognition across and within languages with AoA as a modulating factor. We use a masked priming paradigm (Forster and Davis, 1984), where a sequence of hash-marks (#####) precedes the presentation of primes to function as a forward mask and prime words are presented very briefly (50 ms in the current study). With this design, participants are unaware of the fact that they saw the prime words, which minimizes the effects of episodic or strategical factors. A total of 59 English speakers and 416 English–Spanish bilinguals participated in the experiments. Bilingual participants were divided into the early bilingual group (EB) and the late bilingual group (LB) based on their age of Spanish acquisition (AoA ≤ 5 versus AoA ≥ 10); while some of the bilinguals were simultaneous bilinguals who learned English and Spanish around the same time, they were more dominant in English. Hence, for the sake of simplicity, we label English (dominant language) as their L1 and Spanish (less dominant language) as L2. A total of four experiments were conducted: Experiment 1 investigated the masked priming effects of prefixes in participants’ L1 (English). Experiments 2 and 3 investigated masked prefix priming across different languages, from L1 to L2 (Experiment 2) and from L2 to L1 (Experiment 3), using cognate prefixes that have the same form and meaning in the two languages (e.g., pre in English word pretext and Spanish word predecir). Lastly, Experiment 4 tested masked prefix priming in participants’ L2.

6. Experiment 1. L1–L1 priming

In Experiment 1, we tested prefix priming with English speakers. Both primes and targets were presented in English with the stimulus onset asynchrony (SOA) of 50 ms.

7. Methods

7.1. Participants

Fifty nine English speakers (26 males, 30 females, 3 other responses; age: mean = 34.9, SD = 12.91) were recruited via the online recruitment platform Prolific. One participant who reported to have learned English at age 16 was excluded from data analysis. The remaining 58 participants were residing in the United Kingdom (N = 38) or the United States (N = 20) at the time of participating in the study. They all reported English to be their first language and dominant language as indicated by their daily use of English, which was 96.5% (SD = 13.97) on average. Most participants (N = 42) did not speak other languages than English; 16 participants reported some knowledge of Polish (N = 2), French (N = 5), German (N = 2), Portuguese (N = 1), Spanish (N = 4), or Punjabi (N = 2).

7.2. Stimuli

Stimuli consisted of 240 prime-target pairs with target words and 120 prime-target pairs with target pseudowords. All prime and target words were English words. Prime words paired with word targets were either Related or Unrelated to target words in one of four dimensions; Related primes were identical to target words (“Identical,” e.g., hang-HANG) or shared the same prefix (“Prefix,” e.g., distract-DISSUADE), word-initial letters (“Orthography,” ignite-IGNORE), or meaning (“Semantic,” e.g., elect-VOTE). Unrelated primes did not have any orthographic or semantic overlap with target words and were bimorphemic prefixed words (e.g., unbend) for the Prefix condition and monomorphemic for Identity, Orthography and Semantic conditions. Word targets were verbs, adjectives, or nouns consisting of bimorphemic prefixed words for Prefix conditions and monomorphemic for the other three conditions. Each prime type was formed into 60 word-pairs with 30 Related primes and 30 Unrelated primes. All primes were presented in lowercase letters and targets in uppercase letters. See Appendix A for the full stimulus set.

The lexical characteristics of the target words and prime words across conditions are summarized in Table 1. Log word frequency per million obtained from CLEARPOND (Marian et al., 2012) was used as a measure of word frequency. This resource was used because it provides word frequency measures for Spanish as well as English so that it allows a more even comparison between the two languages in subsequent cross-language priming experiments. Here we also report log word frequency per million obtained from British National Corpus (BNC; Davies, 2004) as well as log HAL frequency measures obtained from the English Lexicon Project (Balota et al., 2007) for reference. The frequency measures obtained from CLEARPOND and BNC show a relatively high correlation (r = 0.67). Semantic similarity between prime words and target words was calculated as the cosine distance between word vectors using the pre-trained GloVe word embeddings (Pennington, 2014). All unrelated primes had a cosine distance <0.37; this threshold follows previous studies (Grainger and Frenck-Mestre, 1998; Rastle et al., 2000). Finally, the orthographic overlap between prime and target words was quantified with length-normalized Levenshtein distance (Schepens et al., 2012), with the value ranging from 0 to 1 (1 = perfect overlap). Related primes in Prefix and Orthography conditions had initial letters overlapping with target words (2–5 letters in Prefix condition, 2–3 letters in Orthography condition), resulting in the average Levenshtein distance of approximately 0.45. The number of overlapping letters between target words and prime words in other conditions was kept minimum, hence the average Levenshtein distance was 0.10 or below.

Table 1. Lexical characteristics of target and prime words used in Experiment 1 (SD in parentheses)

* Normalized to length.

In addition to target words, 120 target pseudowords were constructed, 30 of which resembled prefixed words, such that half of them were generated by combining existing prefixes and pseudowords and the other half were generated by combining non-existing prefixes and real words. Other pseudoword targets were generated with the Wuggy tool (Keuleers and Brysbaert, 2010). Prime-target pairs with word targets were separated into two different lists; those with pseudoword targets were the same across lists. Consequently, each participant saw 120 word pairs with word targets (15 items per condition) and 120 pairs with pseudoword targets.

7.3. Procedure

Participants first completed a language background questionnaire. Then they were assigned one of the two lists for the experiment. They used their own computer to perform the experiment implemented in an online platform (https://pavlovia.org) with PsychoPy (version 2020.1.3). They completed a practice session of 15 trials prior to the main session. In each trial, a forward mask (#######) was presented in the center of the monitor for 30 frames (approximately 480 ms at 60 Hz ¹), followed by a prime word presented for three frames (approximately 50 ms). Then a target word was presented for which participants performed a lexical decision task by pressing keyboards to the question “is this a real word or not” as fast as they can (‘z’ for ‘yes’ and ‘m’ for ‘no’). Items were randomly presented in a single block.

A frame-by-frame analysis of pilot data collected over this platform confirmed stimulus timing accuracy. Specifically, we recorded the pilot experiment on an uncontrolled laptop with a 60 Hz frame rate and analyzed the stimulus presentation duration, which revealed that 199 of 200 trials were at 80.5% (or within 8.3 ms) of the intended duration. See also Angele et al. (2022) and Cayado et al. (2023) who demonstrate the feasibility of testing masked priming in online platforms including Prolific and Gorilla.

7.4. Data analysis

Prior to the experiment, the accuracy cut-off was set to 70% for participants and 50% for items. No participant or item was removed based on this criterion. Also, reaction times that are too fast (less than 200 ms) or slow (greater than 2000 ms) were excluded from data analysis, which resulted in removing 2.5% of all data points. For statistical analysis, accuracies were analyzed with a generalized mixed effects regression model using the glmer function in R. This model had Condition (Identity, Prefix, Orthography and Semantic) and Relatedness (Related versus Unrelated) and their interaction as fixed effects and centered Target word frequency and Prime word frequency as fixed covariates.

As reaction times are positively skewed, they were inverse-transformed based on the lambda value (−1.19) examined by the box cox test (Box and Cox, 1964) using the boxcox() function in the MASS package (Venables and Ripley, 2002). The transformed RTs were analyzed with a linear mixed effects regression analysis with the lmer function in the lme4 package (Bates et al., 2015). Fixed effects included Relatedness (Related versus Unrelated) and Condition (Identity, Prefix, Orthography and Semantic) and their interaction. In addition, Target word frequency, Prime word frequency, Target word length, and Prime word length were included as covariates. All continuous variables were centered and dummy coding was used for Condition with Prefix as the reference for both analyses, such that the intercept is the mean of Prefix condition and each contrast compares Prefix and Identity (Condition 1), Prefix and Orthography (Condition 2), and Prefix and Semantic conditions (Condition 3). Results from the most complex model that reached convergence are reported, which included random intercepts for items and participants for both accuracies and RTs. The model simplification process was completed by removing one random slope at a time until the model converged. Confidence intervals and p-values were computed using the Kenward–Roger approach (Kenward and Roger, 1997) by setting the df_method to “Kenward” in the model_parameters() function in parameters package (Lüdecke et al., 2020). All post-hoc analyses were conducted using emmeans() function in emmeans package (Lenth, 2023). Effects were considered statistically reliable for p-values less than 0.05.

8. Results

Table 2 shows by-subject accuracy rates and mean reaction times (RTs) for word targets. Mean RTs for each condition with standard errors are plotted in Figure 1A.

Table 2. By subject accuracy rates and mean RTs for word targets in Experiment 1 (SD in parentheses)

^* p < 0.05 in pairwise comparisons of RT.

^† Prefix priming that is statistically different from orthographic priming (p < 0.05).

Figure 1. Mean RTs and standard errors for target words in Experiments 1–4.

Note. Asterisks indicate p < 0.05.

Accuracy rates were generally high in all conditions, and no significant difference was found between Related versus Unrelated primes in Prefix condition (Odds ratio = 1.124, 95% CI = [0.664, 1.903], p = 0.664), nor did it interact with other conditions (ps > 0.156).

The linear mixed effects regression model with reaction times as a dependent variable showed statistically reliable Prefix priming (Relatedness; β = 0.040, 95% CI = [0.013, 0.068], p = 0.003). Its size was smaller than that of Identity priming (Relatedness × Condition1; β = 0.040, 95% CI = [0.003, 0.078], p = 0.034) but larger than that of Orthography (Relatedness × Condition2; β = −0.054, 95% CI = [−0.094, −0.014], p = 0.008). The size was not statistically different from Semantic priming (Relatedness × Condition3; β = −0.026, 95% CI = [−0.063, 0.012], p = 0.178). In addition, the main effects of Target word frequency (β = 0.068, 95% CI = [0.042, 0.094], p < 0.001), Prime word frequency (β = 0.018, 95% CI = [0.004, 0.033], p = 0.011), and Target word length (β = −0.022, 95% CI = [−0.034, −0.010] p < 0.001) were statistically reliable, such that reaction times were faster for shorter and more frequent target words and also when target words were preceded by more frequent prime words. (See Supplementary Materials for full summary of the model).

Post-hoc pairwise comparisons were conducted to probe priming effects for each of Identity, Orthography and Semantic conditions. Identity priming effects (β = 0.071, 95% CI = [0.043, 098], p < 0.001) were statistically reliable, but not Orthographic priming effects (β = −0.007, 95% CI = [−0.041, 0.028], p = 0.711) or Semantic priming effects (β = 0.014, 95% CI = [−0.012, 0.039], p = 0.290).

9. Discussion

Experiment 1 tested masked prefix priming effects in English with English dominant speakers in comparison to identity, orthography, and semantic priming effects. The results show that only Identity and Prefix conditions show reliable masked priming effects (Identity: 32 ms; Prefix: 29 ms). These results replicate previous studies on prefix priming in monolingual settings (Chateau et al., 2002; Domínguez et al., 2006, 2010; Giraudo and Grainger, 2003) and confirm that prefixes in morphologically complex words are identified and facilitate the subsequent access within the same language.

Importantly, the null priming effects in Orthography and Semantic conditions suggest that the prefix priming effects are not a result of form or semantic overlap but rather of morphological overlap. The statistical distinction between prefix priming and orthographic priming is a replication of Domínguez et al. (2006, 2010), confirming that these two effects are statistically separable with our methods. A comparison between prefix priming and semantic priming has not been reported in previous studies, and the interaction in the current study did not reach statistical significance. Yet, we tentatively suggest that semantic overlap may have a minimal contribution, if any, to prefix priming, given that its priming effects are numerically small (5 ms compared to 29 ms for prefix priming) and statistically unreliable.

In short, the results from Experiment 1 demonstrate robust masked priming effects of prefixes in one’s dominant language. In the following three experiments (Experiment 2–4), we examine prefix priming effects between L1 and L2 and within L2.

10. Experiments 2–4

Experiments 2–4 tested prefix priming in the directions of L1–L2 (Experiment 2), L2–L1 (Experiment 3), and L2–L2 (Experiment 4) by early and late English-Spanish bilinguals to examine whether cognate prefixes in different languages prime each other as well as within one’s second language.

11. Methods

11.1. Participants

A total of 414 English–Spanish bilinguals were recruited via Prolific. All of them were living in the United Kingdom (N = 152) or in the United States (N = 262) at the time of participating in the study. Among them, 135 participants (77 females, 57 males, 1 other response) participated in Experiment 2 (L1–L2), 137 participants (74 females, 61 males, 2 other response) in Experiment 3 (L2–L1) and 142 participants (71 females, 70 males, 1 other response) in Experiment 4 (L2–L2). They were then further grouped into early bilinguals (EB) or late bilinguals (LB) based on their age of acquisition. The former group learned English and Spanish during early childhood (at or before age of 5), while the latter group learned English as their native language and Spanish a second language at or after age of 10 (Heyer and Clahsen, 2015; Veríssimo et al., 2018). Table 3 shows the details of the language background of the groups in each experiment.

Table 3. Language background of participants in Experiments 2–4

11.2. Stimuli

Each experiment stimuli consisted of 240 prime words and 120 target words that were either English or Spanish according to the corresponding language pair of the experiment. The majority of English prime and target words were the same as in Experiment 1 with some adjustments. The prime-target word pairs were non-cognate translation equivalents of each other (“Identity”), or shared the same prefix (“Prefix”), word-initial letters (“Orthography”), or meaning (“Semantic”). Each pair was matched with Unrelated prime-target pairs with no orthographic or semantic relation that were bimorphemic for Prefix condition and monomorphemic for all other conditions. The lexical characteristics of prime words were measured in the same manner as described for Experiment 1. For prime-target pairs that contained Spanish words, semantic similarity was measured based on English translation equivalents of the Spanish words. See Table 4 for example stimuli and a summary of lexical characteristics of target words and prime words for each experiment.

Table 4. Lexical characteristics of target and prime words used in Experiments 2–4 (SD in parentheses)

* Normalized to length.

In addition, 120 Spanish pseudowords were constructed in the same way as English pseudowords used in Experiment 1. Participants in Experiment 2 (L1–L2) and Experiment 4 (L2–L2) saw 120 Spanish pseudowords along with 120 Spanish word targets (15 items per condition) following English or Spanish primes, respectively, whereas those in Experiment 3 (L2–L1) saw 120 English pseudowords and 120 English word targets (15 items per condition) after Spanish primes.

11.3. Procedure

The procedure was the same as in Experiment 1. At the end of each experiment, participants additionally completed Lextale-Esp (Izura et al., 2014) implemented in Psychopy as a measure of their Spanish proficiency ². This test consists of 60 Spanish words and 30 pseudo-words presented one at a time. Participants were asked to decide whether each string of letters is a Spanish word or not.

11.4. Data analysis

For each experiment, participants with accuracy rates less than 70% and items with accuracy rates less than 50% were removed from data analysis as well as reaction times less than 200 ms and greater than 2000 ms. ³

Accuracies were analyzed with a generalized mixed-effects regression model with binomial family. Fixed effects included Relatedness (Related versus Unrelated), Condition (Identity, Prefix, Orthography, and Semantic), and Group (EB versus LB) and their interactions, and centered Target word frequency and Prime word frequency as fixed covariates.

Reaction times were transformed according to the lambda value from the box cox test (Box and Cox, 1964) using the boxcox() function to account for positive skewness. This resulted in inverse-transformation in Experiments 2 and 3 and log transformation in Experiment 4. The transformed RTs were analyzed with a linear mixed-effects regression model with Condition, Relatedness, and Group and their interactions as fixed effects. In addition, Target word frequency, Prime word frequency, Target word length, and Prime word length were centered and included as fixed covariates. For both analyses, the Condition variable was dummy coded with Prefix as the reference level; therefore, each contrast compares Prefix and Identity (Condition 1), Prefix and Orthography (Condition 2) and Prefix and Semantic conditions (Condition 3). Results from the most complex model that reached convergence are reported, which included random slope for Relatedness for item and random intercepts for items and participants for accuracies, and random slopes and intercepts for both accuracies and reaction times. As in Experiment 1, confidence intervals and p-values were derived from the Kenward–Roger approach (Kenward and Roger, 1997) using the model_parameters() function in parameters package (Lüdecke et al., 2020). The alpha was set to 0.05.

12. Experiment 2 (L1–L2 priming) results and discussion

Experiment 2 was conducted to investigate cross-linguistic masked priming effects of prefixes from one’s dominant language (L1, English) to non-dominant language (L2, Spanish).

12.1. Results

Three participants and five items were excluded from data analysis due to low accuracy rates (<70% and <50%, respectively). Following this, reaction times less than 200 ms or greater than 2000 ms were excluded, which accounted for 10.0% of the data.

Participants’ Lextale-Esp scores were computed as percentage of correctly answered word trials minus incorrectly answered pseudoword trials. The mean score for the EB group was 58.6 (SD = 12.19) and for the LB group 57.5 (SD = 13.70) with no significant difference between groups (t (134.21) = 0.51, p = 0.612). All participants fell under the B2 level (upper intermediate) or lower based on their scores (Lemhöfer & Broersma, 2012).

Table 5 shows accuracies and reaction times by the two groups (Also see Figure 1B for reaction times). In the statistical analysis, accuracies were significantly lower in the Prefix condition compared to other three conditions (all p < 0.001), and for less frequent target words (Odds ratio = 2.810, 95% CI = [1.833, 4.308], p = 0.001). Importantly, the effect of Relatedness in the reference (Prefix) condition was statistically significant (Odds ratio = 1.307, 95% CI = [1.040, 1.643], p = 0.022), indicating that accuracy rates were higher when primes and targets shared the same prefix; post-hoc analysis showed that this effect was statistically significant only for the EB group (Odds ratio = 1.441, 95% CI = [1.032, 2.011], p = 0.032) and not the LB group (Odds ratio = 1.156, 95% CI = [0.829, 1.613], p = 0.393) although the interaction between Relatedness and Group did not reach statistical significance (Odds ratio = 0.849, 95% CI = [0.542, 1.330], p = 0.476).

Table 5. By subject accuracy rates and mean RTs for word targets in Experiments 2–4 (SD in parentheses)

* p < 0.05 in pairwise comparisons of RT.

^† Prefix priming that is statistically different from orthographic priming (p < 0.05).

The linear mixed effects regression model showed that prefix priming effects on reaction times did not reach statistical significance (Relatedness; β = 0.023, 95% CI = [−0.001, 0.047], p = 0.060). This effect showed a three-way interaction with Condition 2 (Prefix versus Orthography) and Group (β = 0.063, 95% CI = [0.005, 0.123], p = 0.035). The main effects of Target word frequency (β = 0.076, 95% CI = [0.046, 0.106], p < 0.001) and Target word length (β = −0.044, 95% CI = [−0.056, −0.031], p < 0.001) were also statistically significant, indicating faster reaction times for more frequent and shorter target words. Post-hoc analyses were conducted separately for each group to further investigate the three-way interaction. For the EB group, both Identity priming effects (β = 0.042, 95% CI = [0.019, 0.065] p < 0.001) and Prefix priming effects (β = 0.037, 95% CI = [0.006, 0.069], p = 0.019) were statistically reliable as well as the difference between Prefix and Orthographic priming effects (β = −0.058, 95% CI = [−0.100, −0.016], p = 0.007). Priming effects of Orthography and Semantic conditions were not statistically reliable (all p > 0.122). For the LB group, none of the priming effects reached statistical significance (all p > 0.129), neither was the difference between Prefix priming effects and Orthographic priming effects statistically reliable (β = 0.013, 95% CI = [−0.035, 0.060], p = 0.600).

In summary, only the EB group shows statistically reliable Identity priming effects and Prefix priming effects. In addition, a dissociation of Prefix priming effects and Orthographic priming effects is observed only for the EB group.

12.2. Discussion

Experiment 2 was conducted to test whether prefixes prime from a first to a second language. To this end, 135 English-Spanish bilinguals were recruited, who are further split into two groups based on their age of acquisition of Spanish – early bilinguals (N = 69) and late bilinguals (N = 68). The results show statistically reliable priming effects for prefixes as well as translation equivalents only for the EB group who acquired their second language before the age of 5. Also, the three-way interaction of Group, Relatedness and Condition 2 (Prefix versus Orthography) indicates that the prefix priming effects are dissociable from orthographic priming effects for the EB group, unlike the LB group.

These results suggest that at least for early bilinguals, cognate prefixes in two languages form connected representations. Models on the representation of bilingual lexicon differ in how exactly these connections might be instantiated. For RHM, it could be that cognate prefixes are also stored in separate lexicons, as word representations are, and share conceptual links as well as lexical links to each other. When a morphologically complex word that contains a cognate prefix is presented in one language, the word is decomposed into its root and prefix, which then activates the prefix in the other language via conceptual links and lexical links, thus leading to facilitative cross-language prefix priming effects.

An alternative explanation is shared representation without morphological decomposition. This position is based on distributed approaches which includes the BIA and BIA+ models. In these models, morphology does not form discrete representations in the mental lexicon (Jared et al., 2017; Plaut and Gonnerman, 2000; Stevens and Plaut, 2022). Rather, morphology is an emergent property due to consistent form-meaning mappings. According to these approaches, morphological priming occurs when the consistent form-meaning mappings facilitate recognizing a word. If words in two languages are integrated at multiple levels as assumed in BIA and BIA+ models, it follows that cognate prefixes in two languages may share the same nodes at both orthographic and semantic levels such that activating a prefix in one language speeds up recognizing a word that has the same prefix in the other language. The current study does not aim to directly tease apart these two approaches to the bilingual lexicon. Still, a remaining question is whether cognate prefixes also show asymmetric priming, that is, whether priming effects are smaller, if any, in the L2–L1 direction. This is tested in Experiment 3.

A separate contribution concerns the absence of prefix priming for late bilinguals; this suggests that the representation of cognate prefixes may vary depending on the age of L2 acquisition. The lack of robust priming effects for late bilinguals may be due to either a weaker connection between L1 prefix and L2 prefix or their less sensitivity to L2 morphological structure. We examine these different possibilities further in Experiment 3 and Experiment 4.

Finally, we also observe different patterns between the two groups in translation priming. While such AoA effects on L1–L2 translation priming are not predicted from bilingual lexicon models reviewed thus far, several studies have reported AoA effects on bilingual word recognition, such as in L2–L1 translation priming (Sabourin et al., 2014) and within-L2 categorical priming (Kotz, 2001; Kotz and Elston-Güttler, 2004). Also note that the stimuli used for translation priming in the present study were non-cognate pairs, whose priming effects are known to be less consistent than cognates (e.g., Kirsner et al., 1980; Davis et al., 2010) as they are only conceptually mediated in contrast to cognates that are either lexically mediated (Kroll and Stewart, 1994; Kroll and Tokowicz, 2001, 2005) or share connections at the orthographic level (Van Heuven et al., 1998). Hence, one possible factor for the absence of translation priming from late bilinguals is that late exposure to L2 results in less automated connections at the conceptual level, but this warrants further investigation.

13. Experiment 3 (L2–L1 priming) results and discussion

In Experiment 3, primes were presented in participants’ L2 (Spanish) and targets were presented in their L1 (English).

13.1. Results

Data from one participant from the EB group were excluded due to low accuracy (<70%). Also, reaction times faster than 200 ms or slower than 2000 ms were discarded, which accounted for 3.0% of the whole data.

Average Lextale-Esp scores were 56.5 (SD = 12.37) for the EB group and 59.0 (10.75) for the LB group, with no significant difference between groups (t(133.79) = −1.22, p = 0.226).

Table 5 shows by subject accuracies and reaction times for each condition. The logistic regression model indicated that accuracies were higher for Related primes in Prefix condition than Unrelated primes (Odds ratio = 1.537, 95% CI = [1.154, 2.045], p = 0.003), and this effect did not interact with Group (Odds ratio = 0.833, 95% CI = [0.509, 1.365], p = 0.469). Post-hoc analysis that examined this effect separately for each group indicated statistical insignificance for both the EB group (Odds ratio = 1.350, 95% CI = [0.970, 1.880], p = 0.075) and the LB group (Odds ratio = 1.689, 95% CI = [0.935, 3.050], p = 0.082).

Reaction times are plotted in Figure 1C. We observe statistically significant Prefix priming effects with the main effect of Relatedness (β = 0.039, 95% CI = [0.017, 0.061], p < 0.001) that are statistically larger than Identity priming effects (Relatedness × Condition1; β = −0.031, 95% CI = [−0.061, −0.001], p = 0.042), but these effects are not dissociable from Orthographic or Semantic priming effects as Relatedness does not interact with Condition 2 (β = 0.012, 95% CI = [−0.039, 0.063], p = 0.650) or Condition 3 (β = 0.023, 95% CI = [−0.028, 0.073], p = 0.377). Reaction times were also faster for short target (β = −0.020, 95% CI = [−0.033, 0.006], p = 0.004) and prime words (β = −0.007, 95% CI = [−0.012, −0.002], p = 0.009). There are no significant interactions concerning Group (p > 0.209).

In the post-hoc analysis, the EB group showed statistically reliable Prefix priming effects (β = −0.042, 95% CI = [0.012, 0.072], p = 0.006), but these effects did not differ from Orthographic priming effects (β = −0.030, 95% CI = [−0.072, 0.013], p = 0.169) and Semantic priming effects (β = −0.040, 95% CI = [−0.082, 0.001], p = 0.055). Also, other conditions (Identity, Orthography and Semantic) did not show statistically significant priming effects (all p < 0.240). The LB group showed a similar pattern: Prefix condition had statistically reliable priming effects (β = 0.036, 95% CI = [0.009, 0.064], p = 0.010), but did not differ from priming effects in Orthography or Semantic conditions (all p > 0.164). No statistically reliable priming effects were found in other conditions (all p < 0.164).

In summary, both the EB group and the LB group showed facilitatory priming effects in Prefix condition, but these effects were not statistically dissociable from priming effects stemming from Orthographic or Semantic overlap.

13.2. Discussion

In Experiment 3, prefix priming was investigated from participants’ L2 to L1. Unlike Experiment 2, cross-language prefix priming effects for the EB group are not statistically dissociated from orthographic priming effects, nor do translation equivalents show robust priming effects. The LB group showed a similar pattern, where priming effects for Prefix and Orthography conditions do not interact. Given the similar Levenshtein distance between related versus unrelated prime words and target words in Prefix and Orthography condition, the absence of interaction leaves it unclear whether the observed prefix priming effects are purely morphological or are due to form overlap. Taking together the results from Experiments 2 and 3, we find asymmetry in both Identity and Prefix priming effects for the EB group, where the priming effects are reliable in the L1–L2 direction but not in the L2–L1 direction.

Asymmetric priming effects for translation equivalents (Identity condition in the current study) align with previous studies (e.g., Jiang, 1999; Smith et al., 2019; Wang, 2013). Models of the bilingual lexicon provide explanations for such asymmetry that are based on weaker representations of L2 primes, either because they have less substantial connections to concepts (RHM) or they are harder to recognize due to low resting activation level (BIA, BIA+). Another explanation comes from Smith et al. (2019), who attribute the asymmetry to L2 targets having more room for priming due to their overall slower response times. In that study, L2–L2 identity priming yielded larger priming effects than L1–L1 identity priming among unbalanced Hebrew-English bilinguals with mean AoA of 6.5. The authors also observed larger priming effects of translation equivalents for L1–L2 than L2–L1. If the stronger connection of L1 words to concepts is the only factor for the asymmetry, then L1–L1 identity priming effects should be larger than L2–L2 identity priming. Therefore, the results suggest that relatively lower resting levels of L2 targets also play a role in the asymmetry — because L2 words are generally less activated for unbalanced bilinguals, they benefit more from priming. While these findings provide a convincing explanation for the asymmetry, it remains to be tested whether the early bilinguals with younger AoA as in the current study (less than 5) also have lower resting levels for L2 than L1 words.

As for prefixes, there are two possible reasons for the lack of purely morphological priming effects in the L2–L1 direction for both early bilinguals and late bilinguals. It can either be the case that prefixes in L2 primes are identified but do not prime those in L1 as is the case for translation equivalents, or they are not identified in the first place, especially when they are only briefly presented (50 ms in the current experiment). In Experiment 4, we aim to tease apart these two possibilities by testing prefix priming effects within participants’ L2. If prefixes in L2 masked primes are identified, we expect to find robust prefix priming effects, thereby rejecting the second possibility.

14. Experiment 4 (L2–L2 priming) results and discussion

Experiment 4 tested prefix priming within one’s L2. Therefore, both primes and targets were presented in Spanish.

14.1. Results

One participant from the EB group and seven participants from the LB group were removed due to low accuracy rates (<70%). Five items with low accuracy rates (<50%) as well as reaction times faster than 200 ms or slower than 2000 ms were also excluded from data analysis. The data removal based on the reaction times accounted for 14.6% of the data.

The mean scores for Lextale-Esp were 58.1 (SD = 13.95) for the EB group and 58.1 (14.97) for the LB group. The difference between groups was not statistically significant (t(142.24) = −0.02, p = 0.982).

See Table 5 for by subject accuracy rates and mean RTs. Reaction times are also plotted in Figure 1D. Analysis on accuracy data revealed no statistically reliable effect of Group, Condition, or Relatedness (all p > 0.190).

The output of the linear mixed effects regression model for reaction times showed statistically reliable Prefix priming effects (β = −0.025, 95% CI = [−0.050, −0.000], p = 0.046) but these effects did not differ from other priming effects including Orthography (Relatedness × Condition 2; β = 0.020, 95% CI = [−0.011, 0.051], p = 0.210) and Semantic (Relatedness × Condition 3; β = 0.019, 95% CI = [−0.012, 0.049], p = 0.227) condition. In addition, target frequency (β = −0.073, 95% CI = [−0.099, −0.046], p < 0.001) and target length (β = 0.032, 95% CI = [0.023, 0.044], p < 0.001) had reliable effects on the reaction times.

Post-hoc analyses examined priming effects separately for each group. The EB group showed statistically significant Identity priming effects (β = −0.026, 95% CI = [−0.050, −0.002], p = 0.035) but not Prefix priming effects (β = −0.029, 95% CI = [−0.063, 0.004], p = 0.086). Orthography and Semantic conditions did not show statistically significant priming effects (all p < 0.732). For the LB group, none of the priming effects in the four conditions reached statistical significance (all p > 0.096).

14.2. Discussion

In Experiment 4, prefix priming within one’s non-dominant language was examined. The results indicate that prefix priming effects for both the EB and the LB groups are not statistically reliable, indicating that they may not be able to rapidly identify prefixes in multimorphemic words that are visually presented very briefly in their less dominant language. This result further suggests that the prefix priming effects observed in Experiment 3 may be more likely due to form and/or semantic overlap between prime and target words than morphological overlap.

The current results show that prefixes do not yield robust masked priming effects in one’s less dominant language in contrast to the results from one’s dominant language (Experiment 1). These results also contrast with L2 root priming studies (e.g., Diependaele et al., 2011) that report robust morphological root priming effects in participants’ L2 (e.g., viewer-VIEW). In short, bilinguals with an intermediate-level L2 proficiency may not be able to identify prefixes as opposed to roots in masked L2 complex words, regardless of their AoA. These results are in line with the Word and Affix model proposed in Beyersmann and Grainger (2023) according to which the processing mechanism of affixes is distinct from that of roots as affixes, but not roots, are subject to positional constraints. In this sense, the development of processing affixes may be slower than that of roots similar to L1 morphological development among young children (e.g., Dawson et al., 2018; Hasenäcker et al., 2020). Future studies may examine whether L2 speakers with higher proficiency than those in the current study show robust prefix priming in their L2 to determine whether increased proficiency enhances affix processing.

Results thus far from each experiment have been analyzed separately. The next section quantifies the relative size of prefix priming effects per experiment and per group in a single linear mixed effects model.

15. Comparison of prefix priming between experiments

Qualitatively comparing significant versus non-significant results does not entail a statistically reliable interaction (e.g., Gelman and Stern, 2006). Thus, to better evaluate the similarities and differences across experiments we fit a linear mixed effects model that aggregates Prefix data from Experiments 2, 3, and 4. This allows us to examine whether the magnitude of prefix priming across the experiments and groups is different. Data from Experiment 1 were not included in this analysis because they do not have the group factor. The statistical model included Relatedness (Related versus Unrelated), Experiment (Experiment 2, 3, or 4), and Group (EB or LB) and their interactions as fixed effect and Target word frequency, Prime word frequency, Target word length, and Prime word length as covariates. Binary and continuous factors were centered. The factor Experiment was coded using simple contrast coding with Experiment 2 as reference, such that the intercept is the grand mean of all the three experiments, while the two contrasts compare Experiment 2 and Experiment 3, and Experiment 2 and Experiment 4, respectively. The most complex random effects that converged included random slopes for Relatedness and random intercepts for both participant and item.

The results showed statistically reliable main effect of Relatedness (β = −0.037, 95% CI = [−0.063, −0.012], p = 0.004), which means that prefix priming effects are robust across the three experiments and two groups. The two-way interactions between Relatedness and Experiment 2 versus 3 and Relatedness and Experiment 2 versus 4 were not statistically significant (all p > 0.431), indicating that the size of prefix priming effects is not statistically dissociable between Experiment 2 and Experiment 3 and between Experiment 2 and Experiment 4.

16. General discussion

The main goal of the current study was to examine the representation of prefixes within and across L1 and L2 using the masked priming paradigm. In Experiment 1, we replicated robust prefix priming in one’s dominant language as reported in previous studies (Ciaccio et al., 2020; Chateau et al., 2002; Dominguez, 2006, 2010; Giraudo and Grainger, 2003). Priming effects of orthographic or semantic overlap, on the other hand, are not statistically reliable, indicating that priming effects of prefixes cannot be attributed to those factors. In other words, in one’s dominant language, prefixes form independent representation units in the mental lexicon that are not reduced to orthographic or semantic relations, and these units facilitate subsequent access as do roots. These results provide a foundation to probe prefix priming across different languages (Experiment 2 and Experiment 3) and within one’s less dominant language (Experiment 4). The rest of this section discusses the results from Experiment 2–4. Separate analyses in each experiment suggest that prefix priming effects are less reliable and not dissociable from orthographic priming in certain situations. When aggregating across experiments however we do not find significant interactions with experiment. We therefore proceed to discuss these apparent patterns with caution.

17. Cross-language prefix priming

Experiment 2 tested prefix priming in the L1–L2 direction, where primes and targets are in different languages but share prefixes that have the same form and meaning. Experiment 3 tested the same effect in the opposite L2–L1 direction. While the comparison between experiments indicates no significant difference in the size of prefix priming between Experiments 2 and 3, separate analyses for the two experiments show that for early bilinguals, these cognate prefixes are primed independently from orthographic overlap when primes are presented in L1, but not when primes are presented in L2. Late bilinguals, on the other hand, do not show evidence of purely morphological priming effects in either direction.

To our knowledge, this is the first time to show cross-language prefix priming effects in the L1-L2 direction. These results suggest that among the early bilinguals, cognate prefixes are mapped onto a shared representation. While we do not attempt to tease apart the two bilingual models introduced in the Introduction (RHM and BIA+), we acknowledge that this claim has different implications for each model when implemented. Within the framework of RHM, this can be explained by postulating that prefixes are represented in the same way as monomorphemic words; those cognate prefixes in L1 and L2 are connected by lexical links and also are connected to a single concept node by conceptual links. When an L1 prefixed word is encountered, for example, it is decomposed into a prefix and a root, each of which then activates its linked concept and L2 counterpart, enabling cross-linguistic prefix priming as well as root priming. According to the BIA+, cognate prefixes in the two languages are connected at the orthographic and semantic levels. Within this framework, the consistent mapping of form and meaning of these prefixes in the two languages yield between-languages prefix priming without a separate process of decomposition.

The results are also in line with previous literature on language transfer, specifically of morphology (Marks et al., 2023; Ramírez et al., 2011; Wang et al., 2022), and convergence (Baptista et al., 2016; Muysken, 2000, 2013; Weinreich, 1953). For instance, in Ramírez et al. (2011), English learners with L1 Spanish and those with L1 Chinese outperformed each other on the component of English morphology that is similar to their L1: derivational morphology for Spanish learners of English and compounds for Chinese learners of English.

In a similar vein, according to the convergence hypothesis, particularly the isomorphic hypothesis, the features that are common in the two languages are prioritized and learned faster (Muysken, 2000, 2013; Weinreich, 1953). One piece of supporting evidence for this hypothesis comes from Baptista et al. (2016), where English speakers learned an artificial language. After one session of a learning phase, participants who were exposed to an artificial language that have morphemes similar to English (“nat” for negation and “iss” for plurality) produced them more accurately than those exposed to the reverse condition (“iss” for negation and “nat” for plurality) and the novel condition (“plick” for negation and “mut” for plurality). This suggests that L2 morphemes that have a similar form and function to those in L1 may map onto them, thus gaining advantage during the learning process.

One crucial point worth noting is that the cognate prefix priming reported in the study is seemingly contradictory with prior results for cognate root priming, which was not dissociable from orthographic priming between Hebrew and French (Voga and Grainger, 2007). However, the participants in Voga and Grainger (2007) come from populations more similar to late bilinguals than early bilinguals in the current study based on the description that they were “taught French as a second language during school years (p. 940).” Future research with early bilinguals is necessary to examine whether they show reliable cognate root priming effects that are greater than form based priming effects.

In short, by using the priming paradigm with bilinguals, the present study provides more direct evidence that such morphemes form shared morphological representation among individuals who acquired two languages in a naturalistic setting. On the other hand, the early bilingual group did not show reliable prefix priming effects in L2–L1 direction (Experiment 3). The lack of its interaction with those in the L1–L2 direction (Experiment 2) calls for careful interpretation. Yet, the results from Experiment 4 where participants did not show uniquely morphological L2–L2 prefix priming effects suggests that such asymmetry arises because early bilinguals are not able to rapidly identify prefixes from L2 masked words. Nevertheless, it is also possible that even when L2 words are presented for sufficient time, cross-language prefix priming effects are not observed, which would then indicate that prefixes, even when fully recognized, do not prime in the L2–L1 direction. Such asymmetry is predicted by the discussed bilingual models, if we assume that prefixes are represented in a similar way to words (i.e., RHM: weaker conceptual links between L2 prefix and concept, BIA+: lower resting levels of L2 prefixed words). Further research would be necessary to test this possibility.

We do not observe any reliable cross-language prefix priming effects from the late bilinguals, both in the L1–L2 and L2–L1 directions. These results, especially in the L1–L2 direction, contrast with some assumptions of prominent models of the bilingual lexicon, specifically that novice bilinguals will show a stronger asymmetry in cross-language priming. With the current method that requires very fast recognition of prefixes in L2 words in order for any priming to occur, the results suggest that late bilinguals may struggle to achieve the same level of prefix recognition as early bilinguals (c.f., Babcock et al., 2012; Basnight-Brown and Altarriba, 2007; Veríssimo et al., 2018). One limitation of this method is that it does not rule out the possibility that late bilinguals show cross-languages prefix priming effects when they are allowed sufficient time to process prefixes. Therefore, it may be premature to conclude solely from the current study that cognate prefixes are stored completely separately for late bilinguals. Implementing a research design that complements the current methods, such as presenting prime words for longer duration or using a long-lag priming paradigm where participants perform a lexical decision task on prime words as well, would be a useful next step.

Finally, although the current results suggest differences between early bilinguals and late bilinguals in the way they process cognate prefixes, factors other than age of acquisition may also contribute to the difference. While the L2 proficiency measured by Lextale-Esp is comparable between the two groups across the three experiments, the amount of L2 daily use is significantly greater for early bilinguals than late bilinguals in all three experiments (all p < 0.001). This may be in part inevitable as half of the early bilinguals (52%) in the current study reported to be heritage Spanish speakers and thus listed family members or parents as their primary source of Spanish education, whereas only three late bilinguals were in this category. Most late bilinguals (87%) reported that they learned Spanish from school, private institute, or tutoring. A study that compares late bilinguals with different percentage of L2 daily use, such as those living in English speaking countries and those living in Spanish speaking countries, would be able to disentangle the effects of age of acquisition from language exposure.

18. L2 prefix priming

In addition to between-languages prefix priming, we also tested within-language prefix priming in one’s dominant language (Experiment 1) and in one’s less dominant language (Experiment 4) to examine whether the latter case also yields reliable priming effects. In contrast to robust prefix priming effects found in Experiment 1, these effects were not observed in Experiment 4 for both early bilinguals and late bilinguals.

As in the case of Experiment 3, however, the statistical comparison of effect sizes across experiments did not reveal statistically significant interaction of priming effects between Experiment 2 and Experiment 4, which again warrants caution in interpreting the results. Keeping that in mind, we tentatively suggest that these findings indicate that the processing mechanisms of prefixes may be different in L1 versus L2. This is partially similar to what Heyer and Clahsen (2015) report on the processing of derived words by L1 versus L2 speakers. In their study, in contrast to L1 speakers who showed priming effects only for morphological overlap (e.g., darkness – DARK), L2 speakers showed priming effects for both morphological overlap and orthographic overlap (e.g., example – EXAM). These results are interpreted as L2 learners’ higher reliance on the orthographic relationship than the morphological relationship when processing morphologically complex L2 words. While the current results are similar to that study in that purely morphological L2 priming effects are not observed, they are different in the sense that we did not find orthographic priming effects, either. One possible reason for such discrepancy is that the stimuli in Orthography condition in the current study have a smaller number of overlapping letters (2–3 letters) compared to those used in Heyer and Clahsen (2015) (3–7 letters).

19. Conclusion

In summary, we tested prefix priming within and between languages across four experiments: L1-L1 (Experiment 1), L1–L2 (Experiment 2), L2–L1 (Experiment 3), and L2–L2 (Experiment 4) with AoA as a modulating factor. Following the robust prefix priming effects within one’s L1 observed in Experiment 1, a key finding of the current study is comparable prefix priming effects from one’s L1 to L2 by early bilinguals (Experiment 2). Such effects were not dissociated from orthographic priming in the L2–L1 direction and not observed within their L2. We also find an effect of AoA and possibly the percentage of L2 daily use on cross-language prefix priming as late bilinguals did not show robust prefix priming in any of the directions as well as within the L2. Based on these results, we conclude that cognate prefixes map onto the same representation for early bilinguals, but possibly not for late bilinguals.