Introduction
Not all words are created equal; some words are easier to learn than others and this can be due to a number of features such as concreteness, word type, frequency, cognate status, length, and phonology (De Groot & Keijzer, Reference De Groot and Keijzer2000; Lotto & De Groot, Reference Lotto and de Groot1998; Morra & Camba, Reference Morra and Camba2009; Vidal, Reference Vidal2011). For example, new foreign language (FLFootnote 1) words are learned more easily if their phonology follows the phonotactic rules of the native language (Ellis & Beaton, Reference Ellis and Beaton1993; Kaushanskaya et al., Reference Kaushanskaya, Marian and Yoo2011; Morra & Camba, Reference Morra and Camba2009). Another important word-related feature that can impact learning is cross-linguistic orthographic similarity, or how similar FL words look compared to native language words (Bartolotti & Marian, Reference Bartolotti and Marian2017; Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; Ellis & Beaton, Reference Ellis and Beaton1993). What are the mechanisms then that support learning of words with familiar and unfamiliar orthographies? Prior research has highlighted various cognitive mechanisms with a role in FL word learning, for example, short-term and/or working memory (Bisson et al., Reference Bisson, Kukona and Lengeris2021; Martin & Ellis, Reference Martin and Ellis2012; Morra & Camba, Reference Morra and Camba2009), as well as phonological abilities (Bisson et al., Reference Bisson, Kukona and Lengeris2021; Hu, Reference Hu2012; Morra & Camba, Reference Morra and Camba2009; Vijayachandra, Reference Vijayachandra2007). The following section highlights the current findings regarding the role of orthography in FL word learning. Then, the cognitive mechanisms supporting word learning are reviewed. The final section introduces the current study that investigates the interaction between cross-linguistic orthographic similarity (word-related feature) and cognitive abilities (individual differences).
The role of orthography in FL learning
Research on the impact of orthography on FL word learning has not been extensive (Simon & van Herreweghe, Reference Simon and van Herreweghe2010); however, several studies suggest it is an important factor (Bartolotti & Marian, Reference Bartolotti and Marian2017; Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; Ellis & Beaton, Reference Ellis and Beaton1993). The cross-linguistic orthographic regularity of FL words is typically measured through lexical variables (e.g., n-gram probabilities, n-gram frequencies, or neighborhood density) or psycholinguistic variables (e.g., judgments/ratings of wordlikeness, typicality, similarity, etc.; Dijkstra et al., Reference Dijkstra, Miwa, Brummelhuis, Sappelli and Baayen2010; Tokowicz et al., Reference Tokowicz, Kroll, De Groot and Van Hell2002). Therefore, it is a measure of how closely the FL words follow the spelling patterns of the native language, or how similar/typical the FL words look compared to native language words. For example, the Welsh word llygad (meaning “eye”) would score low on an n-gram probability measure and would be judged dissimilar or atypical to English words, as words in English never start with the bigram ll. Ellis and Beaton (Reference Ellis and Beaton1993) used average bigram frequencies and minimum bigram frequency to predict native language to FL and FL to native language translation accuracy. Their measures of cross-linguistic orthographic regularity were not significant predictors in their causal path analyses; however, they found a correlation of .30 between minimum bigram frequency and native language to FL translation. Using a similar classification of cross-linguistic orthographic regularity (neighborhood size, positional segment and bigram frequencies), Bartolotti and Marian (Reference Bartolotti and Marian2017) found that orthographically wordlike nonwords were easier to learn during a picture-word paired-associate learning task. Once they controlled for phonological wordlikeness however, their wordlike advantage only remained significant during written word production (as opposed to a meaning recognition task). Another study found some impact of FL orthotactic probabilities on FL word learning (Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016). Similarly to studies by Ellis and Beaton (Reference Ellis and Beaton1993) and Bartolotti and Marian (Reference Bartolotti and Marian2017), part of the target item selection process involved calculating bigram and trigram probabilities, however Bordag et al. (Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016) also collected ratings of wordlikeness, and only targets judged either atypical or typical on a scale of 1 to 6 (the end points of a typicality continuum) were included. Target items (German pseudowords) were learned through either an incidental or intentional learning phase where participants either read texts without their knowledge of the word learning aspect of the study or read definitions of novel words for a follow-up test. All participants then completed a self-paced reading task where the reading time on the target items was compared in semantically plausible versus implausible sentences. Results showed orthotactic probabilities only impacted reading times in the intentional learning group. Therefore, orthographic similarity, as measured through lexical or psycholinguistic variables seems to predict different aspects of word learning linked to the recall or processing of the written forms of the FL words in some learning contexts only (i.e., intentional learning).
Cognitive mechanism supporting word learning
As well as word-level predictors, individual differences are known to impact FL word learning (Bisson et al., Reference Bisson, Kukona and Lengeris2021; Hu, Reference Hu2012; Kapa & Colombo, Reference Kapa and Colombo2014; Martin & Ellis, Reference Martin and Ellis2012; Morra & Camba, Reference Morra and Camba2009; Vijayachandra, Reference Vijayachandra2007). In Bisson et al. (Reference Bisson, Kukona and Lengeris2021), a composite score including short-term and working memory was a significant predictor of FL word learning. The tasks for phonological and visuo-spatial short-term memory included a storage element only (digits and red dots on a grid, respectively) whereas the working memory tasks involved both storage and processing (Conway et al., Reference Conway, Kane, Bunting, Hambrick, Wilhelm and Engle2005; Martin & Ellis, Reference Martin and Ellis2012). These four tasks shared the underlying requirement to encode serial information about the to-be-remembered items and this was proposed as an important aspect of FL word learning. For example, the syllables that compose a FL word must be recalled in the correct order otherwise this may lead to communication or comprehension errors. As well as short-term and working memory, other important language learning skills include phonological abilities (Bisson et al., Reference Bisson, Kukona and Lengeris2021; Hu, Reference Hu2012; Vijayachandra, Reference Vijayachandra2007). In children this is sometimes referred to as phonological awareness or sensitivity and measured using tasks such as rhyme and alliteration awareness, phoneme and vowel deletion as well as syllable and rhyme discrimination (Hu, Reference Hu2012; Morra & Camba, Reference Morra and Camba2009; Vijayachandra, Reference Vijayachandra2007). It is thought to be important for the perception, manipulation and encoding of speech and nonspeech sounds (Melby-Lervåg et al., Reference Melby-Lervåg, Lyster and Hulme2012; Morra & Camba, Reference Morra and Camba2009; Silbert et al., Reference Silbert, Smith, Jackson, Campbell, Hughes and Tare2015). Although prior research has shown that phonological abilities are important for children language learners, Bisson et al. (Reference Bisson, Kukona and Lengeris2021) expanded this to adult language learners using phonological and acoustic discrimination tasks. These tasks focused on fine-tuned discrimination abilities (Lengeris & Hazan, Reference Lengeris and Hazan2010; Silbert et al., Reference Silbert, Smith, Jackson, Campbell, Hughes and Tare2015) and showed that being able to encode precise phonological representations is an integral part of FL learning.
Cognitive mechanisms and orthography
As discussed so far, both individual differences and word-level variables are important predictors of language learning. Much prior research has investigated these in isolation, but interactions between the two may be important to consider (Housen & Simoens, Reference Housen and Simoens2016). In their taxonomy of second language learning difficulty, Housen and Simoens (Reference Housen and Simoens2016) explain that feature-related difficulties, context-related difficulties, and learner-related difficulties are all important to study for the field of language learning to evolve, but this will be restricted unless we also investigate how they interact. Recent research showed an interaction between context-related difficulty (learning conditions) and learner-related difficulty (working/short-term memory; Bisson et al., Reference Bisson, Kukona and Lengeris2021), but there is no prior research to our knowledge investigating the interaction between feature-related and learner-related difficulties.
The current study
The aims of the current study are therefore to (1) expand the findings on the role of orthography in FL learning and (2) to investigate the interaction between orthography (feature-related variable) and cognitive abilities (memory and phonological abilities; learner-related variables). To classify the FL word’s orthography, judgments of familiarity, similarly to the typicality ratings used in Bordag et al. (Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; but see also see Dijkstra et al., Reference Dijkstra, Miwa, Brummelhuis, Sappelli and Baayen2010; Tokowicz et al., Reference Tokowicz, Kroll, De Groot and Van Hell2002) were collected. A group of native English speakers judged pairs of FL words as to which one was more similar to native language words (the comparative judgment approach; Bisson et al., Reference Bisson, Gilmore, Inglis and Jones2016, Reference Bisson, Gilmore, Inglis and Jones2020; Jones et al., Reference Jones, Bisson, Gilmore and Inglis2019). The word’s phonology was controlled for (as in Bartolotti & Marian, Reference Bartolotti and Marian2017) through judgments of phonological familiarity. The raw data for this study was obtained from Bisson et al. (Reference Bisson, Kukona and Lengeris2021). For individual differences the study focused on the cognitive abilities found to be significant predictors in Bisson et al. (Reference Bisson, Kukona and Lengeris2021), which included a “memory” composite score (short-term and working memory) and a “phonological abilities” composite score (phonological and acoustic abilities).Footnote 2 Because individual differences in vocabulary knowledge are linked to better language learning and processing skills, this was controlled in the current study (Bisson et al., Reference Bisson, Kukona and Lengeris2021; Mainz et al., Reference Mainz, Shao, Brysbaert and Meyer2017; Morra & Camba, Reference Morra and Camba2009). Prior research on the role of orthography in word learning measured learning through recall or processing of FL word form. Here prior research was expanded by measuring learning at the level of meaning recall. Based on prior research, it was predicted that orthography would be a significant predictor of meaning recall: The meaning of words with more unfamiliar orthographies would be more difficult to recall. As mentioned earlier, prior research has not addressed the interaction between orthography and cognitive abilities. However, based on the research of Bisson et al. (Reference Bisson, Kukona and Lengeris2021) who found an interaction between cognitive abilities and learning conditions, it was predicted that cognitive abilities would interact with orthography such that participants with better cognitive abilities would be less affected by unfamiliar orthography. In other words, cognitive mechanisms would compensate for the difficulty of the items.
Method
Ethical considerations
Ethical approval was granted by the Faculty of Health and Life Sciences Research Ethical Committee at De Montfort University. All participants gave informed consent to take part in the studies.
Participants
The final participant sample included 132 native English speakers from a UK Midlands University (20 males, M age = 20.29, SD = 3.96). Participants generally reported having some knowledge of FLs and a moderate level of fluency (M number of FLs = 2.39, SD = 1.20; M fluency = 3.59, SD = 1.71 on a scale ranging from 1 = poor to 7 = fluent). One bilingual participant, as well as four participants with prior knowledge of Welsh were excluded. Two participants were excluded for achieving lower than chance on the letter-search task (incidental learning) and one because they reported having cognitive difficulties. A further 12 participants did not complete Session 2 and were therefore excluded, and 7 participants were removed due to technical difficulties during one of the tasks.
Another sample of 16 native English speakers was recruited to provide familiarity data (4 males M age = 21.23). These participants were also students at a UK Midlands University and they had no prior knowledge of Welsh.
Material
Three lists of 40 depictable Welsh words were used for the purpose of the learning and recall tasks (see Stimuli.txt on https://doi.org/10.21253/DMU.14170832.v1). This included auditory and written word forms as well as a picture depicting the meaning of the words. Auditory word forms were recorded by a Welsh native speaker, and pictures were selected from Brodeur et al. (Reference Brodeur, Dionne-Dostie, Montreuil and Lepage2010) or Moreno-Martínez and Montoro (Reference Moreno-Martínez and Montoro2012). One list of words was used for the incidental learning task and one list was used for the intentional learning task (one list was use for an additional task in Bisson et al., Reference Bisson, Kukona and Lengeris2021) counterbalanced across participants. All computer tasks were presented using PsychoPy (Peirce et al., Reference Peirce, Gray, Simpson, MacAskill, Höchenberger, Sogo, Kastman and Lindeløv2019).
Procedure
Participants completed all tasks over two sessions on consecutive days (Session 1: incidental learning followed by intentional learning tasks, language background questionnaire, short-term memory tasks, phonological and acoustic tasks, and Session 2: meaning recall, working memory tasks, native language vocabulary knowledge; see Figure 1). Participants also completed additional tasks not included here (see Bisson et al., Reference Bisson, Kukona and Lengeris2021).
Figure 1. Graphical representation of the individual differences task, learning and recall tasks as well as the familiarity judgment tasks (the latter completed by a separate sample of participants). Pictures and faces taken from Brodeur et al. (Reference Brodeur, Dionne-Dostie, Montreuil and Lepage2010), Moreno-Martínez and Montoro (Reference Moreno-Martínez and Montoro2012), and Burton et al. (Reference Burton, White and McNeill2010).
Learning tasks
Participants completed both an incidental and an intentional learning task. For incidental learning, participants were asked to complete a letter-search task (as in Bisson et al., Reference Bisson, van Heuven, Conklin and Tunney2013, Reference Bisson, van Heuven, Conklin and Tunney2014, Reference Bisson, van Heuven, Conklin and Tunney2015). Participants were not informed about the upcoming vocabulary test and they were not asked to learn the meaning of the words, thus creating the conditions in which incidental learning could occur (see Hulstijn, Reference Hulstijn and Robinson2001). On each trial, participants first saw a blank screen for 500 ms followed by a letter presented for 1 second, and finally the written word appeared. Participants were instructed to indicate with a button press whether the letter they saw was present in the written word. Even though these were not necessary for the letter-search task, at the same time as the written word onset, participants were also exposed to the auditory word form and to a picture depicting the meaning of each word. The written word and picture remained onscreen until the end of the trial and the auditory word form was played once. Stimuli presentation duration was set to 3 seconds to allow for encoding of information without being too lengthy. Participants first completed 10 practice trials with feedback before completing three blocks of 80 trials without feedback. Each Welsh word was presented twice in each block, once with a letter that was in the word and once with a letter that was not. The order of the trials in each block was randomized. For the intentional learning task, participants were now specifically instructed to learn the meaning of 40 Welsh words. The stimuli were presented exactly in the same manner as in the incidental learning, except that there was no letter to search and no buttons to press. However, participants were now informed that they would be tested on the words the next day. Each learning task lasted 20 minutes.
Meaning recall task
Participants were presented with each Welsh word from the incidental and intentional learning tasks in random order. They were asked to type the English translation for each word. Participants were encouraged to try to answer each trial, but they could also just press “enter” to proceed to the next trial. Welsh auditory word forms were presented once on each trial, however the written word forms remained on screen until participants completed their answer. There was no time limit to complete this task and on average participants took approximately 5 minutes to complete it.
Short-term/working memory tasks
Participants completed both a verbal and a visuo-spatial short-term and working memory task. For the verbal short-term memory task, participants were presented with digits from 1 to 9 and upon the presentation of a question mark had to recall the digits in the correct order using the keyboard. For the visuo-spatial short-term memory, the to-be-remembered stimuli was a red circle on a 3 × 3 grid and participants recalled using mouse clicks on an empty grid. For both verbal and visuo-spatial working memory tasks, in between each storage element described in the preceding text, participants had to judge whether pairs of faces were from the same or different people (Hubber, Reference Hubber2015). Spans ranged from three to nine elements presented in random order and participants completed three trials of each span length. For each memory task, the proportion of correctly recalled items in the correct order was calculated on each trial (Conway et al., Reference Conway, Kane, Bunting, Hambrick, Wilhelm and Engle2005).
Phonological/acoustic abilities
A same-different A-X discrimination task was used (see Lengeris & Hazan, Reference Lengeris and Hazan2010; Silbert et al., Reference Silbert, Smith, Jackson, Campbell, Hughes and Tare2015) with FL (/pita/ - /peta/) and native language (/beat/ - /bit/) continuums for phonological abilities as well as a nonspeech continuum (F2) for acoustic abilities. Participants had to indicate on each trial whether two words or sounds (beeps) were the same or different. Stimuli varied in their similarity to the end point of the continuums in nine steps such that on some trials, stimuli were easier/harder to discriminate. There was one block of 32 trials (16 “same” and 16 “different” trials) for each continuum and the order of trials and blocks was randomized.
Language background
Participants completed 72 trials of the Peabody Picture Vocabulary Test (PPVT; Dunn & Dunn, Reference Dunn and Dunn2007) and received one point for each correct answer. The most difficult items (157–228) were used. Participants also completed a self-reporting language background questionnaire where they were asked to indicate what FLs they knew and to rate their fluency in each of reading, written, listening, and speaking on a scale of 1 = poor to 7 = fluent (see Bisson et al., Reference Bisson, Kukona and Lengeris2021).
Familiarity judgments
The comparative judgment approach (Bisson et al., Reference Bisson, Gilmore, Inglis and Jones2016, Reference Bisson, Gilmore, Inglis and Jones2020; Jones et al., Reference Jones, Bisson, Gilmore and Inglis2019) was used to rate the Welsh words from most unfamiliar to most familiar. For orthographic familiarity, participants were shown two of the 120 Welsh words on each trial on the left and right side of the screen (see Figure 1). They were asked to select which one looked more similar to an English word. Following this, participants rated the words for phonological familiarity using the same method. They heard 2 of the 120 items one after the other on each trial and were asked to indicate which one sounded more similar to an English word by pressing “1” or “2.” Each participant received a different trial file to ensure that each item was compared against different items across participants. Each participant completed 120 judgments in random order and each item was judged 32 times across participants.
Results
The data and analysis script for this study are available at https://doi.org/10.21253/DMU.14170832.v1.
Orthographic and phonological familiarity judgments
The judgments were highly reliable (Scale Separation Reliability index = .90 and .82 for orthographic and phonological familiarity judgments, respectively; Verhavert et al., Reference Verhavert, De Maeyer, Donche and Coertjens2018). Using the Bradley–Terry statistical model (Firth, Reference Firth2005) a z-score parameter estimate was calculated for orthographic and phonological familiarity for each item. Figure 2 shows a ranking order from most unfamiliar (low score) to most familiar (high score) item. For example, the three items with the lowest ranking and hence deemed most unfamiliar were “ymennydd,” “cyfrifiadur,” and “cwmwl” (brain, computer, and cloud, respectively) and the three items with the highest ranking were “crys,” “haul,” and “clust” (shirt, sun, and ear, respectively).
Figure 2. Z-score rank order for the orthographic judgments with standard error of the estimates.
Cognitive predictors
Table 1 shows the means (SD) for the raw scores in each task. A principal component analysis (see Principal_Component_Analysis.docx on https://doi.org/10.21253/DMU.14170832.v1) was conducted to group the cognitive tasks together into a memory composite (visuo-spatial and verbal short-term and working memory tasks) and a phonological abilities composite score (phonological abilities in L1 and FL and acoustic abilities). A z-score was then calculated for each task and these were averaged together to create the memory and phonological abilities composite scores for the mixed-effects model in the following text. The vocabulary test score loaded on the same factor as the memory and acoustic abilities tasks. This is likely due to the vocabulary test using acoustic abilities and short-term/working memory abilities to process each target word and keep it active while selecting an answer. These were therefore regressed from vocabulary test score and the residuals were used as a measure of vocabulary knowledge for the mixed-effects model mentioned in the following text (see Bisson et al., Reference Bisson, Kukona and Lengeris2021; Morra & Camba, Reference Morra and Camba2009). Using residuals in an analysis is a contested practice in psycholinguistics (see Wurm & Fisicaro, Reference Wurm and Fisicaro2014), therefore a second analysis using the unresidualized vocabulary test score (see Table 4) is also provided, and we come back to this issue in the “Discussion” section.
Table 1. By-participant correlation matrix
Note: Phono = phonological, VSTM = verbal short-term memory, SSTM = visuo-spatial short-term memory, VWM = verbal working memory, SWM = visuo-spatial working memory, vocab = vocabulary, α = Cronbach’s alpha.
* = p < .05;
** = p < .01;
*** = p < .001.
Learning scores
The raw accuracy scores in the recall task were used for the mixed-effects model using the binomial family. However, by-participant and by-item mean recall accuracies were calculated for the correlation matrices (see Tables 1 and 2).
Table 2. By-item correlation matrix
Note: Phono = phonological, ortho = orthographic.
*** = p < .001.
Mixed-effects models
Analyses were conducted using R (R Core Team, 2019) version 3.6.2 glmer function of the lme4 package (version 1.1-27.1) with participants and items as random factors and maximum likelihood estimation. The model started with a maximal random effect structure with random slopes for all repeated-measure variables (Barr et al., Reference Barr, Levy, Scheepers and Tily2013), and the “bobyqa” optimizer (Linck & Cunnings, Reference Linck and Cunnings2015). As the fit was singular, the model was simplified by removing the random slopes for the fixed-effects contributing the least variance (ibid.). The model investigated the main effects of memory, phonological abilities, and cross-linguistic orthographic similarity and controlled for cross-linguistic phonological similarity and native language vocabulary (using residuals). The model also included interactions between both cognitive abilities (memory and phonological abilities) and orthography. The following R syntax was used in the final model: glmer(accuracy~memory*Ortho_estimates + phono*Ortho_estimates + Phono_estimates + ZRE_vocab + (1 + Ortho_estimates|participant) + (1 + ZRE_vocab|item), data = data, family = binomial, control = glmerControl(optimizer = “bobyqa”)). As can be seen from Table 3, all main effects were significant, but the interactions were not.
Table 3. Final model showing main effects and interactions
Note: Phono = phonological, STWmemory = Short-term and working memory composite score, Est = estimates, Ortho = orthographic; ZRE_vocab = standardized residual vocabulary test score; number of observations = 10,560, number of participants = 132, number of items = 120, VIFs < 1.34.
As explained earlier, the use of residual is criticized, and as such, a model using the same parameters as mentioned previously but with the unresidualized native language vocabulary test score is provided in the following text. As can be seen in Table 4, except for the main effect of memory, which is now only approaching significance, all other main effects remain significant and the interactions are not significant.
Table 4. Final model using unresidualized vocabulary test score
Note: Phono = phonological, STWmemory = Short-term and working memory composite score, Est = estimates, Ortho = orthographic; Z_vocab = standardized vocabulary test score number of observations = 10,560, number of participants = 132, number of items = 120, VIFs < 1.34.
Discussion
The first aim of the current study was to investigate the impact of orthography on FL word learning. The results supported prior research, as even after controlling for how different a word’s phonology is, it remained that how a word looked was important (Bartolotti & Marian, Reference Bartolotti and Marian2017; Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; Ellis & Beaton, Reference Ellis and Beaton1993). In particular, participants performed better on the meaning recall task for words that were judged to have a similar orthography to native language words. This result expanded prior research as learning was measured at the level of meaning recall whereas prior research assessed the impact of orthography on recall and processing of word form (Bartolotti & Marian, Reference Bartolotti and Marian2017; Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; Ellis & Beaton, Reference Ellis and Beaton1993). Having a dissimilar orthography impacted participant’s ability to recall the meaning of the words either because it increased the difficulty in encoding new word form representations or because it increased the difficulty in linking these to meaning representations (or both). Future studies could probe the locus of the effect further by using a word form recognition task.
The second aim of the study was to investigate the interaction between orthography (a psycholinguistic variable) and cognitive abilities (individual differences variables). Here results showed no significant interactions. However, this could be due to how individual differences and word-level variables were conceptualized in this study (memory and phonological abilities for the former and cross-linguistic orthographic similarity for the latter) as well as how they were measured. Therefore, future studies should consider other individual differences and psycholinguistic variables as well as interactions between the two. For example, prior research investigated the role of phonological (phonotactic) similarity on FL learning (Ellis & Beaton, Reference Ellis and Beaton1993; Kaushanskaya, et al., Reference Kaushanskaya, Marian and Yoo2011; Morra & Camba, Reference Morra and Camba2009), but this research could be expanded to include the interaction between this and phonological abilities (an individual differences variable). Similarly, combinations of native/foreign languages that are more phonologically/orthographically distant (see Gamallo et al., Reference Gamallo, Pichel and Alegria2017) may require more input from cognitive mechanisms to support learning. For language families that are closely related (e.g., similar phonology and orthography), long-term language knowledge can support learning (Majurus et al., Reference Majerus, Poncelet, Ven der Linden and Weekes2008) and word processing (Akamatsu, Reference Akamatsu1999), and therefore there may be less reliance on cognitive mechanisms. In the current study, the combination of Welsh FL and English native language was used. The Welsh alphabet includes eight digraphs (ch, dd, ff, ng, ll, ph, rh, and th), which is dissimilar to the English alphabet and may have contributed to the difficulties experienced by participants with orthographically dissimilar words. However, Welsh is not considered a difficult language to learn because of its transparent orthography. Every letter is pronounced and the grapheme to phoneme mapping is consistent as each letter corresponds to one sound (Davis, Reference Davis2014; Ellis & Hooper, Reference Ellis and Hooper2001). It would therefore be important to investigate other combinations of languages with more/less orthographic distance and also for language combinations with more/less transparent orthographies. Furthermore, the current study did not investigate interactions with learning situations (context-related difficulties in Housen & Simoens, Reference Housen and Simoens2016), and this should be addressed in future research. Bisson et al. (Reference Bisson, Kukona and Lengeris2021) found, for example, that memory was more important for intentional than incidental learning and this finding could be expanded by considering word-level predictors.
This study used familiarity judgments which can be qualified as a psycholinguistic measure compared to, for example, more lexical measures such as orthotactic probability or neighborhood density. Judgments are quick to obtain and do not necessitate lengthy calculations. In addition, they allow a glimpse into how native English speakers react to FL words’ orthography. The comparative judgment approach is novel in psycholinguistics, but it is ideal for situations in which the criteria is difficult to define (Bisson et al., Reference Bisson, Gilmore, Inglis and Jones2020). In the present study, participants judged the words as to their familiarity or similarity with native language word orthography. A Likert-scale could have been used instead, for example, from 1 = very unfamiliar to 7 = very familiar (similarly to Bordag et al., Reference Bordag, Kirschenbaum, Rogahn and Tschirner2016; Dijkstra et al., Reference Dijkstra, Miwa, Brummelhuis, Sappelli and Baayen2010; Tokowicz et al., Reference Tokowicz, Kroll, De Groot and Van Hell2002). However, this kind of measure seems arbitrary. For example, how unusual does a word need to be to rate a 2 on the scale? Conversely, with comparative judgment a decision is made as to which out of two words is more familiar. It is therefore not necessary to decide the degree of unfamiliarity of each word, but rather to make a holistic judgment. This method has been used successfully in many domains, such as educational assessments and research (Bisson et al., Reference Bisson, Gilmore, Inglis and Jones2016, Reference Bisson, Gilmore, Inglis and Jones2020; Humphry & Heldsinger, Reference Humphry and Heldsinger2019; Jones et al., Reference Jones, Bisson, Gilmore and Inglis2019; Pollitt, Reference Pollitt2012). Importantly, the current study shows it can also be applied in psycholinguistics.
As mentioned in the “Results” section, residual vocabulary test scores were used instead of the raw vocabulary test scores for the main analysis. This decision was based on the principal component analysis, which indicated that the native language vocabulary test score loaded onto the same factor as the memory tasks and, to a smaller extent, the acoustic task. This is probably due to some extent to the way native language vocabulary knowledge is tested in the PPVT. During the PPVT, participants hear a word such as “terpsichorean” once. They then have to keep this word active in memory while searching the four choices of pictures to select the correct depiction of the word. Therefore, to perform well on a vocabulary test such as the PPVT, one requires good short-term/working memory and acoustic abilities. It would therefore be important in future research to use a vocabulary test that relies less on additional cognitive abilities to perform the test and therefore is a purer measure of actual vocabulary knowledge. As using residuals in multiple regression analysis is a contentious issue in psycholinguistics (Wurm & Fisicaro, Reference Wurm and Fisicaro2014), a second analysis was also provided with the unresidualized vocabulary test scores. The results of the two analyses were very similar overall confirming the main effect of orthographic similarity on FL word learning. However, the composite score for memory only approached significance in the unresidualized analysis. Based on all the prior research showing a role for short-term and/or working memory in word learning (e.g., Kapa & Colombo, Reference Kapa and Colombo2014; Martin & Ellis, Reference Martin and Ellis2012; Morra & Camba, Reference Morra and Camba2009), this is probably due to vocabulary test used in the current study as well as its relationship with cognitive abilities as explained in the preceding text. Finally, this study involved secondary data analysis and it would therefore be important to replicate the results on another dataset.
In conclusion, the current study is unique in that it investigated both individual differences and word-level predictors concurrently (the interaction between learner-related and feature-related variables mentioned in Housen & Simoens, Reference Housen and Simoens2016). However, results did not show an interaction between orthography and the cognitive abilities (short-term/working memory and phonological abilities) included in this study. The current study expanded prior research showing that orthography impacted word learning even at the level of meaning recall. It is hoped this study will encourage researchers to pursue similar lines of enquiries (e.g., with combination of more/less distant languages and interactions between variables) to refine models of FL word learning.
Acknowledgments
This work was supported by the Experimental Psychology Society and De Montfort University.
Data availability statement
The experiment in this article earned an Open Data badge for transparent practices. The materials are available at https://doi.org/10.21253/DMU.14170832.v1
Open access
