1. Introduction
Words are the basic building blocks of language knowledge, and their acquisition is critical in language learning. By adulthood, the mental lexicon includes more than 20,000 words (Altmann, Reference Altmann1997; Nation, Reference Nation2001; Nation & Waring, Reference Nation, Waring, Schmitt and McCarthy1997) that can be effortlessly and effectively utilized in both listening and speaking (Levelt, Reference Levelt2001). Numerous factors can influence word learning, including (1) individual-related factors, such as age (Bishop et al., Reference Bishop, Barry and Hardiman2012; Halberda, Reference Halberda2003; Marcotte & Ansaldo, Reference Marcotte and Ansaldo2014) and cognitive abilities (e.g., attention abilities: Dixon & Salley, Reference Dixon and Salley2006; symbolisation: Goodwyn et al., Reference Goodwyn, Acredolo and Brown2000); (2) word-related factors, such as the morphophonological structure of the learned word (Aitchinson, Reference Aitchison2003; Marslen-Wilsonand, Reference Marslen-Wilson, Garrod and Pickering1999) and its semantic relationships with other words (Webb, Reference Webb2019); (3) condition-related factors, such as repetition (Horst et al., Reference Horst2013), frequency (Kachergis et al., Reference Kachergis, Yu and Shiffrin2017; Webb, Reference Webb2019), contextual diversity (Johns et al., Reference Johns, Dye and Jones2016), overheard speech (Shneidman et al., Reference Shneidman, Arroyo, Levine and Goldin-Meadow2013), physical environments (Breitfeld & Saffran, Reference Breitfeld and Saffran2023), and the provision of external feedback (Metcalfe et al., Reference Metcalfe, Kornell and Finn2009; Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005). As opposed to individual-related factors, such as cognitive abilities, condition-related factors can be controlled by the instructor (whether that’s an educator, clinician, or parent). Therefore, it is particularly important to gain more insight into the effect of these factors on the learners’ achievements. This knowledge has the potential to significantly enhance word training protocols for language enrichment, second language acquisition, and pathological populations.
Clinicians, educators, and parents often provide feedback to learners with the primary intention of motivating them, yet few, if any, are aware of the potential influence of different feedback types, each of which can yield distinct effects on the learning process (e.g., Butler et al., Reference Butler, Godbole and Marsh2013; Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). Moreover, studies that examined the effect of feedback on word learning reported inconsistent results (e.g., Eraut, Reference Eraut2006; Fazio et al., Reference Fazio, Huelser, Johnson and Marsh2010; Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005), possibly due to inadequate control over intervening variables, including the specific type of feedback administered. In the present study, we aimed to thoroughly examine the effect of the type of feedback on word learning, also addressing the possible effect of the word’s morphophonological structure (existing versus pseudo morphophonological patterns). To this end, we trained 5-year-olds on new, artificial words, providing either verification feedback, corrective feedback, corrective feedback after verification feedback, or no feedback at all. Half of the artificial words were constructed from existing Hebrew morphophonological patterns, and the other half was based on pseudo-Hebrew morphophonological patterns.
In the process of word acquisition, the learner associates a phonological structure and a meaning (Levelt, Reference Levelt1992), learns the communicative role of the word (Grassmann, Reference Grassmann and Matthews2014; Vulchanova et al., Reference Vulchanova, Salda, Baggio, Pirrelli, Plag and Dressler2020), and links all the representations of the word, including its syntactic and morphological networks (Horst & Samuelson, Reference Horst and Samuelson2008), as well as its orthography (Perfetti & Hart, Reference Perfetti and Hart2002) and its motoric schema (Redford, Reference Redford2019). This multiplicity of representations creates optimal word storage in the lexicon and consequently enables efficient retrieval and use of the word (Levelt, Reference Levelt1992, Reference Levelt2001). The process of lexical access during speech production requires, therefore, the selection of an appropriate lexical item (referred to as a “lemma”) from the mental lexicon, along with the retrieval of the word’s morpho-phonological codes, culminating in using its motoric schema (Redford, Reference Redford2019).
Multiple factors were suggested to influence the learning of words. Word learning has been shown to depend on repeated exposure to the word (Perfetti & Hart, Reference Perfetti and Hart2002), and its subsequent repeated retrieval (Nakata, Reference Nakata2017), supporting the notion that “practice makes perfect” (Friedrich, Reference Friedrich2002). The speaker’s morpho-phonological knowledge was also suggested to help recognize the morphological units integrated into the words to decipher the meaning of new words, store them in memory, and retrieve them when necessary (Botwinic & Adam, Reference Botwinic and Adam2012; Deacon et al., Reference Deacon, Whalen and Kirby2011; Marslen-Wilson, Reference Marslen-Wilson, Garrod and Pickering1999). In Hebrew, knowledge about derivational morphology becomes even more critical, as it is a Semitic language in which most spoken and written words are built from roots and patterns (Ravid, Reference Ravid and Shimron2003). Hebrew-speaking children assemble words from roots and morphological patterns already by the age of three years old, reaching 98% success in assembling new words during school years (Berman, Reference Berman and Shimron2002). Thus, by this age, children acquire a wide and varied lexicon using morphological links, while about half of the words in the acquired lexicon are words that are learned through inference and mapping according to the morphological rules of the language (Berman, Reference Berman2004).
Both reinforcement and feedback can be provided by an educator, clinician, teacher, parent, or even a computer to enhance word learning. However, while reinforcement is primarily used to encourage the learner to continue learning (Sutton & Barto, Reference Sutton and Barto2018), feedback is used to inform the learner of the correctness of his or her response (Eraut, Reference Eraut2006; Hattie & Timperley, Reference Hattie and Timperley2007; Sadler, Reference Sadler1989). The accepted notion is that feedback is crucial for learners to be able to evaluate and monitor their responses (Özdener & Satar, Reference Özdener and Satar2009). Effective feedback should ideally be non-judgemental, supportive, and accurate (Shute, Reference Shute2008). Such feedback enables the learner to correct errors while preserving correct responses (Eraut, Reference Eraut2006). The ability to process performance feedback and adjust behavior accordingly shows significant improvement in children as they develop (Crone et al., Reference Crone, Zanolie, Van Leijenhorst, Westenberg and Rombouts2008; Eppinger et al., Reference Eppinger, Mock and Kray2009).
Studies that investigated the effect of feedback on learning in different modalities have revealed inconsistent results. Specifically, in many studies, the feedback was reported to enhance learning (Amitay et al. Reference Amitay, Halliday, Taylor, Sohoglu and Moore2010; Butler & Roediger, Reference Butler and Roediger2008; Liu et al., Reference Liu, Lu and Dosher2010; McCandliss et al., Reference McCandliss, Fiez, Protopapas, Conway and McClelland2002; Seitz et al., Reference Seitz, Nanez, Holloway, Tsushima and Watanabe2006; Zaltz et al., Reference Zaltz, Ari-Even Roth and Kishon-Rabin2017). However, others reported external feedback to be redundant (Dobres & Watanabe, Reference Dobres and Watanabe2012; Zaltz et al., Reference Zaltz, Ari-Even Roth and Kishon-Rabin2010) or even to interfere with the learning (Schmidt, Reference Schmidt1991). These varying reports can be attributed, at least in part, to factors including the learner’s age (Zaltz et al., Reference Zaltz, Ari-Even Roth and Kishon-Rabin2017), the timing of feedback delivery (whether it is immediate or delayed; Metcalfe et al., Reference Metcalfe, Kornell and Finn2009; Opitz et al., Reference Opitz, Ferdinand and Mecklinger2011), the frequency with which the feedback is given (high-frequency feedback may potentially interfere with the learning process; Schmidt, Reference Schmidt1991), its valence (Arbel et al., Reference Arbel, Fitzpatrick and He2021), the type of feedback, and the nature of the task being acquired (Bangert-Drowns et al., Reference Bangert-Drowns, Kulik, Kulik and Morgan1991; Butler et al., Reference Butler, Godbole and Marsh2013).
Feedback is typically categorized based on the level of information it provides (Goodman et al., Reference Goodman, Wood and Hendrickx2004). In verification feedback, the learner is informed of whether the answer is correct or incorrect. In corrective feedback, the learner is provided with the correct answer. In “try again” feedback, the learner is informed that the answer was incorrect and is encouraged to try again. In “error flagging” feedback, the learner is directed to the errors in his or her answer. In elaboration or explanation feedback, the learner receives an explanation regarding the correct answer or the reason why his or her answer was incorrect (Butler et al., Reference Butler, Godbole and Marsh2013; Shute, Reference Shute2008). The variable found to have the greatest impact on learning is the amount of information it provides (Shute, Reference Shute2008). The information included in feedback is crucial as it allows learners to correct mistakes (Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005) and reinforce correct responses (Butler & Roediger, Reference Butler and Roediger2008). Therefore, a key goal of feedback research is to identify the optimal content of feedback to enhance learning.
There is some evidence to suggest that the impact of feedback type may also vary based on the nature of the learned language skill. For instance, explanation feedback was found to enhance performance on new inference questions to a greater extent than corrective feedback when acquiring concepts from a text, despite both types of feedback resulting in similar performance improvements with repeated questions (Butler et al., Reference Butler, Godbole and Marsh2013). Furthermore, in a more recent study, 8-year-old children demonstrated more substantial advantages in learning and applying repeated items of an artificial morphological rule (AMR) through corrective feedback in contrast to verification feedback, regardless of whether the former was delivered independently or immediately after verification feedback (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). This result was attributed to the idea that corrective feedback offers repeated exposure to items that require memorisation, and as a result, contributes to enhanced learning. The combination of corrective feedback following verification feedback further allowed explicit discovery of the nature of the AMR, which enables its explicit generalisation to new items. The authors proposed that the use of combined feedback types captured the children’s attention and motivated them to find the correct answer. In this heightened state of alertness and focus, the children became more responsive to the beneficial effects of corrective feedback. We hypothesized that word learning resembles learning repeated items within the context of the AMR, and thus anticipated a similar positive effect for corrective feedback, with an even greater impact from the combined verification and corrective feedback on children’s learning.
Although both verification and corrective feedback have been suggested to improve word learning compared to no-feedback conditions (e.g., Metcalfe et al., Reference Metcalfe, Kornell and Finn2009; Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005; Sippel, Reference Sippel2019), very few studies have directly compared the effects of these two types of feedback on the learning (Özdener & Satar, Reference Özdener and Satar2009; Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005). Pashler et al. (Reference Pashler, Cepeda, Wixted and Rohrer2005) noted a slight advantage in favor of corrective feedback over verification feedback in a web-based study. Özdener and Satar (Reference Özdener and Satar2009) found that explanatory feedback was more effective than confirmation feedback using Computer Assisted Language Learning (CALL).
The current study used an artificial language paradigm that was equally new to all participants, thus affording a controlled learning environment (Ferman & Karni, Reference Ferman and Karni2010; Gomez, Reference Gomez2002; Nevat et al., Reference Nevat, Ullman, Eviatar and Bitan2017). The purpose of the study was twofold: (1) to compare the effect of four feedback conditions on preschool children’s ability to identify and produce (name) 12 artificial words. The feedback conditions were no feedback, verification feedback, corrective feedback, and verification + corrective feedback. These feedback types were chosen as a follow-up to Ferman et al.’s study in Reference Ferman, Amira Shmuel and Zaltz2022, aiming to test the hypothesis that word learning resembles learning repeated noun–verb items within the context of an AMR. While Ferman et al. study (Reference Ferman, Amira Shmuel and Zaltz2022) examined the effect of three types of feedback on the ability to learn an AMR, the current study added another learning condition: no feedback. This addition enabled us to explore whether feedback, regardless of its type, influences word learning. (2) To investigate a possible interaction between the type of feedback provided and the type of artificial words: with or without known morphophonological patterns. We hypothesized that (a) all feedback types would boost the children’s learning compared to not providing feedback at all, and (b) the feedback that provides more information would be superior to those that convey less information. Therefore, corrective feedback was expected to produce larger learning gains compared to the verification feedback, and a combination of corrective feedback following verification feedback was expected to be superior to all other feedback types, especially for words that do not constitute existing Hebrew morphophonological patterns. No feedback at all was expected to result in the least improvement in performance.
2. Method
2.1 Participants
Sixty-three native Hebrew-speaking children (mean age = 5.09 ± 0.33 years, range: 4.08–5.67 years, 32 females), participated in the study. This age group was chosen based on a pilot study that demonstrated that 5-year-old children can stay engaged in a word-learning “computer game” for at least 45 minutes, including several breaks (see also Ferman & Bar-On, Reference Ferman and Bar-On2017). To recruit participants for this study, we reached out to mainstream kindergartens located in the central region of Israel within areas characterized by a medium to high socio-economic status. Only children whose kindergarten teachers affirmed their overall competence at a medium to high level, with no identified deficits, were selected for participation.
In addition, to ensure cognitive and language skills within the normal range for age, standardized cognitive and language assessments were administered to all participants. The Raven’s Colored Progressive Matrices test (Raven, Reference Raven1998) was used to assess nonverbal cognitive abilities. The Aston test (Hebrew norms) was used to measure verbal working memory. Vocabulary and grammar subtests from the Katzenberger diagnosis for preschoolers (a standardized Hebrew language test; Katzenberger & Meilijson, Reference Katzenberger and Meilijson2014) were used to assess lexicon and grammatical abilities. The Rapid Automatized Naming Pictures test (Wolf & Denckla, Reference Wolf and Denckla2005) (Hebrew norms) was used to measure naming speed, given its relevance to word production. All the children obtained standard scores within the range of mean ± 1.5 standard deviations in all cognitive and language tests.
The children were randomly assigned into four groups: The first group included 16 children (mean age = 5.15 ± 0.35 years) who were provided with only verification feedback and were labeled the verification group. The second group included 16 children (mean age = 5.01 ± 0.26 years) who were provided with only corrective feedback and were labeled the corrective group. The third group included 16 children (mean age = 5.07 ± 0.36) who were provided with corrective feedback after verification feedback and were labeled the ver + cor group. One child from this group participated in only two practice sessions and thus was excluded, resulting in a final sample of 15 children for the ver + cor group. The fourth group included 15 children (mean age = 5.16 ± 0.34 years) who did not receive any feedback during their practice; this group was labeled the no-feedback group.
Informed consent was obtained from all the parents of the children who participated in the study. The study received approval from Tel Aviv University Institutional Review Board (IRB #2012/21).
2.2 Materials
Twenty-four artificial words were used in the study. All the words were artificial in their meaning, representative picture, and morphophonological pattern. The words included invented meanings in two semantic categories: professionals (12 words) and devices (12 words), each with a representative picture that was painted by a dedicated painter. The words were constructed from Hebrew pseudo-roots. To investigate the effect of familiarity with the morphophonological pattern on learning, half of the artificial words were constructed from existing Hebrew morphophonological patterns, while the other half was based on pseudo-morphophonological patterns (artificial patterns). These patterns were constructed by changing the consonants and vowels in existing morphological patterns in Hebrew. For example, Table 1 presents 12 artificial words, half with existing patterns and half with pseudo-patterns.
One of the two lists of artificial words
To increase the statistical power of the comparison of the two types of words (namely, words constructed with morphophonological patterns and words constructed with pseudo-patterns), the 24 words were divided into two lists, 12 words each (6 professionals and 6 devices, 6 with existing patterns, and 6 with pseudo-patterns). The two lists were equally balanced in terms of morphophonological and phonotactic complexity, and each list was administered to half the children in each group. This approach increased the total number of words for each type (those with existing patterns or pseudo-patterns) from 6 to 12.
The artificial words were organized into separate lists based on existing Hebrew morph-phonological patterns and pseudo-morphological patterns. Each list consisted of six words, each appearing twice, totaling 12 artificial words per list. The words were prerecoded by a female native Hebrew speaker. Two PowerPoint presentations were created for this study, each used to present and practice one of the lists.
2.3 Procedure
Each participant took part in one session of background assessment and three sessions of word practice. Background assessment lasted approximately 60 minutes and included an examination of nonverbal intelligence, verbal working memory, and language abilities. Each practice session lasted approximately 45 minutes. The session was divided into two phases (Figure 1). The first phase was the presentation phase, wherein the participants were exposed to the artificial words incorporated in a heard story about “Mr. Catti who went out to visit a surprise house and discovered many surprises there.” Then, for each word, the participants viewed a picture representing the word, heard the target word, and received an explanation of its meaning. For example, they looked at a picture (Figure 2A) and heard: “In this corner, Mr. Catti found a miSBeMeT. Mr. Catti can press the button and the miSBeMeT will slip the socks onto his feet. Here is a miSBeMeT.”
Study design. Overall, three word-learning sessions were conducted. The first and second sessions were spaced 1–2 days apart and the second and the third sessions were spaced one week apart. (A) practice list with words featuring existing morphophonological patterns. (B) practice list with words featuring pseudo-patterns.
Sample panels for the artificial word ‘miSBeMeT’: (A) presentation, (B) identification task, and (C) naming task.
In the second phase of each practice session, the participants practiced artificial words using two tasks: an identification task and a naming task. In the identification task, a panel displaying four pictures representing four artificial words was shown to the child, who was then asked to select the correct picture. For instance, the participant was presented with a panel featuring four pictures (as shown in Figure 2B) and heard the prompt, ‘Where is a miSBeMeT?’ The expected response was to point to the lower-right picture. In the naming task, a child was presented with a single picture representing one of the artificial words (as shown in Figure 2A) while hearing the following recording: “What do you see here?” The child’s objective was to articulate the target word accurately in response. The response was considered correct only when the word was pronounced flawlessly.
Each session consisted of the following components, which were repeated twice within a session: (1) presentation phase, (2) identification task featuring existing patterns, (3) identification task with words featuring pseudo-patterns, (4) naming task with words featuring existing patterns, and (5) naming task with words featuring pseudo-patterns. As a result, each word list, whether containing words with existing patterns or pseudo-pattern words, was alternately practiced twice within each task during a session. This means that each target word was practiced eight times in a session: two times in each list * two times in each task * twice (two lists for each task) in a session. The order of word types and tasks remained consistent across all practice sessions for all participants. The first and second practice sessions were spaced 1–2 days apart, and the second and third sessions were spaced one week apart, to evaluate retention.
Three types of feedback were employed, regardless of whether the answer was correct or incorrect: 1. Verification feedback: Auditory feedback of “correct” or “incorrect” was provided after each answer, 2. Corrective feedback: Auditory feedback included the phrase “the correct answer is…” after each answer (both correct or incorrect), and the correct answer was circled in the identification task, and 3. Ver + cor feedback: Corrective feedback followed verification feedback, with auditory feedback saying “correct/incorrect, the correct answer is…” after each answer, and the correct answer was circled in the identification task. Additionally, a fourth group received no feedback during training and was thus termed as the No feedback group.
2.4 Apparatus
The PowerPoint presentation, encompassing both word presentation and practice, was displayed on a laptop using the computer’s speakers. The sessions were conducted in a quiet room at the participants’ homes. Accuracy, measured as a percentage of correct responses (%correct), was documented by the researcher for each list.
2.5 Statistical analysis
Sample size justification: A recent study examining the effect of feedback type on AMR learning across ten training sessions reported a substantial feedback*session interaction effect (f = .67; Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). Given that the current study included only three training sessions, we conservatively assumed an interaction effect half the size of that reported by Ferman et al., representing a moderate effect size (f = .34). Power analysis was conducted through a series of simulations intended to estimate the power to detect a significant interaction effect of this size with a power of .80 and a Type I error probability of .05. Simulations were carried out using SAS software (Version 9.4, SAS Institute Inc., Cary, NC, USA). The outcomes of the model were assessed for statistical significance using SAS PROC MIXED with a repeated measures design accommodating the autoregressive structure of the data (AR(1)). The simulation results indicated that a minimum of five participants per group is required to achieve a power of .80, allowing for robust detection of the group by session interaction effect. However, to account for the expected intersubject variability in children’s performance, the sample size was increased to include 15–16 children per group.
Data were analyzed using a mixed-effects linear model (PROC MIXED, SAS Institute Inc., Cary, NC, USA). The dependent variable was the percentage of correct responses. Fixed effects in the model comprised three categorical variables (task: identification/naming, with performance averaged across two trials per session; pattern: existing pattern/pseudo-pattern; and group: no feedback, verification feedback, corrective feedback, combined feedback), one of them defined as a CLASS variable, as well as one continuous variable (session number: 1, 2, 3). The model was specified to evaluate the impact of these factors on percent correct, controlling for individual variability. All possible interactions were also specified in the model. To account for the repeated-measures design, we implemented a first-order autoregressive covariance structure [TYPE = AR(1)] with subject as the repeated measures unit. This covariance structure assumes that measurements taken closer in time are more highly correlated than those taken further apart, which is appropriate for our longitudinal design. Degrees of freedom were estimated using the between-within method (DDFM = BW), which partitions the residual degrees of freedom into between-subject and within-subject portions. The model was estimated using restricted maximum likelihood (REML). The full model specification in SAS syntax was
3. Results
Table A1 details the mean performance (± 1 standard error) across the four study groups in the identification and naming of the existing pattern and pseudo-pattern words. Results from the task * pattern * group * session mixed-effects model showed a significant main effect of task [F(1,59) = 176.34, p < .001, f = 1.73], indicating better performance on the identification task compared to the naming task (Figure 3).
Box plots for performance on the identification and naming tasks. Box limits include the 25th to 75th percentile data, and the continuous line within the box represents the median. Bars extend to the 10th and 90th percentiles. Black dots represent outliers. The mean is shown by the dashed line within the box.
There was also a significant effect of pattern [F(1,59) = 5.87, p = .018, f = .32], indicating better performance on words with existing morpho-phonological patterns compared to words with pseudo-patterns (Figure 4).
Box plots for performance on words with existing morpho-phonological patterns and words with pseudo-patterns. Box limits include the 25th to 75th percentile data, and the continuous line within the box represents the median. Bars extend to the 10th and 90th percentiles. Black dots represent outliers. The mean is shown by the dashed line within the box.
A significant effect was found for session [F(1,661) = 874.13, p < .001, f = 1.15]. This effect was qualified by a significant session X group interaction [F(3,661) = 14.57, p < .001, f = .26] (Figure 5). In order to reveal the source of the interaction effect, a series of six post-hoc partial interaction tests were performed, in which the interaction between session and every pair of groups was assessed. Using Bonferroni correction, the maximum Type I error probability for each analysis was set at .008. The analysis showed significant interactions between the session and no-feedback and corrective feedback groups [F(1,327) = 41.08, p < .001, f = .35], and between session and no-feedback and ver + cor feedback groups [F(1,327) = 16.44, p < .001, f = .23]. These interactions revealed a faster learning rate, i.e., more efficient learning for the groups who received corrective feedback (either with or without verification feedback) compared to the no-feedback group.
Visual presentation of the Group * Session interaction revealed in the statistical analysis. Mean performance (± 1 standard error) over the three practice sessions (mean two lists in each session) is shown for the four study groups: no feedback, verification feedback, correction feedback, and combined ver + cor feedback. Ver + cor = corrective feedback after verification feedback. **p < 0.01, ***p < .001.
An additional analysis of the session * group interaction was conducted by performing post-hoc comparisons between feedback conditions within each session. Bonferroni correction was applied, with the Type I error rate set at .003 for each analysis. At Session 1, the ver + cor feedback group performed significantly better than both the no-feedback group [t(59) = 4.33 (p < .001)], and the verification feedback group [t(59) = 3.27 (p = .003)]. At Sessions 2 and 3, both the ver + cor and corrective feedback groups outperformed the no-feedback and verification feedback groups [Session 2: t(59) = 6.65, 5.06, 4.97, and 3.34, all p’s ≤ .001; Session 3: t(59) = 6.33, 5.39, 3.97, and 3.05, all p’s ≤ .003].
Additionally, there was a significant interaction between session and verification and corrective feedback groups, [F(1,327) = 16.03, p < .001, f = .22], suggesting more efficient learning for the group that received corrective feedback than verification feedback. All other main and interactive effects in the model did not reach statistical significance (all p’s > .05).
4. Discussion
The purpose of the present study was to investigate the effect of various feedback conditions – no feedback, verification feedback, corrective feedback, and a combination of verification followed by corrective feedback – on children’s ability to learn new (artificial) words. Including the no-feedback condition enabled an examination of whether feedback, regardless of its type, affects young children’s word learning. We also explored the possible interaction between the type of feedback provided and the familiarity of the morpho-phonological pattern of the learned words.
The results clearly show that practice over the three sessions resulted in significant gradual learning of the artificial words, in both identification and naming tasks, regardless of the presence or type of feedback and regardless of the word type. These findings follow the notion that young children can recognize phonological patterns in the speech input and track those patterns to extract and name words using unguided, implicit, statistical learning (e.g., Saffran et al., Reference Saffran, Pollak, Seibel and Shkolnik2007). In the current study, the children also learned to map the phonological patterns to their meanings, aligning with existing literature that indicates children can acquire this skill at a very young age (Bion et al., Reference Bion, Borovsky and Fernald2013).
The results indicate that not all types of feedback influenced children’s ability to learn new (artificial) words. Specifically, verification feedback did not significantly aid in learning these artificial words among young children. Although the verification feedback group performed somewhat better than the no-feedback group (Figure 5), this difference was not statistically significant. Verification feedback, which indicates whether an answer is correct or incorrect, can direct learners’ attention and encourage them to actively seek the correct answer (Metcalfe et al., Reference Metcalfe, Kornell and Finn2009). However, our findings reveal that such feedback had a minimal effect on the learning of artificial words, likely because it did not provide sufficient information or stimulate the cognitive processes necessary for identifying and producing new, unfamiliar words (Levelt, Reference Levelt2001).
The primary discovery of the study indicates that providing corrective feedback, whether following verification feedback or standalone, resulted in the most notable performance enhancements compared to solely receiving verification feedback or no feedback at all. This was observed in the capacity of 5-year-old children to identify and produce new artificial words. The advantage of corrective feedback over verification feedback is to be expected when considering that corrective feedback offers more comprehensive information. Whereas verification feedback merely indicates whether the response is correct or incorrect, with no hint or guidance towards the correct answer, corrective feedback provides the target word, directly supporting word learning. Additionally, the multimodal nature of corrective feedback, which includes both auditory and visual components conveying the same linguistic information, may further enhance its effectiveness by reinforcing the learning process. From a behaviorist learning perspective, corrective feedback can be seen as a consequence (i.e., reinforcement) that increases the likelihood of a response occurring again in the future (Sutton & Barto, Reference Sutton and Barto2018).
The positive impact of corrective feedback on children’s word learning may also be attributed to the repeated exposure it provides to the words they are trying to learn. Repeated exposure is widely acknowledged for facilitating learning across diverse domains (Brutus et al., Reference Brutus, Donia and Ronen2013; Loui & Wessel, Reference Loui and Wessel2008; Montoya et al., Reference Montoya, Horton, Vevea, Citkowicz and Lauber2017), including language learning (Bybee, Reference Bybee2008; Tachihara & Goldberg, Reference Tachihara and Goldberg2022), especially in the context of word learning (Bisson et al., Reference Bisson, van Heuven, Conklin and Tunney2014; McGregor et al. Reference McGregor, Friedman, Reilly and Newman2002; Wible et al., Reference Wible, Kuo, Chien and Taso2001). This phenomenon is linked to the understanding that the learning of various language components (Perruchet, Reference Perruchet2019; Saffran et al., Reference Saffran, Pollak, Seibel and Shkolnik2007), including words (Romberg & Saffran, Reference Romberg and Saffran2010; Saffran, Reference Saffran2001), relies on statistical mechanisms. According to the statistical learning model, the brain detects recurring stimuli and patterns in the environment, extracts these patterns, and learns them. It is acknowledged that a substantial database plays a crucial role in facilitating statistical learning (James et al., Reference James, Witten, Hastie and Tibshirani2021; Schapiro & Turk-Browne, Reference Schapiro and Turk-Browne2015). In line with this notion, the recurrent exposure to corrective feedback in the present study likely enhanced the memory of the artificial words. Our study mirrors child-directed speech (CDS), where caregivers refine and expand children’s speech to reinforce proper language use (Rowe & Snow, Reference Rowe and Snow2020; Snow, Reference Snow1972). Parental expansions and recasts have been proposed to offer children critical linguistic feedback that facilitates the acquisition of syntax and vocabulary (Snow, Reference Snow1972). Similarly, in our study, corrective feedback functioned as enriched input, supplying children with the correct target words. This suggests that our findings may be relevant to natural language learning strategies within the context of CDS.
This interpretation is reinforced by a previous study that emphasized the advantage of corrective feedback over verification feedback in word learning (Pashler et al., Reference Pashler, Cepeda, Wixted and Rohrer2005), as well as in the learning of repeated items governed by an AMR (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). In future research, it may be worthwhile to investigate whether the effectiveness of corrective feedback, in terms of facilitating faster learning rates, primarily stems from the increased exposure to words. This can be achieved by comparing the impact of corrective feedback to a control condition that lacks feedback but involves an equal number of word exposures on word learning.
The results further indicated that adding verification feedback before corrective feedback did not enhance performance compared to providing corrective feedback alone, contrary to the findings reported by Ferman et al. (Reference Ferman, Amira Shmuel and Zaltz2022). This finding supports the view that corrective feedback, on its own, is the most effective type of feedback for word learning in young children. It may also highlight the interaction between the type of feedback provided, the language task to be learned, and the learner’s age. Specifically, while the combination of verification and corrective feedback enhanced AMR learning in 8-year-old children (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022), it did not improve word learning in 5-year-olds. This discrepancy may be due to the fact that verification feedback with a negative valence can harm children’s motivation to learn, especially when it is frequently provided during the early stages of learning when children are more prone to making errors. Although motivation was not directly assessed in the current study, this hypothesis is supported by prior research showing that negative feedback may reduce the motivation to learn in individuals with low self-esteem (Brockner et al., Reference Brockner, Derr and Laing1987). To address the potential influence of individual differences on feedback processing, future studies may benefit from exploring a within-subject design, wherein each participant will learn artificial words under all four feedback conditions. Such an approach could offer additional insights into how individual learners respond to different feedback types and further clarify the conditions under which verification feedback may or may not be beneficial. Additionally, a within-subject design could enhance the applicability of findings to clinical contexts by identifying patterns of response variability across learners.
As expected, the learning of artificial words that were based on Hebrew morpho-phonological patterns was more efficient than the learning of those based on pseudo-Hebrew morpho-phonological patterns (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). This distinction was evident in the identification task for all types of feedback but became apparent in the naming task only when corrective feedback was provided. In line with the previously discussed limited impact of verification feedback only in the identification task, this finding suggests that knowing established morpho-phonological patterns is sufficient for supporting word identification but insufficient for supporting word naming (production), which is a more demanding task (Levelt, Reference Levelt1992, Reference Levelt2001).
5. Conclusions
The outcomes of the current research have both theoretical and practical implications that can be applied to language instruction. The results clearly show that although 5-year-old children can learn new (artificial) words without any feedback, the provision of corrective feedback greatly enhances their learning to identify and name novel words, regardless of the word’s morphophonological structure. This is likely achieved by reinforcing the implicit (statistical) memory processes of the target words. Interestingly, the combination of corrective feedback after verification feedback does not boost performance beyond the provision of only corrective feedback in word learning at 5 years of age.
Considering the outcomes of the current research alongside prior studies (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022), it becomes clear that there is no one-size-fits-all feedback method that universally optimizes all learning processes. The results of the current study suggest that corrective feedback, by providing repeated exposure to the target information, is likely to improve the acquisition of language elements that depend heavily on memory. This includes recurring patterns as observed in Ferman et al., (Reference Ferman, Amira Shmuel and Zaltz2022), and words, as indicated by current research. A previous study demonstrated that the combination of verification feedback followed by corrective feedback is the most effective in encouraging the discovery of an artificial linguistic rule, and thus its generalisation to new items representing this rule (Ferman et al., Reference Ferman, Amira Shmuel and Zaltz2022). The researchers indicated that verification feedback likely initiates explicit processes aimed at comprehending the rule. This increased awareness and attention render learners more open to receiving the correct answer through subsequent immediate corrective feedback. Considering these findings, we propose that the pivotal factor impacting the effectiveness of feedback is the alignment between the processes initiated by the provided feedback and the processes essential for learning a specific task.
Future research endeavors could enhance the scope of this investigation by examining the impact of various feedback types such as error flagging or elaboration on the acquisition of new words as well as other language tasks. Additionally, since fatigue tends to increase over time in experimental tasks, future studies could benefit from counterbalancing task order and morphophonological patterns. This approach would help isolate and clarify the potential influence of each variable on learning outcomes. Furthermore, it is important to exercise caution when applying the conclusions drawn from the current findings to individuals of different age groups than those studied, to atypical individuals, and to learning environments that differ from those examined in this study. Expanding the study to different age groups, children from lower socioeconomic backgrounds, and atypical populations, such as children with language impairments, would provide a more comprehensive understanding and practical implications of feedback’s effects on language learning.
Data availability statement
The data that support the findings of this study are publicly available on the Open Science Framework (OSF) at the following link: https://osf.io/va25t/?view_only=816bdf76d302478484068b07a24325d8.
Acknowledgements
The authors wish to acknowledge all the undergraduate students who assisted in the data collection. The authors also express special gratitude to the participating children and their parents for their invaluable contribution to this study.
Competing interests
The authors declare none.
Disclosure of use of AI tools
During the preparation of this work, the authors used ChatGPT to improve language and readability. After using this tool, the authors reviewed and edited the content as needed and therefore take full responsibility for the content of the publication.
Appendix A
Mean performance (± 1 standard error) across the four study groups in identification of the existing pattern words, naming of the existing pattern words, identification of pseudo-pattern words, and naming of pseudo-pattern words (Ver + cor = corrective feedback after verification feedback)
Open access
