Abstract
Word learning usually takes place when there is background noise in the environment and these noise levels can hamper word learning. There is substantial evidence to suggest that orthography promotes word learning. This study tests the hypothesis that orthography could mitigate the impact of noise on word learning by reducing the cognitive load and providing stable scaffolding for phonology. Word learning occurred in a self-directed online learning paradigm which reflects the increasingly common learning context for university students. Participants were 125 native English-speaking university students (mean age = 28.21, SD = 4.41; 65 female). They were taught 16 novel words online across three training blocks that included repetition and picture naming. Orthography was manipulated between participants (present vs. absent), and background noise was manipulated within participants (quiet vs. noise). Learning was assessed through picture naming and spelling tasks before and after training. Noise impacted learning of word meanings, but not word forms (phonology and spelling) while orthography facilitated the learning of word forms but not word meanings. Further, there was greater orthographic facilitation for word phonology retrieval for words learned in noise suggesting that orthography may reduce the negative effects of noise on some aspects of word learning. This study demonstrated orthographic facilitation of word learning in a self-directed online learning paradigm with implications for naturalistic learning. Further, it suggested the need for further examination of orthography as a tool to reduce the negative impact of noise on word learning.
Introduction
Following the COVID-19 pandemic, there has been a global shift in university teaching with a significant rise in online or distant learning (García-Morales et al., 2021). As a result, university students are more likely to learn in shared spaces with high levels of background noise (e.g., at home, cafes). Noise is known to interfere with individuals’ abilities to perform the complex linguistic tasks that are crucial for processing and learning new information like words (Marrone et al., 2015; Neidleman et al., 2015). As discussed by Salins et al. (2024), noise may impact word learning in two ways. First, background noise may mask the spoken forms of words by obscuring auditory input (i.e., masking). Second, the process of selectively attending to the auditory signal of interest (while supressing background noise) could increase cognitive load leaving fewer cognitive resources to engage in encoding, processing, and storing the new information in memory (Beaman, 2005; Brungart et al., 2001; Grenzebach & Romanus, 2022; Kahneman et al., 1973; Larsby et al., 2005).
Learners show increased learning for the spoken forms (phonology) and meanings (semantics) of words when they are encountered with their written forms (orthography), a phenomenon known as ‘orthographic facilitation’ (Ricketts et al., 2009). Salins et al. (2024) demonstrated that orthographic facilitation mitigates the impact of noise on word learning. In their lab-based study, 60 undergraduate students (18–35 years) learned novel spoken forms in quiet or in noise, and with or without orthographic forms, with picture naming and picture word matching tasks used to index learning. A negative impact of noise on learning was observed, but was reduced for novel words learned with orthography.
The current study aimed to replicate findings from Salins et al. (2024) in an adapted paradigm where word meanings were learned in addition to spoken forms. Words were also learned in an online self-learning task to establish whether existing findings would generalise to such learning. Word learning here refers to the acquisition of phonological forms and meaning for novel referents (He & Arunachalam, 2017; Rosenthal & Ehri, 2008). We did not assess long-term consolidation in the current study, although research in children suggests longer-term benefits of orthographic facilitation lasting between 2 weeks and 2 months (Ricketts et al., 2021; Salins et al., 2024).
As in Salins et al. (2024), 16 words were learned in quiet or in noise, and with or without orthographic forms. Word learning was assessed using a picture naming task during learning and at post-test. Learning of word meanings was assessed using verbal definition recall, and spelling using a dictation task. This study was pre-registered on the OSF platform (https://osf.io/9zv5k/). Based on previous studies of orthographic facilitation, we predicted that (a) the presence of orthography would promote orthographic learning, evidenced in more accurate spelling to dictation in the orthography-present than in the orthography-absent group; (b) better word learning in quiet than in noisy conditions, evidenced by better picture naming accuracy for words learned in quiet than in noise. This hypothesis deviates slightly from the pre-registration due to an error in the OSF version which implied an effect of noise only in the orthography-present group; (c) orthography would reduce the impact of noise on picture naming and spelling to dictation, evidenced as a significant interaction of orthography group and listening condition, with a smaller impact of noise in the orthography-present condition. Our fourth hypothesis was exploratory and predicted that the orthography present group would perform better on the definition recall task than the orthography absent group. We also explored the impact of noise, and the interaction of noise and orthography for definition recall.
Method
A priori sample size estimation was conducted using the simr package in R (Green & MacLeod, 2016). Power was estimated for both the main effect of noise and the critical interaction effect indexing orthographic facilitation under noisy conditions in the picture-naming task, based on model parameters derived from Salins et al. (2024). Simulation results indicated 100% power to detect the main effect of noise with a sample of 70 participants, and approximately 90% power to detect the interaction effect with 90 participants.
The planned sample size was increased beyond these estimates for two reasons. First, the present study includes fewer learning trials (16 items) than the dataset used for simulation (24 items), reducing the total number of observations and thus statistical power. Second, adult data were not available to simulate the orthographic facilitation effect for the definition-recall task. Prior research suggests that orthographic facilitation for semantic outcomes may be smaller and less consistent than for naming (e.g., Colenbrander et al., 2019).
Participants
Participants, all native speakers of English, were recruited via Prolific (www.prolific.com), an online research study recruitment and administration website. The study was advertised to adults aged 18 to 35 living in the United Kingdom with no diagnosed literacy difficulties, language-related disorders or hearing difficulties. In total, 135 adults met the eligibility criteria and were aged 19 to 35 (M = 28.37; SD = 4.37), with 68 male participants and 67 female participants. However, three participants were rejected from the study as they did not pass the sound and microphone checks. This study was granted ethical approval by the Royal Holloway Ethics Committee and participants gave informed consent before tasks were completed.
Materials
Experimental Stimuli
Sixteen 6-letter bi-syllabic novel words from Salins et al. (2024) were used as stimuli in this study. These novel words, modified from stimuli used by Gilchrist and Allen (2015) and from the ARC Nonword Database (Rastle et al., 2002), were paired with pictures of novel inventions and accompanying experimental sentences from Wegener et al. (2018; e.g., Rick put his dirty socks in the raitep to clean them). The evidence to date does not suggest a systematic association between spelling-sound consistency and the degree of orthographic facilitation experienced (e.g., Colenbrander et al., 2019). Therefore, while our stimuli varied in the degree of consistency, we did not intentionally manipulate consistency or include it as a factor in our analysis. These words were split into two-word lists and the allocation of each of the two lists to either the noise or quiet condition was counterbalanced across participants. The written spellings of the items were presented above the images in Arial regular font at 30-point size in the orthography present condition, but this was omitted in the orthography absent condition. The auditory stimuli were recorded by a female British English speaker. Audition and Matlab were used to embed the spoken stimuli in noise (multi-talker babble) at 10 dB signal-to-noise ratio (SNR) and to equalise loudness across stimuli. In our previous work, we used an SNR of 0 dB, where the signal and noise are at the same level and found minimal learning of new words where teaching was in person. In this online study, we were concerned that learning would be lower overall. Therefore, we adopted 10 dB SNR where the signal is 10 dB louder than the noise to ensure that learning was possible, and in line with previous learning studies in noise (Sato et al., 2011).
Background Measures
Standardised tests of vocabulary and reading were used to obtain an overall profile of the participants. The vocabulary assessment was administered first, followed by the two reading measures (Table 1).
Background Measures.
Note. aTOWRE scaled scores are presented; in the OP group n = 9 missing TOWRE7.47 Phonemic Decoding Efficiency and 1 missing TOWRE Sight Word Reading. The OA group is missing n = 2 TOWRE SW and 16 TOWRE PDE due to recording errors. bBPVS = British Picture Vocabulary Scale; TOWRE = Test of Word Reading Efficiency; SD = standard deviation.
Vocabulary
The British Picture Vocabulary Scale (BPVS; Dunn et al., 2009) was used to measure vocabulary. In this task, participants heard a word and were required to indicate its corresponding referent from an array of four pictures. The standardised published version of this task is designed for children up to 16 years. The task was adapted for use with adults by removing sets 1 to 8 and was administered online due to the COVID-19 pandemic. The number of words correctly recognised was recorded.
Reading
The Test of Word Reading Efficiency (TOWRE; Wagner et al., 2011) and then Sentence Verification Task (Garvin & Krishnan, 2022) were used to assess reading. In the TOWRE sight word reading and phonemic decoding efficiency (nonword reading) subtests, participants read a practice list first and then read as many test items as they could in 45 s. Participants were first given the practice items with the instructions in the manual and asked to press the “next” button to read the list out loud. Following this, instructions were provided on the screen for the test words. Participants were timed and audio recorded within Gorilla and scored offline by two research assistants to calculate the total number of words or non-words read correctly in each subtest was recorded. In the Sentence Verification Task, participants were given 90 s in total to read and assess the truthfulness of a sentence. There are 80 possible sentences, and they were given a maximum of 3 s per sentence to indicate whether they thought it was true or false. Participants received 1 point for each sentence correct, with a maximum possible score of 80.
Procedure
Participants completed the experiment independently, within a single online session (Figure 1). They were first exposed to an experimental paradigm in which they learned 16 words. During training, the presence of orthography was manipulated between participants, such that one group learned all words without spellings (orthography absent condition) and the other group learned all words with spellings (orthography present condition). Alongside this, the presence of noise was manipulated between participants, with all participants learning half of the words in quiet and half in noise (eight words in quiet and eight in noise). Following the learning phase, post-tests were used to assess learning. Finally, they completed background measures to assess their pre-existing vocabulary knowledge and reading ability—this was a slight deviation from the pre-registration to support streamlined structure within programming. All tasks were presented using the Gorilla Experiment Builder (Anwyl-Irvine et al., 2020) and administered online via Prolific.
Procedure.
Sound and Recording Check
At the start of the experiment, checks were administered to ensure adequate sound delivery and recording. First, participants were asked to adjust the volume on their device to their preferred level, press the play button and enter the word they heard into the box provided. If the participant provided an accurate response, a recording check was conducted where participants were prompted to allow microphone access on their devices, record a sentence and check that it played back. If a participant had either inadequate sound or recording abilities, they were rejected from the experiment (n = 3). If they passed both the sound and recording check, they proceeded with the tasks.
Experimental Word Learning Paradigm
Experimental Procedure
Participants were randomly allocated to either the orthography present or the orthography absent condition. Within the two conditions, they were also randomly allocated to list 1 or 2. These lists were counterbalanced for listening conditions, that is, items presented in quiet and noise were balanced between the two lists.
Training
The word learning paradigm in this study is similar to that used in previous studies of orthographic facilitation (Ricketts et al., 2009, 2011; Salins et al., 2024). Participants were informed that they would be learning the names of fictional inventions. The training phase consisted of three blocks, each including a repetition trial followed by a naming trial. The stimuli were presented randomly in each trial in each block. In the repetition trial the image of the invention was presented on the screen (either with or without the written spelling depending on which orthography condition) along with an audio recording consisting of the name of the invention and it’s function. Participants were asked to repeat the name of the invention. For example, “Diana put the best orange on the valtem to juice it. Say valtem.” The participant then repeated the “invention” name aloud. This was followed by feedback that reiterated the name of the invention, for example, “This is a valtem.” Once all 16 of the “inventions” had been introduced in a repetition trial, the images were re-presented and participants were asked to recall the name of the “invention” in response to the images they had seen in the repetition trials. After each naming trial, the image and its auditory stimulus were presented as a form of feedback. Each item was seen four times by participants in each block. All responses were audio recorded using the recording function in Gorilla and later transcribed and assessed for accuracy by research assistants. The outcome measure for the training phase was picture naming accuracy (the number of “inventions” named correctly per learning block for each condition).
Post-Tests
After the training blocks were complete for all 16 “inventions,” post-tests to assess vocabulary learning were administered in the following order: (a) picture naming task; (b) meaning recall task; (c) spelling task. The picture naming post-test assessed the mapping from semantic to phonological information. Participants were shown the “inventions” one at a time, then had to press the record button on the Gorilla task and verbally recall the name of the “invention.” Responses were transcribed and coded for accuracy (0 = no response/incorrect response; 1 = correct response). The definition recall task assessed the mapping from phonological to semantic information. Participants were instructed to verbally recall everything they could remember about the “inventions” one at a time (e.g., “Tell me all you can remember about raitep”) and had to press the record button on the Gorilla task to record their response. Responses were transcribed and coded for accuracy (0 = no response/incorrect response; 1 = partially correct response; 2 = fully correct response). And finally, the spelling task assessed the mapping from phonological to orthographic information. Participants heard a recording of the name of the “inventions” one at a time and had to type the written name of the “invention.” Reponses were coded for accuracy (0 = no response/incorrect response; 1 = correct response).
Results
The binomial data (picture naming during training and at post-test) were analysed using generalised linear mixed effects models (GLMER) using the lme4 package in R (Bates et al., 2015; Kuznetsova et al., 2017). The models were fit by maximal likelihood with a full random structure, and an iterative process was followed for random structure reduction to ensure model convergence (Barr et al., 2013). For both models, sum contrasts were applied to the factors, and type III Wald chi-square tests were computed using the ANOVA function available in the car package for R (Fox & Weisberg, 2019). The ordinal meaning recall data were analysed using the cumulative link mixed models (CLMM) using the ordinal package in R (Christensen, 2023). The model was fit by maximal likelihood with a full random structure. Post hoc comparisons were conducted using the emmeans package in R (Lenth et al., 2018).
Training Phase
Performance on the picture naming task during the three blocks of training is depicted in Figure 2. The binomial model was fit with orthography group (present vs. absent), listening condition (noise vs. quiet), and training block (1 vs. 2 vs. 3) as predictors of orthographic facilitation.
Picture naming accuracy during training.
The analyses revealed a significant main effect of orthography condition, χ2(1) = 15.75, p < .001, such that items in the orthography present condition were learned more accurately than items in the orthography absent condition. No other effects were significant (see the online Supplemental Material A).
Testing Phase
Three post-tests (picture naming, definition recall, and spelling) were analysed separately and Holm-Bonferroni correction was used to correct for multiple comparisons across the three tasks.
Picture Naming
Performance on the picture naming post-test is depicted in Figure 3. The GLMER (binomial) revealed a significant main effect of orthography, χ2(1) = 29.37, p < .001, such that the orthography present group recalled words more accurately than the orthography absent group. There was a significant interaction between orthography group and listening condition χ2(1) = 4.89, p < .05 such that in the orthography-absent group, the words in quiet were recalled significantly better than the words in noise. There was no significant difference between performance in the noise and quiet conditions within the orthography present group (Supplemental Material B).
Picture naming accuracy at post-test.
Meaning Recall
Figure 4 depicts the proportion of trials of the meaning recall task, on which participants scored 0 (incorrect), 1 (partially correct) or 2 (correct). The CLMM was fit with orthography condition and listening condition as predictors of meaning recall score. A detailed summary of the model is provided in Supplemental Material C. There was no main effect of noise or orthography, or interaction of the orthography and listening conditions.
Meaning recall accuracy at post-test.
Spelling
Performance on the spelling post-test is depicted in Figure 5. The GLMER (binomial) revealed a significant main effect of orthography, χ2(1) = 81.24, p < .001, such that the orthography present group spelt words more accurately than the orthography absent group. No other effects were significant (see Supplemental Material D).
Spelling recall accuracy at post-test.
Discussion
This study investigated whether noise impacts learning of new words, and whether the impact of noise is mitigated by orthographic support. We investigated this in an online, self-directed learning context. Noise impacted the learning of new words, as indexed by a definition recall, but not in picture naming or spelling tasks. This indicates that noise disrupted the learning of word meaning, but not word forms (phonology, orthography). Orthographic support was observed for picture naming and spelling to dictation (i.e., form learning) but not for definition recall (meaning).
Learning in noise did not appear to affect performance on picture naming during training. However, it significantly reduced picture naming accuracy at post-test for adults learning new words without access to spellings. In contrast, the absence of this effect in the orthography-present condition suggests that written forms may mitigate the adverse impact of background noise on word learning in adults. These findings diverge somewhat from those of Salins et al. (2024), who reported a significant negative impact of noise on picture naming during both training and post-test. The attenuated noise effect observed in our study may be attributable to methodological differences. Specifically, our stimuli were presented at a 10 dB SNR, a more favourable listening condition compared to the 0 dB SNR used by Salins et al. (2024). Additionally, the online format of our study meant that we could not control participants’ environments. It is plausible that unmeasured ambient distractions—such as conversations with family members or background noise in public spaces—may have influenced learning outcomes. While this enhances ecological validity for adult online learners, it also raises the possibility that orthographic support mitigated unmeasured environmental noise, which may have contributed to the pattern of results observed. Despite this limitation, the study successfully replicated the orthographic facilitation effect in adults, even in an unsupervised, self-directed online learning context.
Consistent with prior research (Salins et al., 2024), we found that words were learned more readily with orthographic support, and that orthography enhanced phonological recall (picture naming) both during training and at post-test and spelling, extending evidence of orthographic facilitation to virtual learning environments. However, this benefit did not extend to meaning recall, aligning with mixed findings in the literature (e.g., Colenbrander et al., 2019) and suggesting that access to written forms may not aid adults’ semantic learning in online settings. This lends support to the view that it is important to distinguish different aspects of word learning (i.e., phonological, orthographic, and semantic) as they are predicted by different variables (Deacon et al., 2019; Ricketts et al., 2011).
With universities increasingly employing flipped classroom or similar instructional strategies, online self-guided learning occurs quite frequently for university students. Our findings suggest that providing written words on the screen for key concepts may to assist with better learning and retrieval in university students. Our study extends and replicates the lab findings of orthographic facilitation in an online self-guided learning setting. An important future direction will be to examine orthographic facilitation in classroom instruction settings to increase generalisability of these findings.
In sum, our study replicates the negative impact of noise on meaning recall but not picture naming or spelling. Further, we replicated the orthographic facilitation effect for form learning but not meaning of words. Both these findings emphasise the importance of separately examining different aspects of word learning because they are differently impacted by variables (Ricketts et al., 2009). An important future direction will be to examine the impact of noise on word learning in younger children who are more susceptible to the detrimental effects of background noise (Erickson & Newman, 2017). Further, a stronger manipulation of noise in a more naturalistic environment, such noise generated during group discussions in a classroom may provide useful insights into the utility of orthography as a mitigator of the negative impact of noise on word learning.
Supplemental Material
sj-docx-1-qjp-10.1177_17470218261442538 – Supplemental material for Does Orthography Mitigate the Impact of Background Noise on Word Learning?
Supplemental material, sj-docx-1-qjp-10.1177_17470218261442538 for Does Orthography Mitigate the Impact of Background Noise on Word Learning? by Andrea Salins, Courtney Hooton-Robinson and Jessie Ricketts in Quarterly Journal of Experimental Psychology
Footnotes
Funding
The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: This research was funded by the Experimental Psychology Society New Graduate Research Bursary and Research Initiative Funding from the Department of Psychology, Royal Holloway, University of London. AS was supported by the Australian Research Council grant FL220100061.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Data Availability Statement
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
