Young learners’ receptive L2 English vocabulary knowledge in relation to extramural English exposure at the onset of formal instruction in Norway

2.1 Definitions of key concepts

Several terms are used in the literature to refer to the informal learning of L2 English outside educational institutions. In the present study, we adopt the term extramural (‘outside the walls’) English as defined in Sundqvist and Sylvén (2016). EE involves voluntary learner-initiated (typically interest-driven) informal activities that are not connected with school(ing). EE activities can be both digital and non-digital and may encompass both incidental and intentional learning, but the former is much more common. Another important concept here is engagement, since learners themselves choose whether to engage in EE or not. There appears to be relatively strong agreement on a definition given in Fredricks et al. (2004), which says that engagement comprises three interrelated dimensions (or aspects), having to do with behavior, cognition, and affect/emotion (see, e.g., Mercer 2019, for an insightful discussion). Some suggest adding a social dimension to the concept (e.g. Svalberg 2009), whereas others, like Mercer (2019: 646), “take a situated view of learners meaning all aspects of cognition and affect are socially situated and behavior typically involves others in social settings”.

Related to self-regulated vocabulary learning, Schmitt (2008) emphasizes that any engagement that leads to being exposed to words, paying attention to or manipulating them, are beneficial for learning, and there are two key factors that will affect vocabulary learning: repetition (number of encounters with a word) and the quality of attention at each encounter (Webb and Nation 2017). To shed light on these key factors, Uchihara et al. (2019) have shown that incidental word learning is an incremental process which necessitates several encounters for creating a mapping of word form and meaning, but exactly how many encounters are needed is difficult to establish; a range of 8–10 encounters has been suggested in studies predominantly investigating learning from reading (see, e.g., Pellicer-Sánchez and Schmitt 2010; Webb 2007). However, some words are likely learned through considerably fewer repetitions, by being salient in some way to the learner. Conversely, some words may require many more repetitions, due to a host of factors. A further caveat is that different aspects of word knowledge may require different numbers of encounters (e.g., receptive versus productive skills).

It is assumed in vocabulary research that vocabulary consists of three dimensions: Vocabulary size/breadth, vocabulary depth, and vocabulary fluency/activation (Gyllstad 2013). Meara (1996) has argued that the basic dimension is size/breadth, and that learners with large vocabularies are more proficient in a wide range of language skills. Regarding L2 vocabulary, it is generally agreed that learning can happen through two main processes. Words may be learnt intentionally, through deliberate study of the form, meaning and use aspects of words and lexical phrases. As mentioned, vocabulary learning can also happen incidentally by sheer exposure to words and phrases whilst the focus of the learner is on comprehension, for example, when listening to the radio. Thus, incidental learning is a by-product of repeated exposure to language input. It stands to reason that a learner heavily engaged in EE is predicted to develop a larger vocabulary in comparison to a learner with a markedly more restricted exposure, on the assumption that no intentional word learning takes place. In line with these ideas, it follows that Sundqvist and Sylvén (2016) and other EE scholars (e.g., Liu et al. 2024) have advocated the benefits of engagement in EE for learning, suggesting that EE activities can be a means of increasing learners’ L2 vocabulary in particular (Peters et al. 2019; Webb 2015) “because of their potential for incidental learning” (Puimège and Peters 2019: 945). While a scoping review covering 206 informal L2 learning studies published between 2000 and 2020 shows that vocabulary is a preferred research objective, it also reveals that young learners (<10 years) are clearly the least investigated learner group (Kusyk et al. 2025). Thus, the present study contributes to filling that gap.

Next, we first present EE research involving young learners, defined as age 12 and younger. Whereas some of the studies cover connections between EE and several aspects of L2 English proficiency, our focus is specifically on findings that concern EE and L2 English vocabulary learning. We then present our research questions.

2.2 The connection between EE engagement and L2 English vocabulary knowledge

Young learner EE research generally reveals positive findings related to L2 English, in particular in technologically-advanced settings where children have easy access to the internet and the target language (English) through various digital devices. A review of the literature shows that these studies are all European; we present them here starting with the youngest learners and ending with those at the end of primary school (age 11/12).

In the Netherlands, L2 learning (English, French, or German) can be introduced early in the first grade when children are 4 years old. Most schools offer English. In this context, Unsworth and colleagues (2015) investigated factors that affect L2 English learning: formal instruction, input quantity and teachers’ language proficiency, and EE. They compared 168 learners divided into three groups receiving early instruction (<60 min/week, 60–120 min/week, or > 120 min/week) and a control group with no instruction. To measure vocabulary skills, the Peabody Picture Vocabulary Test 4, PPVT^TM-4 (Dunn and Douglas 2007) was used, and for EE (“Out-of-school exposure”), a parental questionnaire. The EE variable did not correlate with vocabulary scores in the first post-test administered after 1 year of instruction, but did so in a second post-test after 2 years of instruction (PPVT mean scores ranged from about 10 to 32). However, EE was not a significant predictor in a regression analysis with children’s vocabulary scores as outcome variable. The return rate of the parental questionnaire was low (67 %), which might explain this finding, and the authors also suggest the participants might have been too young for EE to be a significant predictor of their English skills.

In Sweden, Sylvén (2022) used the same vocabulary test when investigating 7 children in kindergarten and 13 in first grade (exact ages undisclosed, but Swedish 1st-graders are aged 6–7). The purpose was to explore EE and proficiency, and to evaluate research instruments. The results showed a very large individual variation as regards both EE and vocabulary knowledge. On average, the 1st-graders spent 317 min/week on EE (SD = 265), ranging from 75 min/week to 983. The mean for the PPVT^TM-4 (Dunn and Douglas 2007) in grade 1 was 59 (SD = 30.4). Notably, the participant reporting the highest engagement in EE scored at a level equivalent to that of a native speaker aged 14–16. A recent study from Spain which also involved 1st-grade learners (N = 176) adopted the same vocabulary test (Soto-Corominas et al. 2024). The focus here was on the effects of (a) a content and language integrated learning (CLIL) approach and (b) different sources of individual difference (including EE) on learners’ receptive and productive L2 English skills. The findings showed that participants who reported more engagement in EE activities and participants with more educated mothers performed better; the median vocabulary test score at the end of first grade was 25 (CLIL learners) and 27 (non-CLIL).

An early EE study from Iceland found that 7- and 8-year-olds (N = 182) picked up a great deal of English incidentally (Lefever 2010). These children had not had any formal English instruction in school and reading, listening, and conversational skills (a subsample, n = 51) were examined. It is noted that some children were “capable of spontaneous interaction and had a strong desire to express themselves [in English]” (p. 92) and while there were no gender-differences for reading and listening, the boys scored significantly higher than the girls on conversational skills (the effect size was large: phi = 0.44). The author interviewed the parents of 10 children and found that some children used English words with one another during play, others were frequent gamers, and all were frequently exposed to English through media.

In Denmark, Hannibal Jensen (2017) researched young learners (8–10 years, N = 107) to map EE activities (language diary, parentally guided) and vocabulary knowledge. Seven EE-activities were listed in the diary: gaming, listening to music, reading, talking, watching television, writing, and other. The most time was spent on gaming (rank 3 for girls, rank 1 for boys), music (rank 1 for girls, rank 3 for boys), and watching television (rank 2 for both genders). No significant difference was observed between boys’ and girls’ vocabulary scores. However, playing digital games with both oral and written English input and gaming with only written English input were significantly related to vocabulary scores, as measured by the PPVT^TM-4 (Dunn and Douglas 2007), and especially so for boys (Hannibal Jensen 2017).

Hubbe Emilsson (2021) also studied learners aged 8–10 years (N = 24), but in Sweden. Information on EE activities was gathered using a learner and a parent questionnaire. Vocabulary knowledge was assessed through the Picture Vocabulary Size Test (PVST, Anthony and Nation 2017). Top-reported activities were gaming, watching YouTube, and watching English-speaking films. Correlations at 0.64–0.80 were found for reported EE activities and PVST scores. No significant gender difference was found for vocabulary.

In another study from Spain targeting English CLIL programs in primary school, the aim was to analyze the impact of CLIL on the four language skills alongside three more factors: EE, socioeconomic status, and non-verbal intelligence (Lázaro-Ibarrola 2024). The participants (N = 171, aged 10–11) were divided into two groups depending on the intensity of the CLIL program (High or Low). The results revealed that engagement in EE was very frequent in both groups, but more so in the High Intensity group (mean: 8.4 h/week). Music was the most popular activity. In addition, there were several positive associations between EE and learners’ scores, and speaking appeared to be particularly influenced by EE.

In a large-scale project focusing on early language learning in Europe (ELLiE), data were drawn from more than 1,400 children from Croatia, England, Italy, Netherlands, Poland, Spain, and Sweden, all with an early introduction of formal English instruction (Enever 2011). A sub-study of ELLiE examined the effect of out-of-school variables on participants’ listening and reading skills when in the fourth year of English instruction (N = 865, age 10–11) (Lindgren and Muñoz 2013). The results showed that cognate linguistic distance was the strongest predictor for both listening and reading, but this was followed closely by EE. Parents’ educational levels had an effect, but only on reading scores. Viewing films in English was the most predictive EE activity.

In a Swedish study including 11- and 12-year-olds (N = 86), the researchers examined the relationship between EE gaming on the one hand and vocabulary knowledge, listening comprehension, and reading comprehension on the other (Sylvén and Sundqvist 2012). The Young Learner Vocabulary Assessment Test (Sylvén and Sundqvist 2016) was used and a comparison of the vocabulary scores for three ‘gamer groups’ showed that the frequent gamers (cut off: ≥ 5 h/week) scored higher than the moderate gamers, who outperformed the non-gamers (the same pattern was found also for the other tests). None of the examined background variables could explain the between-group differences, so gaming turned out to be an important EE variable. It was also the most popular EE activity, followed by watching TV, and listening to music. A one-week language diary with 7 predetermined activities and one open option was used.

In Flanders, the Dutch speaking part of Belgium, French is introduced in the 5th grade whereas English instruction starts late, in the 7th or 8th grade. Three large-scale studies carried out in the 6th grade (De Wilde et al. 2020a, 2020b; Puimège and Peters 2019) provide overwhelming evidence that these young learners (age 10–12) know a lot of English, despite never having had a single lesson in school. For example, results showed that they know more than 3,000 English word families (Puimège and Peters 2019) and many performed at the A2 level (according to the Common European Framework of Reference, Council of Europe 2020) for listening, speaking, reading and writing (De Wilde et al. 2020a).

Summing up, there is without a doubt a connection between young learners’ EE engagement and L2 English vocabulary learning. The affordances of multimodal input through audio-visual materials are prominent in all reviewed studies and YouTube, launched in 2005, has grown increasingly popular over time and almost seems to have turned into an EE activity of its own (see also Kusyk et al. 2025; Sundqvist 2024).

2.3 Research questions

The following research questions (RQs) were postulated for the present study:

3.1 Participants

Sixty-one first grade children participated in the study, as shown in Table 1. They came from five classes from two schools (denoted S02 and S04) in Norway. School S04 was located in a city (>100,000 inhabitants), and School S02 in a small town (<10,000 inhabitants). In Norway, the Norwegian Directorate for Education publishes statistics about all schools. As part of this, a so-called school contributor index (SCI) can be computed. In general terms, it signifies the difference between how students at a school perform and how they are expected to perform whilst controlling for certain conditions (previous school results, parents’ education level, household income and immigration background). If the performance matches the expectations, the school has the same index as the mean for all schools in Norway, which is 0. If the index is positive, the school’s contribution is above average for the country, and if negative, it is below average. Based on the SCIs for 2021/2022, School S02 was associated with a SCI of −0.3 whereas School S04’s SCI was 3.5.

The language situation in the home of the participants is provided in Table 2. All except two of the care-takers of the children reported using predominantly Norwegian with their child. Specifically, one third stated that Norwegian was the only language used, and another 20 % reported use of Norwegian and another language (not English).

English was used in combination with Norwegian or another language by one third of the participants.

3.2 Data collection instruments

The study made use of a mixed-methods approach, whereby predominantly quantitative test and questionnaire data were complemented with qualitative diary and interview data. In total, five instruments were used: A learner questionnaire of extramural English activities (EEQ), a parental questionnaire (PQ), a shortened version (PVST-S) of the Picture Vocabulary Size Test (PVST) (Anthony and Nation 2017), a Language Diary (LD), and Student Interviews (SI). These instruments are described below.

3.3 Procedures

The collection of data took place in the first semester of the school year, in October-November, preceded by researcher participation in parental meetings in June and August of the same year. During these meetings, parents were informed about the project aims and the implications for their children’s participation. Parental consent and the PQ were collected during these meetings.

Data collection involving the children commenced with an in-class introduction of the researchers, an age-appropriate account of the project, and information about what it meant for participants to participate (10–15 min). Data were collected one-on-one with each participant, while the rest of the class continued with their regular lessons.

The researcher escorted each participant to a separate data collection room. The participant was given the choice to sign a consent form by coloring a pre-printed smiling face (consent) or a frowning face (no consent) on a sheet of paper. All children consented, and the researcher commenced with the data collection.

Each participant first sat the PVST-S. The PVST-S items were shown on slides on a computer. For each item, the researcher read the target word out loud twice. Participants were then asked to point at the picture out of four that best represented the meaning of the word. Participants also had the option to say that they did not know the answer, by pointing at the image of a question mark (see example of item slide in Appendix E, which was used as a practice item).

For the EEQ, the researcher presented each question in an age-appropriate way, for example by breaking it down into shorter questions. The goal was to simulate a conversation with the participants, creating a more natural environment for them, since young learners are particularly vulnerable in assessment situations (McKay 2006). The researchers filled out the online form on the spot according to the participants’ answers.

Each participant spent approximately 30 min with a researcher. After completion, participants received a Language Diary with their names written on it. Participants were reminded to hand over the diaries to their parents, who had been asked beforehand to fill them out with their child once per day for a week.

4.1 Vocabulary scores

The scores from the PVST-S are presented in Table 3. The total mean score was just over 12 (out of 31). The reliability (Cronbach’s alpha) for the PVST-S data was an acceptable 0.79. As a benchmark, Hubbe Emilsson (2021) reported a reliability of 0.85 for the full 96-item test. As longer tests (more items) typically lead to a higher reliability, having a slightly lower value with only 1/3 of the items is a positive outcome. The sampling ratio of 16:1,000 in the original PVST would have afforded an extrapolation by multiplying a score by 62.5 for the most frequent 6,000 words (Anthony and Nation 2017). However, our use of a modified, shorter version resulted in a lower sampling rate (see Gyllstad et al. 2015). As a majority of words came from the 1–2 K bands (see Appendix A), we have also reported the mean score for these 20 high frequency words in Table 3. The sample ratio 20:2,000 thus rendered a multiplication value of 100, and an extrapolated mean vocabulary level of 862 words.

The Item facility (IF) values of the 31 words are visualized in Figure 1. The items varied between 0.89 and 0.07 (89 % and 7 %, respectively, of the testtakers who answered correctly). The easiest item was cream and the most difficult canary. To find out whether the frequency of the word was related to the IFs, the IFs for the words were plotted against the frequency band number that the words were taken from. Figure 2 visualizes the level of the correlation (−0.357).

Figure 1:

Item facility values for the 31 items in PVST-S (N = 61).

Figure 2:

Item facility values (IFs) (Y-axis) as a function of frequency band (X-axis) for the 31 Items in PVST-S (N = 61).

Trend-wise, Figure 2 shows how a lower frequency band rank for a word corresponds to a lower IF. It should be noted, though, that words from frequency band 1 displayed a wide range of IFs.

One factor known to affect L2 vocabulary knowledge is cognateness between the target language and a learner’s first language. Cognates, defined as translation equivalents with high orthographic overlap, can be identical or similar. There were 12 cognate words (indicated in Appendix A), none of them identical, only similar. The mean IF value for the cognates was 0.44 (SD = 0.22) compared to 0.36 (SD = 0.20) for the non-cognates, and a t-test comparison of these means yielded a non-significant result (mean difference = 0.073, t = 0.951, p = 0.359). The Levenshtein distance (Levenshtein 1966) metric can be used to indicate how close two words are orthographically, by counting the minimal number of substitutions, insertions, and deletions to edit one string of letters into another of any length. As comparing the phonetic representations would be more ecologically valid for young learners who are not yet literate (see De Wilde et al. 2020a), we also computed values for the phonetic representations. Correlations were then run between the Levenhstein distance values and the IF values to see if degree of similarity affected the IF, using Kendall’s Tau for small data sets and many tied ranks. The correlation for orthographic form was τ = −0.112, p = 0.625, whereas it was τ = −0.148, p = 0.536 for the phonetic form. Consequently, the similarity between the cognate words did not have any statistically significant effect on how easy they were for the learner group.

4.2 The relation between extramural English and vocabulary knowledge

Figure 3 shows the participants’ self-reported exposure to the sources in the EEQ: English-speaking TV, English material on YouTube, and English videogames. Eleven children reported exposure to no sources (18 %), 15 to one source (25 %), 18 to two sources (30 %) and 14 to all three sources (23 %). Data was not available for three students.

Figure 3:

Children’s self-reported number of EE sources exposed to (N = 58).

Out of the 61 participating learners, 38 submitted their language diary. The reported time spent on EE overall was tallied and correlated with the PVST-S scores (see Table 4). All activities except Writing (no time reported) showed positive relations to the PVST-S scores. Two correlations were statistically significant: time spent watching TV and total time spent on EE activities.

The activities that most learners engaged in were Listening to music and Watching TV. The large standard deviations illustrate substantial variation in time reported. For many learners, no time was reported on many activities, mainly Reading and Other.

To further address RQ2, a mixed-effects model (MEM) based on the whole participant sample was used. MEM provides many advantages over traditional analysis of variance, such as accounting for random variation in items and participants in one analysis, being robust against violations of homoscedasticity and sphericity, and combining random and fixed effects in the same analysis (Linck and Cunnings 2015). We used R version 4.4.0 (R Core Team 2021) together with R Studio (Version: 2023.12.1). We employed a mixed effects logistic regression model. The model used item score on the PVST-S as the binary dependent variable (correct = 1, incorrect = 0) using the glmer function with a binomial family under the lme4 package (Version 1.1–35.1; Bates et al. 2015). The optimizer ‘bobyqa’ was used.

Table 5 reports the MEM and its parameters. The model included the EEQ index, the PQ index and gender as the primary fixed effects. We also included maternal education level, word AoA, and word frequency band as covariates. These were included as previous research has indicated that they are relevant factors (De Wilde et al. 2020b; Soto-Corominas et al. 2024). Random intercepts were specified for items and participants, and participants were nested in Class and School. Further specification of the random-effects structure led to a failure to converge.

Estimates are on the log-odds scale, but exponentiated values (Exp β) are also reported in the right-most column in Table 5. As to the fixed effects, the children’s own EEQ index failed to reach significance. However, there was a statistically significant effect for the PQ index of parents’ reported child EE activity on the children’s vocabulary score. The estimate of 0.380 for PQ means that a 1 unit increase in PQ index is associated with a 0.380 increase in the log-odds of an item score being 1, compared to it being 0.

If exponentiated this number yields the odds ratio (Exp β) of 1.462, as seen in Table 5. Additional significant effects were observed for gender, whereby the odds of answering a PVST-S item correctly were 42 % lower for girls compared to boys, if keeping all other variables constant (given an odds ratio of Exp (−0.537) = 0.584). Thus, gender had an effect on the participants’ receptive word knowledge. Equally significant was the co-variate of AoA (i.e., the age at which native speakers of English are estimated to have acquired a word). Word frequency was also a negative estimate, indicating a lower score on the PVST-S as an effect of decreased word frequency.

4.3 The learners’ perceptions about English and English language learning

A final analysis was based on the interview data. The interviews served two purposes. One purpose was to shed light on the learners’ perceptions about English and their own learning and use of English through a thematic analysis. Another purpose was to further shed light on the relationship between amount of EE and word meaning recognition knowledge for a subsample of eight learners. Table 6 provides information on the subsample of interviewees, appearing as pseudonyms.

Interviews were transcribed verbatim and subjected to thematic analysis (Braun and Clarke 2022) with the aim to explore RQ3. First, emerging patterns in the material were identified from familiarising ourselves with the dataset through listening to audio recordings and reading transcripts multiple times. Based on these, potential codes were developed, and the entire dataset was then coded systematically. The codes were developed into themes, which in turn were reviewed to ensure that they highlighted the patterned meaning across the dataset in relation to the research question.

Two themes were developed in the thematic analysis. First, English was perceived to be a lingua franca; English could be used to communicate with people who could not speak Norwegian, as argued by Marcus: “if someone is new here, coming to the class, and can’t do Norwegian, then we can talk to them”. Such reflections were offered as answers to questions about reasons for learning English, but not all the participants answered the question. For some the concept of English as a language, or of language in general, seemed to be difficult to grasp. “English” was frequently associated with the school subject.

The second theme was related to perceptions about English language learning. The young learners agreed that they had known some English before starting school, but they had a difficult time remembering where this knowledge originated. Some answered at home and some the kindergarten, but they could not answer who had actually taught them. Cole explained that English one day had appeared in his repertoire, when he was 3 or 4 years old: “I just thought that it was English, then I started talking it”. Neither Cole nor the other young learners suggested EE as a source of English learning when asked explicitly about this, but some of them still offered reflections about incidental learning from EE, specifically from music. Another interviewee was Bro. When asked why he listened to English songs instead of Norwegian songs at bedtime, Bro offered incidental English learning as a reason: “I want to learn English while I am sleeping in my brain”.

In the same vein, Celine explained that she listened to English music rather than Norwegian because she wanted to learn English. She also said that English songs were more fun, as did Leah and Tagg. Tagg also said that English songs were “more real”, suggesting perceived authenticity of English in relation to music. Apart from this metalinguistic awareness related to English music, the participants did not seem to deliberately target English-language content. Rather, they chose digital platforms, games and content based on their interests, and these happened to be in English.

Information about participants’ EE exposure obtained in the interviews suggested that there is no straightforward relationship between amount of EE and word meaning recognition. Celine, for instance, had very positive attitudes to English and was able to produce whole sentences in English in the interview. However, her PVST-S score was just slightly above the mean (14 points). Tagg’s LD showed that she engaged in considerable amounts and varying types of EE, but her PVST-S score was low (5 points). Bro and Cole, who had very high PVST-S scores (25 and 24 points), stood out among the interview participants as avid gamers. In the interviews, they enthusiastically listed games that they engaged with, and they both reported to game with family members. Although there is not enough evidence to draw inferences about influence from gaming on word meaning recognition from these individual cases, the concurrence is interesting and merits further study.