Implicit Foreign Language Learning: How Early Exposure and Immersion Affect Narrative Competence

Abstract

This study investigates how short-term naturalistic immersion shapes the development of evaluative narrative competence in Japanese junior high school students learning English as a foreign language. While prior second language acquisition (SLA) research has established the benefits of input-rich environments, little is known about how implicit learning during brief immersion experiences supports higher-order storytelling skills. To address this gap, we analyzed students’ performance on a standardized problem-solving task and a storytelling task before and after a one-month homestay abroad. Results showed significant post-immersion gains in narrative complexity, with longer stories, greater use of causal and evaluative devices, and increased diversity of expression. Regression analysis revealed that the age of first English exposure strongly predicted outcomes: early starters demonstrated broader and more sophisticated use of evaluative strategies than later starters. These findings suggest that short-term immersion can substantially enhance narrative competence, particularly for learners with early exposure, while highlighting the need for tailored pedagogical interventions to help later starters capitalize on implicit learning opportunities.

1. Introduction

Research in second language acquisition (SLA) has long debated the extent to which learning is driven by implicit, unconscious processes versus explicit, conscious ones. While there is broad consensus that both input and output are essential, theoretical perspectives diverge on how learners internalize new forms and what conditions most effectively promote acquisition.

The Cognitive-Interactionist framework situates learning at the intersection of internal cognitive processes and external communicative experiences. M. Long’s (1996) Interaction Hypothesis argued that language acquisition is fostered when learners engage in ‘negotiated’ interaction that makes input comprehensible and provides opportunities for feedback. Building on this, Ellis’s (2005) Noticing Hypothesis—based on Schmidt’s (1990) noticing construct—emphasized that learners must consciously register specific linguistic features for input to become intake. Ortega (2009) synthesized these perspectives, proposing that successful acquisition depends on the dynamic interplay between cognitive processes such as attention and awareness and environmental factors such as social interaction and communicative feedback.

In contrast, Krashen’s (1985, 2013) Input Hypothesis advanced a more unidirectional view, maintaining that acquisition is primarily driven by exposure to comprehensible input, with little role for conscious learning. This position sparked decades of debate over the balance between implicit and explicit processes in SLA. More recent theories have reframed the discussion in terms of the distinction between implicit knowledge—unconscious, procedural, and automatized—and explicit knowledge—conscious, declarative, and rule-based (Ellis et al., 2009; Ellis & Roever, 2021; Suzuki, 2017).

Contemporary refinements such as the Ordered Development Hypothesis (Lichtman & VanPatten, 2021) suggest that acquisition proceeds through predictable developmental stages, which cannot be easily bypassed or accelerated by direct instruction. This insight reinforces the limits of purely explicit teaching while highlighting the enduring need for rich, contextualized language exposure. Meanwhile, pedagogical applications such as Task-Based Language Teaching (TBLT) have translated the cognitive-interactionist view into practice by creating conditions for negotiation of meaning, attention to form, and feedback (M. Long, 2015; Loewen & Sato, 2018).

Taken together, the evolution of these hypotheses underscores a central principle of the Cognitive-Interactionist approach: language learning is most effectively facilitated when learners are exposed to meaningful, communicatively embedded input and are simultaneously engaged in cognitive processes of noticing, awareness, and interaction. This theoretical foundation explains why effective SLA pedagogy must balance opportunities for implicit development with carefully scaffolded conditions for explicit noticing and reflection.

A central question concerns how age of exposure interacts with different learning environments. The Maturational Hypothesis suggests that children and adolescents are more adept at acquiring language implicitly through natural exposure, while adults rely increasingly on explicit learning (Lichtman, 2016). Evidence from formal classroom contexts shows that explicit instruction benefits adults and older adolescents (Goo et al., 2015; Muñoz, 2006; Norris & Ortega, 2000; Spada & Tomita, 2010), whereas children often make greater gains in oral proficiency during immersion or study-abroad contexts rich in implicit input (Llanes & Muñoz, 2013; Llanes & Serrano, 2017). Importantly, a linguistic threshold appears necessary for learners to fully benefit from naturalistic input (Leonard & Shea, 2017; A. Y. Long & Solon, 2021; Pérez-Vidal & Juan-Garau, 2011), suggesting that starting age and baseline proficiency jointly mediate outcomes.

Despite this growing body of work, little is known about how implicit-rich immersion environments foster higher-order discourse abilities such as narrative competence. Narratives are a powerful measure of spontaneous language use because they require real-time meaning-making and are believed to draw on implicit knowledge (Bowles, 2011; Ellis & Roever, 2021). Beyond simple event descriptions, effective storytelling also involves evaluation—the narrator’s ability to highlight significance, convey mental states, and create coherence through reasoning and interpretation (Labov & Waletzky, 1967). While prior study-abroad research has generally reported quantitative gains in linguistic aspects of learners’ second language (L2) narratives—such as vocabulary (Serrano et al., 2011; Zaytseva et al., 2021), grammar (A. Y. Long & Solon, 2021), and fluency and accuracy (Leonard & Shea, 2017; Llanes & Muñoz, 2013; Llanes & Serrano, 2017)—it has yet to delve into their evaluative dimensions. To our knowledge, only a few studies, even in the broader field of SLA, have explored the evaluative aspects of children’s L2 narrative development. These studies show that as children’s cognitive development advances with age, their narrative abilities become more complex, and they employ evaluative devices—linguistic tools that convey attitude and perspective—in increasingly varied and sophisticated ways (Kupersmitt et al., 2014; Viberg, 2001). As they refine their storytelling, children also begin to align with culturally shaped narrative conventions (Chen & Yan, 2011). However, expressing and integrating such evaluative elements appropriately within a narrative requires adequate L2 proficiency. Thus, while discourse abilities generally develop from early childhood through adolescence, producing higher-order narratives as richly evaluative as those in the first language remains challenging even for children who are regularly, though not extensively, exposed to an L2 at home or in school (Kang & Uchikoshi, 2022; Montanari, 2004; Viberg, 2001). Although these studies each touch on related areas of L2 narrative development, they do not examine implicit learning effects in short-term naturalistic immersion contexts.

The present study investigates this previously underexplored issue involving implicit learning during short-term naturalistic immersion and its effects on the development of higher-order discourse skills in Japanese junior high school learners of English as a foreign language (FL). Building on the work by Kano and Quay (2023), we combine a standardized problem-solving task with a detailed analysis of evaluative devices in a frog story narrative task (Mayer, 1969). Specifically, we examine whether a one-month immersive experience abroad enhances learners’ ability to articulate solutions in English and enriches their use of evaluative devices in storytelling. Additionally, we test whether the age of first exposure to English predicts outcomes, given its theorized role in shaping implicit learning capacity. Accordingly, our research questions (RQ) are as follows:

• Can Japanese junior high school students articulate logical and relevant solutions to problems in a FL before and after a short homestay abroad?
• Does the starting age of implicit English exposure predict children’s ability to use evaluative devices in L2 narratives?

These questions are designed to assess how implicit learning can foster the ability to produce complex, meaningful discourse, a critical skill for academic and social success. By focusing on the evaluative dimensions of narrative competence, this study contributes to ongoing debates on implicit learning, age effects, and immersion in SLA.

2. Materials and Methods

2.1. Participants

Fifty-five Japanese junior high school children (34 girls and 21 boys), aged 12;4–14;2 (in year;month), participated in this study. All were recruited through Labo Teaching Information Center (LABO), a private educational organization in Japan that provides informal English exposure and summer homestay opportunities. LABO emphasizes naturalistic interaction over explicit instruction (thus providing an ideal context for investigating our research questions). Participants came from monolingual Japanese, middle- to upper-middle-class families residing in Tokyo, Kanagawa, or Shizuoka. Written informed consent was obtained from the parents of all participants, and assent was received from the children themselves. Participants were assigned two-letter pseudonyms as shown in

Table 1. Children’s starting age of exposure to English and age at the pre-test.

#	Part	Sex	SAE	Age	#	Part	Sex	SAE	Age	#	Part	Sex	SAE	Age
1	MK	F	2;8	13;3	20	MM	M	4;4	12;6	39	RE	M	5;6	12;5
2	SY	M	2;11	12;7	21	TC	F	4;4	13;7	40	AK	F	5;6	12;7
3	SK	M	3;0	14;1	22	YZ	F	4;5	12;5	41	SU	F	5;6	12;9
4	HN	F	3;2	13;2	23	RT	M	4;5	12;5	42	RK	M	5;6	13;2
5	EA	M	3;3	13;3	24	KZ	M	4;5	13;3	43	YN	F	5;8	13;1
6	FN	F	3;3	13;4	25	ME	M	4;5	13;3	44	YM	M	5;9	12;5
7	ST	M	3;6	12;5	26	SO	F	4;6	12;4	45	KT	M	5;9	12;6
8	HA	F	3;6	12;6	27	HH	F	4;6	12;8	46	SR	F	5;9	14;0
9	MD	M	3;8	13;0	28	AY	F	4;8	12;10	47	KM	F	6;0	12;5
10	TY	M	3;9	12;5	29	SA	F	4;11	12;5	48	SH	M	6;2	12;5
11	MA	F	3;11	12;5	30	MO	F	4;11	12;7	49	CH	F	6;2	12;5
12	YT	M	4;0	12;5	31	RY	M	5;0	12;5	50	YA	F	6;2	12;7
13	HK	F	4;2	12;7	32	HI	F	5;2	12;4	51	YK	F	6;4	13;3
14	NT	F	4;2	12;9	33	HO	F	5;2	12;5	52	YO	F	6;6	12;6
15	AS	F	4;2	13;1	34	RO	M	5;2	12;8	53	AN	F	6;6	13;2
16	SN	F	4;2	14;2	35	YD	F	5;5	12;4	54	NN	F	6;8	12;6
17	MS	F	4;2	13;2	36	SI	F	5;5	12;5	55	YS	M	6;9	12;5
18	TM	F	4;4	12;6	37	YU	F	5;5	13;1	Mean			4;8	12;8
19	MT	M	4;4	12;6	38	HR	M	5;6	12;5	SD			1.08	0.46

to ensure confidentiality.

Information on children’s background and details about the program come from parental questionnaires and tutor interviews. Children began attending LABO between the ages of two and six, receiving approximately two hours per week of implicit English exposure through songs, games, and stories conducted mainly in English. Role-play activities were performed bilingually, with support from trained Japanese-English bilingual tutors. Groups typically consisted of 10–15 peers of similar age. This experiential approach contrasts with the participants’ formal English classes at school (introduced in Grade 5), which they reported as relatively easy and less beneficial.

As part of the LABO program, all participants completed a one-month homestay in the United States or Canada. They lived with host families, typically middle- to upper-middle-class households in rural areas or small towns, where opportunities to interact with other Japanese speakers were minimal. No formal language lessons were provided; students were immersed in daily family life, interacting primarily with host siblings of similar age.

Table 1 lists participants in order of their starting age of English exposure (SAE; M = 4;8, SD = 1.08), along with their age at pre-test (M = 12;8, SD = 0.46) and sex. For analysis, participants were categorized as “early starters” (SAE ≤ age 4;6) or “late starters” (shaded, SAE > age 4;6) to examine potential developmental patterns. Age 4;6 was chosen to divide the participants into two groups because it is close to both the mean SAE (4;8) and the median (4;7). Since there were no participants aged 4;7, the late starter group from age 4;8 reflects a clear separation in the overall SAE distribution.

2.2. Assessments and Procedures

Data were collected immediately before and after the one-month homestay. To minimize test–retest effects, participants were not informed in advance that the same tasks would be repeated. All sessions were conducted individually and audio-recorded for later transcription and coding.

2.2.1. Problem-Solving Task

To address RQ1, all participants completed the Oral Proficiency Assessment for Bilingual Children (OBC; Canadian Association for Japanese Language Education, 2000) before and after their homestay. The OBC is a standardized interview test designed to evaluate the oral proficiency of bilingual children, particularly Japanese children living abroad, in both Japanese (their heritage language) and English (their second language, L2). Modeled after the ACTFL Oral Proficiency Interview, the OBC consists of a three-tiered oral proficiency assessment: basic/linguistic, communicative, and academic/cognitive. We focused on the academic/cognitive task, in which participants viewed a picture card shown in Figure 1 depicting environmental damage (a crying earth surrounded by pollution) and were asked to (a) explain the problem and (b) propose possible solutions.

Responses were evaluated and quantified according to the following:

• Whether a cause is given to question (a), “Why do you think the earth is crying?”, and a solution is provided to question (b), “What do you think should be done for this problem?” (1 point when answered and zero when not);
• Qualitative assessments of coherence, adequacy of content, vocabulary appropriateness, and utterance length using a five-point scale (1 = poor to 5 = excellent) for a possible total score of 20.

Inter-rater reliability was established on 20% of responses by an independent trained rater. Agreement reached 88% with discrepancies resolved by discussion.

2.2.2. Storytelling Task

To address RQ2, we used the picture book, Frog, Where Are You? (Mayer, 1969), a widely adopted tool for eliciting spontaneous narratives. This book tells the wordless story of a boy and his dog searching in the woods for their pet frog. Their search leads them to encounter various animals before finding their lost pet and returning home. In both pre- and post-tests, participants were given five minutes to examine the book before narrating the story in English. Narratives were transcribed and coded for evaluative devices based on a coding scheme adapted from Nakamura (2009, 2019) with seven categories that help narrators convey meaning and complexity beyond simple descriptions:

• Frames of mind: Expressions describing a character’s mental and affective states (e.g., emotion terms, mental state verbs);
• Character speech: Direct or indirect speech that adds vividness and dramatization;
• Negative qualifiers: Negations highlighting events that did not happen;
• Causal connectors: Words like “because” that create cause-and-effect relationships;
• Hedges: Distancing devices that reflect uncertainty (e.g., “maybe”);
• Onomatopoeia: Words that vividly mimic sounds;
• Enrichment expressions: Adverbial phrases that act as intensifiers (e.g., “very”) and connectives that signal unexpected or contrastive events (e.g., “however”).

To quantify the quality of pre- and post-immersion narratives, we calculated both the frequency and diversity of these evaluative devices. Frequency was determined by dividing the total number of evaluative devices by the total number of clauses, including those with elliptical expressions. Diversity was calculated by counting the number of different evaluative categories used in each narrative (maximum 7). Again, 20% of transcripts were double-coded by a second rater. Inter-rater reliability exceeded 90% agreement.

2.3. Statistical Analyses

Analyses were conducted in R (version 4.5.1; R Core Team, 2025). Paired-sample t-tests were used to compare pre- and post-immersion scores for the storytelling task as they account for the dependency of the pre- and post-test data from the same individuals, thereby controlling for subject-level variability. Effect sizes (Cohen’s d) and 95% confidence intervals were calculated to assess the magnitude of gains. A multiple regression analysis was also performed, as this statistical model allows us to test the predictive role of SAE, narrative length, and frequency of evaluative devices on post-immersion diversity scores.

3. Results

3.1. Performance on Problem-Solving Task

Figure 2 shows that both early and late starters improved in identifying causes of environmental problems and proposing solutions after the homestay, with early starters performing considerably better. Cause identification increased from 26% to 74% among early starters (n = 27) and from 7% to 32% among late starters (n = 28). Solution proposals rose from 7% to 22% for early starters but remained rare among late starters (0% → 3.5%). Identifying causes from the visual cues shown in Figure 1 was generally easier than suggesting solutions, which required greater conceptual reasoning and vocabulary.

Qualitative assessment scores (based on coherence, content adequacy, vocabulary, and utterance length) are shown in Figure 3. Significant gains in mean scores occurred for both groups, with larger and more variable improvements among early starters. Statistical results are summarized in

Table 2. Statistical results of qualitative assessments before and after homestay.

Group	n	Pre-Immersion M (SD)	Post-Immersion M (SD)	T	p	d
Early starters	20	2.1 (3.74)	5.4 (3.3)	5.73	<0.001	0.93
Late starters	9	0.6 (1.3)	3.4 (0.53)	7.62	<0.001	2.28
Combined	29	1.66 (3.23)	4.79 (2.88)	6.41	<0.001	0.93

for the sake of clarity.

Notably, the effect size (d) of 2.28 for late starters is very large, indicating that the homestay had an exceptionally strong impact on the qualitative skills of the 9 participants who completed this task. The most significant finding is that while the homestay program was highly effective for both groups, it had a disproportionately larger effect on the Late Starters. The participants who completed the problem-solving task in this group, which started with a lower average score, experienced a much greater magnitude of change, as evidenced by their exceptionally large effect size. The results strongly suggest that the homestay program is an effective tool for improving the qualitative skills assessed, with a particularly powerful impact on those who begin with lower proficiency. Despite these gains, most participants (86%) achieved only one point per criterion (out of 5), suggesting that post-immersion responses reflected greater willingness to attempt answers rather than consistently higher-quality output.

Qualitative examples further illustrate these findings. Example 1 comes from MA, an early starter who scored at her group’s median, who answered the first question and demonstrated improved ability to articulate her thoughts after her homestay.

Example 1. 

Early starter, post-test.

“Many trees are nothing because people cut a lot. And people use bus and truck and many bad, bad, not good, not clear gas, so earth is crying.”

In the post-test, she used simple vocabulary and causal connectors to describe deforestation and vehicle exhaust. Her self-corrections (“many bad, bad, not good, not clear gas”) indicate a developing ability to monitor and refine her spontaneous speech for clarity. Nevertheless, despite having studied environmental pollution at school in Japanese, MA could not provide a response in English with adequate content or use a variety of related vocabulary, culminating in a total score of four out of 20. Example 2 comes from YD, a late starter scoring at her group’s median, who attempted to answer both questions in the post-test (which she could not do before the homestay).

Example 2. 

Late starter, post-test.

“Because cut tree and many smokes. So hot, hot earth. Um, maybe I think we don’t cut tree and little smoke.”

While her response lacked complete sentences, her use of a hedge (“maybe”) and a mental state expression (“I think”) showed an attempt to structure her thoughts more complexly. She scored zero for response length and earned one point for each of the other three criteria for a total score of three.

Notably, one early starter, HN, showed a dramatic improvement as she provided no response in the pre-test (score: 0) but was able to produce coherent explanations with concrete examples post-immersion (e.g., citing tree cutting, gas emissions, and environmentally friendly solutions such as hybrid cars) in the post-test (score: 12). This highlights that immersion can support not only greater output quantity but also integration of academic content knowledge into English responses as shown in Example 3.

Example 3. 

Early starter, most substantial improvement.

“The earth is crying because, um, the people is always do not good for nature, such as cut, cut many trees and car, gas, and use very much gas. We should more kindly for earth. For example, um, using, using LED or not riding, and riding hybrid car or electricity car. Then, I think it won’t earth crying.”

3.2. Performance on Storytelling Task

After the homestay, participants’ narratives showed significant and meaningful improvements in their use of evaluative language.

Table 3. Narrative performance before and after homestay.

Measure	Pre- (M, SD)	Post- (M, SD)	Change	t(df)	p	Cohen’s d	[95% CI]
Frequency (%)	54.5 (33.1)	69.6 (29.9)	15.1	4.05 (54)	<0.001	0.48	[7.60, 22.54]
Diversity (0–7)	2.15 (1.38)	3.40 (1.44)	1.25	5.90 (54)	<0.001	0.90	[0.83, 1.68]
Length (no. of clauses)	12.05 (9.66)	20.25 (11.7)	8.2	7.56 (54)	<0.001	0.75	[6.03, 10.37]

Note. Means and standard deviations are reported for pre- and post-immersion measures. Cohen’s d values are accompanied by 95% confidence intervals to indicate the precision of effect size estimates. All intervals excluded zero, confirming reliable medium-to-large effects: frequency showed a moderate effect (d = 0.48), while diversity (d = 0.90) and length (d = 0.75) reflected large effects.

reveals that the average frequency of evaluative devices rose from 54.5% to 69.6%, a change that reflects a moderate effect (d = 0.48, 95% CI [7.60, 22.54]). Participants also demonstrated a marked increase in the diversity of these devices, with average scores rising from 2.15 to 3.40. This was the strongest improvement observed, reflecting a large effect (d = 0.90, 95% CI [0.83, 1.68]). Additionally, narrative length saw a large effect as well (d = 0.75, 95% CI [6.03, 10.37]), with students producing an average of eight more clauses per story. Importantly, 84% of our participants improved in frequency, diversity, or both, while those who did not show greater evaluative variety nonetheless produced longer and more event-rich narratives.

Overall, the findings demonstrate that immersion led to meaningful gains in narrative performance. Increases in frequency and diversity of evaluative devices and narrative length were not only statistically significant but also represented medium-to-large effects, with the largest gains observed in diversity. These results indicate that even short-term immersion can foster both richer and more elaborated storytelling, suggesting substantial pedagogical benefits for learners’ expressive repertoires. These findings align with cognitive-interactionist accounts of L2 development, which emphasize the role of input, output, and attention in promoting language growth. The significant improvements in evaluative frequency, diversity, and narrative length suggest that immersion not only increased learners’ opportunities for meaningful interaction but also heightened their awareness of discourse-level features. In particular, the large effect observed for diversity resonates with theories highlighting the importance of noticing and form–function mapping in expanding learners’ expressive repertoires. These results therefore provide empirical support for the view that intensive communicative environments foster both structural complexity and pragmatic richness in learner narratives.

3.3. Predictors of Post-Immersion Narrative Performance

A multiple regression analysis (

Table 4. Multiple regression analysis predicting post-immersion diversity of evaluative devices.

Predictor	B	SE	β	t	p	[95% CI]
(Intercept)	2.89	0.91	—	3.18	0.003	[1.06, 4.72]
Starting age of English exposure	−0.04	0.01	−0.33	−3.13	0.003	[−0.06, −0.01]
Narrative length (no. of clauses)	0.07	0.01	0.55	5.09	<0.001	[0.04, 0.09]
Frequency of evaluatives	0.02	0.005	0.40	4.16	<0.001	[0.01, 0.03]

Model summary. F(3, 51) = 22.74, p < 0.001; R² = 0.55; Adjusted R² = 0.55. Note. Standardized coefficients (β) reflect relative predictive strength. Narrative length was the strongest predictor (β = 0.55), followed by frequency of evaluative expressions (β = 0.40) and earlier English exposure (β = −0.33). Confidence intervals excluded zero for all predictors, indicating stable and robust effects.

) revealed that three factors significantly predicted post-immersion diversity of evaluative devices. Earlier starting age of English exposure (SAE) predicted greater diversity (β = −0.33, p = 0.003), narrative length was the strongest predictor (β = 0.55, p < 0.001), and frequency of evaluative expressions also contributed substantially (β = 0.40, p < 0.001). Together, these variables explained 54.7% of the variance in evaluative diversity (F(3, 51) = 22.74, p < 0.001, adjusted R² = 0.55), representing a large effect size. Of practical importance, narrative length was the strongest predictor, nearly twice as influential as age of exposure. The predictors showed both statistical significance and meaningful effect sizes, thus underscoring the robust predictive value of these factors.

Inspection of the unstandardized coefficients further illustrates the magnitude of these relationships. Each one-year delay in English onset was associated with a 0.04 decrease in diversity (95% CI [−0.06, −0.01]). Conversely, each additional clause in the narrative predicted a 0.07 increase in diversity (95% CI [0.04, 0.09]), and each additional evaluative expression was linked to a 0.02 increase (95% CI [0.01, 0.03]). Importantly, the 95% confidence intervals for all predictors excluded zero, strengthening confidence that the observed effects are robust rather than due to sampling error.

In summary, children who began English earlier, told longer stories, and deployed evaluative expressions more frequently demonstrated the greatest variety of evaluative strategies. The model’s strong explanatory power indicates that these factors jointly capture much of the variability in evaluative diversity across learners.

3.3.1. The Use of Evaluative Devices by Early Versus Late Starters

As Figure 4 illustrates, both groups significantly increased their use of six evaluative categories after the homestay. Consistent with the pre-immersion data, character speech and frames of mind remained the most frequently used categories across all narratives. However, a closer look reveals different patterns of improvement for each group. While both groups improved in all categories, their biggest gains occurred in different areas. Early Starters showed particularly large increases in their use of enrichment expressions and causal connectors, which are crucial for adding logical flow and complexity to a narrative. In contrast, Late Starters made their most significant strides in categories like character speech, negative quantifiers, and onomatopoeia. This finding aligns with the results from Table 4, suggesting that children who started learning English earlier leveraged their month abroad to develop a greater variety of evaluative devices.

3.4. Illustrative Examples

The individual narratives of MA and YD (featured earlier in the problem-solving task as representing the typical narrative improvement patterns of their respective groups) reinforce these findings. In Example 4, MA, an early starter, refined her use of evaluative devices after the homestay. She incorporated more frequent enrichment expressions and negative quantifiers (Mike and James are looking for Hunter everywhere; Mike and James wake up, but Hunter is not here) to highlight key events and added causal connectors (because and so) to express cause-and-effect relationships. Most notably, she began to use direct character speech, making her story more dynamic.

Example 4. 

Narratives produced by MA, an early starter.

Pre-immersion:

The boy, his name is John, and the dog’s name is Mike, and the frog. They stay together. At the night, the frog outside over the window. The next morning, John and Mike woke up, but the frog is out. They are, John and Mike are searching a frog. John and Mike went looking for the frog outside. John and Mike call the frog. There are, is an animal. But, but they couldn’t reach the frog. They are fall, fall off in the pond. After that, they look, they find the frog by the tree. The frog is his family, come back, come back his family. After that, John and Mike and the frog come back to their house.

Post-immersion:

The boy’s name is James. James has one dog and one frog. The dog’s name is Mike, and the frog’s name is Hunter. One day, night, Hunter is going outside. Next morning, Mike and James wake up, but Hunter is not here. Mike and James are looking for Hunter everywhere. Mike and James are looking for Hunter outside and call Hunter, but can’t find. Mike and James are looking for Hunter in the tree, but can’t find. And Mike and James climbed the big rock. “Where is Hunter?” But deer, Mike and James found the deer. And Mike and James fall out in the pond. But Mike and James can hear a nice song because Hunter’s family sings a song. James and Mike could find Hunter, so Mike and James and Hunter come back home.

In Example 5, YD, a late starter with more limited proficiency, also showed clear improvements. While her pre-test narrative was a simple monologue, her post-test narrative included multiple perspectives (from the boy, dog, and other animals) and a conversation between characters. The addition of negative quantifiers and character speech enriched her storytelling, even if her overall proficiency remained lower than MA’s.

Example 5. 

Narratives produced by YD, a late starter.

Pre-immersion:

Once upon a time, a boy and frog and dog in this house. One day, in the, etto [um]. “Where is frog? Frog? Frog? Frog? Frog? Oh, ouch. Frog? Frog? Oh, no! Shh. Oh, frog! Your family? Bye.

Post-immersion:

Once upon a time, there is a boy, dog, and frog in the house. “Good night, frog. Good night, dog. I go to the bed. Oh, no! Frog is gone! Frog, where is a frog? Do you know frog?” “No.” “Where are you, frog? Frog, frog, frog, are you in the hole?” “This is my house.” “Frog, are you in the tree?” “No, this is my home.” “I’m sorry. Frog, frog!” “Be careful. Don’t touch me!” “Oh, no! Shh, be quiet.” “I got it.” “Oh, this is your girlfriend? This is your child? Good-bye, frogs!”

These examples highlight how immersion supports both linguistic expansion and the emergence of higher-order discourse strategies, albeit with differences by starting age of English exposure.

4. Discussion

This study found that a one-month naturalistic immersion can significantly enhance the FL skills of junior high school students, particularly their narrative competence as measured through a storytelling task. The results showed that participants’ stories had significant improvements in three key areas:

• Narrative Length: The average story length increased by eight clauses, a gain with a large effect size (d = 0.75), demonstrating that students were able to produce more elaborated stories.
• Frequency of Evaluative Devices: The average frequency of evaluative devices—linguistic ways of adding meaning and complexity beyond simple descriptions—rose from 54.5% to 69.6%. This included using character speech, causal connectors, and enrichment expressions.
• Diversity of Evaluative Devices: Participants showed the most significant improvement in the diversity of evaluative devices used, with the average number of different categories rising from 2.15 to 3.40. This change had a large effect size (d = 0.90), indicating that the homestay helped students expand their expressive repertoire.

Our study also used a problem-solving task to assess the children’s ability to articulate solutions in English. Post-immersion, both early and late starters showed significant gains in their qualitative assessment scores, indicating improved coherence, content, and vocabulary. Notably, some late starters, who completed this task and began with lower scores, showed an exceptionally large effect size (d = 2.28) in their improvement, suggesting the homestay was particularly impactful for them.

4.1. Implicit Learning and Immersion

Our findings are consistent with and extend existing SLA theory, particularly regarding the role of implicit learning and the age of exposure in naturalistic settings. The substantial gains in narrative competence—evidenced by increased frequency, diversity, and length—support the notion that rich, authentic input, as provided by the homestay experience, can facilitate significant language development outside of formal classroom instruction. Our results align with the Noticing Hypothesis (Schmidt, 1990; expanded in Ellis, 2005) as the demands of real-world communication likely pushed learners to consciously attend to and internalize new linguistic forms, even without explicit instruction. The particularly large gains in diversity and length for both groups underscore how immersion environments can foster the acquisition of more complex, higher-order language skills that are crucial for coherent and engaging storytelling. Our participants, immersed in authentic home and community contexts, encountered precisely the type of comprehensible input espoused by Krashen’s (1985, 2013) Input Hypothesis as crucial for driving acquisition. The significant improvements in storytelling suggest that immersion provided opportunities for learners to consolidate implicit knowledge, evident in their ability to produce spontaneous, content-rich narratives.

The gains observed in evaluative devices also support the Ordered Development Hypothesis (Lichtman & VanPatten, 2021), which argues that acquisition progresses through predictable stages not easily accelerated by explicit instruction. Our findings imply that immersion created conditions that allowed learners to advance naturally in narrative complexity, beyond what formal classroom instruction alone had fostered.

4.2. The Role of Age of Exposure

A key contribution of this study is the demonstration of how the age of first English exposure interacts with a short-term immersion experience to shape narrative development. The finding that early starters, despite having lower baseline proficiency on some measures, ultimately showed broader and more sophisticated use of evaluative devices, suggests that their longer exposure to the language, even if limited, provided a foundation for more robust implicit learning. This highlights the differential impact of immersion based on a learner’s prior experience and challenges a one-size-fits-all approach to language immersion programs.

Consistent with the Maturational Hypothesis (Lichtman, 2016), early starters benefited more substantially from immersion than late starters. Regression analysis confirmed that earlier age of first exposure predicted greater diversity of evaluative devices, suggesting that a longer history of implicit engagement with English prepared learners to capitalize on immersion experiences. This resonates with the linguistic threshold hypothesis (Leonard & Shea, 2017; A. Y. Long & Solon, 2021; Pérez-Vidal & Juan-Garau, 2011), which proposes that a certain baseline proficiency is necessary to fully benefit from naturalistic input. Early starters, having already internalized foundational forms, were better positioned to expand into evaluative and discourse-level skills.

For practitioners, these findings suggest that short-term immersion can be a powerful supplement to classroom instruction, particularly for learners with early exposure to the L2. At the same time, later starters may require scaffolding—for example, targeted vocabulary support or pre-departure training—to maximize their benefits from immersion. Language learning programs should therefore be designed with learner profiles in mind.

4.3. Limitations and Future Directions

While our results offer indicative evidence for the effectiveness of short-term immersion, they also point to several limitations and directions for future research. A primary limitation is the study’s small sample size and lack of a non-immersion control group, which limits the generalizability of the findings and our ability to isolate the distinctive effects of the homestay from other variables. Furthermore, our participants were recruited from a specific extracurricular program (LABO), which may attract families with higher motivation and resources. Future research could address these limitations by incorporating larger, more diverse cohorts and by using longitudinal designs to track the long-term retention and development of these narrative skills. Future research could also examine:

• Classroom-based interventions that replicate immersion-like input for learners unable to go abroad;
• Cross-linguistic comparisons, investigating whether similar age effects and narrative outcomes appear in learners of languages other than English;
• Mixed-methods approaches, combining quantitative analysis with ethnographic observation of how learners engage in real-life communication during immersion.

5. Conclusions

Despite inevitable limitations, this study makes three main contributions to SLA research. First, it extends prior work on immersion by focusing not only on fluency and accuracy but also on evaluative narrative competence, a higher-order discourse skill often overlooked in SLA studies. Second, by integrating both problem-solving and storytelling tasks, the study captures learners’ ability to engage in academic reasoning as well as narrative evaluation, offering a more holistic view of language development. Third, it highlights how starting age interacts with immersion outcomes, underscoring the importance of tailoring pedagogical approaches to learner backgrounds.

By addressing our two RQs, we discovered two key implications from the results. First, immersion offers a valuable supplement to classroom instruction by providing the implicit, meaning-focused input necessary for developing advanced narrative skills. Second, L2 program designs should be sensitive to learner profiles: early starters may progress rapidly, while later starters may benefit from targeted scaffolding to bridge gaps in vocabulary and discourse strategies.

While limited by the short duration and the absence of a control group, this study contributes new evidence to debates on implicit learning and age effects. Future research should examine the long-term durability of immersion gains and explore how classroom-based interventions might replicate the benefits of naturalistic exposure for learners who cannot go abroad. By connecting age, immersion, and evaluative narrative development, this study contributes a new dimension to SLA theory and practice. We hope our findings provide valuable insights for educators and language-immersion program designers, paving the way for more effective and engaging FL learning experiences.