The significant performance gap between native (L1) and foreign language (L2) acquisition has led to two prominent hypotheses: the Fundamental Difference Hypothesis (FDH), suggesting fundamentally different learning mechanisms, and the Linguistic Coding Deficit/Differences Hypothesis (LCDH), proposing shared core language functions. This study challenges both hypotheses by introducing the Learning Condition Hypothesis (LCH), proposing that learning conditions (naturalistic vs. classroom) are a key factor. The researchers examined a large sample of 636 Chinese-descent undergraduates who learned Chinese (L1), English (L2), and one of three L3s (French, German, or Spanish). This sample allowed for an examination of language learning across a continuum of learning conditions, from naturalistic L1 acquisition to formal classroom instruction for L2 and L3. The study aimed to evaluate the FDH, LCDH, and LCH by examining the association between proficiency levels in these three languages, considering various non-language factors that might contribute to language learning success.

Existing research supporting the FDH highlights the large performance gap between L1 and L2, often attributed to a critical period for language acquisition. Studies supporting the LCDH demonstrate correlations between L1 and L2 performance, implying shared core language functions. However, these studies often focused on typologically similar languages and specific metalinguistic tasks, potentially limiting the generalizability of their findings. This study addresses these limitations by examining a wider range of languages and using more comprehensive proficiency measures, thus offering a more rigorous test of the existing hypotheses. The study introduces the LCH as an alternative explanation, positing that the learning context significantly impacts language proficiency.

The study enrolled 636 undergraduate students of Chinese descent who learned Chinese (L1), English (L2), and one of three L3s (French, German, or Spanish). L1 and L2 proficiency were assessed using Hong Kong Diploma of Secondary Education (HKDSE) scores. L3 proficiency was measured using a combination of classroom performance and laboratory-based assessments, including narrative production, lexical access, and pronunciation judgments. These measures were combined to obtain an L3 Global score. Non-language factors such as musical background, socioeconomic status (SES), nonverbal IQ, age, gender, anxiety, and motivational factors were also collected. Data were analyzed using general linear models (GLM), structural equation models (SEM), and support vector regression (SVR) machine learning to determine the association between L1, L2, and L3 proficiency levels while controlling for non-language factors. The methodology included detailed assessments of L3 proficiency using various laboratory and classroom-based measures (narrative production, lexical access, pronunciation ratings, and classroom exam scores), followed by principal component analysis (PCA) to derive an L3 Global score. The researchers also utilized the Modern Language Learner Questionnaire to assess motivational and affective factors influencing L3 learning.

Bivariate correlations revealed a significant positive association between L1 and L2 proficiency and between L2 and L3 proficiency, but not between L1 and L3. Multiple linear regression models confirmed these associations, even after controlling for non-language factors. SEM analysis showed significant paths from L1 to L2 and from L2 to L3, but not from L1 to L3. The LCH model demonstrated the best fit compared to models representing FDH and LCDH. Machine learning (SVR) analysis further supported the findings, showing that L2 proficiency was a stronger predictor of L3 proficiency than L1 proficiency. The analysis of separate models for low and high proficiency learners, and across different L3 languages, did not alter the core finding supporting the LCH. Importantly, the study showed that the association between L2 and L3 remained significant even with different assessment methods used for L2 and L3, suggesting that the associations were not simply artifacts of measurement similarities.

The findings strongly support the LCH, demonstrating that similar learning conditions between languages contribute to stronger associations in learning success. The significant association between L1 and L2, acquired under relatively similar conditions (though one is more naturalistic than other), and between L2 and L3, both learned in a classroom setting, further strengthens this conclusion. The absence of an association between L1 and L3, acquired under significantly different learning conditions, underscores the importance of learning context. The study's results are consistent with research on experience-related neural adaptation in bilingualism, suggesting that early language learning may establish neural pathways that later influence learning under more formal settings. The results do not support a simple critical period account of language learning since the L2-L3 association remains despite the substantial age difference between their acquisition. Although the study found certain associations that might be interpreted through existing theories like the Typological Primacy Model (TPM), Cumulative Enhancement Model (CEM), L2 Status Factor, Dynamic Model of Multilingualism (DMM), Revised Hierarchical Model (RHM), and Linguistic Proximity Model, the core findings support the primacy of learning conditions over other theoretical factors.

This study provides compelling evidence for the Learning Condition Hypothesis, showing that learning contexts significantly influence the relationship between proficiency levels in L1, L2, and L3. The strong association between L2 and L3 proficiency, even after controlling for numerous factors, highlights the importance of considering learning conditions when examining language learning. Future research should explore a wider range of non-language variables and investigate diverse learning populations, such as heritage speakers and immigrant learners, to enhance the generalizability of these findings. Further, systematic investigation of age of acquisition as a confounding variable is needed.

While the study included several non-language variables, some factors like different types of memory systems (procedural/declarative, working memory), language aptitude, and genetic variations were not considered and could impact the results. The study primarily focused on learners who acquired L2 and L3 in formal classroom settings, limiting the generalizability to learners with different learning experiences. The potential confounding effect of age of acquisition also needs further investigation in future studies.