The acquisition of possessive agreement in L3 German: L2 English proficiency, not L1 Polish, predicts cross-linguistic influence

1 Introduction

The acquisition of possessive agreement has long been recognized as a particularly challenging area for adult second language (L2) learners. Most research has focused on the English third-person singular possessive pronouns his and her, which are especially prone to error. Such errors have been attributed either to transfer of first language (L1) agreement mechanisms (e.g. White et al., 2007; Antón-Méndez, 2011; Jesús-Ortiz and Calvo-Ferrer, 2023) or to gender interference from the possessee noun (Lago et al., 2023). By contrast, possessive agreement has received little attention in third or additional language (L3/Ln) acquisition.

The present study examines cross-linguistic influence (CLI) in acquiring third-person singular possessive pronouns in intermediate to upper-intermediate L3 German among L1 Polish speakers with L2 English. Classroom observations suggest this group struggles with possessive agreement in German, potentially due to structural differences from Polish—a language in which reflexive possessive pronouns remain neutral with respect to the possessor—and this difficulty is also well-documented in L2 research on possessive pronouns (e.g., Antón-Méndez, 2011; White et al., 2007). To investigate this, the study adopts a subtractive language group design, comparing L3 learners with L1 English learners of L2 German (Westergaard et al., 2017), thereby isolating the effects of L1 Polish. As English proficiency is virtually universal among Polish learners, making it impossible to find L1 Polish speakers of L2 German without English, the study also considers the role of L2 English proficiency.

This research contributes to L3 acquisition by addressing the understudied phenomenon of possessive agreement and exploring two key factors underpinning cross-linguistic influence: the role of L1 and the impact of L2 proficiency.

2 Possessive agreement in L2 and L3/Ln acquisition

Variations among languages in how they handle agreement rules pose notable difficulties for L2 learners. A well-documented example involves the English possessive pronouns his and her, which have proven particularly troublesome for French and Spanish learners to master, even after extended exposure to English (e.g., Antón-Méndez, 2011; White et al., 2007).

In a study examining the production of these pronouns by learners of English with different L1s—Italian, Spanish, and Dutch—(Antón-Méndez 2011) observed that learners' native language patterns affected their English usage. Specifically, while both English and Dutch require agreement with the possessor, Spanish and Italian instead align agreement with the possessee. The research found that Italian and Spanish learners were more prone to errors when the possessed noun was animate, often using possessive forms that matched the gender of the possessee rather than the possessor. Antón-Méndez interpreted these results as evidence that learners' L1 processing mechanisms continue to operate automatically during L2 production.

Lago et al. (2019) provided further evidence by examining how bilinguals process L3 German possessive pronouns. Their study involved English-Spanish bilinguals learning L3 German, with either English or Spanish as L1. While German and English encode possessor agreement, Spanish only encodes possessee agreement. The results showed that both L1 and L2 influenced comprehension. L1 effects were stronger in real-time measures (reading times, judgment latencies), with participants' processing patterns aligning more closely with their L1 grammar. In turn, the L2 affected only grammaticality judgment accuracy, suggesting that deeply ingrained L1 routines guide online comprehension, while L2 influence emerge in more explicit, metalinguistic tasks.

(Długosz 2024) examined how native and non-native grammars influence both offline and online comprehension of reflexive possessive pronouns in L3/L4 Swedish among L1 Polish speakers, with or without prior knowledge of non-native German. In a self-paced reading task, learners were sensitive to both possessor and possessee agreement violations regardless of language background. However, in a grammaticality judgment task, they struggled with both types of agreement. L4 learners showed higher judgment accuracy than L3 learners, indicating facilitation from their L3 German knowledge.

To date, only Lago et al. (2019) and (Długosz 2024) have directly investigated possessive agreement in multilingual speakers. Both studies reported facilitative CLI from a previously acquired non-native language on L3/Ln, with some evidence of L1 effects in real-time processing.

3 The role of structural similarity and L2 proficiency in L3/Ln acquisition

Recent research increasingly suggests that structural similarity between the L3/Ln and previously acquired languages plays a crucial role in predicting CLI (e.g., Kolb et al., 2023). This factor has received particular attention within the framework of the Linguistic Proximity Model (LPM) proposed by Westergaard et al. (2017). According to the LPM, third or additional language (L3/Ln) acquisition proceeds incrementally, property by property, and allows for both facilitative and non-facilitative CLI from one or both previously acquired languages as a result of grammar co-activation.

The main driver of CLI is structural similarity – abstract parallels between the L3/Ln and the learner's L1 and/or L2, especially in later acquisition stages. Non-facilitation may arise when the parser misanalyses input due to activation of a lexically similar but structurally dissimilar language, allowing lexical overlap to override morphosyntactic similarity. The LPM also highlights interacting factors such as proficiency, dominance, and complexity.

The primacy of structural similarity is likewise shared by the Scalpel Model (Slabakova, 2017), which introduces further variables such as structure frequency, misleading input, and processing complexity, all of which may modulate CLI. Other L3/Ln acquisition models adopt alternative perspectives, attributing CLI primarily to overall typological proximity (Rothman, 2015) or to cognitive similarity between the L2 and L3 (Falk and Bardel, 2011).

With regard to the impact of L2 proficiency on L3 acquisition, it has been proposed that learners must surpass a certain L2 proficiency threshold for CLI from L2 to occur (Jaensch, 2009). Although this factor remains largely understudied in research on L3 grammar, preliminary evidence supports positive associations between general L2 proficiency and L3 attainment. For example, (Aribaş and Cele 2021) found that higher L2 German proficiency facilitated the acquisition of articles in L3 English (among speakers of L1 Turkish, an articleless language). Similarly, Tasseva-Kurktchieva and Fahey (2025) reported that higher L2 Spanish proficiency was associated with greater accuracy in gender and number agreement in L3 Portuguese (among L1 English speakers).

Beyond general L2 proficiency, some studies have defined this construct through learners' knowledge of specific L2 structures. (Sánchez 2020) found an inverse relationship between L2 syntactic proficiency and CLI: low proficiency in verb-final and inversion structures in L2 German led to overgeneralisation of discontinuous verb placement in L3 English. Eibensteiner (2023) showed that L3 Spanish learners with stronger L2 English aspect knowledge outperformed those with weaker knowledge when the languages shared similar form-meaning pairings.

Taken together, these findings suggest that structural similarity and L2 proficiency jointly modulate CLI in L3 acquisition, an interaction that remains underexplored, particularly in the domain of possessive agreement.

4 Linguistic background

German possessive determiners encode features of the possessor in their stem (e.g., sein, “his” vs. ihr, “her”) and agree with the possessed noun in gender, number, and case via inflection. Unlike some other languages, German does not have a dedicated reflexive possessive form. As a result, third-person possessives (sein, ihr) are ambiguous between subject-oriented (coreferential) and non-coreferential readings (Zifonun, 2005).

This ambiguity in the third-person singular likewise characterizes the English possessive pronouns his and her. Unlike Polish or German, however, English does not display agreement with the possessee noun. All possessive pronouns can be coreferential with the subject, with third-person forms agreeing in natural gender with the possessor and first- and second-person forms agreeing in person rather than gender.

Finally, in Polish, the reflexive possessive pronoun swój is used across all persons. It inflects according to the gender of the possessee noun but remains neutral with respect to the gender and person of the possessor. In the third person, the non-reflexive pronouns jego (“his”) and jej (“her”) cannot refer back to the subject (but see Żychliński et al., 2025). Because Polish permits referential null subjects, determined by subject-verb agreement, the pronominal possessor can be omitted depending on discourse context.

To summarize, in the third person singular, the focus of this study, the possessive pronoun agrees in natural gender with the possessor in both German and English, but not in Polish (see Table 1).

Table 1. Possessive marking in German, English, and Polish (third person singular).

In German and English, third-person possessives are ambiguous between coreferential and non-coreferential readings.

In Polish, the reflexive possessive swój enforces coreference with the subject.

1, possessor features; 2, agreement with possessed noun; Polish forms encode only (2), English only (1), and German both.

5 The study

The aim of this study is to examine the role of L1 Polish and of L2 English proficiency in the acquisition of possessive agreement in L3 German. The primary research questions are as follows:

• Are L3 German learners' difficulties with possessive agreement attributable to CLI from L1 Polish, which exhibits structural dissimilarity?
• Does L2 English proficiency predict the acquisition of possessive agreement in L3 German? Given that difficulties in producing possessive pronouns may stem from gender mismatches between the possessor and the possessee noun (Lago et al., 2023), the study also addresses the following secondary research question:
• Do L2/L3 German learners experience greater difficulty with possessive agreement when the possessor and possessee nouns differ in gender?

5.1 Participants

Participants were 66 university students majoring in German, 39 in Poland (M_age = 20.6; 25 females, nine males, and five non-binary individuals) and 27 in the United States (M_age = 19.9; 15 females, 11 males, and one non-binary individual). For the United States group, the initial sample comprised 46 participants; individuals reporting prior knowledge of Spanish or another Romance language were excluded to ensure an L2 (rather than L3) German group. All participants were first exposed to German after the age of six. Participants in Poland (henceforth, L3 learners) had Polish as their L1, English as L2, and German as L3, while participants in the United States (henceforth, L2 learners) had English as L1 and German as L2. All L3 learners were exposed to English at an earlier age than to German (M = 6.2, SD = 2.4; M = 12.5, SD = 3.6, respectively).

Participants' proficiency in German ranged from A2 to B2 (according to CEFR) and was assessed using the LexTALE test (Lemhöfer and Broersma, 2012: 329). The mean score was 59.2 (range: 43.8–82.5) for L3 learners and 58.9 (range: 45.0–86.3) for L2 learners, with the difference between the two groups being statistically non-significant (p = 0.561, Mann–Whitney U-test). Self-reported proficiency in English in the L3 group was 8.1 on a 1–10 scale (range: 3–10). In the L3 group, German and English proficiency were largely independent, r₍₃₇₎ = −0.09, p = 0.576.

5.2 Materials

The task included 16 sentences categorized into two groups by gender (mis-)match between the possessor and the possessee: eight congruent, for example, Frau Müller *(feminine)* begrüßte ihre Nachbarin *(feminine)* (“Mrs Müller greeted her neighbour”), and eight incongruent, for example, Frau Schäfer *(feminine)* vermisste ihren Enkel *(masculine)* (“Mrs Schäfer missed her grandson”). Within each category, half of the possessors were masculine, while the other half were feminine. The possessee nouns were simple lexical items referring to family members or professions.

The items were equally divided into felicitous and infelicitous conditions (see Lago et al., 2019). Felicitousness was manipulated across two lists (A and B) using a Latin square design, such that each item appeared in only one condition per list but in both conditions across the full set of items. The presentation order within each list was randomized for each participant. Infelicitous items were created by manipulating gender agreement between the possessor and pronoun, for example, #Frau Bauer *(feminine)* warnte seinen *(masculine)* Sohn (“Mrs Bauer warned his son”). The pronoun was always congruent with the possessee. The task also included 79 unrelated fillers intended for a separate study. Table 2 presents example items by felicitousness and gender match.

Table 2. Example items by felicitousness and gender match between the possessor and the possessee.

5.3 Procedure

The entire experiment was designed and conducted using the Qualtrics platform, with the link shared at participants' universities. Participation was voluntary, and each participant completed the study individually during their free time. All instructions were provided in English. Participants were randomly assigned to either list A or list B.

The acceptability judgment task began with written instructions outlining the procedure. Participants were informed that they would be presented with a series of sentences in German and asked to judge whether each sentence was acceptable or unacceptable. Acceptability reflected conformity to the default local agreement interpretation (see Lago et al., 2019). The task began with four unrelated practice items. Participants read the items one at a time, with no time limit imposed. After making their judgments, they were asked to correct the items marked as unacceptable by clicking on individual words.

5.4 Results

For infelicitous items, participants had to both reject the sentence and mark the offending element (the possessor or pronoun). Marking one correct element earned a score of 1. If they rejected an item but marked an irrelevant element (e.g., the verb), they scored 0. For felicitous items, participants scored 0 if they rejected the sentence and marked the possessor or pronoun, but 1 if they marked another element, indicating their judgment was unrelated to possessive agreement. Marking was required to ensure that judgments were tied to possessive agreement rather than to unrelated aspects of the sentence. One response was excluded because the participant marked all elements, making their intent unclear.

A logistic mixed-effects model was fit to predict participants' accuracy in Jamovi (version 2.4) with the GAMLj module (Gallucci, 2019). Fixed effects included Felicitousness, Gender match, L2/L3 German proficiency (LexTALE scores), Group, the interaction between Group and Felicitousness, and Possessor gender for control. A random intercept and random slope for Felicitousness were specified by participant: Accuracy ~ Felicitousness + Gender match + Possessor gender + L2/L3 German proficiency + Group + Felicitousness:Group + (1 + Felicitousness | Participant).

Continuous predictors were standardized, and categorical variables were coded using treatment (dummy) coding. The model used a binomial distribution with a logit link and was estimated with the bobyqa optimizer. Model fit was acceptable (AIC = 897.82; BIC = 947.42; marginal R² = 0.39; conditional R² = 0.64). The model converged successfully and showed no overdispersion (χ²/df = 0.64). A by-item random intercept was tested but excluded because it produced a singular fit (zero variance). An inspection of by-item accuracy indicated relatively uniform performance across stimuli (range: 0.59–0.77), with no evidence that the results were driven by a subset of items. Detailed parameter estimates for fixed effects are available in the OSF repository.

Omnibus tests of fixed effects showed significant effects of Felicitousness (χ² = 68.58, p < 0.001) and L2/L3 German proficiency (χ² = 8.10, p = 0.004). The estimated marginal means for L2/L3 German proficiency indicated moderate to high accuracy levels, ranging from 0.69 (−1 SD) to 0.86 (+1 SD), with a mean of 0.79 (SE = 0.04, 95% CI [0.71, 0.85]). Effects of Gender match (χ² = 0.47, p = 0.493), Possessor gender (χ² = 0.06, p = 0.812), and Group (χ² = 1.16, p = 0.282) were not significant, nor was the interaction Felicitousness × Group (χ² = 0.12, p = 0.731). Figure 1 plots model-predicted accuracy by Felicitousness and L1.

To explore the role of self-reported L2 English proficiency within the L3 group, an additional model was fit to the data from L3 learners, keeping other fixed and random effects constant. Omnibus tests of fixed effects revealed a significant main effect of Felicitousness (χ² = 44.82, p < 0.001) and L2 English proficiency (χ² = 5.26, p = 0.022), with higher proficiency associated with greater accuracy. Estimated L2 English proficiency scores ranged from 0.72 (1 SD below the mean) to 0.89 (1 SD above the mean), with a mean estimated proficiency of 0.82 (SE = 0.04, 95% CI [0.73, 0.89]). Effects of L2/L3 German proficiency (χ² = 1.67, p = 0.197), Gender match (χ² = 0.08, p = 0.782), and Possessor gender (χ² = 0.45, p = 0.502) were not significant. Figure 2 plots accuracy by L2 English proficiency.

In summary, the L2 and L3 groups showed no differences in judgment accuracy and performed similarly regardless of gender matching between possessor and possessee. Participants found rejecting infelicitous items harder than accepting felicitous ones, and overall accuracy improved with higher L2/L3 German proficiency. Moreover, L3 learners' performance in German increased with greater L2 English proficiency.

6 Discussion

The starting point for this study was the observation that learners of German often experience difficulties with possessive agreement, as evidenced by classroom practice and L2 research on possessive pronouns (e.g., Antón-Méndez, 2011; White et al., 2007). The study did not detect a statistically reliable difference between proficiency-matched L1 Polish-L2 English learners of L3 German and L1 English learners of L2 German. This suggests that the acquisition of possessive agreement is inherently challenging, and that L1 Polish may not be responsible for learners' errors with possessive pronouns. Accordingly, structural dissimilarities between L1 Polish and L3 German do not appear to play a significant role in the learning process, contrary to the findings reported by Lago et al. (2019) and (Długosz 2024). It should be noted, however, that the present study employed an untimed judgement task, and L1 effects may still emerge in online L3 processing.

At the same time, the present results revealed that increasing L2 English proficiency exerted a facilitative influence on the acquisition of possessive agreement in L3 German, aligning with previous studies reporting similar associations (Aribaş and Cele, 2021; Tasseva-Kurktchieva and Fahey, 2025). Importantly, in the L3 group, L2 English and L3 German proficiency were not correlated, allowing the effect of L2 proficiency to be isolated. At lower proficiency levels, participants performed less accurately and displayed greater variability than those at higher levels, who showed increased accuracy and consistency (see Figure 2). This finding is interpreted within the framework of the Linguistic Proximity Model (Westergaard et al., 2017), as evidence of facilitative CLI from L2 English arising from structural similarities between English and German. Arguably, stronger L2 representations are more easily co-activated, thereby enhancing the likelihood of facilitative CLI where structural similarity between L2 and L3 exists. In addition, the present findings indicate a linear relationship between L2 proficiency and L3 attainment, rather than evidence for a specific proficiency threshold.

Finally, participants performed similarly regardless of gender mismatch between the possessor and the possessee noun, suggesting that potential processing costs associated with such mismatches do not necessarily manifest in offline judgement tasks.

Although participants' performance improved with increasing L2/L3 German proficiency, suggesting a stepwise convergence on the target, overall accuracy for infelicitous items remained below chance. This indicates that the acquisition of possessive agreement is inherently challenging in adult language learning. One possible explanation lies in the feature complexity of third-person singular possessive pronouns, which require semantic gender agreement with the possessor, as well as grammatical gender, case, and number agreement with the possessee noun. Alternatively, when binding occurs [e.g., Frau Müller *(feminine)* besuchte ihre *(feminine)* Schwester, (“Mrs Müller visited her sister”), the pronoun's gender feature must correspond to that of its antecedent. Learners' representations may therefore contain underspecified feature values, leading them to interpret the pronoun as a general third-person singular possessive.

Finally, the findings can also be interpreted within the framework of the Shallow Structure Hypothesis (Clahsen and Felser, 2006), which posits that adult language learners rely more heavily on lexical-semantic cues during parsing and less on syntactic information. Indeed, the infelicitous items used in this, albeit offline, study can be processed without the need for gender-feature checking, consistent with the predictions of this hypothesis.

The study provides new evidence that when a grammatical property does not pattern identically in L1 and L3, its acquisition in L3 is not necessarily hindered. Specifically, the findings do not support the claim that the difficulties experienced by L1 Polish learners of German with possessive agreement stem from structural differences between the two languages. Instead, the source of errors in the use of possessive pronouns appears to lie elsewhere.

An important contribution of this study concerns the role of L2 proficiency. The findings revealed a positive relationship between general L2 English proficiency and L3 German attainment. Future research should examine grammatical properties that do not align between L2 and L3 to determine whether structural similarity interacts with L2 proficiency, or whether the influence of L2 proficiency is more general in nature.

The present findings should be interpreted with caution. While the total sample (n = 66) is broadly comparable to previous work on possessive agreement (e.g., Lago et al., 2019: n = 65; Długosz, 2024: n = 30), the relatively small number of items per condition may have limited statistical power and increased susceptibility to response variability, particularly in an offline accuracy paradigm. The use of an offline measure may have further reduced sensitivity to subtle processing differences relative to studies employing online methods. Importantly, the descriptive pattern does not align with the predicted direction, as the L3 group did not show lower accuracy than the L2 group. Taken together, these findings do not provide evidence for the predicted negative effect of L1 Polish on L3 German. At the same time, given the absence of statistically reliable effects and the limited sensitivity of the measure, they should not be taken as strong evidence for an advantage of the L1 Polish group. In addition, L2 English proficiency was assessed via self-report, which may have introduced additional variability. Future studies should combine more sensitive online measures with objective proficiency assessments.

Generative AI statement

The author(s) declared that generative AI was used in the creation of this manuscript. The author(s) declared the use of the generative AI tool GPT-4o (OpenAI) for English language editing only.

Any alternative text (alt text) provided alongside figures in this article has been generated by Frontiers with the support of artificial intelligence and reasonable efforts have been made to ensure accuracy, including review by the authors wherever possible. If you identify any issues, please contact us.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Footnotes

Edited by: John Matthews, Chuo University, Japan

Reviewed by: Panagiotis Kenanidis, University of Erlangen Nuremberg, Germany

Nadja Wulff, University of Education Freiburg, Germany