The effect of L2 German on grammatical gender access in L1 Polish: proficiency matters

Previous research has shown that bilinguals process nouns that have the same grammatical gender in their two languages faster than nouns that differ in gender between L1 and L2. This finding, referred to as the gender congruency effect, has so far only been documented in L2. Hence, the aim of the present study was to examine whether late unbalanced bilinguals would also show gender congruency effects in their L1. To that end, 44 L1 Polish/L2 German bilinguals were tested in a gender decision task in Polish, which included gendercongruent and gender-incongruent nouns. The results revealed a robust gender congruency effect in L1, which was limited to bilinguals with very high L2 proficiency. This indicates that bilinguals activate grammatical gender information in L2 when accessing gender in L1, providing that they are highly proficient in L2. More broadly, the study demonstrates that foreign language knowledge can affect native language performance in exclusively native contexts. Finally, as a first attempt to examine grammatical gender access in Polish, this study shows that feminine gender is accessed faster compared to masculine and neuter, suggesting that the ending -a is the most reliable gender cue in Polish.


Introduction
A key finding of recent research on the bilingual lexicon, and on bilingualism more generally, is that both languages are always active, even when only one language is required. Much of the discussion pertains to lexical access, that is, the process of entering the lexicon to retrieve information about words, which is said to be nonselective in nature (for discussion, see Libben and Goral 2015). Non-selective lexical access implies that when a bilingual produces or comprehends a word, the translation equivalent of this word in the other language becomes active, and certain of its linguistic properties affect production and comprehension in the intended language. For example, it is now well established that cognateswords that share form and meaning across languagesare accessed in the lexicon faster than non-cognates (e.g., Szubko-Sitarek 2011). However, research on the bilingual lexicon increasingly goes beyond form and meaning and focuses on lexical-syntactic information related to words, such as grammatical gender.
A substantial number of studies have demonstrated that the gender values of the native language (L1) interact with those of the L2 in language processing (e.g., Salamoura and Williams 2007;Bordag and Pechmann 2008;Lemhöfer et al. 2008;Paolieri et al. 2010;Morales et al. 2011;Klassen 2016a;Paolieri et al. 2020; for an extensive overview, see Sá-Leite et al. 2019). To illustrate, when bilinguals are asked to describe a picture in their L2, and the object depicted has the same gender value in the L1 and L2, they describe it more accurately (as reflected in error rates) and more quickly (as reflected in response time; e.g., Bordag and Pechmann 2008, for Czech-German bilinguals;Lemhöfer et al. 2008, for German-Dutch bilinguals; Klassen 2016a, for Spanish-German bilinguals).
Although the interactions between the two gender systems of a bilingual have been studied very extensively, research thus far has analysed learner performance exclusively in L2, thereby ignoring the possibility of an interaction in the reverse direction. Therefore, the issue of whether the weaker L2 may affect gender processing in the dominant L1 has not yet been elucidated. The present study was designed to bridge this research gap by addressing the following research question: do gender values learned in L2 affect gender access in L1? Or more broadly: does foreign language knowledge affect grammatical gender processing in the dominant native language?
The paper is organized as follows. Section 2 gives an overview of previous research on gender access in bilinguals. Section 3 introduces gender assignment principles in Polish and German. The body of the study is presented in Section 4, and Section 5 discusses the findings.

Gender access in bilinguals
In the mental lexicon, grammatical gender is represented either as part of each noun entry at the lemma level (Carroll 1989) or as a gender node to which all nouns of a given gender are linked (Schriefers and Jescheniak 1999). There is ample research to suggest that the gender value of a target noun becomes automatically activated for processing when the noun is accessed from the mental lexicon, irrespective of the presence of an agreement context (e.g., Alario et al. 2008;De Martino et al. 2011).
Research on grammatical gender access during comprehension has mainly focused on the role of the morphological transparency of nouns, that is, on statistical regularities between noun endings and their gender. In that vein, Gollan and Frost (2001) have proposed two distinct mechanisms for accessing grammatical gender. The first mechanism derives gender from its correlation with the noun ending (the form-based route to gender), while the second mechanism incorporates an abstract representation of grammatical gender without any involvement of gender-correlated form markers (the lexical route to gender). For example, in the Polish sentence lampa jest duża 'the lamp FEM is big', the noun lampa has the ending -a, which is strongly associated with the feminine gender and provides an important cue facilitating gender access. In contrast, in the Polish sentence kiwi jest kwaśne 'the kiwi fruit NEUT is sour', the noun kiwi does not provide any formal cues to gender. Research has shown that participants are usually faster and more accurate when deciding the gender of transparent nouns than nouns whose endings are uninformative of gender (Hernandez et al. 2004;Spalek et al. 2008).
Although the research reviewed above was concerned with gender access in native speakers, a large body of studies on gender processing in bilinguals has arrived at similar conclusions. Importantly, the obligatory access to grammatical gender information during bare noun processing has also been observed in adult L2 learners (Paolieri et al. 2019(Paolieri et al. , 2020. In research on gender processing in bilinguals, much interest has centred on interactions between the two gender systems. The common approach has been to analyse learner performance in L2 regarding gender-congruent nounsthat is, nouns whose translation equivalents in L1 share the same gender valuein comparison to gender-incongruent nounsthat is, nouns whose translation equivalents in L1 have a different gender. The vast majority of studies adopting this approach have observed shorter reaction times and higher accuracy in response to gender-congruent nouns both in production and comprehension, a finding commonly referred to as the "gender congruency effect" (for an extensive overview, see Sá-Leite et al. 2019).
Studies on gender processing in bilingual speakers inform theories and models of grammatical gender representation in the bilingual mind. Observation of gender congruency effects in L2 has led scholars to propose the "gender-integrated representation hypothesis", which posits that L1 and L2 have a shared grammatical gender system (Klassen 2016a;Salamoura and Williams 2007). Under this view, lexical items that have the same gender value activate the shared gender node. During processing, a noun in L2 receives additional activation from the shared gender node activated by both L1 and L2 lexical items. This boosted activation of the same gender node facilitates the response in L2. If lexical items do not match in gender, they activate different nodes, but these nodes are still common to both languages. In this case, the shared gender node inhibits the response because lexical items in L1 and L2 activate different gender nodes which compete for selection and impede the response. If a gender value only exists in L2, the response is less inhibited (for discussion, see Klassen 2016b). Although the gender-integrated representation hypothesis predicts interference from L1 to L2 and vice versa, the reverse direction, from L2 to L1, has not yet been explored.
However, studies looking at how bilinguals process lexical features other than gender in L1 have revealed some effects of the weaker L2 (e.g., Bice and Kroll 2015;Brenders et al. 2011;Linck et al. 2009;Malt et al. 2015;Poarch and van Hell 2012;Szubko-Sitarek 2011;van Hell and Dijkstra 2002). Importantly, these effects are contingent on L2 proficiency. As Schwartz (2015: 345) puts it: "When processing is in the dominant language, greater proficiency in the non-target language is associated with greater effects of non-selectivity."

Gender assignment in Polish and in German
Polish distinguishes between masculine, feminine, and neuter. The three gender values are not equally distributed. According to Stefańczyk (2007: 48), masculine includes approximately 50 % of all Polish nouns and is thus the most frequent gender category, followed by feminine with a frequency of approximately 40 %. Neuter is least frequent in Polish as it only covers 10 % of nouns. Nouns ending in a consonant are masculine (e.g., stół 'table'), nouns ending in -a and -ść are feminine (e.g., lampa 'lamp'), and nouns ending in -o and -e/ę are neuter (e.g., oko 'eye'). These gender-to-ending regularities apply to the vast majority of Polish nouns, meaning that grammatical gender assignment is highly predictable based on the morphological shape of the noun in the nominative singular. Semantic factors, such as natural gender, also contribute to gender assignment, but the morphophonological cues play a crucial role in this process in Polish.
Like Polish, German classifies nouns into masculine, feminine, and neuter. According to Bauch (1971), 50 % of German nouns are masculine, 30 % are feminine, and 20 % are neuter. Gender assignment is assumed to be arbitrary to a large extent, with some semantic and morphological regularities which primarily apply to monosyllabic nouns (Köpcke 1982;Köpcke and Zubin 1996). These regularities differ with respect to their reliability. Most of them must be considered probabilistic, and only a small number of them, such as suffix regularities, are deterministic in nature (Hohlfeld 2006: 129).

The study
The aim of the study was to examine to what extent the weaker L2 affects gender access in the dominant L1, depending on proficiency in the L2. Given the previous studies showing L1 effects on gender access in L2 on the one hand, and L2 effects on lexical processing in L1 on the other, it is predicted that, in this study, L2 will have an effect on gender access in L1. This effect, however, is expected to be more pronounced at higher L2 proficiency levels, as previously evidenced by research on cross-language activation in bilinguals (e.g., Poarch and van Hell 2012;van Hell and Dijkstra 2002).

Participants
Two groups of adult Polish-German bilinguals participated. The first group included 20 speakers who were highly proficient in German (C2 level according to the Common European Framework of Reference for Languages). They were between 28 and 38 years old (M = 33), and 16 were female. All of them were university teachers of German at a Polish university. They had studied German philology in Poland and completed at least one term at a German university as exchange students in the past. The participants occasionally used English, predominantly in reading. Their average performance in the German version of LexTALE (Lemhöfer and Broersma 2012) was 90.5 % (SD = 7.3). In what follows, these participants are labelled "speakers" rather than "learners" due to their very high proficiency in German and the fact that they used German on a daily basis in the classroom and in their research.
The second group consisted of 24 learners who were less proficient in German (upper-intermediate level according to the Common European Framework of Reference for Languages). They were between 19 and 22 years The effect of L2 German on grammatical gender access in L1 Polish old (M = 20), and 17 were female. All of them were students of German and English and had completed three terms of Applied Linguistics at a Polish university. In the German LexTALE, they scored a mean of 71.5 % (SD = 12.7).
All participants acquired German after the age of 7 (M = 10.4, SD = 2.12). They knew English to different levels of proficiency (although this fact does not play a role in this study because English does not encode grammatical gender). The participants were not informed about the purpose of the study. Therefore, they were tested in a purely native language context.

Materials
This experiment used a gender decision task, which requires explicit retrieval of gender information from the lexicon (e.g., Bobb et al. 2015;Radeau and Van Berkum 1996). In gender decision, the participant's task is to indicate the grammatical gender of a noun. Sixty Polish non-cognate nouns were selected for the experiment based on their German translations available on the Goethe-Zertifikat A1 list (Goethe-Institut 2004; 53 nouns) and the Goethe-Zertifikat A2 list (Goethe-Institut 2016; seven nouns)the intent was to use nouns whose German translations are very common and thus most likely known by the participants. Six conditions with 10 nouns each were created by combining the factors of gender and congruency with German. Gender-congruent nouns had the same gender in Polish as their German translations (e.g., list MASC and Brief MASC 'letter'), while gender-incongruent nouns had a different gender in Polish from their German translations (e.g., czas MASC and Zeit FEM 'time').
All feminine and neuter nouns had endings which were predictive with respect to grammatical gender in Polish (i.e., -a for feminine and -o or -e/ę for neuter), with one exception for feminine gender (pomoc 'help'). All masculine nouns ended in a consonant, which is a reliable gender cue in Polish. All nouns were inanimate, with one exception for feminine gender (ryba 'fish'). All nouns are included in Appendix A.
The six conditions were matched as closely as possible on number of letters, similarity to German, frequency in Polish, and the log of the frequency in German. Polish frequency was determined based on the National Corpus of Polish (https://nkjp.pl) using the PELCRA search engine. German log frequency was imported from the CELEX database (Baayen et al. 1995). Formal similarity was measured using an AWSM tool (https://awsm-tools.com) which calculates text resemblance based on Levenshtein distance and length of source/target nouns. The characteristics of the nouns are summarized in Table 1. A two-way ANOVA showed no effect of gender or congruency, and no significant interaction between them (see Appendix B). This means that the six conditions were perfectly matched in terms of factors that may potentially influence response time in the gender decision task.

Procedure
The testing session included a gender decision task in Polish, the LexTALE in German (Lemhöfer and Broersma 2012), and a background questionnaire to gather information on the participants' demographics. The participants were tested individually and received gift cards for a bookstore for their participation. The procedure for the gender decision task was as follows. The participants were first presented with a fixation point (+) for 500 ms on a computer screen. Then, they were visually presented with a Polish bare noun (e.g., dom 'house') for 2000 ms or until a key-press response was registered. The participants were to make a decision about the gender of the target noun and press one of the response keys labelled "M" (masculine), "F" (feminine), or "N" (neuter) as quickly and accurately as possible. The inter-trial interval was 1,000 ms. The task began with written instruction in Polish explaining the experimental procedure and a practice block. The order of the items was randomized across participants. Stimuli presentation and data recording were controlled by PsyToolkit, a web-based software for programming and running reaction-time experiments (Stoet 2010(Stoet , 2017.

Data analysis and results
Analyses of variance were run on response time, on both participant and item means, with the factors of gender (masculine, feminine, neuter) and congruency (congruent, incongruent). In the analysis of participants, the factors were repeated-measures factors. For the items, a two-way analysis of variance was conducted. The Bonferroni test was used as a post hoc test.
Responses that fell outside the 2 s response window were excluded from the analyses, as were responses that were more than two standard deviations away from the participant and item mean. This led to the exclusion of 5.3 % of the data of the high-proficiency speakers and 3.9 % of the data of the upper-intermediate learners. Error rates were not analysed because accuracy scores were always over 97 %.

High-proficiency speakers of L2 German
In order to test the extent to which gender access in L1 Polish was affected by L2 gender values at very advanced levels of L2 proficiency, first gender decision latencies in the group of high-proficiency speakers were considered. Table 2 shows their mean response times.
First, the results of the analysis by participants are reported. The multivariate Pillai's trace test showed main effects of gender and congruency, but no interaction between them. The participants responded faster to feminine (M = 844.61 ms) than to masculine (M = 894.02 ms; p = 0.014) and neuter nouns (M = 887.10 ms; p = 0.027). There was no difference between masculine and neuter nouns (p = 1.000). Response times for gender-congruent nouns (M = 838.09 ms) were shorter than for gender-incongruent nouns (M = 912.40 ms; p < 0.001). The results by participant are graphically depicted in Figure 1. The analysis by items confirmed the main effects of gender and congruency. Response times for feminine nouns (M = 827.65 ms) were shorter than for masculine (M = 895.75 ms; p = 0.002) and neuter nouns (M = 905.28 ms; p < 0.001). Response times for gender-congruent nouns (M = 858.26 ms) were shorter than for gender-incongruent nouns (M = 894.20 ms; p = 0.024). No interaction between gender and congruency was found in the analysis by items. The results of the analyses are reported in Table 3.

Upper-intermediate learners of L2 German
In order to test the extent to which gender access in L1 Polish was affected by L2 gender values at lower levels of L2 proficiency, gender decision latencies in the group of upper-intermediate learners should be considered. Table 4 shows their mean response times.   First, the results of the analysis by participants are reported. The multivariate Pillai's trace test showed a main effect of gender, but no effect of congruency and no interaction between gender and congruency. The participants responded faster to feminine (M = 802.17 ms) than to masculine (M = 968.77 ms; p < 0.001) and neuter nouns (M = 971.33 ms; p < 0.001). There was no difference between masculine and neuter nouns (p = 1.000). The results by participants are illustrated in Figure 2.
The analysis by items confirmed the main effect of gender. Response times for feminine nouns (M = 796.75 ms) were shorter than for masculine (M = 954.52 ms; p < 0.001) and neuter nouns (M = 959.74 ms; p < 0.001). The effect of congruency was not significant, nor was the interaction between gender and congruency. The results of the analyses are reported in Table 5.

Discussion
The results reveal a gender congruency effect in the L2-to-L1 direction, but only in the group of high-proficiency speakers of L2 German, who responded faster to gender-congruent than to gender-incongruent nouns. This effect was observed both in the analysis by participants (74.3 ms) and by items (−35.9 ms). Based on the partial eta squared value (η p 2 ) of 0.60, the effect size can be considered very large (Cohen 1988). The observed gender congruency effect could have been facilitated by the symmetric nature of the Polish and German gender systems. However, interactions between two gender systems have also been attested in language pairs with asymmetric gender systems (e.g., Klassen 2016a). The lack of any noticeable gender congruency effect in the upper-intermediate learners indicates that L2 proficiency plays a crucial role in predicting the influence of the L2 on gender access in the L1. Note that the upper-intermediate learners scored a mean of 71.5 % (SD = 12.7) in the German LexTALE, which corresponds to   (Lemhöfer and Broersma 2012). Moreover, they had completed three terms of applied linguistics at a Polish university, where they were intensively exposed to German on a daily basis. Clearly, this was not sufficient for L2 influence to occur. Given the difference between the high-proficiency speakers and upper-intermediate learners, the study lends further support to the hypothesis that L2 proficiency is an important modulator of cross-language activation (Schwartz 2015).
To the best of the author's knowledge, this study is the first to examine the gender congruency effect in the L1. By flipping the direction in which the cross-language activation was considered in previous research, this study adds new evidence to the debate on the non-selectivity of the language processing system of bilinguals. The results demonstrate that processing the dominant L1 may be non-selective with respect to the L2. Therefore, the study confirms previous findings showing non-selectivity in processing certain lexical features in the L1, but it also extends the non-selective access hypothesis to grammatical gender.
The results of the study are in line with the gender-integrated representation hypothesis, which posits that the L1 and L2 have a shared grammatical gender system (Klassen 2016a;Salamoura and Williams 2007). Note that a shared system of linguistic representations would imply not only L1-to-L2 but also L2-to-L1 interactions, which have not been addressed in previous research on grammatical gender. Therefore, the study is novel in providing evidence for a shared grammatical system based on the data from L1 processing.
An important aspect of this study is that the effect of L2 German on gender access in L1 Polish was observed in a purely native language context. The participants in this study were not informed about the purpose of the research, so they were in monolingual mode when deciding the gender of the nouns in their L1 Polish. This provides further evidence for the theoretical position that the language processing system of bilinguals is nonselective in nature (for discussion, see Libben and Goral 2015). However, a high level of L2 proficiency seems to be required in order for cross-language activation to occur.
The gender congruency effect observed in this study (and many others of this kind) could be a result of L2 instruction. Since all participants had learned German in instructed settings, they had metalinguistic knowledge of gender. Crucially, they knew that gender is a classification system for nouns. There is evidence that heritage speakers who learned gender in naturalistic settings do not show any gender congruency effects (Di Pisa and Marinis 2021). To address the relationship between language instruction and gender congruency effects, future studies should involve naturalistic L2 learners.
These results may also be considered in the context of gender access in Polish, which has not yet been addressed in the literature. The results suggest that retrieval of feminine gender is less demanding, compared to retrieval of masculine and neuter gender, as evidenced by shorter response times. Importantly, faster gender decisions in response to feminine nouns were observed in both participant groups, meaning that they reflect a general pattern of gender access in native Polish. Therefore, it is plausible to assume that the ending -a constitutes the strongest gender indicator which Polish native speakers can depend on when assigning gender to nouns, arguably because it is the only noun ending that shows high availability (i.e., how often a specific cue is available in the lexicon) and reliability (i.e., how often a specific cue is associated with a given gender class) in parallel (Długosz in press). At the same time, the lack of differences between masculine and neuter nouns with respect to gender decision latencies suggests that the endings -o and -e/ę associated with neuter gender and consonant endings associated with masculine gender are equally predictive of gender in Polish. The patterns observed are thus in accord with models that assume a form-based route to gender (e.g., Gollan and Frost 2001).
It was the aim of the study to make a first step towards a better understanding of L2 effects on grammatical gender processing in L1 by using a gender decision task. Nevertheless, future research should undoubtedly explore this topic using other tasks to test whether the non-selectivity of gender access also holds true for contexts that do not require explicit retrieval of gender information from the lexicon. The effect of L2 German on grammatical gender access in L1 Polish