Hypotheses

Various accounts have been put forward attempting to explain observed differences between L1 and L2 language processing, such as the interface hypothesis (Sorace and Filiaci, Reference Sorace and Filiaci2006; Sorace, Reference Sorace2011) or the shallow structure hypothesis (Clahsen and Felser, Reference Clahsen and Felser2006, Reference Clahsen and Felser2018). Despite making different predictions about when difficulties for L2 speakers arise, what these accounts have in common is that they describe L1-L2 differences in terms of difficulties in applying information during online processing. Cunnings (Reference Cunnings2017) proposes that during the cue-based memory retrieval processes, similarities among the cues may interfere and lead to differences in weighting of the cues, which in turn may result in differences during processing. Cunnings’s approach fits particularly well with the idea of the form-specific account that different syntactic, pragmatic, and discourse-level cues are weighted to render each pronoun’s referent (cf. Kaiser and Trueswell, Reference Kaiser and Trueswell2008; Kaiser, Reference Kaiser2017). Since languages and individual pronouns may differ in how they weight cues, bilingual speakers may have more weighting options available. Moreover, discourse-based cues seem to generally be weighted more strongly in L2 than L1 processing (cf. Schimke and Colonna, Reference Schimke and Colonna2016). This could actually be explained in terms of cue-weighting as follows: Prosodic focus marking results in increased attention to focused referents during processing and memory retention (Hert et al., Reference Hert, Järvikivi and Arnhold2024; Káldi and Babarczy, Reference Káldi and Babarczy2021). Therefore, using attention, an explicit link can be established to pronoun resolution: A prosodically focus-marked referent may receive a boost in its memory representation, which in turn will make establishing the link between it and a pronoun easier. In L1 processing, this does not determine the ultimate likelihood of establishing such a link, since pronoun resolution is generally determined by grammatical characteristics of individual pronouns, e.g. subject pronouns like English he and German er are linked to preceding subject referents (Foraker and McElree, Reference Foraker and McElree2007; Bly- thing et al., 2021; Hert et al., Reference Hert, Järvikivi and Arnhold2024). For L2 speakers, for whom cue-weighting is generally expected to be challenging, the effect of focus marking can be expected to be even stronger, and it is possible that, unlike L1 speakers, L2 speakers are generally more likely to select focused referents in pronoun resolution.

Thus, the first overarching hypothesis tested in the present study is that L2 speakers are more sensitive to information structure than L1 speakers in processing pronouns. In particular, we predict them to be more likely to select focused entities as pronoun referents than L1 speakers. Our second overarching hypothesis, based on previous findings (e.g., Ellert, Reference Ellert2010; Lozano, Reference Lozano2018; Wilson, Reference Wilson2009), is that proficiency will affect L2 speakers’ pronoun processing, such that more proficient L2 speakers will be more similar to L1 speakers than less proficient ones. Further, we predict proficiency to be more important than type of L1 in the L2 speakers group, since L1 type does not necessarily lead to a native-like performance (see Polio, Reference Polio1995) and even native speakers under attrition behaved more similar to L2 speakers than to native controls (Polio, Reference Polio1995; Tsimpli et al., Reference Tsimpli, Sorace, Heycock and Filiaci2004). The third overarching hypothesis to be tested is therefore that proficiency is more important for L2 speakers’ performance than whether their L1 is a pro-drop or non-pro-drop language.

More detailed predictions are derived in the individual sections for the two experiments below. Both experiments were approved by the Research Ethics Board 2 of the University of Alberta (study ID Pro00105075).

Experiment 1: Interpretation of er and der in L2 speakers

In experiment 1, we investigate whether L2 speakers’ referential choice can be aided by focusing on possible referents. That is, can their referent preference be biased toward one referent if prosody explicitly marks that referent as focused in the discourse context (see section 3.1.2 for an example)? In accordance with the first overarching hypothesis, we predict that the answer is yes, and that focused referents will be chosen more often than those that are not focused.

Further, we want to examine whether L2 speakers are sensitive to the different referential functions of the subject pronouns er and der. Previous L2 research suggests that L2 speakers differ from L1 speakers in their resolution of the demonstrative pronoun der, but are more alike to L1 speakers with the personal pronoun er (e.g., Ellert et al., Reference Ellert, Roberts, Järvikivi, Spiegel and Krafft2011). Therefore, we predict interpretation of the personal pronoun er in L2 speakers to be similar to L1 speakers, i.e. they will show a preference for the subject referent, but we predict a less clear preference for the demonstrative pronoun der.

As for proficiency, we predict that L2 speakers’ referent selection interacts with their level of proficiency, resulting in an increasingly more native-like performance with increasing levels of proficiency, as per the second overarching hypothesis. This means, more proficient L2 speakers should select the subject referent more often than less proficient ones, especially for er. Finally, in accordance with the third overarching hypothesis, we predict that proficiency will have a stronger effect on the results than whether participants’ L1 is a pro-drop language.

Method

Design and materials were identical to those used with L1 speakers in experiment 2 reported in Hert et al. (Reference Hert, Järvikivi and Arnhold2024).

Participants

In total, 80 participants with various L1sFootnote ³ (see Table 1) completed the experiment via Prolific for monetary compensation (£9/h). For the analysis, we excluded 15 participants based on their high error rate with filler items (more than 60% incorrect), resulting in 65 participants (age range: 19–62, mean: 33, sd: 12.4). Participants indicated they learned German in language classes in school or university. At the time of testing, participants spent all or the majority of their lives in their respective L1 country.

Table 1. L1s and number of speakers for experiment 1

Materials

We designed a comprehension task where participants listened to short dialogues. Two experimental factors were manipulated: (i) whether the dialogue contained an unaccented personal pronoun er or an unaccented demonstrative pronoun der; (ii) which possible referent for the pronoun was focused (subject, object; see Table 2 for an example). Information structure was manipulated in a twofold way, coupling prosodic focus marking with changes in the context that licensed the prosodically indicated information structure. A total of four conditions were tested, and we used ten sentences per condition, which resulted in 40 experimental dialogues.

Table 2. Example dialogue with critical manipulation in all four conditions. Prosodic focus marking in italics, unaccented ambiguous pronoun in bold. Contexts were identical for all conditions except where indicated with slashes and condition names in brackets. Note that information structure was manipulated prosodically in the critical sentence, as well as in the preceding context

The dialogues were recorded using Shure SM10A headset microphones in a sound-attenuated booth by two native speakers of German, one female and one male. The female speaker (the second author, a prosody researcher) recorded all the introductions and critical sentences (A-turns in Table 2) for the experimental items, while for the filler items, it was the male speaker. The speaker uniformly produced a single falling accent on the focused constituent, while the rest of the sentence remained unaccented (represented as H* L-% in GToBI notation Grice et al., Reference Grice, Baumann, Benzmüller and Jun2005, Reference Grice, Baumann, Ritter and Röhr2017).

The experimental items were distributed across four lists in a Latin square design. Additionally, we constructed 40 filler items that were the same across all lists. These fillers also contained four possible referents, but unlike the experimental items, we did not include any ambiguous pronouns.

Note that, like prior research on German pronouns, we only use masculine third-person singular pronouns in the experimental items. This is because the feminine third person singular pronouns sie and die are the same in nominative and accusative cases, and are additionally homonymous with the third person plural and the polite second person singular pronouns, which would induce additional ambiguity. The possible referents in all target items were grammatically masculine occupation names, which are interpreted as referring to men in German (Horvath et al., Reference Horvath, Merkel, Maass and Sczesny2016), meaning all of them were equally suitable referents for the target pronouns.

Procedure

The experiment was created with the jsPsych framework for carrying out online experiments (version 7.2.1, de Leeuw, Reference de Leeuw2015). The participants were given a brief written explanation of the tasks they were about to complete. First, participants filled out a questionnaire about their language background. Additionally, we included the German LexTALE (Lemhöfer and Broersma, Reference Lemhöfer and Broersma2012) as a measure of L2 speakers’ vocabulary knowledge (see Table 3). Afterwards, a screen with instructions appeared, asking participants to carefully listen to the dialogues. They were also given the chance to check their speakers’/headphones’ volume before starting the task.

Table 3. LexTALE scores (raw) for L2 speakers, including range, mean, and standard deviation

While listening to the dialogues, participants saw the names of the four mentioned referents on the screen. Following each dialogue, they saw a question on the screen probing to which of the two target referents, subject or object, the pronoun referred (see last row in Table 2). We also included the other two referents as possible responses to ensure that participants paid attention during the experiment. Participants gave their answer by clicking on one of the names on the screen. The positions of the referents’ names on the screen were randomized for each list. Halfway through the experiment, participants were given a break.

Results

We performed generalized linear mixed-effects regression modeling (GLMER) using the lme4 package (version 1.1-35.1, Bates et al., Reference Bates, Mächler, Bolker and Walker2015) in the software R (version 4.3.0, R Core Team, 2023) to analyze the participants’ responses. The models included a binomial dependent variable coding whether the participant chose the subject or the object as the referent of the pronoun. We therefore excluded 111 responses choosing a distractor referent, i.e. 4.27% of the data, leaving 2489 data points for analysis. We added a three-way interaction for Condition (subject focus vs object focus), Pronoun (er vs der), and LexTALE (centered). L1 type (a binary variable coding whether participants’ L1 was a pro-drop language) was also included as a fixed effect (a more complex model with a four-way interaction did not converge). Opting for a backward-fitting procedure, we excluded fixed factors one by one to see whether they significantly contributed to the model fit. Neither L1 type nor any interactions between Pronoun, Condition, and LexTALE affected the model’s fit significantly and therefore the fixed effect L1 type and the interactions were excluded from the final model shown in Table 4, which only indicates significant effects of Condition, Pronoun, and LexTALE. Moreover, the model contained by-Participant random slopes for Pronoun, as well as a random intercept for Item. Models with more complex random effects structures did not converge.

Table 4. Fixed effects for best fitting generalized linear mixed-effects model of referent choice for er and der

As illustrated in Figure 1, the subject referent was chosen significantly less often in the object focus condition compared to the subject focus condition (the intercept in Table 4). With regards to the pronouns, participants selected the subject referent significantly less often for der than for er. Moreover, participants’ LexTALE score predicted their referent choice significantly. Higher scores correlated with an increase in subject preference.

Figure 1. Referent choice for er and der by condition, with error bars for standard error.

In addition, we compared the subject preference for the two pronouns in both prosody conditions to chance level using one-sample Wilcoxon signed rank tests and found the difference to be significant only for er in subject focus, where subject preference was significantly above chance level, and der in object focus conditions, where it was significantly below chance level (for both p < 0.05).

Experiment 2: Interpretation of er and ihn in L1 and L2 speakers

In experiment 2, we test L1 and L2 speakers’ referential choice for the subject pronoun er and object pronoun ihn. We manipulated whether pronouns were unaccented or accented, as for the subject pronoun in (1), to investigate whether prosody would affect referent selection and if L1 speakers would differ from L2 speakers (see section 4.1.2 for full example and all pronoun conditions).

1. (1) Der Arzt bringt den Koch mit einer Clownsnase zum Lachen, als er/ER die Musikerin mit der Kamera filmt.

   “The doctor makes the cook laugh with a clown’s nose, when he/HE filmed the musician with the camera.”

As discussed in section 1.1, unaccented personal subject pronouns are preferably resolved toward the preceding subject referents (e.g., Abashidze et al., Reference Abashidze, Gagarina and Bittner2023; Bader and Portele, Reference Bader and Portele2019b; Bouma and Hopp, Reference Bouma and Hopp2007; Colonna et al., Reference Colonna, Schimke and Hemforth2012; Hert et al., Reference Hert, Järvikivi and Arnhold2024), whereas for object pronouns, a preference for object referents (Sauermann and Gagarina, Reference Sauermann and Gagarina2017) or subject and object referents (Abashidze et al., Reference Abashidze, Gagarina and Bittner2023) was observed in L1. Therefore, for L1 speakers, we expect a clear preference for linking the unaccented subject pronoun to the subject referent. For the unaccented object pronoun, if its interpretation is indeed affected to the same extent by multiple factors as proposed by Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023), then we expect no preference for either referent. With respect to accented pronouns, previous findings suggest that whether or not the accent leads to a reversal in referential choices depends on alternatives being explicitly available in the previous discourse (Mozuraitis and Heller, Reference Mozuraitis and Heller2017). Our experimental items include alternative referents, which should enable the reversal of accented subject pronouns. For the object pronoun, we assume no preference following Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) and Sauermann and Gagarina (Reference Sauermann and Gagarina2017), and accenting the pronoun should therefore not lead to a reversal in referent preference.

Turning to L2 speakers, considering the research presented in section 1.2, we assume referent choice to be similar to that of L1 speakers with unaccented pronouns, but the preference may not be as pronounced as with L1 speakers. Regarding accented pronouns, in line with our first hypothesis, experiment 1 showed that L2 speakers are highly sensitive to focus marking, where it helps them with the task of resolving ambiguity of pronouns, even though this can lead to non-native like preferences. But will the effect of information structure marking be equally pronounced when it applies to the pronoun itself rather than highlighting one of the possible referents, and thus not directly aiding with ambiguity resolution?

In addition to the accent, we included an order of mention manipulation for the object pronoun ihn. This was done so that we could target possible effects of parallel position (cf., Abashidze et al., Reference Abashidze, Gagarina and Bittner2023; Sauermann and Gagarina, Reference Sauermann and Gagarina2017). Note that while Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) and Sauermann and Gagarina (Reference Sauermann and Gagarina2017) have looked into the effect of parallel position as well, they actually did not manipulate order of mention of the pronoun.

Lastly, in relation to previous work on L2 pronoun resolution mentioned in section 1.2, we also considered the role of proficiency and included a measure of vocabulary knowledge. Following previous findings (e.g., Ellert et al., Reference Ellert, Roberts, Järvikivi, Spiegel and Krafft2011; Lozano, Reference Lozano2018; Wilson, Reference Wilson2009), we assume that L2 speakers’ performance correlates with their level of proficiency. This means the higher their score on the proficiency measure, the closer their performance on referent selection should be to that of the native speakers, as stated by our second overarching hypothesis. Moreover, following the third hypothesis, we expect proficiency to be of greater importance for pronoun resolution than for L2 speakers’ L1.

Methods

Participants

A total of 249 participants were recruited via Prolific as well as from the University of Kaiserslautern. After excluding German bilingual participants and participants who scored less than 60% correct on the non-ambiguous filler items (see below), data from a total of 220 L1 (n = 113; age range: 18–69, mean: 31, sd: 10.47) and L2 (n = 107; age range: 19–66, mean: 30, sd: 10.37) participants were analyzed. L2 participants varied in their L1s (see Table 5; pro-drop: n = 94). All L2 participants learned German in school or university languages classes. Prolific participants received monetary compensation (£9/h), participants from the University of Kaiserslautern received a 10€ gift card as compensation for their participation.

Table 5. L1s and number of speakers for L2 participants in experiment 2

Materials

For the experimental items, we created 42 mini stories containing either a subject or an object pronoun; see example in (2). The first sentence introduced a feminine referent, followed by two masculine referents. The second sentence contained additional information about the first sentence, but did not include any of the referents. The third sentence was made up of a main clause, which repeated the two masculine referents, one as subject the other as object referent (marked with italics in (2)), and a subordinate clause which contained an ambiguous pronoun (bold), either subject (see 2a) or object pronoun either in first- or second-mention position (see 2b and 2c). Additionally, the pronouns were either unaccented or accented (upper-case letters).

1. (2) Die Fotografin, der Buchhalter und der Radiosprecher haben für Silvester eine kleine Feier ge- plant. Es gibt auch ein Feuerwerk. Der Buchhalter überwacht den Radiosprecher beim Zünden der Raketen, als

“The photographer (feminine), the bookkeeper (masculine) and the radio announcer (masculine) have planned a small party for New Year’s Eve. There will also be fireworks. The bookkeeper supervises the radio announcer firing the rockets, when”

1. a. er/ER die Fotografin achtsam anstupst. (subject pronoun) “he/HE carefully nudges the photographer.”

2. b. ihn/IHN die Fotografin achtsam anstupst. (object pronoun first) “the photographer carefully nudges him/HIM.”

3. c. die Fotografin ihn/IHN achtsam anstupst. (object pronoun second) “the photographer carefully nudges him/HIM.”

The mini stories were recorded using a Shure SM10A headset microphone in a sound-attenuated booth by a native speaker of German. The speaker (the second author, a prosody researcher) uniformly produced a contrastive rising accent on the accented pronoun, which would be represented as L*H following Féry (Reference Féry1993) and as L+H* in GToBI notation (Grice et al., Reference Grice, Baumann, Benzmüller and Jun2005, Reference Grice, Baumann, Ritter and Röhr2017). Accented pronouns were approximately four times longer in duration than unaccented pronouns. Figure 2 shows an example of a pronoun in unaccented (Figure 2a) and accented (Figure 2b) conditions.

Figure 2. Prosodic contours for unaccented and accented pronouns.

The target items were distributed across six lists following a Latin square design. Additionally, we constructed 28 filler items that were the same across all lists. These fillers also contained three referents, but unlike the experimental items, they had two feminine and one masculine referent and were followed by a question asking about the referent of a full noun phrase instead of a pronoun.

As in experiment 1, the target items only contained masculine pronouns due to the syncretism of the feminine pronouns with other forms. Also, we again used occupation names, which are grammatically unambiguously gendered. Therefore, the feminine referent was not a possible referent for the pronoun due to grammatical gender mismatch (and because the feminine referent was already present in the same clause as the other argument of the verb and could therefore only be referred to by a reflexive pronoun). Like previous research, we are interested in whether the pronoun is resolved toward the subject or object of the preceding sentence (i.e. the two masculine referents). We opted to include a third referent in the discourse context not to make this choice more difficult, but to make the use of accented pronouns more natural. While the feminine-marked referent is not a possible referent for the pronoun itself, it is a plausible member of a set of alternative referents with which the pronoun referent is contrasted, licensing the use of contrastive accent on the pronoun.

Procedure

The experiment was created using jsPsych (version 7.2.1, de Leeuw, Reference de Leeuw2015). As in experiment 1, participants carried out a questionnaire and LexTALE (see Table 6), followed by the main task. While listening to the stories, participants saw the names of the three mentioned referents on the screen. Following each story, they saw a question on the screen asking them to choose the subject, object, or another masculine referent that was not mentioned in the story as the referent of the pronoun. For subject pronouns, we asked a subject question (e.g. “Who filmed the musician?”), and for the object pronouns, we asked an object question (e.g. “Who did the musician film?”). The positions of the referents’ names on the screen were randomized for each list.

Table 6. LexTALE scores (raw) for L1 and L2 speakers, including range, mean, and standard deviation

Results

We performed GLMER using the lme4 package (version 1.1-25, Bates et al., Reference Bates, Mächler, Bolker and Walker2015) in the software R (version 4.3.0, R Core Team, 2023) to analyze responses. The models included a binomial dependent variable coding whether the participant chose the subject or the object as the referent of the pronoun. We excluded answers that selected a distractor (0.84%), which resulted in a total of 9162 observations. Fixed effects included a 4-way interaction between Language_Type (L1, L2 pro-drop or L2 non-pro-dropFootnote ⁴ ), Pronoun (er, 1_ihn, or 2_ihn), Prosody (unaccented or accented), and the centered LexTALE score. Selecting a backward-fitting procedure, we excluded fixed factors one by one to see whether they significantly contributed to the model fit. For random effects, as more complex random effects structures led to convergence issues (we initially included the interaction term as a by-participant random slope), we followed Sonderegger’s (Reference Sonderegger2023) steps for non-convergence issues. After some nonintrusive steps (e.g. changing the maximum number of iterations for the optimization), which did not help with model convergence, we ultimately dropped one random effect at a time to obtain the maximal model possible. The final model included a by-Participant random slope for Pronoun and a random intercept for Item. LexTALE did not contribute to any significant interactions and only remained as a fixed effect. The final model in Table 7 shows an effect for LexTALE, as well as for the interaction of Language_Type, Pronoun, and Prosody.

Table 7. Fixed effects for best fitting generalized linear mixed-effects model of referent choice for er and der

Both Language_Type and Pronoun consist of three levels. In order to see where the significant difference between the different levels of factors involved in the interaction are, we ran a pairwise comparison using emmeans (version 1.10-0, Lenth, Reference Lenth2024) applying Bonferroni adjustment. Table 8 shows the significant differences among the factor level combinations.

Table 8. Multilevel comparison of the interaction term Language * Pronoun * Prosody of the generalized mixed-effects model. Positive estimates indicate a higher bias for the subject referent for the left factor level combination in the pair

As can be seen in Figure 3, overall, there was a preference for the subject referent for both pronouns with all groups. However, within L1 speakers, there was a difference for er; the subject referent was selected significantly less often for accented than unaccented er (see Table 8). L1 speakers also selected the subject referent significantly less often for accented er compared to accented first- and second-mention ihn. While there were no significant differences between the two L2 speaker groups, only L2 speakers with pro-drop L1s showed significant differences from the L1 group: For unaccented first-mention ihn and accented second-mention ihn, L1 speakers chose the subject referent more often than L2 speakers with pro-drop L1s. Finally, the positive effect of LexTALE in Table 7 indicates that the overall subject preference was larger for higher-proficiency speakers.

Figure 3. Referent choice for L1 and L2 speakers, accented and unaccented for er and ihn, with error bars for standard error.

General discussion

The two experiments in this study investigated the effects of prosodic focus marking in the preceding context on referent choice for the subject pronouns er (personal) and der (demonstrative), as well as the effects of accents on the subject pronoun er and the object pronoun ihn (personal) in L1 and L2 speakers. We additionally varied the position of the pronoun ihn, allowing us to disentangle potential effects of grammatical role parallelism from positional parallelism.

Our first overarching hypothesis was that L2 speakers are more sensitive to information structure than L1 speakers in processing pronouns. In the first experiment, we asked if L2 speakers’ referent selection would be affected if the referents in the previous discourse were prosodically marked as focused, hypothesizing that focus marking may guide L2 speakers’ attention toward the focus-marked referent, which would render this referent more available in memory and in turn lead the referent to be selected more often. The results supported this hypothesis. We found that L2 speakers preferred the focused subject referent with the personal pronoun er, and chose the focused object referent more often for the demonstrative der. However, the results suggest that clear preferences only emerge when information structure and pronoun effects combine. Interestingly, while converging effects may help to achieve a more native-like performance, this is not guaranteed, as the two factors are not simply additive (as they appear to be in L2), but show a complex interplay in L1 processing, in line with the form-specific account (Kaiser and Trueswell, Reference Kaiser and Trueswell2008). Thus, L1 speakers prefer the subjects and topics as referents for er and objects and non-topics as referents for der, but subjecthood is the determining factor for the default pronoun er, whereas the effect of information structure is stronger for the more marked der (e.g., Bader and Portele, Reference Bader, Portele, Gattnar, Hörning, Störzer and Featherston2019a, 2025; Hert et al., Reference Hert, Järvikivi and Arnhold2024; Kaiser and Trueswell, Reference Kaiser and Trueswell2008; Kaiser, Reference Kaiser2011c; Portele and Bader, Reference Portele and Bader2016). By contrast, our findings point toward L2 learners weighting information structural (discourse-based) cues more heavily than grammatical role cues overall, in line with Cunnings’s account (2017) and our hypothesis.

In experiment 2, we investigated referential choice for unaccented and accented subject pronoun er and object pronoun ihn in L1 and L2 speakers. The results showed an overall preference to link both pronouns to the subject referent, which was only reduced with accented er for L1 speakers. This suggests that L1 speakers interpreted the accent on this pronoun as a signal for a topic shift, i.e. for a reversal of the usual preference of er to be resolved toward the subject (and topic) of the preceding sentence, but L2 speakers did not. How does this fit with our hypothesis that L2 speakers are more affected by information structure than L1 speakers?

One possible explanation for the absence of reversal in L2 follows from Robenalt and Goldberg (Reference Robenalt and Goldberg2016)’s findings that only highly proficient L2 speakers can take competing alternative forms into account during language processing. Thus, when less proficient L2 speakers encounter an accented pronoun, they are still unable to consider resolving it toward the alternative referent – a process that is acquired over time (cf. Robenalt and Goldberg, Reference Robenalt and Goldberg2016). However, L2 proficiency was notably higher for participants in experiment 2 (mean scores of 71.25% for non-pro-drop and 69.99% for the pro-drop group) than in experiment 1 (44.00%), yet information structure affected participants in experiment 1, but not in experiment 2.

Therefore, another explanation is more likely: We suggest that our manipulation of information structure in experiment 1 affected L2 participants because it provided a salient cue that (seemingly) helped them resolve the ambiguous pronoun, even if this led to a non-native like result. Thus, when participants encounter an ambiguous pronoun, they have to pick one referent with which to associate it. As argued above, focus marking in experiment 1 highlighted one possible referent, leading participants to resolve the ambiguous pronoun toward this referent. Even though focus marking did not necessarily help L2 speakers by allowing them to resolve pronouns the way that native speakers do, it allowed them to make a choice, resolve the pronoun, and move on. In contrast, accenting the pronoun in experiment 2 did not have the same effect because it did not help resolve the ambiguity. For L1 speakers, who already have a stable preference for resolving unaccented er, accenting the pronoun signals a contrast, which allows them to adjust their normal interpretation of this pronoun. As results of both of our experiments show, L2 speakers do not have an equally strong subject preference for unaccented er, meaning they do not start from the same point as L1 speakers and are therefore unable to follow the same process. Instead, they are still faced with the basic task of pronoun resolution—pick a referent—and unlike highlighting one of the possible referents, accenting the pronoun itself does not help with this task.

In fact, it is possible that accounts of pronoun resolution in terms of accessibility, prominence or salience (e.g. Ariel, Reference Ariel, Sanders, Schilperoord and Spooren2001; Arnold, Reference Arnold2001; von Heusinger and Schumacher, Reference von Heusinger and Schumacher2019; Grosz et al., Reference Grosz, Weinstein and Joshi1995b) may work well for L2 pronoun resolution, precisely for the reasons why they do not work for explaining the behavior of L1 speakers. L1 speakers have established (grammatical) preferences, which differ for individual pronouns like personal er vs demonstrative der. There is therefore no need to simply pick the referent that currently happens to be at the forefront of their minds (i.e. the most accessible/prominent/salient) to deal with the apparent ambiguity. In contrast, L2 speakers do not have the preferences of individual pronouns and the differences between them firmly acquired. In the absence of the detailed knowledge of different weights of various grammatical factors, they seem more likely to assign the most accessible/prominent/salient referent to any pronoun they encounter. If this is true, it is not clear that L2 speakers are necessarily following information structure as such, in the sense that they would decode its marking and compute its implications. Instead, they may simply be affected by emphasis, in other words, by the fact that focus marking makes one referent perceptibly more prominent, which in turn renders the referent more distinct in working memory (Birch et al., Reference Birch, Albrecht and Myers2000; Foraker and McElree, Reference Foraker and McElree2007; Norberg and Fraundorf, Reference Norberg and Fraundorf2021). Thus, even though both L1 and L2 speakers are seemingly affected by information structure, it is possible that they are affected by it in different ways. For L1 speakers, effects of information structure are grammatically driven, part of their linguistic knowledge—e.g. that der prefers non-topic referents whereas er prefers subjects. For L2 speakers, prosodic focus marking as a cue may be easier to access than form-based preferences and subjecthood, as these would involve accessing grammar. This fits with results from experiment 1, suggesting that less proficient participants were more likely to be guided by information structure marking than more proficient participants. It may be easier to rely on overt information structure marking as a cue, especially at earlier stages of L2 acquisition, since it is readily, perceptually available and its use would not need access to grammar (cf. Clackson et al., Reference Clackson, Felser and Clahsen2011; Felser and Clahsen, Reference Felser and Clahsen2009; Jacob et al., Reference Jacob, Şafak, Demir and Kırkıcı2019). So, what has previously been defined as an influence of L2 learners’ L1 may actually be more accurately described as an over-reliance on prominence. This would then also explain why in some studies where both the L2 and L1 shared a referent preference pattern, L2 learners did not benefit and their performance deviated from native speakers’ (e.g. Polio, Reference Polio1995).

Our second overarching hypothesis was that more proficient L2 speakers would be more similar to L1 speakers (see also Ellert et al., Reference Ellert, Roberts, Järvikivi, Spiegel and Krafft2011; Lozano, Reference Lozano2018; Polio, Reference Polio1995). In line with this hypothesis, there was a significant effect of proficiency, with higher LexTALE scores leading to an increase in the subject referent choices for both pronouns in both experiments. Still, in both experiments, L2 speakers differed from the L1 group, although pairwise comparisons were only significant for L2 speakers from the pro-drop group in experiment 2. This is in line with previous research as reviewed in section 1.2 and suggests that pronoun resolution is a challenging task, reflecting the complexity of cue-weighting for individual pronouns as described in the form-based account Kaiser (Reference Kaiser, Branco, McEnery and Mitkov2005).

Even though the effect of proficiency was the same in both experiments, the second experiment seems to show a more native-like performance than the first experiment. The difference in performance may be attributed to the overall level of proficiency being higher in experiment 2 than in experiment.

1. We therefore assume that proficiency plays an important role in achieving a more native-like pattern of pronoun interpretation. In fact, if we take proficiency to reflect language dominance more generally, then this could also explain why speakers under L1 attrition perform less native-like (Tsimpli et al., Reference Tsimpli, Sorace, Heycock and Filiaci2004) when compared to native speakers. This would predict that language dominance affects even native-like speakers’ performance during pronoun resolution.

The results further support our third hypothesis that proficiency is more important for L2 pronoun resolution than learner’s L1, whether L1 is a pro-drop language or not. In contrast to the effect of proficiency, which was significant and consistent across both experiments, L1 background did not have a significant effect in experiment 1. Experiment 2 showed some significant differences between L1 speakers and pro-drop L2 speakers, whereas there were no differences between and L1 and non-pro-drop L2. While this seems to suggest slight differences between the two L2 groups, there were no significant differences between them. Note that the group size for the L2 non-pro-drop speakers is a lot smaller than the other L2 or L1 groups, which may have obscured some possible differences. However, Figure 3 shows very similar patterns for both groups, and both groups had very similar mean LexTALE scores (cf. Table 6), meaning this factor at least did not obscure any possible effect of L1 type. Thus, on the basis of the present evidence, proficiency indeed seems to be a stronger predictor of L2 performance than whether the L1 is a pro-drop language.

To be clear, we are not arguing that proficiency is the only predictor of L2 performance. We likewise acknowledge that proficiency may itself be predicted or modulated by other factors, such as length of residence in a German-speaking country or quantity and quality of L2 input (see e.g. Xu et al., Reference Xu, Case and Wang2009; Granena and Long, Reference Granena and Long2013; Paradis et al., Reference Paradis, Rusk, Duncan and Govindarajan2017). Here, we were specifically interested in the relative influence of proficiency and the pro-drop status of L1, since both of these factors have been suggested as major influences on L2 pronoun resolution (recall references in section 1.2).

A final aspect of the present results that deserves discussion concerns the L1 speakers’ interpretation of the object pronoun ihn. Regarding unaccented pronouns, referent preference for subject and object pronouns has been mostly described in terms of grammatical role parallelism (Kehler et al., Reference Kehler, Kertz, Rohde and Elman2008; Sauermann and Gagarina, Reference Sauermann and Gagarina2017; Smyth, Reference Smyth1994) or, more recently (Abashidze et al., Reference Abashidze, Gagarina and Bittner2023), in terms of a combination of grammatical role parallelism, positional parallelism, and topic bias. Abashidze et al. argue that this combination would lead L1 speakers not to show a preference for the object pronoun, since the general topic/subject-bias is pitted against grammatical role parallelism. But in contrast to Abashidze et al.’s findings, our participants preferred the subject referent also for object pronouns. Thus, the subject referent was selected regardless of the grammatical role of the pronoun. Therefore, our results do not support grammatical role or positional parallelism. Further, Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) claim that moving the object pronoun into first mention position, as done in their items, would lead to an enhanced topicality effect. However, our results do not support this assumption as there was no difference in preference when the object pronoun was in first mention position compared to when it was in second mention position.

So, how can we explain the different results in our study compared to Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) (and Sauermann and Gagarina, Reference Sauermann and Gagarina2017, who found differences between subject and object pronouns in eye-gaze data)? A possible explanation stems from the experimental items. All the items in Sauermann and Gagarina (Reference Sauermann and Gagarina2017) and the two example items that are shown in Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) contain experiencer verbs and an inanimate second argument in the pronoun-clause, presumably to license the sentence-initial position for object pronouns (e.g. Ihn ermuntern die Farben “He is encouraged by the colors (literally: him encourage the colors)” and Er mag die Farben “He likes the colors”). This may have affected the results, although it is not discussed by the authors: Abashidze et al., Reference Abashidze, Gagarina and Bittner2023 report a bias test for the verbs in the preceding clause, which were apparently relatively prototypical agent-patient verbs such as fangen “catch” or malen malen “paint,” but not for those in the pronoun clause (for the effect of agentivity on pronoun resolution, see Schumacher et al., Reference Schumacher, Dangl, Uzun, Holler and Suckow2016, Reference Schumacher, Roberts and Järvikivi2017, though these studies manipulated agentivity of the preceding referents only). In our materials, both the preceding clause and the pronoun-clause contained similar agent-patient verbs with animate referents as both subject and object. Another difference is that Abashidze et al. and Sauermann and Gagarina used contexts for inter-sentential pronoun resolution (see Colonna et al., Reference Colonna, Schimke and Hemforth2015, for differences between intra- and intersentential pronoun resolutions in German, though this study focused on subject pronouns). Finally, Sauermann and Gagarina (Reference Sauermann and Gagarina2017) and Abashidze et al. (Reference Abashidze, Gagarina and Bittner2023) used contexts containing only two referents in their experimental items, whereas our items included a third referent that differed in gender. As already noted by Patterson and Schumacher (Reference Patterson and Schumacher2021), a limitation in many experiments on pronoun resolution is that the contexts are limited to two potential referents. However, including a third referent can give a more complex base for the interpretation of focus, as there are more alternatives present in the discourse (see Krifka, Reference Krifka2008, for a review on alternatives and focus), and lead to a more detailed picture of how sensitive different pronouns are to the factors involved in pronoun resolution (e.g. grammatical role, information structure). The difference in referent preference for the object pronoun observed in our study may be the result of a “richer” discourse context.

Turning now to accented pronouns, previous research has established (Mozuraitis and Heller, Reference Mozuraitis and Heller2017; Taylor et al., Reference Taylor, Stowe, Redeker and Hoeks2013) that accented pronouns do not always lead to a reversal in referent preference, as had been initially claimed (e.g., Akmajian and Jackendoff, Reference Akmajian and Jackendoff1970). Indeed, our findings for the object pronoun do not show a switch in referent preference for the accented pronouns. However, we do find a switch for the accented subject pronoun. Neither Taylor et al.’s nor Mozuraitis and Heller’s explanations as to why reversal patterns may not be triggered can account for the difference between subject and object pronouns, because their assumptions are not about pronoun type, but about the accessibility of alternatives. Heeding their analyses, we designed experiment 2 to explicitly have alternative referents available, and the fact that we observed a preference switch for the subject pronoun er suggests that this effort was successful. The differences in findings could be partly due to the sentence structure: intra- (current study) vs inter-sentential (Taylor et al. and Mozuraitis and Heller’s studies) pronoun resolution, or there could be true differences between German on the one hand and English and Spanish on the other. Further, our results may actually paint a more detailed picture of the pronoun types’ sensitivity to factors such as information structure and subjecthood (cf. Kaiser and Trueswell, Reference Kaiser and Trueswell2008). This would imply that object pronouns are more restricted regarding the degree to which some of these factors can affect referential links than subject pronouns. Results of our study suggest that object pronouns are more sensitive to subjecthood than to information structure, since changes in information structure (i.e., unaccented vs accented pronouns) did not influence referential choice. This would also explain why there was no effect of “enhanced topicality” (cf. Abashidze et al., Reference Abashidze, Gagarina and Bittner2023) when the object pronoun was first mentioned compared to when it was second mentioned.