Who said it—and did it fit the speaker's age? How age-typicality of named objects improves source memory

Memory for who said what can help native and non-native listeners identify critical information for use in conversations. In this study, source memory for object-speaker associations was tested for listeners differing in language background and nativeness. Native and non-native participants of German first heard the voice of a child and an adult speaker name photographs of objects with age-typical characteristics (e.g., a board book for children vs. a standard hardcover novel) before they had to indicate who had previously named each object. Speakers either consistently named objects typical for their age group, or they randomly named objects from both age groups. Both listener groups associated objects with speakers successfully and showed better source memory in the age-consistent condition over the random condition. Critically, non-native listeners performed comparably to native listeners and benefited equally from age-typical patterns, suggesting that stereotypical knowledge about age can facilitate source memory encoding and retrieval even when processing a second language. The results confirm the role of source memory representations in communication and suggest that memory for object-speaker associations based on salient social categories like age remains robust in non-native listening.

Successful communication requires more than remembering linguistic content; it also requires remembering contextual information, for example about who produced that content. This distinction is commonly captured by the contrast between item memory, i.e. memory for what information was encountered, and source memory, i.e. memory for contextual details such as when, where, or from whom information was obtained (Mitchell and Johnson, 2000). In communicative settings, source memory can play a crucial role in establishing common ground, evaluating credibility, and guiding expectations about future utterances (Clark, 1996; Henkel and Mattson, 2011; Horton and Slaten, 2012). For example, knowing who said something can help establish shared knowledge, knowing that a close friend recommended a restaurant may increase trust in that recommendation, and knowing that a child previously mislabeled an object may lead parents to expect the same mislabeling in later interactions.

Importantly, source memory is shaped not only by episodic encoding of individual events, but also by generalized knowledge structures that guide encoding and retrieval. Humans acquire rich knowledge about social categories and form expectations about which properties, objects, or behaviors are typical for members of these categories (Bigler and Liben, 2006; Klein et al., 2010). When encountering new individuals, these expectations can be recruited rapidly and implicitly, influencing both online processing and later memory. Age is one such particularly salient social category. Speaker age information is extracted quickly from speech by both adults and children (Moyse, 2014) and provides a strong basis for forming expectations about speakers. Objects, too, are often stereotypically associated with age groups: for example, in many Western, literate societies, a board book is more typical for a child, whereas a standard hardcover novel is more typical for an adult. Such age-typicality relations are shaped by shared cultural knowledge and may vary across cultural contexts, potentially providing powerful cues for organizing speaker-related memory. The current study set out to compare native (L1) and non-native (L2) participants' source memory using relevant categorical/stereotypical knowledge about speakers, specifically, knowledge about typical objects associated with speakers of different ages.

In terms of source memory, a growing body of research has shown indeed that adult L1 listeners can reliably encode speaker information alongside item information. For example, McKinley et al. (2017) demonstrated that L1 listeners can accurately remember which conversational partner referred to which object, even when speaker identity was not explicitly task relevant. Such findings suggest that speaker identity can be routinely bound to linguistic representations during communication. Previous research with L1 listeners furthermore suggests that stereotypical relations between speakers and content can enhance source memory. For instance, Wulff and Kuhlmann (2020) showed that age-based stereotypes can bias adult L1 speakers' source attributions, such that the speaker of a statement is best remembered and guessed when the statement is stereotypical of that speaker's age group (e.g., older-typical statements attributed to older speakers).

Whether L2 listeners use such stereotypical knowledge in the same way, however, remains an open question. Source memory is known to be sensitive to limitations in attentional resources (Mulligan, 2004), and second-language processing during language comprehension and production is generally slower, less automatic, and places greater demands on cognitive resources (Cutler, 2012; Sorace and Filiaci, 2006), which may reduce the availability of attentional means for retrieving contextual information such as speaker identity.

Indeed, Beck and Weber (2024) have found first evidence for differences in source memory between L1 and L2 listeners. In their study, English L1 and German L2 adults first heard two English speakers name photographs of objects with colors that are typically associated with those objects in the participants' respective cultural and linguistic contexts (e.g., green for broccoli and red for tomato) before they had to indicate who had previously named the objects. Speakers either consistently named objects of one color, or they randomly named objects of different colors. While both participant groups encoded objects with speakers quite successfully overall, L1 participants, but not L2 participants, showed better speaker memory in the color consistent condition over the random condition. That is, while L1 listeners benefited from a consistent pattern in speaker preferences based on prototypical object color, L2 listeners did not show this source memory advantage, even when the objects were presented in their natural colors and not just in black and white.

The present study builds directly on this work by examining a different type of regularity, one grounded in socially meaningful stereotypical knowledge rather than in perceptual features for which speakers' preferences are usually arbitrary. Specifically, we tested whether age-typical object-speaker associations support speaker memory in L1 and L2 listeners. Age-typicality reflects deeply entrenched social knowledge that adult L2 listeners are expected to share with L1 listeners in a given cultural context. However, if L2 listeners' reduced sensitivity to categorical effects reflects general resource limitations, age-typicality may fail to enhance speaker memory in L2 processing just as it did with perceptual color features in Beck and Weber (2024). Alternatively, if socially grounded stereotypes are particularly salient and robust, they may support source memory equally in L1 and L2 listeners.

To test these possibilities, L1 and L2 German participants completed a speaker memory task closely modeled on Beck and Weber (2024). Participants first heard the voice of a young child and an adult speaker refer to photographs of objects that were either consistently age-typical for the speaker (e.g., a colorful board book for the child and a novel book for the adult) or pseudo-randomly assigned. In a subsequent recognition task, participants indicated which speaker had named each object.

We expected both L1 and L2 listeners to show above-chance speaker memory overall. We further predicted higher recognition accuracy in the age-consistent condition than in the pseudo-random condition, reflecting the facilitative role of age-typical object-speaker associations. Critically, we examined whether this benefit would differ between language groups. A reduced or absent age-consistency effect for L2 listeners would suggest that increased processing demands limit the use of stereotypical knowledge in source memory, whereas comparable effects across groups would indicate that socially grounded age knowledge can support speaker memory even in a second language.

The online experiment was implemented using Gorilla Experiment Builder (www.gorilla.sc; Anwyl-Irvine et al., 2020). It consisted of a German-language listening task (association phase) followed by a source memory task (testing phase), closely adhering to the design of Beck and Weber (2024).

Participants

A total of 113 L1 German participants (18–40 years, M = 24.2; 78 female, 33 male, 2 diverse) and 72 L2 German participants (18–35 years, M = 25.58; 45 female, 26 male, 1 diverse) were recruited from the University of Tübingen via a university-wide email and Prolific (www.prolific.com). All L1 German participants reported German as their native language, although two participants additionally reported a second native language. L2 participants reported a native language other than German (e.g., Spanish, English, Greek) and although they had diverse linguistic backgrounds, they predominantly grew up with Indo-European language backgrounds and Western cultural contexts. They reported a mean German proficiency of 5.5 averaged across all modalities (7-point scale; 1 = poor, 7 = native-like), with separate self-ratings obtained for speaking (5.07), listening (5.8), reading (6), and writing abilities (5.03). In addition, they had a mean score of 72.6/100 on the German LexTALE (Lemhöfer and Broersma, 2012). All L2 participants but one reported having spent time in a German-speaking country. They began learning German at an average of 14.3 years and had been studying the language for an average of 5.8 years.

Materials

Experimental items comprised two photographic image sets depicting the same 34 common, concrete objects: one set of images showed child-typical versions of the objects and the other adult-typical versions (e.g., a child-typical board book vs. an adult-typical novel book). The objects covered a range of everyday categories, including household items, clothing accessories, containers, toys, and personal belongings (e.g., bicycle, cup, backpack, sunglasses). Child-typical versions were selected to reflect objects commonly associated with children through factors such as colorfulness, size, design, or thematic decoration, whereas adult-typical versions reflected more neutral or adult-oriented variants of the same object type. Images were licensed from Canva Pro or produced by the authors, edited as needed, and standardized for size, resolution, and transparent white backgrounds. A rating study with 20 L1 German speakers confirmed the age-typicality of the chosen images. Child-typical images were rated on average with 1.36 and adult-typical images with 5.83 on a 7-point scale (1 = child-typical, 7 = adult-typical; Welch's t = 64.58, p < 0.001).

Names for each of the 34 experimental objects were recorded by a female L1 German adult (36 years) and a female L1 German child (2.11 years). The adult speaker read from a script, whereas the child speaker was prompted with child-typical images. Multiple tokens were recorded per object, and the best-quality recording was selected. To reduce carrier-phrase variability, object words were recorded in isolation and spliced into the standardized carrier phrase das ist ein/e (“that is a”).

Procedure

Association phase

Participants first completed a listening task in the association phase and were asked to pay attention as they would have to answer questions about what they had heard in a subsequent testing phase. They were introduced to the two speakers via names and icons (child and adult), listened to each speaker's self-introduction (“Hi, I am Meli/Nora.”), and could adjust the volume to a comfortable listening level before the association phase started.

In each trial of the association phase, two objects were presented side by side on a screen (Figure 1), and a speaker referred to one object while the other remained unnamed. Participants indicated the position of the named object (i.e., left or right) by pressing a button. To accommodate varying screen sizes for at-home participants, images were displayed as squares occupying one-third of the screen width at full resolution, with a 10% screen-width gap between them. Target position (left vs. right) and the age-typicality of image pairs (same vs. different age group) were counterbalanced across trials, and trial order was randomized for each participant. When a previously unnamed object became the target in a later trial, it was paired with a different object. The age-typicality of individual images was counterbalanced across lists, and each speaker referred to 17 of the 34 objects in a list. In an age-consistent condition, each speaker referred only to objects appropriate for their age (e.g., adult speaker to adult-typical images), whereas in a pseudo-random condition, both speakers referred to objects from both age groups equally in no fixed pattern.

Testing phase

Participants were reminded of the two speakers via their names, icons, and self-introductions. They then viewed the 34 experimental images individually and indicated which speaker had previously referred to each object by clicking the corresponding speaker icon or selecting an “Unsure” option. Image order was randomized, and responses were required to proceed. On average, the association and testing phases took participants 4.8 min to complete (L1: 4.6 min; L2: 5 min).

Statistical analyses were conducted using R (R Core Team, 2025) and the package lme4 (Bates et al., 2015). L1 and L2 participants showed near-ceiling performance in the association phase (99.4 % and 97.2 % accuracy, respectively), indicating successful speaker comprehension and object identification.

In the testing phase, participants from both groups selected one of the two speakers on the vast majority of trials, where “unsure” responses occurred only infrequently (L1: 9.5%; L2: 10.6%) and were coded as incorrect. Responses from the testing phase were used in logistic mixed-effects models (glmer) with Correctness (1 = correct, 0 = incorrect) as the dependent variable. Age Consistency (age-consistent vs. pseudo-random), Language (native vs. non-native), Speaker (child vs. adult), and Image Type (child-typical vs. adult-typical) were included as fixed effects and sum-coded as 0.5 and −0.5; Trial Order was centered around zero. Interactions between Age Consistency and Language, as well as between Speaker and Language, were also tested. Participants and Items were included as random factors. The results are summarized in Table 1, including only factors that improved model fit. P -values for the models were obtained using a Wald z-distribution approximation.

Participants in both language groups performed well above chance in the testing phase [L1: M = 76.37%, t (112) = 20.72, p < 0.001; L2: M = 79.20%, t (71) = 16.15, p < 0.001], with no overall difference between L1 and L2 participants [t (183) = 1.32, p = 0.188]; this pattern remained when accuracy was arcsine-square-root-transformed1. Accuracy also varied with Age Consistency, with higher performance in the age-consistent condition (L1: 80.13%; L2: 83.82%) than in the pseudo-random condition (L1: 72.53%; L2: 73.44%). Results are shown in Figure 2.

A significant effect of Speaker showed better recognition for objects referred to by the adult speaker (L1: 78.6%, L2: 80.9%) than the child speaker (L1: 74.1%, L2: 77.5%). This difference was numerically quite small and may reflect the greater variability and less canonical pronunciation of child speech, which could have hindered encoding and retrieval for L1 and L2 participants alike.

Since Language was neither significant nor interacted with Age Consistency, we conclude that the non-nativeness of L2 participants2 did not disadvantage them relative to L1 participants in overall recognition, nor did age-typicality cues for object–speaker associations affect the two groups differently.

Note that participants in the age-consistent condition could, in principle, have achieved perfect recognition accuracy by relying solely, perhaps even consciously, on stereotypical knowledge about age-typical object-speaker associations, without having encoded information during the association phase. This is because, in that condition, speaker age consistently matched the age-typicality of the objects (e.g., the child speaker named only child-typical images). However, when we excluded the seven participants (4 L1 and 3 L2) who achieved 100% accuracy in the age-consistent condition, the pattern of results remained unchanged: Age Consistency and Speaker were still significant effects, and neither interacted with Language. Moreover, mean accuracy in the age-consistent condition was well below ceiling, and accuracy in the pseudo-random condition was well above the chance level of 50% (the level expected from sole reliance on age-based stereotypes in that condition). Together, these results suggest that encoding of individual object-speaker associations took place in the current experiment.

The present study investigated whether stereotypical knowledge about speakers can support source memory for object-speaker associations in L1 and L2 participants. Participants heard a child or an adult speaker refer to objects that were either consistently age-typical for the speaker or assigned pseudo-randomly, and they later had to indicate which speaker had named each object. The results revealed three main findings. First, both L1 and L2 participants performed reliably above chance, demonstrating successful encoding and retrieval of object-speaker associations. Second, participants in both groups showed better speaker memory in the age-consistent condition than in the pseudo-random condition, indicating that stereotypical knowledge about age facilitated encoding or retrieval of speaker information. Third, the benefit of age-typical patterns did not differ between L1 and L2 participants: language background neither affected overall performance nor interacted with age consistency. Thus, L2 participants were able to use regularities in object-speaker relations as effectively as L1 participants when those regularities were grounded in shared social knowledge. These findings demonstrate that stereotypical knowledge grounded in social categories like age can support source memory encoding and retrieval even when processing a second language. The absence of any difference between L1 and L2 participants suggests that when categorical information is socially meaningful and can be assumed to be shared, L2 listeners can recruit it as effectively as L1 listeners to organize speaker-related memory representations.

An obvious first explanation for the absence of a difference between L1 and L2 listeners could be the proficiency level of the L2 participants in the current study. The L2 participants reported relatively high proficiency with little variation, which may have reduced language-processing costs and allowed sufficient cognitive resources to be allocated to the encoding and retrieval of contextual information. However, this explanation alone appears insufficient. Importantly, the self-reported proficiency levels of the L2 participants were comparable to those reported in Beck and Weber (2024). Nevertheless, in Beck and Weber, German L2 listeners did not show the same benefit from consistent perceptual regularities as L1 listeners. Thus, if the L2 participants in the current study were simply too native-like for group differences to emerge, a similar absence of L1-L2 differences would also have been expected in Beck and Weber (2024). The different outcomes across the two studies therefore suggest that proficiency alone is unlikely to explain the present findings and instead strengthen the interpretation that the nature of the regularity itself, namely socially meaningful age, plays a key role.

However, research on source memory in L2 populations remains scarce, making it difficult to place the present findings within a well-established empirical framework. One of the few studies addressing this issue is Francis et al. (2019), who examined bilinguals' memory for word-context associations defined in terms of visuo-spatial (i.e., screen position), temporal (i.e., list membership), or modality context (i.e., auditory vs. visual presentation). Interestingly, bilingual participants performed better than monolinguals overall, leading the authors to propose that the strength of associative links in memory may depend on language proficiency and experience. One important methodological difference is that Francis et al. examined source memory for contextual features that were intrinsic to the experimental design and did not involve encoding speaker-specific patterns. In contrast, the present study required participants to encode and retrieve associations between objects and individual speakers. Thus, the cognitive processes underlying performance may differ across studies. Source memory for contextual features such as spatial location may rely primarily on perceptual binding mechanisms, whereas source memory for object-speaker associations involves integrating episodic information with social or conceptual knowledge.

At first glance, the present results also appear to contrast with those reported by Beck and Weber (2024), who found that while both L1 and L2 listeners encoded and retrieved object-speaker associations successfully, only L1 listeners benefited from consistent speaker-specific patterns. In their study, speakers consistently referred to objects belonging to a prototypical color category, allowing participants to learn during the experiment that a given speaker preferentially referred to objects associated with one color category. Thus, the critical manipulation was based on experimentally induced regularities in speakers' referential behavior. L1 listeners, but not highly proficient L2 listeners, showed improved source memory in the consistent condition.

One important difference between the two studies thus concerns the nature of the regularity that participants could exploit. In Beck and Weber's experiments, the pattern was based on perceptual (i.e., when the objects were shown in their natural color) or conceptual color categories (i.e., when the objects were shown in black and white). Although prototypical color knowledge is well established (Huettig and Altmann, 2011), the experimentally induced speaker-specific color patterns may have been less obvious and socially meaningful. In contrast, the present study used age-typicality, a socially meaningful and highly salient category that listeners routinely use to form expectations about speakers. It is plausible that such socially grounded stereotypes are more readily activated and more strongly integrated into memory representations than perceptual regularities like color. If so, L2 listeners may fail to exploit weaker or less meaningful patterns but succeed when regularities align with robust social knowledge.

From a theoretical perspective, the findings bear on accounts that attribute L2 difficulties primarily to reduced automaticity and limited processing resources. While such limitations are well documented (Cutler, 2012; Sorace and Filiaci, 2006), the present results indicate that the availability of robust conceptual knowledge can compensate for increased processing demands. In other words, L2 listeners may be selective rather than uniformly impaired in their use of contextual information: they may fail to exploit weak or incidental patterns but succeed when cues are salient, meaningful, and strongly represented in long-term memory.

This interpretation aligns with usage-based and associative approaches to language processing (Francis et al., 2019; Gollan et al., 2008), which emphasize the role of prior experience and knowledge structures in guiding comprehension and memory. If source memory depends on the strength of associations among items, contexts, and conceptual knowledge, then differences between studies may reflect variation in the associative strength of the cues available to listeners. Age stereotypes likely provide stronger and more socially meaningful associative links than color-based speaker regularities learned during Beck and Weber's (2024) experiment, which could explain why L2 listeners benefited in the present study but not in the earlier work.

In sum, the present study shows that both L1 and L2 listeners can use stereotypical knowledge about speakers to enhance source memory for object-speaker associations. Unlike previous findings based on perceptual regularities, age-typical patterns benefited L2 listeners to the same extent as L1 listeners, suggesting that socially grounded conceptual knowledge can offset the additional cognitive demands of L2 processing. Together with prior work, these results indicate that the role of L2 processing difficulty in source memory is nuanced: limitations may emerge when associative cues are less salient or even arbitrary, but not when they are strong, socially meaningful, and widely shared. Expanding research on L2 source memory across different types of associations will be essential for developing a more comprehensive understanding of how listeners build and use speaker-specific representations in multilingual communication.