Glossa Psycholinguistics

1.1 The structure of transfer events

An influential class of linguistic theories (e.g., Gruber, 1965; Jackendoff, 1983, 1990) suggests that transfer events can be analyzed in spatial terms and, thus, bear similarities to motion events. Consider the motion event denoted by the sentence The butterfly flew from the flower to the tree. In this sentence, the subject encodes the Agent, or moving entity (the butterfly), the from-phrase encodes the Source, or starting point, of the motion, and the to-phrase encodes the Goal, or endpoint, of the motion. Adopting the position that spatial representations are fundamental in both language and cognition, with spatial concepts extending into non-spatial domains (Jackendoff, 1983, 1990; Talmy, 2000), this class of theories proposes that participants in transfer events have dual thematic roles: the two animate (often, human) participants involved in the transfer action are simultaneously Agent and Patient (in terms of causality), and Goal and Source (in terms of the transferred object’s spatial movement). For instance, in the Giving event denoted by the sentence Harry gave an apple to Ruby, Harry is the Agent and Source, while Ruby is the Patient and Goal. Conversely, in the Taking event denoted by the sentence Harry took an apple from Ruby, Harry is the Agent and Goal, and Ruby is the Patient and Source. On this theory, Agent/Patient and Source/Goal roles are assigned at different analysis levels, with Agent/Patient assignment determining the mapping of event participants to syntax (with Agents typically being assigned subject position; see also Dowty, 1991) and Source/Goal structure determining the thematic structure and conceptualization of the event (e.g., Jackendoff, 1983). On this approach, the roles of Source and Goal apply to events beyond the spatial (motion) domain. As a result, this and other proposals within this class can naturally accommodate – and, in fact, lead one to expect – similarities in the mental properties of motion and transfer events (see Baker, 1996; Clark & Carpenter, 1989; Goldberg, 1995; Harley, 2002; Pylkkänen, 2008, among others).

Other researchers posit distinct representations for transfer and motion events (Bresnan & Kanerva, 1989; Pinker, 1989; Rappaport Hovav & Levin, 2008; see Rissman & Majid, 2019). These theorists typically reserve the term Goal for entities at the endpoints of motion events and use the term Recipient for the entity receiving the affected (transferred) object in transfer events (see Levin & Rappaport Hovav, 2005). Proponents of these theories point out that spatial Goals differ from Recipients in several respects. For instance, Recipients are typically animate, but spatial Goals do not have to be. Furthermore, spatial Goals are normally reached in motion events (cf. The butterfly flew to the tree) but Recipients do not necessarily end up in possession of the transferred object (one can say Harry gave an apple to Ruby but she didn’t take it). And, even though some kinds of transfer of possession verbs, such as throw and send in English, have been argued to have both a transfer and a caused motion interpretation (and allow either a Recipient or a spatial Goal, respectively), give-type verbs are assumed to have only transfer meanings and, thus, license Recipients exclusively (Rappaport Hovav & Levin, 2008). This last case of events, thus, offers the clearest case where the predictions of the two classes of theories differ.

1.2 Transfer events and the scope of the Goal bias

In the current article, we examine the linguistic encoding of transfer events to test whether these events include a Source-Goal structure, as proposed by some theories in our earlier discussion, but not others. As mentioned already, such events are expected to contain implicit Goals and Sources according to some views (Baker, 1996; Clark & Carpenter, 1989; Goldberg, 1995; Harley, 2002; Jackendoff, 1983, 1990; Pylkkänen, 2008; Talmy, 2000), but not others (Bresnan & Kanerva, 1989; Pinker, 1989; Rappaport Hovav & Levin, 2008). It follows that only the first, but not the second, class of theories expects the linguistic encoding of transfer of possession to be characterized by a signature role asymmetry between Sources and Goals, known within the motion domain as the Goal bias.

The Goal bias refers to the fact that, when people are asked to describe a motion event (e.g., a butterfly flying from the flower to the tree), they tend to mention the Goal (“to a flower”) more often than the Source (“from a tree”) (Chen et al., 2023, 2024; Do et al., 2020; Lakusta & Landau, 2005, 2012; Papafragou, 2010; Regier & Zheng, 2007). This asymmetry is found in both adults and children (Lakusta & Landau, 2012; Lakusta et al., 2016; Papafragou, 2010), is corroborated in naturalistic corpora (Stefanowitsch & Rohlde, 2004), and extends to typologically and modally different languages (e.g., Johanson et al., 2019; Regier & Zheng, 2007; Zheng & Goldin-Meadow, 2002).

The asymmetry between Sources and Goals in language production seems to have conceptual roots, as evidenced by memory tasks. After viewing similar motion events, adults (e.g., Chen et al., 2023, 2024; Do et al., 2020; Lakusta & Landau, 2012; Papafragou, 2010; Regier & Zheng, 2007) and prelinguistic infants (Lakusta & Carey, 2015; Lakusta & DiFabrizio, 2017; Lakusta et al., 2007) are more sensitive to changes to Goals, compared to changes to Sources. These findings suggest that the Goal bias in language reflects a more fundamental aspect of how humans process and remember motion events.

At present, much empirical evidence provides initial support for extending the Goal-Source structure, including the Goal bias, to the encoding of transfer events. In linguistic typology, languages encode Goal-path transfers (e.g., giving) and Source-path transfers (e.g., taking) using different grammatical devices, despite their parallel schematic structures. Cross-linguistic studies have shown that, across languages, transfer events involving Goal paths are more widely encoded in three-place predicates – with the start-state possessor, the transferred object, and the end-state possessor each being realized as one argument of the verb – compared to those involving Source paths (Kittilä, 2006; Margetts & Austin, 2007; Newman, 2005). This bias extends beyond linguistic structure to language production and comprehension. Analyses of sentences describing transfer action in natural corpora have shown that speakers are more likely to re-mention Goals, compared to Sources, in a consecutive sentence (Arnold, 2001). In turn, when asked to identify the referent of an ambiguous pronoun in a previous sentence depicting transfer events, adult comprehenders more readily associate an ambiguous pronoun with the sentence object when that object is a Goal (e.g., Ana sent a text to Liz and then she took a screenshot), than when it is a Source (e.g., Ana received a text from Liz and then she took a screenshot) (Johnson & Arnold, 2023; Langlois et al., 2023).

Elicitation studies in a laboratory setting provide some of the most direct evidence for the presence of a Goal bias in the linguistic encoding of transfer events, as the perceptual and discourse properties of the event components accessible to the speakers are more controlled. In one study, when describing the same videotaped transfer event, participants preferred describing it as a Giving event (e.g., “He gave the flowers to the woman”) rather than a Receiving event (e.g., “She received the flowers from the man”) and included Goal-path prepositional phrases (e.g., “to the woman”) more frequently than Source-path phrases (e.g., “from the man”) (Lakusta & Landau, 2005). Similarly, when shown a video depicting an elephant handing a ball to a rabbit, both adults and preschoolers preferentially described the scene with the verb give, rather than take/receive (Fisher et al., 1994).

Further research on early action role representation suggests a potential conceptual basis for a Goal bias in transfer events. Studies consistently show that a three-participant schema is robust and early for Giving events. For instance, infants react to object removal in Giving events, but not in hugging events, suggesting they recognize the object’s essential role in transfer events (Gordon, 2003). Additionally, 10.5- and 12-month-olds can distinguish between role reversal (when a Giver becomes a Taker) and direction reversal (Schöppner et al., 2006). However, such early sensitivity to the three-participant event structure of Giving does not seem to hold true for Taking events (e.g., Tatone et al., 2015; Tatone et al., 2021). For instance, infants represented Giving events with a three-participant schema, but failed to do so for Taking events: after viewing animated abstract shapes, infants distinguished Giving from Disposing, but not Taking from Acquiring (Tatone et al., 2015). This suggests that, while the presence of a receiver (Goal) in Giving events signals a distinct event type, the presence of the deprived party (Source) in Taking events does not.

However, other considerations cast doubt on the hypothesis that spatial Sources and Goals, as well as the Goal bias, straightforwardly extend to the representation of transfer events. First, recent evidence challenges the presumed similarity between spatial Goals and non-spatial Goals, such as Recipients in transfer events: Ziegler and Snedeker (2018) demonstrated that sentences with spatial Goals (The boy sprayed water on the plant) failed to prime sentences with Recipients (The woman fed the strawberry to the goose), and vice versa. Cross-linguistic studies further support the conclusion that spatial Goals and Recipients should be kept separate: while the two kinds of roles show equally robust clustering across languages, there is no evidence suggesting that Goals are more fundamental or more likely to extend to Recipients than vice versa. Similarly, there is no evidence that Goals emerge earlier than Recipients in the history of different languages (e.g., Bickel et al., 2014; Hartmann et al., 2014; Haspelmath, 2003; Heine, 1990; Malchukov et al., 2010; see Rissman & Majid, 2019, for a review). These findings suggest that people’s representations of spatial Goals and metaphorical Goals (such as Recipients) might be fundamentally distinct, indicating that the Goal bias might not automatically extend to transfer events. Instead, the early robust priority given to Recipients might stem from a separate cognitive system, possibly through a schema representation of socially significant actions.

1.3 Open issues

Currently, it seems clear that we need to broaden the available empirical evidence to assess the extent to which transfer events have a Source-Goal structure (including an asymmetric representation of the two roles). In particular, new empirical data need to address two issues. First, as mentioned already, the two participants in transfer events can also be construed as Agents and Patients (in addition to Sources and Goals). We know that Agents consistently occupy more prominent syntactic positions than Patients (Baker, 1997; Jackendoff, 1990; Levin & Rappaport Hovav, 2005; Pinker, 1989). In language production studies, speakers’ descriptions of events have been found to follow this asymmetry (ünal et al., 2024; Wilson et al., 2014). Furthermore, this Agent bias also sometimes occurs in nonlinguistic conceptual processing of events. For instance, in frame-by-frame self-paced reading of comic book panels, readers spent more time looking at the Agent than the Patient (Cohn & Paczynski, 2013; cf. also ünal et al., 2024).

This well-established Agent bias affects our understanding of the apparent Goal bias in transfer events. While the findings reviewed earlier can be considered as initial evidence for a Goal bias in the linguistic encoding of transfer events, they could be alternatively explained by an Agent bias, namely, the tendency to map the Agent onto the Subject (Dowty, 1991). Specifically, previous studies that examined the Goal bias in transfer events (Fisher et al., 1994; Lakusta & Landau, 2005) looked at the different perspectives taken when describing the same event – e.g., a man giving the flower to the woman. In these scenarios, the Giver was inherently more agentive (i.e., had more active postures, initiated action; Hafri et al., 2013) than the Receiver. Therefore, it is possible that the preference for describing the event as Giving would stem from the fact that people tend to attend to the Givers first and select them as the subject of the sentence, rather than from a bias for a Goal-path perspective.

More generally, if the Agent bias affects transfer event descriptions, we would expect Agents (Givers or Takers) to be mentioned more frequently and more robustly encoded than Patients (Givees or Takees). Notice that the Agent-Patient role assignment cross-cuts the Source-Goal assignment: in Giving events, the Giver is both the Agent and the Source, while the Givee functions as both the Patient and the Goal. However, in a Taking event, the Taker is both the Agent and the Goal, while the Takee is both the Patient and the Source. It remains an open empirical question whether the Goal bias would emerge when Agentivity of the event participants is controlled for.

Second, even if transfer events are characterized by a Source-Goal structure and induce a Goal bias, a critical question is whether that Goal bias would generalize across speakers of different languages, given cross-linguistic variation in transfer event encoding. While the Goal bias in the motion domain has been documented across languages (Johanson et al., 2019; Regier & Zheng, 2007; Zheng & Goldin-Meadow, 2002), experimental studies of Source and Goal encoding in transfer events have been limited to English speakers.

Cross-linguistic research shows that languages typically have more numerous (Svorou, 1994) and semantically richer (Kabata, 2013) devices for encoding Goals compared to Sources. However, languages vary in their morphosyntactic capacity to encode Sources in transfer events, with some languages offering more robust Source-encoding options. Mandarin Chinese exemplifies this linguistic variation: unlike many other languages, it allows Sources in transfer events to be encoded not only as prepositional phrase adjuncts but also as verbal arguments, suggesting a more privileged status for Source encoding.

English permits certain verbs in Double Object Constructions (DOC), where indirect objects can be incorporated into the verb’s argument structure (Gropen et al., 1989). For transfer of possession events, speakers can encode Recipients either through a Prepositional Phrase (1a) or as a direct argument following the verb (1b):

However, English does not allow the incorporation of Sources as indirect objects in its argument structure. For example, as shown in the contrast between (1c) and (1d), the Source can only be introduced by a prepositional phrase, but not as an indirect object in a DOC. Crucially, a Source DOC is allowed in Chinese (Huang, 2007). As shown in (2), the Source can either be introduced by the preposition cong or directly follow the verb in obtaining events (e.g. na ‘take’, tou ‘steal’, mai ‘buy’).

This additional morphosyntactic option for encoding Sources in transfer events raises important empirical questions: How frequently do Chinese speakers employ this construction when describing Taking events, and does the availability of a structure promoting the Source to a core argument status increase overall Source encoding?

1.4 Current study

Our study investigates the linguistic representations of transfer events across languages, with the aim of addressing the open issues raised in 1.3. Within the transfer class, we focus on events of Giving and Taking with animate (human) participants for which different theories make maximally different predictions (see the end of 1.1). We examine the following key questions. First, we ask whether transfer of possession events, long characterized by some linguistic theories as containing Source and Goal paths, exhibit the linguistic Goal bias observed in simple motion events (Experiments 1 and 2). Importantly, we seek to isolate effects of the Goal bias during language production from agentivity effects. As mentioned already, previous evidence for a linguistic Goal bias in transfer events (favoring Giving over Receiving descriptions) is potentially confounded with the tendency to map Agents to Subject position. The preference for Giving descriptions might reflect speakers’ tendency to begin sentences with Agents (who happen to be Givers), rather than a true Goal-path perspective. Notably absent from prior research is the systematic study of Taking events, where the Agent-first preference actually yields a Source perspective (e.g., She took an apple from him rather than He lost an apple to her). In Experiments 1 and 2, we examine descriptions of matched Giving events (Goal = Patient, Source = Agent) and Taking events (Goal = Agent, Source = Patient), treating Source-Goal and Agent-Patient dynamics as independent variables. This design allows us to disentangle any Goal bias from the established Agent bias.

Second, we explore cross-linguistic differences in transfer event descriptions by testing both English (Experiment 1) and Mandarin Chinese speakers (Experiment 2). Mandarin Chinese uniquely allows the Source to be encoded as a verbal argument, providing an opportunity to examine how this additional morphosyntactic option influences event descriptions. If the extent to which the Goal bias is manifested is influenced by the availability of linguistic encoding devices, we would expect it to be weaker or absent for Mandarin speakers. If the bias stems from universal processes in message planning, both language groups should show similar patterns. Together, our experiments bear on theories of thematic roles and event representation as the foundation for language production.

2.1 Methods

2.1.2 Materials

We constructed videos consisting of actions carried out by four characters that were described to participants as forming two pairs of twins. The four characters were played by two actresses. One actress played Ruby (when in a red hoodie) and Grace (when in a green dress). The other actress played Harry (with a hat) and Mitch (with a mustache).

Our critical stimuli were 12 pairs of video clips, each pair depicting a Giving and a Taking event with the same object. These 12 pairs were generated by crossing the all four possible unique character pairings (Harry-Ruby, Harry-Grace, Mitch-Ruby, and Mitch-Grace) with two role assignments (which character served as Agent vs. Patient). As a result, in these critical events, all four characters appeared equally often (6 times each) and with equal chances to appear as Agent or Patient (3 times each). Each character pair also appeared the same number of times. In Giving events, the Agent gave the object to the Patient (Figure 1a). In Taking events, the Agent took the same object from the Patient (Figure 1b). The Patient always performed only minimal actions (i.e., raising hands, looking at the transferred object) and was otherwise static. Each clip lasted two to three seconds.

Figure 1: (a) Sample frame of a Giving event (‘Harry gave an apple to Ruby’). (b) Sample frame of the corresponding Taking event (‘Harry took an apple from Ruby’).

In addition to the critical events, we also created 3 types of Fillers, which did not involve transfer of possession. There were 6 video clips depicting one-participant events (Figure 2a), 6 clips depicting events where one participant was a complete bystander (Figure 2b), and 6 clips depicting joint-action events (Figure 2c). We included a variety of event-participant numbers and dynamics to reassure participants that, during the description task, mentioning only one character in the event was acceptable.

Figure 2: (a) Sample first frame of a one-participant event (‘Harry was opening a package’). (b) Sample first frame of a bystander event (‘Harry was eating an orange’). (c) Sample first frame of a joint-action event (‘Mitch and Grace were folding a blanket’).

In order to compare the linguistic encoding of Giving vs. Taking events within participants, we put the Giving and the Taking version of each event (e.g., the events represented in Figures 1a and 1b) in two separate lists. Each participant was randomly assigned to one experimental list. Therefore, for each video pair, each participant either saw a Giving version or a Taking version. Importantly, each list was organized such that half of the critical trials were Giving events and the other half were Taking events. The filler events, as well as their order among the critical events in the two lists, were identical (see the Appendix for the complete list of critical events). Therefore, participants in the two lists saw an identical set of stimuli, except for the Giving-Taking reversal.

2.2 Results and discussion

2.2.1 Characters mentioned

As can be seen in Figure 3a, participants were highly likely to mention all characters. Participants always mentioned the Agent of each event (i.e., the Source in Giving events and the Goal in Taking events), with no exception. They also always mentioned the Patient, including it 100% of the time when it was the Goal of a Giving event and 95% of the time when it was the Source of a Taking event.

Figure 3: (a) English speakers’ mean proportion of mentioning the Goal/Source character in their linguistic description. (b) English speakers’ mean proportion of encoding the Goal/Source character as a true (formal) Goal/Source in their linguistic description.

With respect to our planned analysis, since Agents were 100% mentioned, to avoid singularity issues, we did not include Agency and its interaction with Role in the model.² In other words, we only compared Sources and Goals mentioned when they were both Patients – whether the same character was more likely to be mentioned when she was given something, compared to when she was deprived of something. Despite the overall high rates of mention, Source characters were still significantly less likely to be mentioned than Goal characters (Est = –2.465, SE = 1.48, p = 0.024). As demonstrated by Figure 3a, the difference between Source and Goal character mention emerged specifically in Patient mentions: while speakers consistently mentioned Agents in both Giving and Taking events and always mentioned Patient Goals in Giving events, they sometimes omitted Patient Sources in Taking events.

3.1 Methods

Experiment 2 was identical to Experiment 1, except that the instructions for the experiment were in Mandarin Chinese. The participants were twenty native speakers of Chinese recruited from the University of Pennsylvania’s subject pool.³ Just as in Experiment 1, we excluded from analysis trials in which the description did not match the video (n = 24).

3.2 Results and discussion

3.2.1 Characters mentioned

The pattern of character mentions paralleled that of Experiment 1. As shown in Figure 4a, Chinese speakers, like English speakers, mentioned Agents more frequently than Patients. Since Agent mentions were, again, categorical, we excluded Agency and its interaction with Role from the model to avoid singularity issues, focusing instead on Patients. The model revealed that Chinese speakers mentioned Patient Goals significantly more often than Patient Sources (Est = 1.399, SE = 0.522, p = 0.007).

Figure 4: (a) Chinese speakers’ mean proportion of mentioning the Goal/Source character in their linguistic description. (b) Chinese speakers’ mean proportion of encoding the Goal/Source character as a true (formal) Goal/Source in their linguistic description.

An additional logistic mixed-effects model predicting character mentions based on Role and Language (English vs. Chinese, sum-coded) confirmed that the Goal bias persisted across both language groups (Est = 1.761, SE = 0.613, p = 0.004). While Chinese speakers showed a tendency to omit Patients more frequently than English speakers, this difference was not statistically significant (Est = 1.246, SE = 0.70, p = 0.076).⁴

4.1 A Goal bias in the linguistic encoding of transfer events

Our results reveal a consistent Goal bias in descriptions of transfer events: participants were more likely to mention a character as a Givee than as a Takee (Experiments 1 and 2). We also found an unequivocal Agent bias, with Agents rarely omitted, consistent with previous findings in causative events (Baker, 1997; Cohn & Paczynski, 2013; Ünal et al., 2024) and the tendency to map Agents onto subjects (Dowty, 1991). By comparing minimally-paired Giving and Taking events, and systematically considering the Agent-Patient role assignments in parallel with Source-Goal assignments, we were able to uniquely demonstrate that the linguistic Goal bias in transfer events exists independently of Agent-to-subject mapping tendencies.

Taken together, our linguistic findings extend the scope of the Goal bias from motion events (Chen et al., 2023, 2024; Do et al., 2020; Papafragou, 2010) to transfer events, complementing previous research showing preferences for Goal/Giving perspectives (Fisher et al., 1994; Lakusta & Landau, 2005) and Goal-oriented reference patterns (Arnold, 2001; Johnson & Arnold, 2023). Particularly, whereas discourse studies document a Goal bias in re-mention once Source/Goal information has been linguistically introduced, our data show a parallel bias during initial event descriptions, when Source/Goal roles are only non-linguistically given and must be selected for encoding. These findings are consistent with theoretical accounts that model transfer events as containing implicit Goal and Source paths (e.g., Jackendoff, 1983, 1990). At the same time, they do not rule out the possibility that transfer events invoke a distinct Recipient role, such that the observed production asymmetry reflects a Recipient bias, rather than the same Goal bias documented for spatial events. These biases could, nonetheless, be related, given the semantic overlap between Goals and Recipients. For simplicity, we adopt the Goal/Source path account as our working explanation.

The attentive reader may wonder whether this conclusion accounts for all available evidence beyond our own studies. Recall that Ziegler and Snedeker (2018) demonstrated that sentences with spatial Goals failed to prime sentences with Recipients, and vice versa. Furthermore, there is no evidence suggesting that Goals are more fundamental or more likely to extend to Recipients than vice versa in cross-linguistic data, and no evidence that Goals emerge earlier than Recipients in the history of different languages (e.g., Bickel et al., 2014; Hartmann et al., 2014; Haspelmath, 2003; Heine, 1990; Malchukov et al., 2010; see Rissman & Majid, 2019, for a review). How can we reconcile these findings, which suggest separate and parallel representations of Recipients and Goals, with our own findings (and other prior work compatible with our data) suggesting an underlying Goal path structure in transfer events?

There are different options for arriving at a theoretical synthesis. One possibility is to reframe the relation between transfer and motion events to avoid the commitment that the observed Source-Goal structure is primarily and essentially spatial, such that its application to the transfer domain is a case of metaphorical extension. In some parts of his work, Jackendoff (1983, p. 210) seems to propose this option: “I am inclined to think of thematic structure not as spatial metaphor but as an abstract organization that can be applied with suitable specialization to any field. If there is any primacy to the spatial field, it is because this field is so strongly supported by nonlinguistic cognition.” On this view, the spatial domain would, then, simply be privileged in terms of the application of this broader organizational principle. Even though currently underspecified, this view could explain the lack of primacy of spatial Goals over Recipients in both cross-linguistic data and the history of different languages (Bickel et al., 2014; Hartmann et al., 2014; Haspelmath, 2003; Heine, 1990; Malchukov et al., 2010), as well as the absence of priming effects across spatial Goals and Recipients (Ziegler & Snedeker, 2018). It would also be consistent with the independent effects of Agents in our own data. This possibility also seems compatible with the proposal that infants process Giving and Taking events differently because of distinct schemas for understanding social interactions (Tatone et al., 2015).

A second, somewhat related, possibility is that event roles in both language and cognition can be located and defined at different levels of event architecture. For instance, the roles involved in motion and transfer events may be representationally distinct (as proposed, for instance, in Rappaport Hovav & Levin, 2008), but connected at some higher level of abstraction. There are various proposals across different linguistic frameworks for hierarchically organizing thematic structure (Dowty, 1991; Van Valin & LaPolla, 1997). For instance, in Dowty’s (1991) system, there are verb-specific roles, more abstract roles – such as Agent and Patient – that cross-cut several verb classes, and, further, prototypical roles (proto-Agents and proto-Patients). Even if this organization exists in the specific cases discussed in our article, however, it does not surface in priming studies of spatial Goals and Recipients (for discussion, see Ziegler & Snedeker, 2018). Additional evidence is needed to arbitrate more decisively between the two accounts, for example, including directly testing cross-structural priming between spatial Goals and transfer Recipients (and vice versa) and individual-difference correlations in Goal/Source mention across spatial and transfer domains.

4.2 The cross-linguistic encoding of transfer events

An important aspect of our findings was that the Goal bias in production appeared consistently across both English (Experiment 1) and Mandarin Chinese (Experiment 2) speakers, revealing a striking pattern that persisted despite significant cross-linguistic differences. While Mandarin Chinese provides speakers with more prominent syntactic positions for Sources, which allow them to encode Sources as incorporated verbal arguments rather than just prepositional adjuncts, this additional linguistic resource did not lead to more robust Source encoding: Mandarin speakers neither mentioned Sources more frequently nor encoded them as true Sources more often than English speakers. Instead, they were slightly more likely to opt to indicate Sources indirectly through possessor mentions.

At the same time, the cross-linguistic similarity in omission patterns is not straightforwardly attributable to what each language “allows”. Both languages provide constructions that can, in principle, express the relevant arguments. For example, English take readily licenses a Source prepositional phrase (e.g., take X from Y), and the discourse context in our elicitation task, which was highly constrained and repetitive with constant participants and a salient Recipient, could plausibly license reduced give descriptions that omit the Givee (e.g., Harry gave a hat). Nonetheless, even when Source expressions were fully licensed, speakers often omitted the Source in take descriptions, whereas Recipient omission in descriptions of giving was unattested in English and rare in Mandarin. While it is open to debate whether the Recipient is grammatically required in both languages, Recipient expression with give is arguably more conventionalized and grammaticalized as the canonical transfer frame than Source expression is with take, and this difference in conventionalization may contribute to the observed asymmetry. Taken together, these considerations suggest that the observed pattern reflects an interplay between language-specific grammatical affordances and speakers’ production choices in the task context, rather than being reducible to grammatical possibility alone.

The cross-linguistic stability in the Source-Goal asymmetry echoes similar findings in the broader literature. For instance, Johanson et al. (2019) documented the persistence of the Goal bias across languages with varying motion encoding strategies, suggesting a deeper pattern that transcends surface-level linguistic differences. If the Goal bias was purely a product of linguistic constraints or conventions, we might expect it to vary more substantially across languages with different encoding options. Instead, its preservation across users of diverse linguistic systems suggests that this asymmetry originates, at least partially, from shared, potentially universal processes that shape how humans design event messages in communication. At the same time, the present data do not allow us to fully disentangle conceptual contributions from language-specific grammatical and discourse constraints, and future work will be needed to more directly arbitrate among these sources.

4.3 Parallel asymmetry in nonlinguistic representation of transfer?

The robust production asymmetry observed in both English and Mandarin is consistent with the possibility that speakers’ linguistic choices reflect underlying nonlinguistic representations of transfer events. At the same time, this inference is necessarily tentative. Establishing a parallel asymmetry in nonlinguistic encoding requires direct evidence from tasks that probe memory and/or attention to event participants.

There are reasons to expect such a parallel. In particular, patterns in event description often covary with memory for event components (e.g., Fausey & Boroditsky, 2011; see Papafragou & Grigoroglou, 2019, for a review). In the spatial domain, a Goal advantage has been reported repeatedly in memory for Goals versus Sources (e.g., Chen et al., 2023, 2024; Do et al., 2020; Lakusta & Landau, 2012; Papafragou, 2010; Regier & Zheng, 2007). Especially relevant are findings suggesting that sensitivity to three-participant structure is more robust for Giving events than for Taking events in infants and adults (e.g., Tatone et al., 2015, 2021; Yin et al., 2022).

However, the link between linguistic asymmetries and nonlinguistic representations may be more complex than a one-to-one mapping. Conceptual biases may shape production while failing to surface in nonlinguistic tasks, depending on event structure, participant properties, and task demands. Prior work has documented dissociations between linguistic and memory asymmetries, including for motion events with inanimate or non-intentional Agents (Lakusta & Carey, 2015; Lakusta & Landau, 2012) and for more complex causal events (Ünal et al., 2021, 2024). For example, Lakusta and Landau (2012) found that although speakers maintained a Goal bias in descriptions of motion events, memory for Sources versus Goals in those same events did not show a corresponding asymmetry. Animacy is especially important when considering transfer events, which (unlike previously studied motion events) depict interactions between two humans. Because animacy strongly shapes event apprehension, the prominence of multiple intentional agents could reduce or alter the expression of Goal and Agent biases in memory, even when production remains Goal-oriented. Methodological choices also matter. Chen et al. (2024) show that Goal advantages in motion events are reliably observed in change-detection tasks, the paradigm used in much of the prior memory literature, but can be attenuated or absent in forced-choice paradigms, where aided retrieval may boost memory for otherwise weakly encoded components, such as Sources. This suggests that null effects in memory need not imply the absence of conceptual asymmetry; they may instead reflect the demands and affordances of the particular task.

Future work is, therefore, needed to assess nonlinguistic role representations in transfer, using converging methods, including direct comparisons between same–different and forced-choice paradigms and manipulations of Givee/Takee animacy. We are currently conducting follow-up studies along these lines. Regardless of the outcome of the future work, the implications for the language–cognition interface will be informative. If a parallel memory asymmetry emerges, it would add to evidence that the relative prominence of event components often aligns across language and event cognition (e.g., Lakusta & Landau, 2005, 2012; Papafragou, 2010; Wilson et al., 2014; Ünal et al., 2021, 2024). If no memory asymmetry emerges, this would provide another case in which linguistic and memory asymmetries diverge, reinforcing the view that language is not an exhaustive representation of conceptual representational resources (e.g., Gleitman & Papafragou, 2005, 2013; Ünal & Papafragou, 2016). Such a pattern would also align with Perkins et al. (2025), who show that although infants represent Taking events in terms of distinct participant roles, they readily accept linguistic descriptions that omit the Takee (e.g., She is pimming the truck). In this scenario, adult production asymmetries may reflect how language-specific argument structure interfaces with event representations, rather than requiring a one-to-one correspondence with memory. Either way, the broader goal is to develop a more fine-grained account of how language production accesses, and sometimes departs from, the nonlinguistic conceptual structure of transfer events.

4.4 Conclusion

In sum, across two experiments with English and Mandarin speakers, we examined how transfer events are linguistically encoded. In both groups, participants’ descriptions showed a Goal bias that mirrors the well-documented asymmetry in spatial event encoding. This convergence has two key implications. First, it provides empirical support for theories positing that Source/Goal path structure generalizes to the linguistic encoding of transfer events. Second, the close alignment between English and Mandarin, despite substantial differences in their lexical and morphosyntactic resources, suggests that the Goal bias in production reflects shared, potentially universal mechanisms shaping how events are mapped onto language.