Language has long been argued to reflect core aspects of human perception and cognition (Slobin, 1982; Jackendoff, 1996, Levelt, 1989, Pinker, 2013 Gleitman & Papafragou, 2012). In linguistic typology, patterns that align with perceptual and conceptual preferences are more widely attested in world languages than those that do not (Culbertson, 2012, Dixon, 1994, Croft, 2002, among others). This alignment is perhaps most apparent in the way we perceive and communicate about events - segmented and structured units of our continuous and dynamic experience. Oftentimes, the relative prominence of different components of an event - for example, the agent or patient of an action, the goal of an action or the instrument to achieve the action - points to parallels between language and non-linguistic event cognition, with some components being more robustly represented compared to others in both domains (e.g., Wilson, Ünal, Trueswell, & Papafragou, 2014; cf. Lakusta and Landau, 2005, Lakusta and Landau, 2012, Papafragou, 2010 Hayward & Tarr, 1995; Munnich, Landau, & Dosher, 2001; Landau & Hoffman, 2005). For instance, just as instruments used to achieve an action (e.g., hit the ball with a bat) are often expressed in optional adjunct positions instead of as obligatory arguments across languages, their visual representations are also less robustly remembered than those of other event components (e.g., the ball) (Ünal, Richards, Trueswell, & Papafragou, 2021).

Despite these findings, there is no consensus about the precise shape of event representation and its interface with linguistic processes. For instance, it remains unclear what exactly it means for an event component to be more robust (or fragile) in both non-linguistic event cognition and language. In the studies cited above, the fragility of an event component was often indexed by less frequent or more peripheral mention in language and/or poorer performance in memory tasks, but no explicit proposal about the nature of such fragility has been made. Relatedly, the connection between the granularity of linguistic construal and that of non-linguistic event representation has rarely been examined since the two kinds of representations have typically been tested separately. Currently, it is an open question whether explicit linguistic encoding might not simply reflect but also affect an existing asymmetry between different components in event cognition. These questions are crucial to understanding how event cognition and language come together and how they diverge (see Papafragou & Grigoroglou, 2019, for a recent review).

In this article, we aim to further our understanding of non-linguistic event cognition and language by addressing the relative robustness of individual event components across the two domains. We focus on a particularly clear test case, the asymmetrical representation of Sources and Goals in motion events, and explore its roots, scope and relation to linguistic encoding.

According to a now extensive literature, Goals (end points) are prioritized over Sources (starting points) in the representation of motion events. As shown by both production studies in a laboratory setting as well as analyses of naturalistic corpora, when describing simple motion events such as a fairy going from a tree to a flower, people mention the Goal (“to a flower”) more often than the Source (“from a tree”) (Do et al., 2020, Lakusta and Landau, 2005, Lakusta and Landau, 2012, Papafragou, 2010, Regier and Zheng, 2007, Stefanowitsch and Rohlde, 2004). Such a linguistic asymmetry has been observed for adults as well as children (Lakusta and Landau, 2012, Lakusta et al., 2016, Papafragou, 2010 Lakusta et al., 2017), in different subtypes of motion events involving Goal and Source paths (e.g., Change of State: Lakusta & Landau, 2005; Attachment/Detachment: Regier & Zheng, 2007, Transfer of possession, Lakusta and Landau, 2005, Chen et al., 2022) as well as in typologically (even modally) different languages (e.g., Regier and Zheng, 2007, Johanson et al., 2019, Zheng and Goldin-Meadow, 2002, Ihara and Fujita, 2000 Lakusta, Yoshida, Landau, & Smith, 2006). Further, this linguistic asymmetry also manifests itself in typological patterns: cross-linguistically, languages have semantically finer-grained and syntactically more prominent devices to encode Goals than Sources (Fillmore, 2006, Ihara and Fujita, 2000, Jackendoff, 1985, Nikitina, 2009, Nam, 2004, Markovskaya, 2006, Regier and Zheng, 2007 Johanson et al., 2019) and children’s earlier acquisition Goal-encoding language than Source-encoding language (Freeman et al., 1981, Bowerman et al., 1995 Pléh, 2010; Landau & Zukowski, 2018).

The asymmetry between Sources and Goals in language seems to have conceptual roots (Lakusta and Landau, 2005, Lakusta and Landau, 2012, Papafragou, 2010, Regier, 1996, Regier and Zheng, 2007). Much evidence suggests that Goals are more accurately encoded in memory than Sources of motion events: adults and 4–5 year olds detect Goal-changes more accurately than Source-changes (e.g., Papafragou, 2010, Lakusta and Landau, 2012, Regier and Zheng, 2007, Do et al., 2020). Even prelinguistic infants prefer to look at an event with a Goal (e.g., a duck going into a bowl) than one with a Source (e.g., a duck going out of a bowl) (Lakusta & DiFrabrizio, 2017) and look longer to a test event where a Goal object has changed from an earlier, familiarization event compared to a test event where the Source object has changed (Lakusta and Carey, 2015, Lakusta et al., 2007).

This parallel asymmetry is often taken as a golden case for a homology between language and non-linguistic cognition: Source, the event role that is less likely to be present in linguistic description, is also less robustly encoded in non-linguistic representation. However, two major issues with regard to the nature of the representation of motion Goals and Sources remain unclear. First, previous literature interprets these findings as demonstrating that Source is less robustly represented as people encode motion events. However, what it means for Source to be less robustly represented is left open. Do people sometimes fail to encode Source entirely and instead represent the motion event as one with only a Goal (e.g., the fairy went to the flower)? This is what linguistic descriptions of events like this might suggest but we lack evidence that the non-linguistic encoding matches this pattern. Alternatively, is Source always encoded in memory, but with a weaker strength or with coarser granularity? Understanding how Source, a well-documented “fragile” component, is represented will shed light on the nature of robustness in event cognition in general.

One way to approach the nature of the Source fragility is to ask whether Source information escapes memory altogether, or is present and can be revived when the retrieval context calls for it. Prior work has not been able to address this question because, to the best of our knowledge, it has always tested recognition memory of Source and Goal via a single test format - the Same-Different task (Do et al., 2020, Lakusta et al., 2007, Papafragou, 2010, Regier and Zheng, 2007, Regier, 1996). In this task, participants view a set of motion videos and later compare new videos to their memory of the previous target event and make a decision (same or different). Surface failure to recognize a Source change in this task could represent unavailability of Source information in the representation, but could also mean that Source information is harder to access (Tulving & Pearlstone, 1966). Given that the same memory trace can be better retrieved when given a retrieval context better matching that of encoding (Godden and Baddeley, 1975, Tulving and Thomson, 1973), the asymmetrical robustness of Sources and Goals may not surface in the same way in different memory tasks where different cues are available.

A second question related to the stability and nature of the Source-Goal asymmetry is how Source-Goal representation might change as a result of linguistic encoding. Previous research suggests that the Goal bias in memory persists even when the experimental design explicitly inhibits linguistic encoding by introducing a verbal shadowing task as people view the visual stimuli (Lakusta & Landau, 2012). However, could the act of describing the event verbally or the act of hearing a linguistic description during encoding affect the Source-Goal asymmetry in memory at a later stage? After all, it has been demonstrated that linguistic labels can affect memory (e.g., Archambault et al., 1999, Feist and Gentner, 2007, Loftus, 1975, Loftus, 1979, Loftus and Palmer, 1974, Loftus et al., 1978, Lupyan, 2008). Within the domain of event cognition, hearing verbs that denote paths of motion before viewing motion events attenuated memory for manners of motion (Skordos et al., 2020). Other studies where participants both provided event descriptions and were tested on event memory have reported that memory of individual event components was stronger when these components were also mentioned (e.g., Do et al., 2020). However, relatively few studies have directly looked at how language production and comprehension modulate event cognition.

To address the two main questions outlined above, the current study tested the stability of the classic Source-Goal asymmetry across different memory tasks and after different levels of explicit linguistic engagement with the visual stimuli.

To examine whether Sources and Goals are consistently remembered asymmetrically, we probed memory for Source and Goal in the Same-different task and compared results to another widely-used recognition test format - the Forced-choice task. In this task, after participants have encoded items in memory, pairs of items - a target and a foil - appear at the same time during the memory test phase, and the participants are asked to select the target. It has long been recognized that, despite the wide use of the two tasks as comparable measures of recognition memory, the assessment of memory can function differently in the two test formats (Deffenbacher, Leu, & Brown, 1981; Bastin & Van der Linden, 2003; Aggleton & Shaw, 1996; O’Reilly & Rudy, 2000; Holdstock et al., 2002, Westerberg et al., 2006; Kroll et al., 2002, Bayley et al., 2008, Khoe et al., 2000 Mayes, Holdstock, Isaac, Hunkin, & Roberts, 2002).1 One critical point of difference between the two test formats is the availability of retrieval cues at test. One possibility is that the direct accessibility to the target at test might boost Goal and Source equally, preserving the Source-Goal asymmetry. Alternatively, the reinstatement of the target could benefit the component that was more weakly encoded (i.e., the Source) to a greater extent, pulling it to the same level of recognition with the more strongly encoded role (the Goal), thus eliminating the bias. The direct comparison of the two memory tasks could thus throw light onto the nature of the Source disadvantage: if Source indeed fails to be encoded at least some of the time, it would not be salvaged by aided retrieval in the Forced-Choice format. However, if it is characterized by a weaker overall trace or coarser granularity of encoding, Source might benefit from the richer retrieval cues in the Forced-Choice format.

To test whether these results would generalize across different contexts of linguistic encoding, we probed memory for Sources and Goals after people viewed events without explicit involvement of language (passive viewing), with the task of describing them (production), or with a verbal description about the events presented to them (comprehension). We also had two more specific aims. First, we wanted to ask whether a linguistic asymmetry in production or comprehension would directly align with the non-linguistic cognitive bias (such that, e.g., Sources that are more poorly remembered would also be less likely to be mentioned). Second, we wanted to probe the effect of language comprehension on the non-linguistic Source-Goal asymmetry. Since Source-only descriptions are unusual in listeners’ linguistic and perceptual experience (Fisher et al., 1994, Lakusta and Landau, 2005), and unexpected information is remembered better (Graesser, 2013, Shapiro and Fox, 2002), hearing a Source-only description might boost Source memory more robustly compared to the effect of hearing a Goal-only description on Goal memory, thus mitigating the asymmetry.

In sum, we conducted three experiments to compare memory for Sources and Goals, probed with either a Same-different task or a Forced-choice task, after participants passively viewed (Experiment 1), viewed and described (Experiment 2) or viewed and heard descriptions of (Experiment 3) the same set of motion events. In a further experiment, we considered and removed alternative interpretations of our findings (Experiment 4). Together, our data can inform theories of event representation and the language-cognition interface (especially those probing representational asymmetries among more and less prominent event components). Our investigation also makes a crucial methodological contribution by highlighting the need for convergent findings from different tasks probing language and memory across multiple contexts to construct valid theories connecting language to non-linguistic event cognition.