Prior experience with unlabeled actions promotes 3-year-old children's verb learning.

Identifying Verb Referents Is a Challenging Task

Verbs typically describe actions, and it is difficult for children to individuate actions in complex events ( Gentner, 1982 ). For example, 3-year-old children struggle to generalize verbs to events that show the referent actions performed by novel actors (e.g., Imai et al., 2008 ; Kersten & Smith, 2002 ), with novel objects (e.g., Imai et al., 2005 ), or with novel instruments (e.g., Behrend, 1990 ; Forbes & Farrar, 1993 , 1995 ). This indicates that children’s semantic representations of verbs include components of action events that are irrelevant to verb meaning (i.e., actors, objects, instruments). In other words, children map verbs to the combination of event components, for example, to a particular actor performing a particular action or a particular action carried out with a particular object (e.g., Imai et al., 2005 ). Thus, if we can help children to individuate action components in complex events, then this could help children to learn verbs with semantic representations that include only the relevant component for verb meaning ( Gentner, 2003 ). However, to achieve this, children need to segment complex events into the different event components (e.g., actors, objects, instruments, and crucially, actions). This study investigates three ways in which this could be achieved for novel intransitive verbs that describe manners of locomotion (actions) performed by adults (actors).

Multiple Exemplars of the Same Action May Facilitate Verb Learning

The first way to help children hone in on actions in complex events is to present them with multiple labeled exemplars that consistently show the components that are relevant to verb meaning, while varying the components that are irrelevant to verb meaning ( Childers, 2011 ; Haryu et al., 2011 ). For example, Childers (2011) taught 2.5-year-olds novel verbs while seeing the experimenter perform target action events (e.g., rolling a ball down a ramp into an opaque box so that the ball disappears from view) followed by either the repetition of these same labeled exemplar, labeled exemplars that repeated only the actions (e.g., rolling a ball down a curved tube), or labeled exemplars that repeated only the results (e.g., covering the ball with a piece of cloth so that it disappears from view). In the test phase, children were asked to enact the novel verb meanings with a set of objects that included the objects used in the target action events (e.g., a ramp), novel objects that could be used to enact the actions (e.g., a curved pipe), and novel objects that could be used to enact the results (e.g., an opaque bag). Children who saw similar labeled exemplars that repeated the actions were more likely to generalize the verbs to novel objects with which the same actions could be performed, and children who saw similar labeled exemplars that repeated the results were more likely to generalize the verbs to novel objects which led to the same results. In contrast, children who saw the repetition of the same labeled exemplar were conservative in generalizing the verbs as they were more likely to recreate the same event using the same objects. Thus, this study shows that when children are presented sequentially with multiple different labeled action exemplars, they can compare those exemplars and extract the consistent component that is shared between those exemplars, which is important for learning the meaning of that verb (i.e., manners, results). This ability to compare exemplars and extract relevant information facilitates children’s verb learning and generalization.

Previous research has shown that children can also learn verbs by integrating information from two different exemplars shown simultaneously ( Snape & Krott, 2018 ). Snape and Krott (2018) taught 3-year-old children novel verbs, while some children saw a single exemplar of an action performed on a novel object, some saw two exemplars simultaneously in which the same action was performed on two different objects, and some saw two identical exemplars simultaneously in which no aspect of the events varied. In all three conditions, each exemplar was always labeled. In a two-alternative forced-choice test, the children were asked to extend the newly learned verbs to one of two events: one that maintained the action but performed on a novel object versus one that maintained the object but performed a novel action. Only children who saw two different labeled exemplars of the same action simultaneously (i.e., when the same action was performed on two different objects) successfully generalized the newly learned verbs to novel events that maintained the actions. This suggests that simultaneously presented exemplars can support 3-year-old children’s verb learning, but only when the content of the exemplars varies (i.e., the action component that is relevant for verb meaning is kept consistent across exemplars, but components irrelevant to verb meaning vary).

Iconic Gestures That Encode Actions May Facilitate Verb Learning

The second way to help children to focus on actions in complex events is to highlight verb referents with iconic gestures (e.g., Goodrich & Hudson Kam, 2009 ; Mumford & Kita, 2014 ; Wakefield et al., 2018 ). Iconic gestures are hand movements that depict features of objects (e.g., shape) or actions (e.g., motion; McNeill, 1985 , 1992 ). The shape and motion of an iconic gesture and its meaning are linked through similarity (e.g., wiggling the index and middle fingers to depict a person walking). As such, iconic gestures can focus children’s attention on components of complex events that are important for verb meaning. For example, Mumford and Kita (2014) taught 3-year-old children novel verbs that could be interpreted as manner verbs (e.g., “to push”) or result verbs (e.g., “to break”). Children saw videos of an actor manipulating material/objects (e.g., sprinkling sand into a square shape on a table surface) with either iconic gestures that highlighted manner (e.g., depicting the manual action of sprinkling) or iconic gestures that highlighted the end-state of the scene (e.g., tracing the square shape that the sand formed) while the experimenter labeled each action event with a novel verb. Children were immediately asked to generalize each novel verb to one of two novel scenes in a two-alternative forced-choice task: one scene showed a manner verb interpretation (e.g., sprinkling powder into a triangle shape) and the other a result verb interpretation (e.g., placing pieces of paper to form a square ). Children who saw iconic gestures highlighting manner when the verbs were taught interpreted the verbs as manner verbs and children who saw iconic gestures highlighting end-state as result verbs. This suggests that iconic gestures can focus children’s attention on different components of complex events, and this influences children’s interpretation of novel verb meanings.

While there is abundant empirical evidence for the beneficial effect of iconic gesture on children’s word learning ( Goodrich & Hudson Kam, 2009 ; McGregor et al., 2009 ; Mumford & Kita, 2014 ; Wakefield et al., 2018 ), the mechanism for how iconic gesture facilitates word learning is unclear. One possibility is that iconic gestures that depict actions merely function as an extra action exemplar, much like the two simultaneous exemplars in the study by Snape and Krott (2018) . Another possibility is that iconic gesture goes beyond merely functioning as an extra exemplar because it schematizes action ( Aussems & Kita, 2019 , 2021 ; de Ruiter, 2000 ; Goldin-Meadow, 2015 ; Kita, 2000 ; Kita et al., 2017 ; Novack et al., 2014 ; Novack & Goldin-Meadow, 2017 ); that is, it provides children with a focused action representation. Iconic gesture is also a communicative signal, which prompts the recipient to search for a matching representation, triggering a top-down search for action.

Prior Experience With Unlabeled Actions

The third way to help children focus on actions in complex events during verb learning is to give children prior experience with unlabeled referent actions. This is the main hypothesis of the current study. The word learning studies discussed so far leave it open whether prior experience with unlabeled actions can promote verb learning. In previous multiple exemplar studies, actions were always labeled on each encounter (e.g., Childers, 2011 ; Haryu et al., 2011 ; Imai et al., 2005 ; Maguire et al., 2008 ; Mumford & Kita, 2014 ); therefore, it is not clear whether children integrate prior experience with unlabeled actions in their semantic representations of novel verbs when they first encounter these action labels.

In a naturalistic verb learning situation, it is plausible that children encounter a referent action (along with many other actions) before they hear the label for that action for the first time. For example, children may have encountered actions like galloping , slithering , leaping , and shrugging before hearing a label for these actions. This is because adults frequently describe specific action events to young children using general all-purpose verbs (e.g., to do , to go ), which are not tied to the specific actions in the events ( Pinker, 1989 ; Rice & Bode, 1993 ). For example, in the nursery rhyme “Itsy Bitsy Spider,” the spider went up the waterspout. Although the verb to crawl is a commonly used verb to refer to a spider’s movement, it does not appear in the nursery rhyme, though many adults would probably intuitively depict the spider’s crawling motion in gesture (i.e., using the hand to represent the spider’s body and the fingers to represent its long legs). Moreover, the age of acquisition of the verb to crawl seems to be quite late, close to 4 years of age ( Kuperman et al., 2012 ). This leaves room for children to gain experience with actions before hearing their specific verb labels, especially when iconic gesture is used to focus children’s attention on those actions. Thus, it is important to investigate whether and how children can take advantage of prior experience with unlabeled actions when they learn the labels for the actions at a later point in time. In the current study, we will emulate this understudied step of the verb learning process experimentally for the first time. In doing so, our study addresses a fundamentally different question than verb learning studies in which children were exposed to multiple labeled action exemplars (e.g., Childers, 2011 ; Haryu et al., 2011 ; Maguire et al., 2008 ). Such studies examined how linguistic representations of verbs change through encounters with multiple labeled exemplars of a referent action. However, our study investigates how nonlinguistic representations of actions can influence how children form initial linguistic representations of those actions when they are labeled with a novel verb.

More generally, most word learning studies to date have focused on isolated label-referent co-occurrences. But recent work by Smith and Yu and colleagues emphasizes that linguistic input is only a small part of input that children receive. In fact, a large amount of input simply involves visual experiences with referents, while label-referent co-occurrences are infrequent ( Clerkin et al., 2017 ; Suanda et al., 2019 ; Yu et al., 2019 ). Our study is in line with the idea that researchers should not just focus their efforts on children’s labeled experiences, which are infrequent, but also on their unlabeled (visual) experiences, which are plentiful. By investigating prior experience in combination with iconic gesture, our study does not only aim to answer the question of how children learn words but also how children make use of different types of visual input they receive.

Prior Unlabeled Experience Versus Delayed Labeling

Prior experience with unlabeled actions, which is the focus of the current study, differs from delayed labeling, in which children hear a label for an action immediately after the referent action has been demonstrated (e.g., Tomasello & Akhtar, 1995 ; Wakefield et al., 2018 ). The key feature of delayed labeling is that children do not see any other actions between seeing the referent action and hearing its label; thus, the referent action is the most plausible referent of the label. For example, previous research has shown that 2-year-old children can link a verb to its referent in delayed labeling situations in a verb learning task, in which a label is given immediately after the referent action is shown, and eye gaze cues direct children’s attention to the apparatus with which the action was performed ( Tomasello & Akhtar, 1995 ). In the study by Tomasello and Akhtar (1995) , a child and an experimenter took turns playing a merry-go-round game. In the following training phase, the experimenter modeled a novel target action with a novel object on the merry-go-round and readied the apparatus for the child’s turn. The experimenter then alternated her gaze between the child and the merry-go-round and provided a delayed language model (“It’s your turn now. Widget , Jason, widget. ”). A control group followed the same procedure except that the child’s turn was preceded by neutral language (“Now it’s your turn, Jason, it’s your turn.”). Children in the control group heard the novel word (e.g., widget ) for the first time in the comprehension test that followed the training phase. In the comprehension test, the experimenter set up the merry-go-round, three familiar objects, the novel object from the training phase, and a second novel object, and asked the children “Show me widget .”. Following this request, children in the experimental group generally performed the target action, whereas children in the control group indicated the objects. Thus, children could make use of delayed labeling (i.e., temporal adjacency cue) in combination with gaze alternation (i.e., nonverbal cue produced by an adult) in a verb learning task.

In contrast to delayed labeling, the key feature of prior experience with unlabeled actions is that children do see other actions between seeing the referent action and hearing its label. Thus, children cannot use temporal adjacency as a cue for linking a novel verb label to its referent action, and they will have to pick out the referent action out of many other actions from memory if they want to structurally align the referent action seen during labeling with the relevant action from prior experience.

Possible Mechanisms

We distinguished three possible mechanisms for how prior experience with unlabeled actions and iconic gestures could promote children’s verb learning. First, children may structurally align ( Gentner, 1982 , 2003 ; Gentner & Markman, 1997 ; Markman & Gentner, 1993 ) an unlabeled exemplar and a labeled exemplar of the same action for verb learning. This recall-event-and-align mechanism suggests that when children encounter an action exemplar and a novel verb label (a) children recall the relevant exemplar (i.e., of the same action) from prior experience; (b) children structurally align ( Gentner, 1982 ; 2003 ; Gentner & Markman, 1997 ; Markman & Gentner, 1993 ) the recalled unlabeled exemplar and the current labeled exemplar, which highlights the action as the shared component between exemplars; and (c) children interpret the highlighted action as the referent of the novel verb. This process requires children to be able to pick out relevant exemplars from memory.

Second, children may make use of prior experience with unlabeled actions in combination with iconic gestures that highlight the actions in the following way. This gesture-for-action-memory mechanism suggests that when children encounter an unlabeled action exemplar and an iconic gesture depicting the action in this exemplar, children’s attention is guided to the action by the information that is schematically depicted in gesture. This helps children to focus on action as a component of the exemplar and create a stable memory representation of the action. When children later go on to encounter a novel action exemplar (i.e., the same action performed in a different context) but now labeled with a novel verb, they can recognize the relevant action from prior experience in this exemplar. This makes the action stand out in the labeled exemplar, and as a result, children interpret the action as the referent of the novel verb.

Third, children may develop a general strategy for focusing on actions in the following way. This gesture-for-general-strategy mechanism suggests that when children encounter an action exemplar and an iconic gesture depicting the action in this exemplar, and this process is repeated for multiple different actions over time, iconic gesture may communicate to children the general strategy to pay attention to actions. Thus, when children encounter a labeled action exemplar, they may use this strategy to focus on actions more generally and as a result they may interpret actions as novel verb referents.

The Current Study

We developed a novel verb learning task, in which children had the opportunity to link their prior experience with unlabeled actions from memory to labeled experiences with the same actions they encountered at a later point. Crucially, children did not encounter an unlabeled action and its label consecutively; children encountered multiple different unlabeled actions and performed a distraction task before any actions were labeled.

In Experiment 1, we manipulated prior experience with unlabeled actions and the gesture type that children saw with these unlabeled actions. The experiment had a prior-experience , label , and test phase . Gesture type was manipulated in the prior-experience phase, where all children were shown a block of six videos of unlabeled action exemplars. While viewing these exemplars, for half of the children, the experimenter produced iconic gestures that depicted the actions in the exemplars (iconic gesture conditions) and for half of the children the experimenter produced interactive gestures ( Bavelas et al., 1992 ) that did not depict any aspect of the exemplars (interactive gesture conditions). Prior experience with unlabeled actions was manipulated in the label phase, where half of the children were taught novel verbs for actions they had seen in the prior-experience phase (relevant exemplar conditions), and half of the children were taught verbs for novel actions that they had not seen in the prior-experience phase (irrelevant exemplar conditions). In the test phase , children’s understanding of the novel verb meanings was tested in a two-alternative forced-choice task. Following the paradigm by Imai et al. (2008) , children could correctly generalize each verb to a novel actor performing the action that was labeled in the label phase (same-action video) or incorrectly to the same actor as in the label phase performing a novel action (same-actor video).

Predictions for Experiment 1

The recall-event-and-align mechanism predicts that children in the relevant exemplar conditions will successfully generalize more verbs than children in the irrelevant exemplar conditions, regardless of whether they see iconic or interactive gestures in the prior-experience phase, and that children in both relevant exemplar conditions will perform above chance.

The gesture-for-action-memory mechanism predicts that children in the relevant-iconic condition will successfully generalize more verbs than children in the other three conditions, and that children in the relevant-iconic condition will perform above chance.

The gesture-for-general-strategy mechanism predicts that children in the iconic gesture conditions will successfully generalize more verbs than children in the interactive gesture conditions, regardless of whether they see relevant or irrelevant exemplars in the prior-experience phase, and that children in both iconic gesture conditions will perform above chance.