Abstract

How do we conjure up novel and unfamiliar entities in our imagination? Thomas Ward and others have suggested that we do so by deriving such entities from ordinary familiar ones. Hybrids, however, pose a challenge to this view since they are not derived from any one single familiar entity. Nevertheless, we argue here that the construction of hybrid entities is indeed governed by principles forming part of our structured imagination. These principles refer to a set of five abstract schemas, defined in terms of properties such as parts, symmetry, and spatial orientation. These schemas, alongside the absence of a schema, together constitute a schematological hierarchy: humanoid (e.g., man) > canoid (e.g., dog) > carroid (e.g., car) > culteroid (e.g., knife) > arboid (e.g., tree) > other (e.g., sponge). When forming a hybrid out of two or more entities, or parents, the overall shape of the hybrid is selected in accordance with the following three principles: (1) coherence: presence of a schema is preferred to absence of a schema; (2) accessibility: a schema corresponding to that of one of the parents is preferred to some other schema; and (3) height: a schema higher on the schematological hierarchy is preferred to a schema lower on the schematological hierarchy. To test these principles empirically, we conducted a large-scale experiment, in which art and design students were given pairs of words denoting familiar objects and asked to draw images of hybrid entities formed from these word pairs. The resulting corpus of 356 hybrids was found to provide strong empirical support for the above three principles. In doing so, it showed how human creativity is not unbound, but rather subject to substantive cognitive constraints, constituting our structured imagination.

1. Introduction

Hybrids, or entities that are built up from two or more parts associated with different constituents—the centaur, formed from the combination of human and horse, is a classic example—are ubiquitous throughout human art, mythology, religion, and culture. Some of the very first attestations of representational visual art are of hybrids, such as the human-animal hunters of the cave art of Sulawesi in Indonesia, dated back to at least 43,900 years ago, recently reported on by Aubert et al. (2019), or the well-known Löwenmensch, or Lion-Man, discovered in a German cave and dating back some 35,000 to 40,000 years. Hybrids play a major role in most of the “classical” civilizations, for example, the Egyptian god Horus and the Hindu Mahabharata creature Navagunjara. Today, hybrids can be found seemingly everywhere, from the woodcarvings of New Guinea as documented by Dirk Smidt (1993), all the way to modern Western popular culture such as BoJack Horseman. The widespread distribution of hybrids across time and space has been noted by, among others, Rudolf Wittkower (1942: 197) and Arnheim (1966: 256) and is also attested to by the various articles in this special issue.

The ubiquity of hybrids raises a fundamental question pertaining to the notion of creativity: How does human imagination create new and unfamiliar entities? This question is of course germane not only in the case of hybrids, which combine two or more familiar entities to produce a novel and unfamiliar entity, but also with regard to the multitudes of other fantastic creatures that feature in literature, art, mythology, science fiction, and elsewhere.

Following McKellar (1957), Murray (1986), and Ward (1994), we consider imagination to be “the deliberate mental generation of some novel entity” (Ward 1994: 2). But how does this process play out in practice? In accordance with the widespread “romantic” approach, the processes involved in creative imagination are individualistic, unexpected, and lying outside our usual, everyday cognitive behavior. Within this approach, each individual's creative productions are completely different from those of every other individual, which leaves us essentially with nothing that can be subjected to systematic investigation. However, in recent years, it has become increasingly clear that such an anything-goes approach cannot be maintained, and that the products of our imagination are subject to a variety of constraints imposed by our cognitive wherewithal. Indeed, there is more and more evidence suggesting that imagination and other related activities are subject to the same cognitive principles and processes that govern other domains of our mental life. This presumption of a kind of overarching cognitive unity lies at the heart of what is now the leading approach to the study of creative cognition, namely, Structured Imagination.

In a series of studies reported on in Ward (1994), Ward and Sifonis (1997) and Ward et al. (2002), subjects were asked to describe an imaginary animal on a distant planet, or to draw such animals, or various other objects such as fruits and tools. In these studies, subjects were encouraged to be as creative as possible and to conjure up entities that are “wildly different” from the familiar ones on earth. And yet, even under such conditions, most subjects imagined entities that were mere variations on familiar existing entities. Some examples of imaginary creatures produced by Ward's subjects are given in figure 1 above.

As evident from examples such as these, the imaginary creatures produced by the subjects were not really “wildly different” after all. Rather, they clearly share several characteristic features of familiar terrestrial land animals: head, torso and appendages, facial organs, and bilateral symmetry. What makes these creatures imaginary is an array of secondary features, details pertaining to size, shape, and other such characteristics.

Thus, for the most part, imaginary entities fail to escape the procrustean bed of familiar cognitive categories. As Olivier Morin and Oleg Sobchuk (this issue: 648) put it, “Fictional animals are rather unimaginative.” Or in Ward's (1994: 4) own words:

When subjects are given the task of generating a novel animal, for example, the label “animal” might lead them to retrieve typical examples of their animal category. They would then use those activated representations as a starting point for the new creation. Because all the models contain information that would at least translate into characteristic features of known category members, they all would predict that newly generated exemplars will possess those characteristic features. That is, the characteristic properties that are so influential in category decision making should also be influential in the development of novel instances. Presumably, a person using imagination would create a novel entity that was similar to the stored representation by projecting characteristic properties of that representation onto the entity.

These studies show clearly that imaginary entities are derived from the forms and representations associated with ordinary familiar entities. Moreover, as shown in Ward (1994), this borrowing from the familiar is true not only for lay subjects in experimental conditions but also for persons working in genres that mandate creativity, such as fantasy and science fiction.

Hybrids, however, pose an apparent challenge to the idea that imaginary entities are derived from more familiar ones. Consider the following examples of a man-hammer hybrid, produced by three different participants as part of a larger experiment conducted with students at the Shenkar College for Engineering and Design, in Ramat Gan, Israel. (This experiment is described in detail in section 4 below.)

In the case at hand, persons asked to draw a man-hammer hybrid do not have at their disposal a familiar category on which to base their drawings. While the component parts, man and hammer are familiar, their combination is not. The absence of a familiar category of man-hammer might suggest that there should be great variation in the forms of the hybrids that are produced, and, at first glance, this is what we encounter, with each of the three images in figure 2 instantiating a quite different configuration of man and hammer parts. Nevertheless, underlying the observed variation is a more fundamental unity. Specifically, each of the three images presents an essentially humanoid gestalt, in which certain human parts are replaced by hammer parts.

A similar state of affairs is evident in the three examples in figure 3, also from the Shenkar College experiment, of dog-flower hybrids. As in the preceding case, participants are asked to conjure up an imaginary entity, and do so by combining dog and flower parts in quite different configurations. Nevertheless, here too there is a fundamental unity, in which the overall gestalt is recognizably canine, and the flower parts are stuck on to the dog in various ways.

Examples such as these show that even when tasked with building a hybrid, our imagination is structured, and our creativity constrained by various cognitive principles. In an alternative universe, one might envisage subjects forming random collages out of, in the first case man and hammer parts, and in the second case dog and flower parts. However, this almost never happens, the reason for this being a strong cognitive preference for coherent gestalts. As Larry Abramson (this issue: 670) puts it, “multiplicity is the prerequisite for any kind of hybridity, but to survive, a hybrid creature—be it biological, linguistic, or visual— needs its multiple parts to work together as some kind of cohesive whole.” Moreover, as we demonstrate below, the choice of the gestalt is, itself, far from random. To begin with, when building a hybrid, the gestalt that is chosen typically corresponds to one of the constituent entities, or parents. Moreover, with greater than chance frequency, the gestalt chosen is that of the parent that is higher on a hierarchy ranging from unstructured entities all the way up to humanoids. Thus, in figure 2, man is higher than hammer, while in figure 3, dog is higher than flower.

The task of drawing a hybrid is thus very different from that of merely drawing an imaginary creature. Whereas Ward's subjects were able to model their productions on a wide range of familiar creatures, there is no obvious prototype on which to base a drawing of a man-hammer or a dog-flower; subjects tasked with producing such hybrid entities must do so from scratch. Nevertheless, as suggested above, their productions are not random but rather rule governed, reflecting a variety of more abstract cognitive principles. Some of these principles may be viewed as instances of Conceptual Combination (Hampton 1997; Estes and Glucksberg 2000); however, others appear to be more specifically associated with the process of constructing hybrid entities. To the best of our knowledge, these principles have not yet been explicitly formulated, let alone put to an empirical test. The goal of this article is thus to formulate a set of such principles and then test them empirically, thereby providing a guide for how to build a hybrid.

2. The Schematological Hierarchy

The construction of hybrid entities is based on five abstract schemas, humanoid, canoid, carroid, culteroid, and arboid. The names of these schemas allude to typical exemplars thereof, humans, dogs, cars, knives, and trees respectively. These five schemas, together with the null case involving the absence of a schema, form a hierarchy: humanoid > canoid > carroid > culteroid > arboid > other. The defining features of the schematological hierarchy are represented in table 1.

As represented in table 1, schemas are defined in terms of four properties: parts, sensory organs, symmetry, and orientation. The schematological hierarchy, and the trajectory from undifferentiated blob through arboid, culteroid, carroid, and canoid all the way up to humanoid, are defined by successive increments in information, or complexity, with regard to these four properties. The lowest schema on the hierarchy, that of arboids, consists of two parts, head and torso, and is endowed with radial symmetry. Examples of familiar objects exhibiting the arboid schema are trees, daisies, jellyfish, and umbrellas. The next schema up, that of culteroids, differs from arboids only insofar as radial symmetry is replaced by bilateral symmetry, generally introducing a distinction between left and right sides. Common objects instantiating the culteroid schema are forks, hammers, worms, and certain flowers such as some orchids. Moving up an additional step on the hierarchy, carroids differ from culteroids in that their typical instantiations have appendages and also canonical horizontal orientation. Examples of carroids include cars, buses, trains, and airplanes. The next schema up, that of canoids, differs from carroids in that it is endowed with sensory organs typically constituting a face that is positioned on the head. Familiar instances of canoids include dogs, robins, tuna, and certain robots, such as the AT-AT walker from Star Wars. Finally, at the top of the schematological hierarchy are humanoids, which differ from canoids in that they have vertical rather than horizontal orientation, and exactly four appendages, a pair of arms and a pair of legs. Common cases of humanoids include humans, chimpanzees, dolls, and many robots, such as C-3PO from Star Wars.

As suggested in table 1, the schematological hierarchy effects a version of a Me-First hierarchy that is reminiscent of the familiar Me-First hierarchy based on ontological categories (Cooper and Ross 1975): in both the schematological and the ontological hierarchies, humans find themselves at the top. Of course, this hierarchy should come as little surprise, given that we humans are mostly concerned with ourselves. However, the schematological and ontological hierarchies are, at least in part, logically independent of each other, as can be seen in table 2.

In table 2, the horizontal axis represents the schematological hierarchy and schematological Me-first, while the vertical axis represents the ontological hierarchy and ontological Me-first. Individual cells indicate the schematological and ontological classifications of selected familiar entities. While not all cells in the table can be filled, sufficiently many are populated to demonstrate the logical independence of these two hierarchies as orthogonal modes of knowledge representation.

Although logically independent of each other, the two hierarchies are structurally analogous to one another. Just as, in the ontological hierarchy, entities higher on the hierarchy may take a wider range of predicates than their counterparts lower down on the hierarchy, so in the schematological hierarchy, entities higher on the hierarchy are associated with a greater number of the relevant properties. For example, just as in the ontological hierarchy die applies to plants, animals, and humans, eat to animals and humans, and talk to humans alone, so in the schematological hierarchy having bilateral symmetry applies to culteroids, carroids, canoids, and humans, having a canonical orientation to carroids, canoids, and humans, having a face to canoids and humans, and having two arms and two legs to humanoids alone. Thus, both hierarchies are based on a notion of richness, with greater elaboration being associated with entities higher up on the hierarchy, culminating in “me.”

Nevertheless, there are significant differences in the respective structures of the two hierarchies. First, while the ontological hierarchy is exhaustive, that is to say, every entity belongs to some ontological category or another, the schematological hierarchy is not: numerous entities do not belong to any of the five schematological categories and are thus assigned to a leftover category of other, for example sponges, grass, telephones, and stones.1 Second, whereas the boundaries between ontological categories are for the most part strictly demarcated, those between schematological categories are in some cases more fuzzy, allowing for various kinds of indeterminacy; to cite just one example, the boundary between humanoids and canoids may be straddled by various animals whose canonical orientation is not clear, either diagonal, such as kangaroos and tyrannosauruses rex, or else alternating between vertical and horizontal, such as certain primates. Schematological categories are thus best conceived of as representing prototypes, to which different entities conform to varying degrees.

In balance, the schematological hierarchy appears to play a more modest role in our mental lives than does its ontological counterpart. This is reflected, inter alia, in the absence of common everyday words in English and other languages denoting schemas, the would-be schematological counterparts of words such as human, animal, and plant, which denote ontological categories. It is also the reason that, whereas the ontological hierarchy is a well-documented and well-understood feature of human cognition, the schematological hierarchy has not yet received due recognition. Still, as we are suggesting, the schematological hierarchy is an important aspect of human cognition. In particular, the schematological hierarchy lies at the heart of structured imagination and provides the foundation for a series of principles governing the ways in which we build hybrids.

3. Using Schemas to Build Hybrids

Let us place ourselves in the position of a person tasked with constructing a hybrid out of two different parents. We are faced with a quandary: How should the hybrid be put together? What should it look like? More specifically, what should we choose for its overall gestalt?

The choice of gestalt may be defined with reference to the schemas of the constituent parents and their relative positions on the schematological hierarchy. Assuming that the parents are associated with different schemas, the following logically possible strategies present themselves regarding the choice of schema:

  • (1) Building Hybrids: Strategies for Schema Selection

    • (a) none

        There is no hybrid schema.

    • (b) higher

        The hybrid schema is that of the schematologically higher parent.

    • (c) lower

        The hybrid schema is that of the schematologically lower parent.

    • (d) above higher

        The hybrid schema is schematologically higher than those of the two parents.

    • (e) in between

        The hybrid schema is schematologically in between those of the two parents.

    • (f) below lower

        The hybrid schema is schematologically lower than those of the two parents.

The above possibilities are illustrated in figure 4, showing some of the hybrids from the Shenkar College corpus.

In (a), the person-fish, the overall gestalt does not conform to any of the schemas provided for by the schematological hierarchy. In contrast, in (b), the person-bull, the overall schema is clearly humanoid, in accordance with the schema of the person, and thus higher on the hierarchy than the canoid schema of the bull. Conversely, in (c), the bird-flower, the overall schema is arboid, in keeping with the schema of the flower, and thereby lower on the hierarchy than the canoid schema of the bird. A different situation is evident in (d), the grass-fork; here the overall schema is canoid, which is higher on the hierarchy than both the culteroid schema of the fork and the lack of any schema associated with the grass. Finally, in (e), the person-bus, the overall schema is canoid, which in this case is lower than the humanoid schema of the person but higher than the carroid schema of the bus. As reflected by their absence in figure 4, the Shenkar College corpus contains no examples of strategy (1f), in which the overall schema is lower than that of both of the parents.

As we shall see in the continuation, this gap in the data is anything but accidental. More generally, the choice of schemas in the construction of a hybrid is not random, but instead can be seen to conform to three general principles:

  • (2) Principles Governing the Selection of Schemas

    • (a) schema coherence

        presence of a schema is preferred to absence of a schema

        (1b)(1c)(1d)(1e)(1f) >1(a)

    • (b) schema accessibility

        a schema corresponding to that of one of the parents is preferred to some other schema

        (1b)(1c) > (1d)(1e)(1f)

    • (c) schema height

        a schema higher on the schematological hierarchy is preferred to a schema lower on the schematological hierarchy

        (1d) > (1b) > (1e) > (1c) > (1f)

The three principles in (2) are couched as preference rules whose effect is to characterize certain choices as of greater optimality, or cognitive appeal, than others. Their interlocking effects are represented in figure 5, in which differently styled lines encircle the schema choices countenanced by each of the three principles.

We shall now consider each of these three principles in turn. The first principle, schema coherence, in (2a), specifies a preference for hybrids with an overall gestalt exhibiting a clear and well-defined schema; as suggested by the dashed line in figure 5, it thus effects a preference for strategies (1b)–(1f) over strategy (1a).

Prima facie, there are good reasons not to expect hybrid entities to be associated with an overall gestalt associated with a clear schema. First, as noted earlier, unlike the case with Ward's imaginary animals, there is no available familiar entity after which a novel hybrid can be modeled. Second, whereas Ward's tasks were conducive to a top-down approach, the intuitive task of building a hybrid is couched in a bottom-up perspective, in which one takes some parts of one entity and some other parts of another entity and puts them together, but with nothing being stipulated about what the outcome of the process should look like. Third, in at least some instances, the process of building a hybrid is faced by conceptual challenges; this is particularly the case when the two parents are cannot be aligned in a way that might be conducive to the construction of a hybrid with a clear and coherent gestalt.

These factors are illustrated in the following image (fig. 6) from the Shenkar College corpus representing an attempt to construct a hybrid from a person and a telephone. People and telephones simply cannot be aligned in a natural way, as a consequence of which the experimental participant asked to create a person-telephone hybrid simply stuck the telephone on top of the person's head, giving it some hair for good measure. However, the image as a whole does not conform to any schema; indeed, it is not even clear that it qualifies as a single hybrid entity rather than a simple spatial juxtaposition of two separate entities.

Nevertheless, despite the above considerations, the overwhelming majority of hybrids that we have observed do in fact exhibit a gestalt corresponding to one of the schemas in the schematological hierarchy; for example, this is true for all of the hybrids in figures 2, 3, and 4 (images (b)–(f)). But this is hardly surprising. Recall that the reasoning behind Ward's notion of structured imagination is that the generation of imaginary entities, such as those in figure 1 earlier, takes familiar existing entities as its point of departure. We propose that the construction of hybrids works similarly, except that the point of departure is not a familiar entity but rather a schema, one of those defined in table 1. In other words, a hybrid is built by selecting a schema and slotting parts of each of the hybrid's parents into their appropriate positions in the schema. How the schema is selected is specified by the next two principles.

Principle (2b), schema accessibility, specifies a preference for a hybrid whose overall gestalt is associated with a schema identical to that of one of the hybrid's parents; as suggested by the double line in figure 5, it thus effects a preference for strategies (1b) and (1c) over strategies (1d), (1e), and (1f). The rationale behind this principle is straightforward; the schemas associated with the hybrid's parents are already present and do not require additional effort to be conjured up. The principle of schema accessibility may perhaps be viewed as a special case of what Ward et al. (2002: 199) had in mind when they wrote that “those items that come to mind most readily are the ones most likely to serve as starting points for the development of novel ideas.”

One may speculate what factors might lead to the construction of a hybrid not in conformity with the principle of schema accessibility. Some possible factors are suggested by consideration of the grass-fork hybrid in figure 4(d), whose overall canoid schema is higher than both the culteroid schema of the fork and the absence of any schema associated with the grass. Essentially, when both of the hybrid's parents are low on the schematological hierarchy and thus structurally impoverished, there is less material to work with and insufficient basis for the two parents’ parts to be aligned—a situation that is thus conducive to the introduction of a completely new and unrelated schema of greater richness. However, as argued below, an additional important factor in many cases is the third of the three principles, namely, schema height.

As formulated in (2c), schema height specifies a preference for a gestalt whose schema is higher on the schematological hierarchy in table 1; as suggested by the fading-out solid black line in figure 5, it effects a scale of preference that may be represented as (1d) > (1b) > (1e) > (1c) > (1f). As noted in the previous section, the schematological hierarchy represented in table 1 may be considered as an instantiation of the so-called Me-first principle, with the humanoid schema situated at top.

However, closer consideration of the ways in which hybrids are constructed suggests that the observable Me-First pattern may in fact be epiphenomenal, a reflection of deeper organizational principles. As pointed out earlier, the schematological hierarchy is defined in terms of richness of information, that is, successive elaborations of structural elements. So, in effect, Me-First is tantamount to More-Structure-First. But why should the degree of structural elaboration play a role? The reason for this can be found in the following deeper principle governing the construction of hybrids:

  • (3) Maximal Structure Preservation

      A hybrid will preferably preserve as many distinctive elements as possible from each of its parents.

The Maximal Structure Preservation principle is a close analogue to the principle of faithfulness that has been proposed in phonological theory to account for word hybrids, or blends, such as smog, formed from smoke and fog (Bat-El 1996; Bat-El and Cohen 2012; Shaw et al. 2014). The idea is simply that we want as many of each parent's parts as possible to be preserved in the hybrid. Of course, the boundary case of maximal preservation would be that in which each parent appears in its entirety, joined somehow to the other parent or parents, but this would fail to constitute a coherent schema—the principle of Maximal Structure Preservation thus coming into conflict with the principle of schema coherence. An optimal resolution of this conflict is thus to select a higher schema, one that provides more slots for distinctive parts of each parent to be inserted, thereby effecting a Me-First pattern in accordance with the principle of schema height in (2c).

To see how this plays out, consider the two person-spoon hybrids from the Shenkar College corpus in figure 7. The schema height principle in (2c) favors the image on the left, with humanoid schema, over the image on the right, with culteroid schema. But comparing the two images shows clearly how the schema height principle is motivated by the principle of Maximal Structure Preservation: whereas both images present one or two more or less complete spoons, only the first image, with its humanoid schema, is able to preserve the head, face, torso, arms, and legs of a human.

A vivid illustration of the principle of Maximal Structure Preservation in action is provided by the following two pig-bus hybrids taken from the same corpus (fig. 8). Again, the schema-height principle effects a preference for the first image, with canoid schema, over the second image, with carroid schema. But note the position of the pigs’ most distinctive feature, their heads. In the image on the left, the head is in its proper place, in accordance with the canoid schema. However, in the image on the right, the experimental participant committed to a carroid schema, the crucial feature of carroids being that they do not have faces—in the case at hand, a snout, ears, and so on. Accordingly, the drawer is left without an obvious location for the pig's head. The solution adopted is to exploit the roundness of the pig's head and to map it on to the similarly round wheel. But this solution is suboptimal for two reasons: first because, with two wheels facing the viewer, there are now two pig heads rather than one; second because the heads are in the wrong place, forming part of the appendages. What this example shows, then, is that once a lower schema, that of the carroid, is selected, the pressure exerted by the Maximal Structure Preservation principle forces the artist to create an image containing strange and unintuitive features that run counter to the selected schema.

In concert, the three principles in (2) effect far-reaching constraints on the choice of schemas in the process of building a hybrid. Still, these three principles are not categorical but mere preferences. Thus, to test their empirical validity, we need to conduct a quantitative investigation of a sufficiently large corpus of hybrid images. Such an investigation is conducted in the next section.

4. The Production Experiment

To explore the cognitive mechanisms governing the construction of hybrids, we performed a large-scale experiment with the collaboration of students at Shenkar College.

Participants in the experiment were given pairs of words denoting familiar objects and were asked to draw images of hybrid entities formed from these word pairs. For example, the images in figure 2 were produced by participants given the words man and hammer, while those in figure 3 were produced by participants given the words dog and flower. The words, written in Hebrew, were presented in random spatial configurations to avoid any possible effects due to linear order and one word appearing before the other. A total of 129 word pairs were distributed amongst participants; each participant received a different set of three word pairs and was tasked with producing a corresponding set of three hybrid images. Resulting was a total of 481 products, of which 477 constituted actual drawings.

These 477 images were analyzed and coded by the two authors working in collaboration. The analysis was detailed and painstaking, with, in extreme cases, up to one hour spent in discussion and debate on a single image. Some of the contents of our deliberations are elaborated on in the following discussion.

The first step of the analysis was to classify the images in accordance with whether they constituted bona fide hybrids. In many cases they clearly did not: sometimes one of the parents was completely absent, while in other cases the two participants were merely juxtaposed side by side. (See fig. 6 above for discussion of a borderline case thereof.) In total, of the 477 images, 121, or 25 percent, were judged not to qualify as hybrids and were accordingly eliminated from further consideration. We were thus left with a total of 356 hybrid images, on which the remaining analysis focuses.

The stimuli, or word pairs, presented to the participants may be classified into five types in accordance with how the two parents relate to each other with respect to the ontological and schematological hierarchies. This classification is presented in table 3 below. In type A, the first parent, P1, is ontologically higher but schematologically lower than the second, P2. For example, grass is ontologically higher than fork (plant vs. inanimate) but schematologically lower than it (no schema vs. culteroid). In type B, the first parent is ontologically higher but schemtologically on a par with the second. For example, cat is ontologically higher than doglike robot but schematologically equivalent to it (both canoid). In type C, the first parent is both ontologically and schematologically higher than the second. For example, person is ontologically higher than hammer (human vs. inanimate) as well as schematologically higher than it (humanoid vs. culteroid). In type D, the first parent is ontologically on a par with the second but schematologically higher than it. For example, dog is ontologically equivalent to jellyfish (both animal) but schematologically higher than it (canoid vs. arboid). Finally, in type E, the two parents are equivalent to each other both ontologically and schematologically. For example, surgeon and butcher are on a par ontologically and schematologically (both human and both humanoid).

The final two columns present a breakdown of the 356 hybrids in the corpus in accordance with the above five types. Note that these numbers represent actual hybrids, not stimuli; in general, as reflected in figures 2 and 3, a single stimulus might have been given to multiple participants. As evident in table 3, the distribution of the five types across the hybrid corpus is uneven. In part, this is a reflection of practical contingencies pertaining to the various combinations of parents. However, it is also a legacy of how the experimental project unfolded over the course of time.2

In the analysis and coding of the corpus, our primary task was to characterize each image in terms of the schema that it embodies. While in some cases this was straightforward, in other instances the task was more challenging; in particular, in many cases, a single image seemed to be ambiguous, possessing alternative readings representing two or even three different schemas.

Consider the images in figure 9. In the first image, that of the person-tree, the person appears to emerge out of the tree, as a result of which much or most of the image can be construed as either part of the person or part of the tree. Are those arms raised aloft or branches; are those legs or a trunk? Rather than come down arbitrarily on one side or another, we decided to characterize the image as ambiguous and associate it with both humanoid and arboid schemas. An even more elaborate situation is presented by the second image, that of the bird-flower. Again, both parents are in competition with each other, and the image can be seen either as a bird, flying upward, with black beak, outstretched wings, and bushy tail, resulting in a canoid schema; or as a flower, suspended facing downward, beneath two fronds, yielding an arboid schema. Moreover, the same image can also be viewed as a person, perhaps a dancer, with pointy hat and outstretched arms, and thereby associated with a humanoid schema. Accordingly, we decided to characterize the image in question as ambiguous between three different schemas, humanoid, canoid, and arboid.3 The prevalence of such ambiguities in our coding of the corpus is represented in table 4. As suggested in table 4, of the hybrid images, 284 (80 percent) were judged to be associated with a single reading representing a single schema. Of the remaining images, 66 (19 percent) were judged to have 2 readings, and 6 (a little under 2 percent) as having 3. Thus, the 356 hybrids in the corpus yielded a total of 434 readings, each associated with its own schema.

The analysis of the corpus may thus be conducted with reference either to individual hybrids or to readings. An overview of the results of our coding in accordance with these two modes of analysis is presented in tables 5 and 6. Tables 5 and 6 present the distributions of the six strategies of schema choice defined in (1), plus three additional strategies making reference to cases where both parents share the same schema: above same, where the schema of the hybrid is higher than that of both of its parents; same, where the schema of the hybrid is the same as that of both parents; and below same, where the schema of the hybrid is lower than that of both parents. Table 5 presents a breakdown of the schemas associated with each of the 356 hybrids, indicating also cases where a single hybrid is associated with two or three different schemas.4 And table 6 presents a breakdown of the 434 schemas associated with the corpus as a whole.

Tables 5 and 6 set the stage for an empirical test of the three principles governing the selection of schemas formulated in (2). Tables 7 and 8, derived from tables 5 and 6, provide evidence for the first principle, schema coherence. As shown in table 7, of the 356 hybrids, 343 (96 percent) are clearly associated with one or more schemas. Of the remaining hybrids, 4 (1 percent) have 2 readings, 1 with a schema and 1 without, while just 9 (3 percent) have no schema whatsoever. Similarly, as demonstrated in table 8, of the 434 readings, 421 (97 percent) are associated with a schema, while a mere 13 (3 percent) are not. Thus, schemas are ubiquitous, the Shenkar College corpus providing overwhelming evidence in support of the principle of schema coherence.

Tables 9 and 10, also derived from tables 5 and 6, provide evidence for the second principle, schema accessibility. In Tables 9 and 10, schemas are classified as either accessible, including not only higher and lower as per (2b) but also same, or as nonaccessible, comprising all others. As shown in table 9, of the 356 hybrids, 293 (82 percent) are exclusively associated with accessible schemas, 30 (8 percent) with a combination of accessible and nonaccessible schemas, and 33 (9 percent) with just nonaccessible schemas. Similarly, as demonstrated in table 10, of the 434 readings, 369 involve accessible schemas while just 65 are associated with nonaccessible schemas. Thus, the Shenkar College corpus also provides strong evidence in support of the principle of schema accessibility.

Finally, tables 11 and 12, also derived from tables 5 and 6, provide evidence for the third principle, schema height. In tables 11 and 12, schemas are ordered vertically in terms of their height on the schematological hierarchy and are grouped to facilitate numerical analysis.

In table 11, the 356 hybrids are grouped into categories H, in which the readings are, in balance, higher on the schematological hierarchy; M, in which the readings are, in balance, mid-range on the hierarchy; and L, in which the readings are, in balance, lower on the hierarchy—plus the residue of hybrids lacking a schema entirely. As evident in table 11, hybrids of category H outnumber hybrids of category L by 207 (58 percent) to 52 (15 percent), thereby providing significant support for the principle of schema height. In table 12, the 434 readings are grouped in two different ways. In the first grouping, higher schemas outnumber lower ones by 224 (52 percent) to 100 (23 percent), while above-higher and above-same schemas outnumber below lower and below same ones by 38 (9 percent) to 0 (0 percent). (Recall the absence of a would-be hybrid (f) in figure 4 instantiating a below-lower schema selection.) Combining these in the second grouping, readings of category H, higher on the schematological hierarchy, outnumber readings of category L, lower on the hierarchy, by 262 (60 percent) to 100 (23 percent). Thus, both groupings provide further significant support for the principle of schema height. In conjunction, then, tables 11 and 12 show that participants in the experiment strongly preferred to choose schemas higher on the schematological hierarchy, in accordance with the principle of schema height.

Of the above cases, the most dramatic manifestations of the principle of schema height are those involving above higher readings, such as the grass-fork hybrid with canoid schema in figure 4 and the bird-flower hybrid with, among others, humanoid reading in figure 9. Two additional examples of hybrids with above-higher readings are given in figure 10. In the first image, although both snake and rifle are associated with culteroid schemas, the experimental subject has somehow contrived to produce an image resembling a tyrannosaurus rex, of a canoid if not a humanoid schema. (Recall the comments on schema indeterminacy toward the end of section 2.) In the second image, although rake is asssociated with a culteroid schema and starfish with no schema whatsoever, the resulting hybrid appears to admit two possible readings, either as a starfish or as a human; in the latter case, the resulting humanoid schema is, once again, higher than that of both the hybrid's parents.5 Cases such as these, in which a schema is selected that is not associated with either of the hybrid's two parents, bear testament to the force of the schema height principle, showing how it may, in some cases at least, take precedence over the competing principle of schema accessibility.

The preceding evidence in support of the schema height principle is subject, however, to an important qualification. Consider, for example, the person-hammer hybrids in figure 2, with humanoid schemas, and the dog-flower hybrids in figure 3, with canoid schemas. All these examples uphold the schema height principle, with the person's humanoid schema being preferred over the hammer's culteroid schema, and the dog's canoid schema being preferred over the flower's arboid schema. However, under an alternative analysis, the choice of schemas could just as easily have been accounted for with reference to the ontological hierarchy. Specifically, the humanoid schema of the person-hammer hybrids could have been chosen because the person is ontologically higher than the hammer, while the canoid schema of the dog-flower hybrids could have been selected because the dog is ontologically higher than the flower.6 So how might we adjudicate between ontologically and schematologically based accounts?

This is where the choice of stimuli, as represented in table 3 earlier, comes into play. The problem with person-hammer and dog-flower is that they are type C stimuli, in which the ontological and schematological hierarchies are in alignment. To tease the two hierarchies apart, it is necessary to examine stimuli of other types, the best of which being type A, in which the two hierarchies are directly in conflict. For example, for the ant-doll stimulus, ant, as animal, is ontologically higher than doll, as inanimate, but doll, as humanoid, is schematologically higher than ant, as canoid. Consider, now, the ant-doll hybrids from the Shenkar College corpus in figure 11. In the first image, the schema is that of the ant, namely, canoid, ontology thereby trumping schematology. In contrast, in the second image, the schema is that of the doll, namely, humanoid, with the schematological hierarchy thus winning out over its ontological counterpart.7 Thus, type A stimuli provide us with a method of teasing apart the relative contributions of the ontological and schematological hierarchies to the choice of schemas.

Tables 13 and 14, corresponding to tables 5 and 6 earlier, present the distributions of schema choice strategy restricted to type A stimuli, here totaling 80 hybrids and 101 readings. Tables 15 and 16, corresponding to tables 11 and 12 earlier, and derived from tables 13 and 14, present the effect of the schema height principle within type A hybrids. As tables 15 and 16 make evident, the effect of the schema height principle is maintained in type A hybrids, even though the ontological hierarchy pulls in the opposite direction. In table 15, hybrids of category H outnumber hybrids of category L by 52 (65 percent) to 17 (21 percent). And in table 16, in the first grouping, higher schemas outnumber lower ones by 49 (49 percent) to 31 (31 percent), and above-higher and above-same schemas outnumber below-lower and below-same ones by 17 (17 percent) to 0 (0 percent); similarly, in the second grouping, readings of category H outnumber readings of category L by 66 (66 percent) to 31 (31 percent).

Nevertheless, comparing the type A figures from tables 1316 to the overall figures from tables 5, 6, 11, and 12, a significant pattern emerges, as shown in tables 17 and 18. Tables 17 and 18 contrast higher and lower schemas in the second line and H and L schemas in the third line. The second and third columns present the figures for the entire corpus and restricted to type A hybrids, while the fourth and fifth columns calculate the ratios of higher to lower and H to L schemas for all of the corpus and restricted to type A hybrids.

That all the ratios are significantly greater than 1.00 bears witness to the consistent across-the-board effect of the schema height principle. Crucially, though, within each line, the ratio for the entire corpus is significantly higher than that for the type A hybrids: 3.18 vs. 2.00, 3.98 vs. 3.06, 2.24 vs. 1.58, and 2.62 vs. 2.13, respectively. What this suggests is that in type A hybrids the effect of the schematological hierarchy in determining the choice of schema is partly counteracted by an opposite effect due to the ontological hierarchy. Thus, both schematology and ontology are of importance in determining the choice of schema, but, in balance, the effect of the schematological hierarchy, as introduced in this article, is stronger than that of the more familiar ontological hierarchy.

5. Conclusion

The results of the production experiment presented in the preceding section provide strong empirical confirmation for the three principles governing schema choice formulated in (2): schema coherence, schema accessibility, and schema height. In doing so, they thus offer strong support for the schematological hierarchy as a central feature of our structured imagination.

To the best of our knowledge, this article represents the first substantive attempt to address the question how hybrids are built. The answer that we provided is that they are constructed in more or less the same fashion as other products of our imagination, that is to say, in top-down fashion, based on abstract schema. While the existence of such schema may not necessarily come as a surprise, ours represents the first proposal to the effect that such schema are not sui generis, but rather organized in a hierarchy that governs the ways in which hybrids are constructed.

Although limited in its scope to hybrids, this article represents a substantive advance in our understanding of the structures that underlie human imagination. Future work, we believe, is likely to show that the principles argued for in this article, and the schematological hierarchy that underpins them, may be relevant not only to the construction of hybrids but to various other domains of our cognitive lives.

This article is the product of years of collaboration by its two authors, which would not have been possible without the generous support of Bernard Comrie and the Max Planck Institute for Evolutionary Anthropology in Leipzig. We would also like to thank Michalle Gal and the Shenkar College of Engineering and Design in Ramat Gan, Israel, for facilitating the hybrid production experiment resulting in the Shenkar College corpus. Some of the material in this article was presented at the Workshop on Visual Hybrids, Shenkar College, Ramat Gan, April 15, 2015; the Tel Aviv Conference on Culture and Cognition, Tel Aviv University, November 27, 2018; and in a course on hybrids that we cotaught at Tel Aviv University in the second semester of 2020–21. We are grateful to participants at those events for helpful comments and suggestions. Work by Yeshayahu Shen was supported by grant no. 1530/16 of the Israel Science Foundation administered by the Israel Academy of Sciences and Humanities.

Notes

1.

Work in progress suggests that the leftover category of other may itself be endowed with internal structure. Specifically, we have reason to believe that there may exist alternative schematological hierarchies, similarly organized around a scale of increasing structural richness, but culminating in entities other than human. One potential candidate for such an alternative schematological hierarchy might be one consisting of various residential structures ranging from simple tents to more complex houses. To the extent that such alternative schematological hierarchies can be empirically motivated, they would seem to point toward another important difference between the ontological and schematological hierarchies: while the ontological hierarchy is linear, the schematological hierarchy may be endowed with branching structure, with two or more different pathways leading upward and outward from the structureless source blob in distinct directions of increasing schematological complexity, such as is associated with humans, houses, and perhaps other entities of our familiar experiential universe. But these questions must remain for future investigations.

2.

In earlier work, represented in Mashal et al. 2014, Shen and Gil 2017, and Gil and Shen 2019, our focus was on the ontological hierarchy; at that time we were not yet aware of the relevance of the schematological hierarchy. We accordingly designed a set of 24 stimuliintended to test the effect of the ontological hierarchy and ran the experiment on a first cohort of participants. In that first experiment, most of the stimuli were of type C, since this is the type characteristic of most culturally familiar hybrids. However, once the significance of the schematological hierarchy became apparent, we designed a second set of 105 stimuli intended to assess the relative importance of the two hierarchies, and then ran the experiment on a second cohort of participants. This time round, pride of place went to stimuli of type A, which were designed specifically to pit the ontological and schematological hierarchies against one another. The figures in table 3, and our subsequent analysis, represent the outcome of pooling the products of both cohorts into a single corpus, based on the combined set of 129 stimuli.

3.

To try and get a feel for whether our deliberations were on the right track, for this particular image we conducted an ad hoc informal survey, in which participants were simply asked what the picture represents, without being told that it was intended to represent a bird-flower hybrid. The responses spanned the entire range of possibilities. Some did indeed recognize the intended bird-flower hybrid, e.g., “bird with a flower tail,” while others simply saw two separate objects, e.g., “a bird doing yoga on top of fuchsia flowers.” Others, though, only noticed a single entity, and these were split between each of the three schemas that we attributed to the image: for example, “bird,” reflecting a canoid schema; “hibiscus,” exemplifying an arboid schema; and “faceless humanoid,” “Rastafarian snowman,” or “cone-hatted gnome doing a ceremonial dance,” associated with a humanoid schema. Ultimately, participants of the informal survey thus lent additional support to our own characterization of the image as ambiguous between the three different schemas.

4.

Note that in two cases, those characterized as above higher–above higher, a hybrid is associated with two different schemas, both of which are higher than the schemas of the hybrid's two parents. For example, in one case, a jellyfish-seaweed hybrid, the two schemas are humanoid and culteroid, both higher than the two schemas of the hybrid's parents, arboid and other. In two other cases, characterized as same-same, a hybrid is associated with twodifferent readings, both associated with a schema occupying the same position on the hierarchy as the schema of the two parents. Both cases are of starfish-house hybrids each with two different readings involving starfish and house gestalts, respectively, whose schemas are thus identical with the schemas of each of the two parents.

5.

Again, as for the bird-flower image in figure 9 earlier, we conducted an informal survey, asking participants what the picture represents. Some of the responses included “costumed dancer,” “man walking on his two legs, strings hanging from the entire length of his sleeves, and his head is atop a very long neck,” and “David after having lost enough weight to attempt flight under his own power,” thereby providing further evidence for the above-higher humanoid reading.

6.

Indeed, this ontologically based analysis was that which we had adopted in our earlier work, as represented in Mashal et al. 2014, Shen and Gil 2017, and Gil and Shen 2019.

7.

It should be acknowledged that the humanoid schema of the second image is an imperfect instantiation thereof, motivated primarily by its vertical orientation. However, the image departs from the prototypical humanoid schema in having more appendages than the expected two arms and two legs.

References

Arnheim, Rudolf.
1966
A Note on Monsters
.” In
Towards a Psychology of Art, Collected Essays
, edited by Arnheim, Rudolf,
255
57
.
London
:
Faber and Faber
.
Aubert, Maxime, et al
2019
. “
Earliest Hunting Scene in Prehistoric Art
.”
Nature
576
:
442
45
.
Bat-El, Outi.
1996
Selecting the Best of the Worst: The Grammar of Hebrew Blends
.”
Phonology
13
:
283
328
.
Bat-El, Outi, and Cohen, Evan-Gary.
2012
. “
Stress in English Blends: A Constraint-Based Approach
.” In
Cross-Disciplinary Perspectives on Lexical Blending
, edited by Renner, V., Maniez, F., and Arnaud, P.,
193
211
.
Berlin
:
Mouton de Gruyter
.
Cooper, William E., and Ross, John Robert.
1975
. “
World Order
.” In
Papers from the Parasession on Functionalism
, edited by Grossman, R. E., San, L. J., and Vance, T.J.,
63
111
.
Chicago
:
Chicago Linguistic Society
.
Estes, Zachary, and Glucksberg, Sam.
2000
. “
Interactive Property Attribution in Concept Combination
.”
Memory and Cognition
28
, no.
1
:
28
34
.
Gil, David, and Shen, Yeshayahu.
2019
. “
How Grammar Introduces Asymmetry into Cognitive Structures: Compositional Semantics, Metaphors, and Schematological Hybrids
.”
Frontiers in Psychology—Language Sciences
. https://doi:10.3389/fpsyg.2019.02275.
Hampton, James A.
1997
. “
Conceptual Combination
.” In
Knowledge, Concepts, and Categories
, edited by Lamberts, K. and Shanks, D.,
133
59
.
London
:
Psychology Press
.
Keil, Frank C.
1979
.
Semantic and Conceptual Development: An Ontological Perspective
.
Cambridge, MA
:
Harvard University Press
.
Mashal, Nira, Shen, Yeshayahu, Jospe, Karine, and Gil, David.
2014
. “
Language Effects on the Conceptualization of Hybrids
.”
Language and Cognition
6
, no.
2
:
217
41
.
McKellar, Peter.
1957
.
Imagination and Thinking: A Psychological Analysis
.
London
:
Cohen and West
.
Murray, Edward L.
1987
. “
Imagination Theory and Phenomenological Thought
.” In
Phenomenological Psychology
, edited by Murray, E. L.,
175
220
.
Pittsburgh, PA
:
Duquesne University Press
.
Shaw, Katherine, White, Andrew, Moreton, Elliot, and Monrose, Fabian.
2014
. “
Emergent Faithfulness to Morphological and Semantic Heads in Lexical Blends
.” In
Proceedings of Phonology 2013
, edited by Kingston, J., Moore-Cantwell, C., Pater, J., and Staubs, R.
New York
:
Linguistic Society of America
.
Shen, Yeshayahu, and Gil, David.
2017
. “
How Language Influences the Way We Categorize Hybrids
.” In
Handbook of Categorization in Cognitive Science
, edited by Cohen, H. and Lefebvre, C.,
1177
1200
. 2nd ed.
Amsterdam
:
Elsevier
.
Smidt, Dirk, ed.
1993
.
Asmat Art, Woodcarvings of Southwest New Guinea
.
Leiden
:
Periplus
.
Ward, Thomas B.
1994
. “
Structured Imagination: The Role of Category Structure in Exemplar Generation
.”
Cognitive Psychology
27
, no.
1
:
1
40
.
Ward, Thomas B., Patterson, Merryl J., Sifonis, Cynthia M., Dodds, Rebecca A., and Saunders, Katherine N.
2002
. “
The Role of Graded Category Structure in Imaginative Thought
.”
Memory and Cognition
30
, no.
2
:
199
216
.
Ward, Thomas B., and Sifonis, Cynthia M.
1997
. “
Task Demands and Generative Thinking: What Changes and What Remains the Same?
Journal of Creative Behavior
31
, no.
4
:
245
59
.
Wittkower, Rudolf.
1942
. “
Marvels of the East: A Study in the History of Monsters
.”
Journal of the Warburg and Courtauld Institutes
5
:
159
97
.