Conversations with Caves: The Role of Pareidolia in the Upper Palaeolithic Figurative Art of Las Monedas and La Pasiega (Cantabria, Spain)

The influence of pareidolia has often been anecdotally observed in examples of Upper Palaeolithic cave art, where topographic features of cave walls were incorporated into images. As part of a wider investigation into the visual psychology of the earliest known art, we explored three hypotheses relating to pareidolia in cases of Late Upper Palaeolithic art in Las Monedas and La Pasiega Caves (Cantabria, Spain). Deploying current research methods from visual psychology, our results support the notion that topography of cave walls played a strong role in the placement of figurative images—indicative of pareidolia influencing art making—although played a lesser role in determining whether the resulting images were relatively simple or complex. Our results also suggested that lighting conditions played little or no role in determining the form or placement of images, contrary to what has been previously assumed. We hypothesize that three ways of artist–cave interaction (‘conversations’) were at work in our sample caves and suggest a developmental scheme for these. We propose that these ‘conversations’ with caves and their surfaces may have broader implications for how we conceive of the emergence and development of art in the Palaeolithic.


Introduction
How central were psychological features deriving from our visual systems to the early evolution of human visual culture?Pareidolia-the psychological phenomenon of seeing meaningful forms in random patterns, such as perceiving faces in clouds-is a universal feature of our visual system.It is likely a consequence of the evolution of our visual system adapting to allow partial or obscured profiles of potential predators to be rapidly identified through the conferral of meaning, and hence to minimize risk (Bednarik 2017, 102;Hodgson 2003;Melcher & Bacci 2008, 351).To achieve this, our visual system constructs a complete picture despite missing visual information, thus often causing us to 'see' things that are not there as it attempts to resolve ambiguous visual cues (Frith 2007, 132;Hong et al. 2013, 79;Ward 2008, 18).This process has been the subject of extensive psychological study, with existing debates regarding the particular cultural mechanisms that may cause pareidolia, e.g.do modern Western people see faces relatively frequently because our visual system has evolved to treat the visual stimuli of faces as 'special' (Carmel & Bentin 2002, 25;Hong et al. 2013, 79;Zhou & Meng 2019, 3) or merely because we have visual expertise in face perception (Harel 2016;Joyce & Cottrell 2004, 127;Rossion et al. 2004, 14521;Tovée 1998Tovée , 1239))?It certainly seems that pareidolia is informed by cultural experience, through the frequent perception of everyday objects or animals (Bracci et al. 2019, 6514;Maranhão-Filho & Vincent 2009, 1117).Thus, one's visual familiarity with certain stimuli (particular animals, faces, or objects) may shape the visual system to perceive ambiguous or fragmented stimuli as being evocative of familiar forms.In modern Western societies, pareidolia frequently manifests as a propensity for perceiving faces or anthropomorphic features in objects and this ability emerges early in human development, perhaps even prenatally (Reid et al. 2017).This may be a consequence of-but certainly triggers-our empathetic response to visual stimuli (Correa Varella 2018;Kato & Mugitani 2015;Proverbio 2017;Zhou & Meng 2020).These psychological responses to pareidolic imagery are not limited to faces; animal-like pareidolia also elicits a response as if the person is viewing an animal in reality (Bracci et al. 2019).Thus, pareidolia is not merely a visual phenomenon, but can elicit visceral emotional responses too.
Previous discussions within archaeology, most notably by Hodgson (2003;2006;2008;2012;2013;2019;Hodgson & Pettitt 2018), have provided a thorough conceptual background for how cave environments may have been conducive to triggering visual responses: either pareidolia or, more potently, hyper-imagery.This discussion has focused on how heightened sensory awareness and the ambiguous nature of visual stimuli, induced by the darkness of caves, would have likely caused Palaeolithic people to experience visual imagery, priming them to depict the same animals they had perceived.There has also been extensive previous discussion pertaining to the integration of the rock support and its role in determining the placement of depictions within a cave, for example with the rock used to frame depictions or add depth and dimensionality to an animal motif (Bahn 2003;Leroi-Gourhan 1971;Lorblanchet 1995;Robert 2017;Sauvet & Tosello 1998).This literature has undoubtedly contributed to understanding the role of psychological responses to evocative features of the rock support, and the morphology of the support itself, in cave-art making.However, questions remain regarding the extent to which pareidolia may have underpinned the making of figurative representations in a particular cave.The explicit focus on pareidolia within this research is intentional; the role of pareidolia in cave-art production has been conceptually discussed in previous literature, but has yet to be directly tested against examples in the archaeological record as part of a systematic and focused study.Other facets of cave-art making, such as sociocultural context, authorship and artistic skill, or sensorial elements such as tactility and acoustics, have been subject to much more extensive research, including systematic and testable studies (e.g.Fazenda et al. 2017;Fernández-Navarro et al. 2022;Jouteau et al. 2020;Rivero 2014;Till 2014;Waller 1993).Arguably, the role of pareidolia has not yet been understood with the same nuance.This is not intended to diminish the contribution of other facets of cave-art making.Rather, our research intends to understand the significance and role of pareidolia as one part of the complex, multifaceted process of cave-art making.
We focused on two caves in Monte Castillo (Cantabria, Spain), Las Monedas and La Pasiega, and evaluated whether, and to what extent, the making of figurative representations within them were influenced by pareidolic responses.These caves were selected due to their geographical proximity, differing chronologies (mid-Magdalenian depictions at Las Monedas, late Solutrean/early Magdalenian depictions in Gallery A at La Pasiega), and the diversity of the art between the two (i.e.depictions drawn in charcoal at Las Monedas, depictions painted in red ochre, usually using finger dots, at La Pasiega).These distinctions allowed us to investigate how pareidolia may have manifested in different chrono-cultural contexts.We developed and tested three key hypotheses, motivated by both visual psychological literature regarding the nature of pareidolic images and conceptual discussions of the conditions that may have triggered pareidolic responses in cave environments.Fieldwork observations, high-resolution photogrammetric models and virtual reality (VR) lighting simulations were used to evaluate each hypothesis.To prevent confirmation bias, all figurative depictions within Las Monedas and all figurative depictions within Gallery A of La Pasiega were evaluated.The results indicated that the extent to which pareidolia influenced the making of a particular depiction varied, and may reflect different 'conversations' occurring between the artist and the cave wall during the process of art making.

Las Monedas
Discovered in 1952, Las Monedas is attributed to the Middle/Late Magdalenian based on stylistic features subsequently confirmed by AMS radiocarbon dates of 13,766-13,248 cal.BP, 14,811-13,791 cal.BP (ibex 16) and 14,076-13,519 cal.BP (horse 20) (Amormino 2000;García-Diez et al. 2021;González & Balbín Behrmann 2007;González Echegaray 1952;Moure Romanillo et al. 1996;Múzquiz & Cabrera Valdés 2000;Ochoa 2017, and references therein;Ripoll Perelló 1972;Ruiz-Redondo et al. 2016).Apart from a lithic collection attributed to the Mousterian and a brief mention of an Aurignacian flint point found among remains of cave bears (Carrión Santafé & Baena Preysler 1998;Ochoa 2017, 298), the lack of Izzy Wisher et al. materials attributable to the Magdalenian suggests that it may not have been habitually occupied when its art was created.Archaeological deposits may be present under the calcite floor in the entrance hall (García-Diez et al. 2021, 311) but it remains possible that, given the proximity of La Pasiega and El Castillo, and the late Upper Palaeolithic deposits in the latter (Cabrera Valdés 1984), these nearby caves were the places of residence (Ortega Martínez & Ruiz-Redondo 2018, 804).
Despite its large size, the art of Las Monedas comprises only 30 figurative and 15 non-figurative images constrained to a small area within the cave (Fig. 1).All figurative depictions appear to be superficially homogenous with the same technique and style used in all depictions, although with some subtle variation in depicting certain features of animals.Horse depictions are the most numerous (50 per cent, n=15) of the figurative representations, followed by ibex (17 per cent, n=5) and reindeer (13 per cent, n=4) (Table 1) Unusually for the Magdalenian art of northern Spain, bison depictions are rare (n=1).All depictions were drawn using a charcoal crayon, mostly detailed outlines occasionally partially infilled or with additional internal lines detailing pelage.Due to the similarities in style and technique, the constrained space of the gallery and the overlapping range of the AMS radiocarbon dates obtained from depictions 16 and 20, it is therefore likely that the art was produced by a limited number of individuals in a brief period of time (García-Diez et al. 2021;Ochoa 2017).

La Pasiega
Discovered in 1911 by Werner and Obermaier, the large system of La Pasiega to the east of Las Monedas contains extensive and varied depictions as well as Mousterian, Solutrean and Magdalenian archaeology, most of which is attributable to the Solutrean (e.g.Balbín-Behrmann & González-Sainz 1994;González Echagaray & Moure Romanillo 1971;González-Sainz et al. 2013, 106;Ochoa 2017;Ochoa et al. 2017, and references therein;Sainz & Balbín Behrmann 2021;Straus 1979).As originally noted in the monograph by Breuil et al. (1913) and systematically demonstrated by Groenen and Groenen (2019), there is an uneven distribution of images; some galleries (e.g.Gallery A) have large concentrations, others (e.g.Gallery D) comparatively few.Artistic techniques vary between galleries; ochre pigments were predominantly used throughout, but the few black depictions present concentrate in specific areas of Galleries A and C. Thematic and stylistic variability is high within and between galleries, all of which suggests that La Pasiega's art has an extensive temporal breadth and was produced in multiple phases in which galleries formed single foci of activity, unlike the cohesive composition of Las Monedas (González-Sainz et al. 2013, 106).
We focused on Gallery A, due to its high number of figurative depictions that are broadly homogenous in style (primarily red ochre outline depictions produced with finger dots/smears) and form a representative sample of the art of La Pasiega, in terms of the animals represented.It is characterized by low ceilings and narrow spaces, which likely restricted the number of individuals that interacted with this space.The high frequency of depictions in Gallery A has been referred to as horror vacui-'fear of empty space' (García-Diez et al. 2018).Superimpositions are rare, further supporting the relative homogeneity of the Gallery A art and implying the depictions may have been produced over a relatively short period.As outlined in Table 2, hinds are the most frequently depicted animal, accounting for 34 per cent of the depictions, followed by horses (30 per cent), stags (14 per cent), aurochs (5 per cent), bison (4 per cent), ibex (3 per cent) and reindeer (2 per cent); indeterminate animals account for the remaining 8 per cent.

Hypotheses
To determine the extent to which pareidolia may have been responsible for the form and placement of figurative depictions in the caves of Las Monedas and La Pasiega, we formulated and tested three hypotheses.First, we reasoned that if pareidolia underpinned the making of figurative depictions within these caves, then the majority of animal depictions would be scaffolded on natural topographic features of the cave wall.Pareidolic responses are triggered by evocative features that the visual system perceives as 'looking like' a meaningful form, and consequently if pareidolia was the motivation behind the placement and form of a particular animal, the depiction would naturally integrate these evocative features.This phenomenon of pareidolia motivating artistic behaviours has been previously documented, where particular forms that trigger a pareidolic response subsequently become integrated into a depiction (Lee 2016).Thus, Hypothesis 1: The majority of figurative depictions should integrate natural topographic features of cave walls.
Secondly, and building on Hypothesis 1, as pareidolic imagery of animals generally does not incorporate detail beyond salient outline form or the natural features that triggered the pareidolic image, we further propose that simpler depictions of animals that are incomplete in form and/or feature no additional details beyond the outline should integrate natural features.By contrast, detailed depictions, i.e. those complete in form and/or featuring internal detail such as pelage, hair, eyes, and particularly those with stylistic features consistent with other contemporaneous depictions, may thus reflect pareidolia having no or minimal influence over the form and placement of depictions.It may therefore be expected that detailed depictions are less likely to be scaffolded onto natural topographic features when compared to simple depictions.Thus, Hypothesis 2: Simpler depictions should have a stronger relationship to natural topographic features of the cave wall than detailed depictions.
Thirdly, to provide an explanation for why certain depictions may have been more strongly underpinned by pareidolia than othersas explored in Hypotheses 1 and 2the likely lighting conditions available in the Palaeolithic must be evaluated.As proposed by Hodgson (2008) and previously discussed, ambiguous visual stimuli that may be caused by fluctuating low light are conditions conducive to  6; 8; 10; 11; 12; 13; 17; 20; 22; 30; 31; 33; 34; 34a The topography of a particular cave wall would have influenced the interaction of light, with flatter topographies likely resulting in more stable lighting conditions, and undulating topographies therefore causing unstable and unpredictable lighting.Thus, there should be a correlation between depictions that have a strong relationship to the cave wall and unstable light conditions, and depictions with a weaker relationship to the cave wall and stable light conditions.As the topography of a particular cave wall would have influenced the distribution of illumination and hence viewing conditions, VR light simulations of 3D photogrammetry models of cave art panels were used to evaluate this hypothesis.Thus, Hypothesis 3: There should be a correlation between unstable and low light conditions and simple depictions with a strong relationship to the cave wall, and between stable and bright light conditions and detailed depictions with a weak relationship to the cave wall.

Fieldwork
Fieldwork was conducted by IW in Las Monedas and La Pasiega during October 2021 to allow for an initial assessment of the relationship between depictions and the topographic features of the cave wall, which was subsequently explored further through VR lighting simulations.Visual observations were made to assess the placement of depictions (e.g.visibility, integration of natural features, degree of truncation/constraint of images by the cave wall's topography), its form (e.g.taxon, internal details, completeness), and artistic techniques (i.e.method of pigment application), in addition to dimensions and orientation.Four specific categories were developed to assess the degree to which depictions related to the cave-wall topography: 1) Direct: depictions directly integrate natural features (e.g. a crack representing the dorsal line) and/or trace natural features (e.g. a dorsal line drawn over a crack); 2) Parallel: lines of the depiction have a close spatial proximity to natural features and follow their contour (e.g. a dorsal line drawn directly underneath a natural crack and follows the crack's profile); 3) Mimic: features of a depiction appear to incorporate the form of a nearby evocative topographic feature, but the Conversations with Caves depiction is not drawn in parallel to this feature (e.g. a dorsal mimics the profile of a nearby crack); 4) Limited/None: there is no clear association between the depiction and the cave wall topography (Fig. 2).These categories were produced to characterize broadly different relationships to the cave wall, enabling an assessment of the extent to which depictions were influenced by topographic features.All depictions recorded were also photographed using a Nikon D3500 DSLR camera with an 18-55 mm AF-P VR lens, using cold LED light sources to illuminate the depictions and minimize the presence of shadow.
In Las Monedas, 28 out of the 30 depictions published by Ripoll Perelló (1980) were recorded (two of the indeterminate depictions could not be located).In Gallery A of La Pasiega, 65 figurative depictions were recorded, with the DStretch© plugin of ImageJ© used to facilitate the identification of faded depictions (Harman 2008).The discrepancy between the number of depictions recorded in Gallery A and those previously reported by Balbín-Behrmann and González-Sainz (1993) may be due to several factors.Some of the figurative depictions previously recorded were engraved-our study focused only on painted depictions, due to the practical limitations of recording engraved motifs using photogrammetry (Rivero et al. 2019).There are also several areas of Gallery A that are spatially constrained and difficult to access during fieldwork, and the area where Gallery A begins and Galleries B and D end is ambiguous and features a few isolated depictions which may have been reported as part of the depictions in Gallery A previously.In any case our study comprises 28 of 30 images for Las Monedas and 65 of 100 for Gallery A of La Pasiega.

3D modelling
To record appropriately the relationship between natural features of the cave wall and depictions, it was necessary to produce 3D models of the art that captured the topography of the cave art.2D images are insufficient for fully capturing the dimensions of the art, significantly distorting depictions and flattening their undulating topography.Consequently, Upper Palaeolithic cave-art research, and rock-art research more widely, has increasingly incorporated 2.5D (e.g.RTI) or 3D modelling (e.g.photogrammetry, TLS) techniques into traditional recording methodologies, to the extent that recording parietal art in this way is now an established approach for the analysis and preservation of the art (e.g.Díaz-Guardamino et al. 2015;Domingo et al. 2013;González-Aguilera et al. 2009;Lerma et al. 2006;2010;Plisson & Zotkina 2015;Rivero et al. 2019;Ruiz López et al. 2019).Photogrammetric models were produced of the art panels in Las Monedas and La Pasiega based on images taken by IW and, for Las Monedas, Dr Blanca Ochoa.Images were sorted into folders for each panel, filtered for quality and imported into Agisoft Metashape using the guided image matching function to increase the likelihood of the software appropriately matching points across images.Any models that appeared to exhibit significant distortion or low confidence levels in the number of points used were omitted from the study.

VR lighting simulations
To evaluate the effect of a flickering low light source on the perception of the natural features of the cave wall, the photogrammetric models were exported as .objfiles and imported into Unity, an open-source 3D gaming-development software (Fig. 3).Unity allows users to build 3D Virtual Reality environments and, notably, facilitates the manipulation of light: ambient lighting can be adjusted, and light sources can be created to emulate the warm, flickering nature of light cast from a small flame.Any 3D models within Unity receive light and cast shadows in the same way as real-world objects, allowing us to explore the real-time effects of a flickering low light source on the perception of cave-art walls.
The art of both Las Monedas and La Pasiega would have received no ambient daylight.Thus, within the VR lighting simulations, the models were illuminated only by a small, flickering light source that cast a warm light of 1850K at source, decreasing to zero at a distance of 1.25 m (diameter of 2.5 m).This is consistent with experimental observations of the lower ranges for light cast from torch technologies made from juniper or pine wood (Medina-Alcaide et al. 2021), likely the only woody sources available during the Older Dryas and late Solutrean-early Magdalenian, corresponding to the art of Las Monedas and La Pasiega respectively.Lower ranges were selected to provide a conservative estimate for the light available, and to accommodate for the likelihood that the torch may have been burning for some time before the art was produced; Medina-Alcaide et al. (2021) demonstrated light quality depletes as fuel is consumed.In Unity, light intensity does not correspond to real-world values; however, a light intensity of 0.5 was deemed suitable for the approximate light intensity produced by a small flame, particularly as this intensity increased to 0.75 and decreased to 0.25 as the light flickered: this defined our parameters.An appropriate light intensity allows for the simulation perhaps to appear more visually realistic, but the specific light intensity is somewhat irrelevant for determining how the light interacts with the surface.The light source and its distance away from the wall were kept consistent for each simulation, to allow for appropriate comparisons to be made across the simulations.
To quantify the visual effects of the VR lighting simulations, relative luminance values were calculated from still images of the simulation at the brightest and the dimmest point of flicker.Relative luminance values (RLV) weigh RGB values proportionately using the equation Y = 0.2126R + 0.7152G + 0.0722B, where Y is relative luminance.This reflects the sensitivities of the eye to particular wavelengths of light, i.e. a particular energy of medium wavelengths, 'green', is perceived as brighter than the same energy of longer wavelengths 'red' and similarly short wavelengths, 'blue', are perceived as dimmer than other wavelengths (Snowden et al. 2006, 33).This produces a value expressed as a percentage, with 100 per cent representing white (brightest) and 0 per cent representing black (darkest).By calculating the RLV at equidistant points across a surface, one can quantify the visual effects of the simulated light from a Palaeolithic torch across the topographic surface of the cave wall and create a profile for how light is interacting with this surface.This can be used to evaluate, for example, whether natural features caused high contrast in RLV, increasing their saliency and drawing visual attention, or if certain features were rendered ambiguous, reflected by low RLV.The dynamic nature of the light source adds another dimension to this, where a topographic feature may be salient in one instance but ambiguous in the next.This disparity in RLV, reflected by a large difference between RLV at the brightest and dimmest points of flicker, may have been most conducive to triggering pareidolic responses-the ambiguity and conflicting visual information would have encouraged the visual system to 'fill in' the missing information.
The Las Monedas photogrammetric models were based on lower-resolution imagery, and hence RLV was calculated at five equidistant points across a depiction.As the La Pasiega photogrammetric models Conversations with Caves had higher resolution, RLV was calculated at specific distances across the cave wall surface: 0 cm, 5 cm, 10 cm, 20 cm, 30 cm, 40 cm and 50 cm from the source of the light.Additionally, an RLV profile was calculated for a blank white wall to provide a comparison for the profiles calculated for the cave-wall models.

Las Monedas
The figurative images within Las Monedas appear to reflect a prominent influence of pareidolia within their making (Table 3).Of the figurative depictions (20/28 analysed), 71 per cent expressed a strong relationship to the natural features of the cave wall.Thirteen of these directly integrate topographic aspects of the wall into the form of the depiction (e.g. using the undulations of the cave wall to represent the back of an animal: Fig. 2A); six depictions appear to have aspects of their form running parallel to topographic features (e.g. a dorsal line being drawn close and in parallel to the curved edge of the cave wall: Fig. 2B); one depiction appears to mimic the shape of a nearby feature (e.g. the form   of the depiction copies that of the natural feature, but not in parallel: Fig. 2C).These results lend support for Hypothesis 1, thus meeting the first requirement for pareidolia having influenced the making of art within this cave.
Our results also show a relationship between simple and incomplete depictions and the cave wall.Of those depictions with strong relationship to natural features on the cave wall, the vast majority (16/20 = 80 per cent; Fig. 4) are simple in style, i.e. lacking additional anatomical detail such as eyes, hair, or pelage.Equally, half of depictions with apparently limited or no relationship to the cave wall (4 out of 8 depictions) bear additional detail.This supports our Hypothesis 2, that the images that were scaffolded onto features of the cave wall should be simple in form and that those that are more detailed lack such a relationship to their supporting surfaces.
In order to compare the relationship of images to RLV results from the lighting simulations of Las Monedas we used averages of the RLV across all simulations of 8.4 per cent, as an overall average RLV, and 14.9 per cent for the average variation in RLV between the brightest and dimmest points in the simulation (see Supplementary Information for the full dataset).We used these as thresholds to evaluate our data, dividing the depictions into four categories: simple depictions with a strong relationship to features on the cave wall; simple depictions with a weak relationship to features on the cave wall; detailed depictions with a strong relationship to such features; and detailed depictions with a weak relationship to such features.Our results (Table 3) show a loose relationship between below average RLV and those images that bear a strong relationship to features of the cave wall (5 out of 13 depictions simulated = 39 per cent) when compared to the other categories.Low average RLV was recorded for only one (out of 4) of the simple depictions with a weak relationship to the cave wall; one (out of 4) of the detailed depictions with a weak relationship to the cave wall; and two (out of 4) detailed depictions with a strong relationship to the cave wall.This appears to lend some support towards Hypothesis 3, albeit weak.

Conversations with Caves
Similarly, the relationship between simple and direct depictions and a high variation in RLV appears to not be distinct, with 46 per cent (n=6 out of 13 depictions simulated) of simplistic depictions with a direct relationship to the cave wall having a high average variation in RLV.Only one (out of 4) of the detailed depictions with a direct relationship to the cave wall also experienced a high variation in RLV.These values are comparable to the depictions with weak relationships to the cave wall; 75 per cent (n=3 out of 4 depictions simulated) of depictions which were detailed with a weak relationship and 33 per cent (n=1, out of 3 depictions simulated) of simplistic depictions with a weak relationship had high variations in RLV.This suggests that unstable light conditions were not a factor that influenced pareidolic responses.These results are therefore not consistent with Hypothesis 3, and suggest that fluctuating light conditions were not responsible for pareidolia in Las Monedas.It appears that, on balance, the particular lighting conditions experienced by artists during the making of these depictions cannot fully explain why certain depictions (i.e.simple depictions scaffolded onto natural features) appear to be heavily influenced by pareidolia, and other depictions (i.e., detailed depictions with weak relationships to the cave wall) were not.This may imply that there were additional facets to the art making process that may have informed the extent to which pareidolia influenced the form of particular depictions.

La Pasiega
The figurative depictions in La Pasiega appear to lend some support to Hypothesis 1, with the slight majority of depictions (55 per cent, n=36 out of 65 depictions) having a strong relationship to topographic features of the cave wall (Table 4; Fig. 5).Of these depictions, 32 directly integrate such features and 9 depictions (5 also directly integrate natural features) have aspects of their form which are drawn in parallel to them.This may suggest that pareidolia underpinned the making of some of these depictions, but there may be additional factors that are further influencing the making of the art.
Our results also show a relationship between simple and incomplete depictions and the cave wall, with 30 out of the 36 depictions (83 per cent) with a strong relationship to topographic features being simplistic in style.However, out of those depictions with a limited or no relationship to the cave wall (n=29), only 7 bear additional detail.This therefore does not seem to support Hypothesis 2; the level of detail of the depictions does not appear to correlate with the relationship between depictions and the cave wall.Since Hypothesis 1 is supported by the La Pasiega data, this may suggest that there are other factors influencing the form of depictions beyond pareidolic imagery.As the majority of depictions in La Pasiega are simpler in style, this may represent a cultural penchant in representing animals by only their most basic and salient outline form.Hinds in Gallery A of La Pasiega (Fig. 6), for example, are predominantly represented by only a red-dotted cervico-dorsal line-a style that is common to hinds depicted during the late Solutrean and early Magdalenian in northern Spain (the so-called 'Ramales school' of depicting hinds, observed in caves such as Covalanas, La Haza, La Garma: Apellaniz 1978;Bicho et al. 2007;Straus 1987;2015).
As with the lighting simulations for Las Monedas, we used two averages for RLV with which to evaluate our data: an overall average across all simulations of 11.4 per cent; and an overall average for the variation between the brightest and dimmest points in the simulation of 18.3 per cent.Again, we divided the depictions into four categories: simple depictions with a strong relationship to features on the cave wall; simple depictions with a weak relationship to features on the cave wall; detailed depictions with a strong relationship to the cave wall; and detailed depictions with a weak relationship to such features.Not all depictions could be appropriately simulated due to low resolution in the 3D models.
There appears to be a loose relationship between simple depictions with a strong relationship to the cave wall and a low average RLV for the figurative depictions at La Pasiega, consistent with the results from Las Monedas; 52 per cent (n=11, out of 21 depictions simulated) had a low average RLV.However, within the other categories a high percentage of the depictions also experienced low average RLV: 50 per cent (n=2, out of four depictions simulated) of detailed depictions with a strong relationship to the cave wall and 67 per cent (4, out of 6 depictions simulated) of detailed depictions with a weak relationship to the cave wall had a low average RLV.By contrast, only 27 per cent (n=4, out of 15 depictions simulated) of the simple depictions with a weak relationship to the cave wall had a low average RLV.It appears, therefore, that there is no clear correlation between low light levels and the influence of pareidolia on the form of a depiction.
Similarly, the results for the variation in RLV appear to not be distinct across all categories, with approximately half of the depictions within any one category appearing to experience high variation in RLV: 48 per cent (n=10, out of 21 depictions Izzy Wisher et al. Table 4. Summary of the results for the figurative depictions in Gallery A, La Pasiega.See supplementary information for the full results from the lighting simulations.RLV could not be calculated for depictions that were small (>10 cm), too faded to identify within the VR simulation, or had low resolution.Note that panel PA11 is situated in a spatially constrained area, and thus a 3D model could not be produced for this panel.Conversations with Caves simulated) of simple depictions with a strong relationship to the cave wall; 40 per cent (6, out of 15 depictions simulated) of simple depictions with a weak relationship to the cave wall; 50 per cent (2, out of 4 depictions simulated) of detailed depictions with a strong relationship to the cave wall; and 50 per cent (3, out of 6 depictions simulated) of detailed depictions with a weak relationship to the cave wall, had a high variation in RLV.This clear lack of relationship between any particular category of depiction and high variation in RLV further suggests that light conditions did not influence, or at least were not the only factor to influence the extent of the role pareidolia had in the making of figurative depictions.While this certainly challenges the assertion made by Hypothesis 3, it is clear that at least some depictions appear to conform to the expectations of pareidolia having a prominent role in art making.For example, depiction PA12.6-a simple representation of a horse head in vertical orientation-directly traces evocative features of the cave wall's topography and experienced low but unstable lighting conditions in the VR simulation.By contrast, there are equally clear examples of depictions that challenge the hypotheses and appear not to have been influenced by pareidolic responses.Depiction PA8.1-a detailed, complete depiction of a hind that does not relate to topographic features of the cave wall-experienced particularly unstable lighting conditions, with a variation in RLV of 36.3 per cent, which one might expect would be associated with pareidolic responses.This alludes to a more complex relationship occurring with the extent to which pareidolia influenced the making of figurative cave art and encourages a deeper consideration of the interaction between the artist, their visual response to light diffusing across the cave wall surface and the subsequent, and perhaps sensorial, interaction with the cave wall.

Discussion and conclusion
In order to evaluate the contribution of pareidolia to the form and position of images in Las Monedas and La Pasiega Caves, we formulated three testable hypotheses and evaluated them using data generated from fieldwork, high-resolution 3D models and simulated light conditions in VR.Hypothesis 1- Rather, it suggests that the integration of wall and image was intentional, and therefore that the pareidolia evidenced in the majority of cases played an active part in the creation of animal representations in these caves.Hypothesis 2-that simpler (outline) depictions should have a stronger relationship with natural features of the caves' walls than those containing more detail-was somewhat supported by our results, albeit with some nuance within them.For both caves, the depictions that directly integrate natural features are overwhelmingly simplistic in style-80 per cent of the Las Monedas depictions and 83 per cent of the La Pasiega depictions that have a direct relationship to topographic features are simple in style.However, the absence of detail does not appear to be directly informed by this relationship; with La Pasiega, for example, 76 per cent of the depictions that have a weak relationship to the cave wall are also simple in style.The results are thus perhaps best interpreted as reflecting a tendency not to add detail to images placed in relationship to natural features.Perhaps this was because the incorporation of natural features was itself a form of detail.This may also be reflecting cultural style.For La Pasiega, hinds are often depicted by representing only the head and dorsal line of the animal, which is characteristic of hind depictions in the late Solutrean of northern Spain.Hypothesis 3, however-simple depictions with a strong relationship to the cave wall should experience low and unstable light conditions, and detailed depictions with a weak relationship to the cave wall should experience bright and stable light conditions-was not supported by the results, and we can reject it.It therefore seems that the reason behind pareidolia having a stronger  4 for relationship to cave wall.) Conversations with Caves influence over certain depictions than others appears not to have any clear relationship to lighting conditions.This has pertinent implications for understanding the specific mechanisms behind pareidolia in caves, and perhaps may reflect a particular penchant for seeking out pareidolic forms in the cave surface, regardless of lighting conditions.The results seem to indicate there was a negotiation occurring between the artist, the cave wall and the form of the animal depiction, with the artist responding to the topographic features of the cave wall in different ways.We thus propose three modes for conceiving of the role of pareidolic responses to cave walls in the making of figurative depictions within La Pasiega and Las Monedas: • Dominant: pareidolia was the dominant factor influencing figurative depictions and thus the cave dominated a two-way conversation, with the artist adopting a more passive 'listening' role.• Collaborative: the artist and cave engage in a twoway conversation in which pareidolia played a partial role in influencing the making of figurative depictions, but this occurred alongside tactile and other sensorial elements of the conversation as well as the artist's own intentions.• Passive: the artist dominates a two-way conversation with the cave, with pareidolia having a minimal or negligible influence over the placement, taxon and form of figurative depictions.
This framework captures the shifting dynamics at play between different agents involved in the cave art making at Las Monedas and La Pasiega.Previous representationalist approaches to Palaeolithic art have tended to assume that only the artist was 'in control' of the creation of their art, i.e. they imposed a predetermined mental representation of the depiction onto a passive surface.In recent years, however, there has been a greater appreciation of the relational nature of art production with sensorial attributes of parietal art making, such as acoustics or tactility, and the materiality of the rock support being understood actively to have shaped this process (DeMarrais & Robb 2013;Fazenda et al. 2017;Jouteau et al. 2020;Pettitt et al. 2014;Sakamoto et al. 2020).Our results further contribute to this relational perspective of Palaeolithic art making, and invite the consideration that this was a nuanced process, with the cave having the potential to exert a strong influence over the form and placement of depictions, yet equally may also have played a quieter role.Within our two sample caves, these different roles of the cave wall within the 'conversation' of art making can be observed.
Art making in Las Monedas and La Pasiega may thus have involved several forms of 'conversations' with the cave walls, mediated through sensory responses such as pareidolia, which varied in terms of the relationship between cave, artist and concepts.This is evocative of recorded examples of art making in ethnographic contexts, for example Aboriginal Australian rock-art making being a process of allowing ancestors to speak through the rock surface, with the artist themselves perceiving their role as passive (Brady & Bradley 2014;Layton 1985;Taçon 1989) or Inuit (Aivilik) figurine making being understood as a conversation with the material and 'releasing' the trapped form (Graburn 1976), with amulets or masks within different Inuit societies also having the capacity to negotiate dialogues with animals (Hill 2013; Oosten 1992).These kinds of perspectives resonate with recent discussions of the relationality and materiality of prehistoric rock art in general (Bacelar Alves 2020; Díaz-Guardamino 2020; Valdez-Tullett 2020), often referred to as the 'ontological turn' in rock-art research (Moro Abadía & Gonzalez-Morales 2020).
For the making of figurative art at Las Monedas and La Pasiega, it seems that, at times, the artist/s assumed a subservient relationship with the cave, passively responding or deliberately seeking guidance from it as to what should be created where, reflected in examples where the topographic features of the cave wall are directly integrated and grafted together with the pigment applied by the artist.For example, several depictions from both Las Monedas (e.g. depictions 5,6,21,25,33) and La Pasiega (e.g.depictions PA2.2, PA5.4,PA7.3, PA9.6, PA12.1, PA12.11) satisfied all three hypotheses; they directly integrate natural features (Hypothesis 1), are simple in style (Hypothesis 2) and their simulated light conditions were unstable and/or low (Hypothesis 3).These depictions perhaps reflect clear examples of pareidolia motivating the theme, placement and form: pareidolia was dominant in their making.
At other times the interaction could be more balanced, utilizing a mixture of topographic, tactile, light and other elements in a two-way discussion of what to depict.This may be the case for depictions that partially satisfied our hypotheses, for example directly integrating natural features and experiencing low light but being detailed in form, or where the depiction is placed in relation to areas of shadow, rather than topographic features, as appears to be the case for depictions 8 and 25 in Las Monedas (Fig. 7).Finally, the artist/s may have adopted a more dominant stance, either overprinting the cave's contribution to the conversation with their Izzy Wisher et al.
culturally informed notions of what to depict and how to depict it, or simply treating the cave wall as a blank canvas for the creation of images contained solely in their imagination.This may be the case for examples of depictions within the two caves that do not satisfy Hypothesis 1, i.e. appear to have no relationship to the topographic features of the cave wall.These conversations thus capture the shifting tensions and multifaceted dynamics of relations occurring between artist, cave wall, light, pigment mediated through sensory responses like pareidolia.
The deeper consideration of the role of pareidolia in the making of Upper Palaeolithic figurative art has important implications for our understanding of how Palaeolithic art was created and why.In terms of our three ways for creating visual culture in caves, the earliest, non-figurative phase of art-created, it appears, both by Neanderthals (Hoffman et al. 2018) and early Homo sapiens (Pike et al. 2012) -can best be seen as a collaborative discussion.While the finger dots, spat disks and hand stencils were often (but not always) associated with concavities, convexities and fissures in the cave walls, these served as a focus of interaction and did not stimulate or constrain the form of the nonfigurative creations (Pettitt 2021).Meaning was achieved through the use of the topography to 'frame' markings derived directly from the bodies of their creators.
This raises questions regarding how the ability or desire to represent 'things' emerged.If our current understanding of the chronological development of Palaeolithic visual culture is correct, it seems that figurative representation occurred after a long nonfigurative phase, perhaps shortly after 40,000 bP in Europe and east Asia, and conceivably elsewhere (Aubert et al. 2014;Conard 2009;Floss 2018;Pettitt et al. 2015).As discussed by Guthrie (2005), Malafouris (2007)   Conversations with Caves our dominant conversation category.If this were the case, seeing an animal in the evocative features of cave walls may have acted as cognitive scaffolding, allowing artists to conceive of and represent animals on a surface.In this sense, the appearance of figurative art reduced the human to a more passive role in the conversation: now, both collaborative and dominant conversations drove the creation of visual culture.Finally, perhaps as artistic behaviour became more established, artists could, should they wish, themselves determine where images were placed and the form they took, with pareidolia consequently becoming a passive agent.This might explain why the most detailed, realistic images and lifelike scenes develop only in the Late Upper Palaeolithic, a period in which the greater majority of Upper Palaeolithic art is found.While future research may allow for this to be robustly tested, we might envisage human art constituting first a collaborative conversation; supplemented later by a dominant conversation where pareidolia scaffolded the emergence of figurative representations; and finally a passive conversation, where images could be conceived of and imposed onto surfaces irrespective of their morphology.

Figure 1 .
Figure 1.Plan of the Gallery of the Paintings in Las Monedas, detailing the position of each depiction (numbers).(Digitally traced and modified after Ripoll Perello 1980.)

Figure 2 .
Figure 2. Examples of the different categories of relationships between depictions and the cave wall, where in A-C the dashed line represents a natural crack.

Figure 3 .
Figure 3. Workflow for creating lighting simulations.A photogrammetry model (A) is first imported into Unity (B), where a light source that captures the properties of a Palaeolithic torch is produced (C).

Figure 4 .
Figure 4. Ortho-images (orthorectified image accounting for distortions in the topography of the cave wall) of depictions that are simple in style and have a strong relationship to topographic features of the cave wall from Las Monedas.(A) depiction 5; (B) depiction 8; (C) depiction 12; (D) depiction 30.

Figure 6 .
Figure 6.Ortho-images of hind depictions depicted in similar styles from La Pasiega, but with varying relationships to the cave wall.(A) depictions PA5.1 (top) and PA5.2 (bottom); (B) depiction PA7.1; (C) depiction PA7.4; (D) depiction PA12.11.(See Table4for relationship to cave wall.) and Dobrez and Dobrez (2013), figurative art may have involved the first human attempt to render a three-dimensional object (i.e. an object in three dimensions moving in time) into a relatively flat, two-dimensional image.Pareidolia may have facilitated this process, resonating with

Figure 7 .
Figure 7. Digital tracings and VR simulation images of depiction 8 (A and B) and depiction 25 (C and D) from Las Monedas.The two depictions appear to be placed in relation to natural areas of shadow, giving the impression of the animals emerging out of the darkness.

Table 1 .
Summary of the figurative depictions in Las Monedas, with the numbers assigned to each depiction corresponding to the numbers in Figure1.(Data from Ripoll Perello 1980, 24.)

Table 3 .
Summary of the results for the figurative depictions of Las Monedas.See Supplementary Information for the full VR lighting simulation results.Note that due to lowresolution models, not all of the depictions were placed in the VR lighting simulation.* refers to depictions that overlap or are spatially close, and thus experienced the same simulated lighting conditions.