This is a postprint of a paper to be published in   Cognition  Kawabe,   Sasaki, Ihaya,   &   Yamada (in press )  When categorization - based stranger avoidance explains the uncanny  valley: A comment on MacDorman & Chattopadhyay (2016)  Takahiro Kawabe   a   , Kyoshiro Sasaki   b,d   , Keiko Ihaya   c   , & Yuki   Yamada   b  a   NTT Communication Science Laboratories, Nippon Telegraph and Telephone Corporation, Japan;   b   Faculty of Arts and Science, Kyushu  University, Japan   ;   c Admission Center, Kyushu University, Japan;   d   Japan Society for the Promotion of Science, Japan  Abstract  Artificial objects often subjectively look eerie when their appearance to some extent resembles a human, which is known  as the uncanny valley phenomenon. From a cognitive psychology perspective, several explanations of the phenomenon  have been put   forth, two of which are object categorization and realism inconsistency. Recently, MacDorman and  Chattopadhyay (2016) reported experimental data as evidence in support of the latter. In our estimation, however, their  results are still consistent with cate gorization - based stranger avoidance. In this Discussions paper, we try to describe  why categorization - based stranger avoidance remains a viable explanation, despite the evidence of MacDorman and  Chattopadhyay, and how it offers a more inclusive explanation   of the impression of eeriness in the uncanny valley  phenomenon.  Keywords:   Uncanny valley; Stranger avoidance; Categorization difficulty; Appearance  M otivation of this commentary  Recently, MacDorman   and Chattopadhyay (2016)  proposed that the consistency of realism among  human morphological features, rather than object  categorization,   is   a   critical   factor   determining  eeriness in the uncanny valley phenomenon (Mori,  MacDorman,   &   Kageki,   2012).   Although   we  understand what they are suggesting, at this stage we  feel that the critical part of their proposal is also  compatible   with   the   categorization - based  explanation,   that   is,   with   categorization - based  stranger avoidance (Yamada, Kawabe, & Ihaya,  2013). Her e, we describe why we believe this and  discuss scenarios where the categorization - based  stranger avoidance can still be one of the factors  contributing to the production of the uncanny valley  phenomenon.  Categorization - based   stranger  avoidance  First, le t us describe how categorization - based  stranger   avoidance   can   explain   eeriness   in   the  uncanny valley phenomenon. Categorization - based  stranger avoidance was first proposed by Yamada,  Kawabe, and Ihaya (2013). In their Experiment 1,  they used morphed images   of real, stuffed, and  cartoon human faces as stimuli and measured both a  likability score and categorization latency to classify  each image into a given categorization class. They  found that the likability score was lowest when the  categorization   latency   was   highest.   In   their  Experiment   2,   they   replicated   the   results   using  stimulus images of real, stuffed, and cartoon dog  faces.  Importantly, in their Experiment 3, they observed  that the likability score did not drop even when the  categorization   latency   i ncreased,   wherein   the  categories the observers needed to judge were both  related to humans (gender or identity).  The results of Experiments 1 and 2 are apparently  inconsistent   with   the   results   of   Experiment   3;  however   all   of   the   results   can   be   consisten tly  interpreted   by   the   categorization - based   stranger  avoidance   Yamada,   Kawabe,   and   Ihaya   (2013)  proposed. In Experiments 1 and 2, images with the  highest   categorization   latency   could   have   an  improbable appearance, such as a cartoon - like real  human face. In   this case, the observers might first  try to categorize the stimuli into a novel class. In  other words, an object with such an improbable  appearance is not categorized into already acquired  classes of objects, is probably judged as a stranger  to be avoided , and, consequently, has a low likability  score. In this scenario, the categorization latency  might increase because the object category inferred  from the stimulus appearance does not exist in a  given set of category classes for the task. On the  other hand , in Experiment 3, images with the highest  categorization   latency   could   have   a   probable  appearance, such as a boyish girl face. In this case,  the observer might try to categorize the stimuli into  a familiar class. That is, an object with such a  probable   ap pearance   is   categorized   into   already  acquired classes of objects, is not judged to be a  stranger to be avoided, and thus does not have low  likability. In this scenario, the object category exists  in a given set of categorization classes. On the other  hand,   because the category classes are too similar to  each other, it may take time for the observer to  discern which category is appropriate for the task.  Correspondence to:  Takahiro Kawabe ,  NTT Communication  Science Laboratories,  Nippon Telegraph and  Telephone Corporation,  3 - 1,  Morinosatowakamiya,  Atsugi, 243 - 0198,  Kanagawa,   Japan  E - mail:  kawabe.takahiro@lab.ntt.co.jp  Telephone: +81 - 46 - 240 - 3592  Facsimile: +81 - 46 - 240 - 4716  Citation:  Kawabe, T., Sasaki, K.,  Ihaya, K., & Yamada, Y.  (in press). When  cate gorization - based  stranger avoidance  explains the uncanny  valley: A comment on  MacDorman &  Chattopadhyay (2016).  Cognition.
This is a postprint of a paper to be published in   Cognition  Kawabe,   Sasaki, Ihaya,   &   Yamada (in press )  In   this   respect,   we   emphasize   that   stranger  avoidance is not driven simply by the categorization  difficulty   that   can   be   quantified   by   measuring  categorization latency. Rather, it is triggered when  an object has an improbable appearance and is  therefore c ategorized into a novel class.  The existence of categorization - based stranger  avoidance is supported by other lines of research.  Using morphed images of a tomato and a strawberry  as   stimuli,   Yamada,   Kawabe,   and   Ihaya   (2012)  investigated how observers ju dged the willingness  to eat. Consistent with Yamada, Kawabe, and Ihaya  (2013), the results showed that the rating score for  the willingness to eat was lowest when the rating  scores for categorization confidence was lowest and  thus   when   the   food   category   wa s   perhaps  improbable (such as a tomato - like strawberry). In  addition,   they   investigated   how   individual  differences in food avoidance behavior, i.e., food  neophobia (Rozin & Fallon, 1980), could modulate  the willingness to eat and found that the rating scor e  for the willingness to eat further decreased for  participant groups with high food neophobia scores  when categorization confidence was lowest. That  the negative impression of foods in morphed images  was enhanced by individual differences in food  neophobi a indicates that the categorization of the  foods in morphed images into a novel class possibly  causes the negative impression of the foods, and it  supports the idea that the difficulty in categorizing  an object into a novel class is one of decisive factors  determining the eeriness impression.  How stranger avoidance explains  MacDorman and Chattopadhyay  MacDorman   and   Chattopadhyay   (2016)   used  morphing images of a 3D model face and real  human   face   and   investigated   the   relationship  between   the   categorization   difficulty   (or  categorization ambiguity) and eeriness impressions.  They found that the eeriness impression did not peak  at   the   m orph   rate   with   the   most   ambiguous  categorization and that it basically decreased as the  morph rate of the real human face with the 3D model  face increased while horizontal shifts of eeriness  impression functions occurred in the morph rate  dimension. On the   basis of the data, MacDorman  and Chattopadhyay tried to rule out categorization -  related   mechanisms   from   the   explanation   for  eeriness in the uncanny valley phenomenon.  On the other hand, we suggest that their results  are not sufficient to deny the invol vement of object  categorization in the uncanny valley phenomenon  and   that   they   are   indeed   consistent   with   the  categorization - based   stranger   avoidance  explanation. As MacDorman and Chattopadhyay  acknowledged, the 3D model was the eeriest among  the stimuli t hey used. That is, it could have the most  improbable appearance of “a real human who has  3D - model   like   facial   features.”   The   stranger  avoidance   theory   predicts   that   the   eeriness  impression will decrease as the morph rate increases  in   the   stimulus   setting   o f   MacDorman   and  Chattopadhyay   because   their   stimulus   images  gradually take on a more probable appearance of “a  real human who has real human - like facial features.”  As the morphing rate increases, and the image can  be better categorized into a familiar clas s. Actually,  this   was   what   MacDorman   and   Chattopadhyay  observed. Cheetham, Suter, and Jancke (2014) also  reported   non - negative   impressions   for   an  ambiguous figure created by morphing between  avatar and human images. The results are also  consistent   with   cat egorization - based   stranger  avoidance: because the avatar originally produces  the most negative impressions, the morphed image  between it and human faces cancels the avatar’s  improbable appearance and the image is more likely  categorized into a familiar cla ss such as human.  MacDorman and Chattopadhyay explained why  they used the limited range of stimulus category in  the   following   way:   “From   the   standpoint   of  experimental control, it is difficult to investigate  transitions along a human similarity dimension ”  (page 192). We appreciate that this is, in a sense, the  proper attitude   for a   scientist who   is   trying   to  appropriately manipulate experimental components  without potential artifacts. However, to rule out the  involvement   of   the   categorization - based  explan ation in a fair manner, they should have used  a stimulus category dimension with a wider range  containing a non - human entity, an ambiguous entity,  and a human being. Although MacDorman and  Chattopadhyay   pointed   out   the   potential  involvement   of   morphing   art ifacts   in   previous  studies, not all of the morphing images used as  stimuli in those studies necessarily had such artifacts  that caused negative impressions. Moreover, recent  studies (Ferrey, Burleigh, & Fenske, 2015; Sasaki,  Ihaya, & Yamada, submitted) sho w that the eeriness  impression is strongest when categorization is the  most ambiguous and possibly the most improbable.  In addition, in Ferrey, Burleigh, and Fenske (2015),  the eeriest impressions occurred when they used  morphed images between different ki nds of animals  as   stimuli.   The   eeriness   impression’s   occurring  between different animal categories is out of scope  for the realism inconsistency hypothesis but is well  explained   by   the   categorization - based   stranger  avoidance.  Two theories:  Independent of   each other?  We have so far argued that the results reported by
This is a postprint of a paper to be published in   Cognition  Kawabe,   Sasaki, Ihaya,   &   Yamada (in press )  MacDorman and Chattopadhyay are also consistent  with the categorization - based stranger avoidance  hypothesis. We do not intend to imply that our  theory is an exclusive explanation of the eeri ness in  the   uncanny   valley   phenomenon.   Actually,   our  theory relies on the perception of object appearance  improbability to determine how novel an object  category   is.   The   evaluation   of   the   appearance  improbability is consistent with the evaluation of the  re alism   inconsistency   (MacDorman   &  Chattopadhyay,   2016)   or   perceptual   mismatch  (Brenton,   Gillies,   Ballin,   &   Chatting,   2005;  MacDorman, Green, Ho, & Koch, 2009; Kätsyri,  Förger, Mäkäräinen, & Takala, 2015). On the other  hand, our theory posits that after reco gnizing the  improbable appearance, the brain categorizes the  object into a novel class and recognizes the object as  a   stranger   to   be   avoided.   We   consider  categorization - based stranger avoidance to be a kind  of hybrid idea between category - based explanation s  and appearance - based explanations such as realism  inconsistency   and   perceptual   mismatch.   Our  definitions of object categorization and improbable  appearance are not new and are possibly similar to  the definitions (i.e., category conflict and feature  atypi cality) proposed in a previous study (Burleigh,  Schoenherr, & Lacroix, 2013).  The uncanny valley phenomenon has been found  in a continuum of stimulus appearances from non -  human to human (Mori, MacDorman, & Kageki,  2012), but there is room to discuss whet her it can be  extended   to   dimensions   not   directly   related   to  human.   The   appearance - based   explanation  emphasizes the deviation of an object’s appearance  from a human one as the decisive factor causing  eeriness   in   the   uncanny   valley   phenomenon  (Brenton,   Gill ies,   Ballin,   &   Chatting,   2005;  MacDorman, Green, Ho, & Koch, 2009; Kätsyri,  Förger,   Mäkäräinen,   &   Takala,   2015).   It   thus  focuses   on   human   likeness   in   the   stimulus  continuum from non - human to human. On the other  hand, the categorization - based explanation do es not  always focus on human likeness. It thus assumes a  more   general   cognitive   function   to   emotionally  evaluate an object on the basis of categorization  (Yamada, Kawabe, & Ihaya, 2012, 2013, Yamada,  Sasaki, Kunieda, & Wada, 2014). Unfortunately, no  study   has   thoroughly   investigated   whether   the  mechanism   underlying   eeriness   impressions   is  different   between   the   continuum   between   non -  human and human categories and the continuum  between two non - human categories. Addressing this  issue may provide more suitable   explanations for the  uncanny valley phenomenon.  Relation with inhibitory devaluation  The eeriness impression can be explained in terms  of an idea that is different from our theory but based  on   object   categorization.   One   of   the   strongest  theories is   inhibitory devaluation (Ferrey, Burleigh,  & Fenske, 2015). Ferrey, Burleigh, and Fenske  (2015) reported experimental data similar to the data  of Yamada, Kawabe, and Ihaya (2013) but gave a  different account of the data. In their theory, when  an object cate gorization is ambiguous, inhibition is  triggered to resolve conflict between competing  stimulus - related representations, and this causes the  inhibitory devaluation. The inhibitory devaluation is  an emotional evaluation phenomenon that generally  occurs in a ttentionally inhibited object (Frischen,  Ferrey, Burt, Pistchik, & Fenske, 2012; Raymond,  Fenske, & Tavassoli, 2003).  We acknowledge that inhibitory devaluation can  explain most of the reported data on the eeriness  impression in the uncanny valley pheno mena. On  the other hand, it does not fully explain the data  reported in Experiment 3 of Yamada, Kawabe, and  Ihaya, wherein they showed that the morphed image  between a male and female facial photograph (or  between different male individuals) did not strong ly  produce eeriness. It is possible that because the  morphed images between a male and female are  categorized into either of familiar classes (male or  female), the devaluation might not occur. For the  theories of both stranger avoidance and inhibitory  deva luation, there are issues to address. For stranger  avoidance, it is necessary to further test when the  difficulty in categorizing an object into a novel class  occurs. For inhibitory devaluation, it is necessary to  examine   what   category   condition   causes   the  devaluation.  Acknowledgments  This study was supported by JSPS KAKENHI Grant  Number JP14J06025 given to K.S., JSPS KAKENHI  Grant Number JP26540067 and JP15H05709 given to  Y.Y., JSPS KAKENHI Grant Number JP26750322 given  to K.I., and Kyushu University   Interdisciplinary Programs  in   Education   and   Projects   in   Research   Development  (#27822) given to Y.Y.  References  Brenton, H., Gillies, M., Ballin, D., & Chatting, D. (2005).  The uncanny valley: does it exist. In: 19th British HCI  Group   Annual   Conference:   workshop   on   human -  animated character interaction.  Burleigh, T. J., Schoenherr, J. R., & Lacroix, G. L. (2013).  Doe s the uncanny valley exist? An empirical test of  the   relationship   between   eeriness   and   the   human  likeness   of   digitally   created   faces.   Computers   in  Human Behavior ,   29 (3), 759 – 771.  Cheetham, M., Suter, P., & Jancke, L. (2014). Perceptual  discrimination   diffi culty   and   familiarity   in   the  Uncanny   Valley:   more   like   a   “Happy   Valley.”  Frontiers in Psychology ,   5 , 1219.  Ferrey, A. E., Burleigh, T. J., & Fenske, M. J. (2015).
This is a postprint of a paper to be published in   Cognition  Kawabe,   Sasaki, Ihaya,   &   Yamada (in press )  Stimulus - category   competition,   inhibition,   and  affective devaluation: a novel account of the   uncanny  valley.   Frontiers in Psychology ,   6 , 249.  Frischen, A., Ferrey, A. E., Burt, D. H. R., Pistchik, M., &  Fenske, M. J. (2012). The affective consequences of  cognitive inhibition: devaluation or neutralization?  Journal   of   Experimental   Psychology:   Huma n  Perception and Performance ,   38 (1), 169 – 179.  Kätsyri, J., Förger, K., Mäkäräinen, M., & Takala, T.  (2015). A review of empirical evidence on different  uncanny valley hypotheses: support for perceptual  mismatch   as   one   road   to   the   valley   of   eeriness.  Frontiers in Psychology ,   6 , 390.  MacDorman,   K.   F.,   &   Chattopadhyay,   D.   (2016).  Reducing consistency in human realism increases the  uncanny valley effect; increasing category uncertainty  does not.   Cognition ,   146 , 190 – 205.  MacDorman, K. F., Green, R. D., Ho,   C. - C., & Koch, C.  T. (2009). Too real for comfort? Uncanny responses to  computer   generated   faces.   Computers   in   Human  Behavior ,   25 (3), 695 – 710.  Mori, M., MacDorman, K. F., & Kageki, N. (2012). The  Uncanny Valley [From the Field]. IEEE Robotics  Automation M agazine ,   19 (2), 98 – 100.  Raymond, J. E., Fenske, M. J., & Tavassoli, N. T. (2003).  Selective attention determines emotional responses to  novel visual stimuli.   Psychological Science ,   14 (6),  537 – 542.  Rozin,   P.,   &   Fallon,   A.   (1980).   The   psychological  categoriz ation of foods and non - foods: A preliminary  taxonomy of food rejections.   Appetite ,   1 (3), 193 – 201.  Sasaki, K., Ihaya, K., & Yamada, Y. (2016). Avoidance of  novelty contributes to the uncanny valley.   Manuscript  submitted for publication .  Yamada, Y., Kawabe,   T., & Ihaya, K. (2012). Can you eat  it? A link between categorization difficulty and food  likability.   Advances in Cognitive Psychology ,   8 (3),  248 – 254.  Yamada,   Y.,   Kawabe,   T.,   &   Ihaya,   K.   (2013).  Categorization difficulty is associated with negative  evaluat ion   in   the   “uncanny   valley”   phenomenon.  Japanese Psychological Research ,   55 (1), 20 – 32.  Yamada, Y., Sasaki, K., Kunieda, S., & Wada, Y. (2014).  Scents boost preference for novel fruits.   Appetite ,  81 (C), 102 – 107.