Perceptual and Category Processing of the Uncanny Valley Hypothesis' Dimension of Human Likeness: Some Methodological Issues

The overall goal of this procedure is to examine effects of categorical perception along the dimension of human likeness of the uncanny valley hypothesis. This is accomplished by first carefully generating morph continua with which to represent the dimension of human likeness. The morph continua are then presented in a forced choice categorization task.

This task is used to determine which stimuli are clearly categorized as avatars and as human, and to define the location of the category boundary. Next, a selection of these images is used in a perceptual discrimination task. This task determines the presence and location of the perceptual discrimination boundary.

The results show that the discrimination boundary is aligned with that of the category. Boundary alignment indicates that the subjective perception of objects along the dimension of human likeness can be understood in terms of the psychological effects of categorical perception. The main advantage of this method is that it allows experimentally controlled, fine-grained manipulation of human-like appearance.

This is important for understanding the subtle relationships between perceived human likeness and uncanny related experience and behavior. The visual demonstration of this method forms the basis for the critical discussion in the accompanying article of this method's use in uncanny valley research Begin by selecting pairs of human and avatar images to be used as the endpoint of each continuum in the morphing procedure. Start with the human endpoint.

Choose an image of an unknown in distinctive face with a neutral expression and direct gaze. No other salient features such as facial hair or jewelry should be present. For the non-human endpoint image, generate an avatar face to match the human face.

Alter each avatar face to match the corresponding human face closely for age and general facial geometry. Match the contrast levels and overall brightness of each pair of parent faces. Here, the skin tone of the avatar parent face is matched to the corresponding human parent face.

Use photo editing software to crop the external features of the faces by overlaying an elliptic form over the image. Next, adjust the final position of the images to ensure alignment of general facial geometry between each pair of endpoint images. When the images are ready, use the morphing software to place control points on the features of the parent faces.

For each face, around 20 points are placed on the mouth, 18 on each eye, 20 on the nose, and eight on each eyebrow. Once these points are placed, the images are ready to be morphed. In this 13 different morphed images are created for each continuum and labeled M zero to M 12.

There are two parent faces and 11 intermediate morphs as seen here, a two alternative forced choice classification task is performed. The participant judges whether each morph image shows an avatar or human face. When ready, the participant views a sequence of trials.

Each trial begins with a fixation point presented for 500 milliseconds, followed by a morph image for 750 milliseconds. The participant responds after each image by pressing the appropriate button. The next trial then begins later.

The avatar Human classification data is analyzed using polynomial regression to describe the shape of the response function. To set up the same different perceptual discrimination task, select two morphs previously categorized as avatars and two as human control for physical differences by selecting morphs that represent equivalent increments of physical change Along each continuum, increments of 33%are used as seen. Here sort the selected morphs into pairs according to three experimental morph pair conditions.

Same in which pairs of morphs are identical, representing no physical or category change within, in which pairs are drawn from within a category and between in which pairs represent different categories. For the avatar category, ensure that the first morph of each face pair in the three conditions is always M four. The same procedure can be applied for morph pairs in relation to the human category, ensuring that the first morph is always M eight.

A same different perceptual discrimination task is performed. The participant indicates by button press whether the faces of each face pair are the same or different in appearance. When ready, the participant is shown a sequence of trials for each trial.

A fixation cross is presented for 500 milliseconds, followed by the first face of a face pair for 500 milliseconds, an inter stimulus interval showing a blank screen for 300 milliseconds. Then the second face of a face pair, the participant then decides whether the faces of the pair are the same or different. An interval of 2, 500 milliseconds follows before the next trial is presented later.

Discrimination accuracy is analyzed for face pairs that cross the category boundary compared with face bears from the same side of the boundary. The data for avatar trials and human trials are treated separately in analysis seen here are representative results for the forced choice categorization task. These are presented in terms of the percentage of responses in which a face is categorized as human.

The solid blue line shows the fitted logistic response curve, the slope of which reflects a sigmoid step function consistent with the presence of a. The blue dash line shows the estimated position of the category boundary, indicating that the maximum uncertainty of 50%in categorization judgments is associated with more position number six. In these results for the forced choice categorization task, the longest mean response latency is associated with morph position number six.

This is consistent with the presence of greatest uncertainty and categorization judgments at this morph position. Discrimination responses in the same different perceptual discrimination task show that accuracy and discriminating between pairs of faces that straddle each side of the category boundary that is in the between conditions is greater than that for pairs of faces drawn from within the human and avatar categories. Consistent with the greater discrimination accuracy at the category boundary, the judgment response times for pairs of faces that straddle the category boundary in the between condition or shorter than those for faces drawn from within a category.

When applying this procedure, the greatest difficulty is in generating multiple morph continua that form an homogenous set of experimental stimuli. Poor control over the experimental stimuli can introduce confounds and artifacts that strongly influence the subjective perception of human likeness. The accompanying article discusses these related issues.