Box 2: Résumé de l’introduction aux chapitres expérimentaux
A la suite de la revue de littérature, nous proposons d’axer la présente thèse autour de trois questions :
1. La perception des expressions faciales se développe-t-elle de manière indépen-dente ou intégrée à la perception des autres dimensions du visage ?
2. L’expérience des visages a-t-elle un rôle pour les nourrissons dans la manière de percevoir les expressions de visages nouveaux ?
3. Le développement de la perception des visages de peur est-il discontinu durant la première année de vie ?
Ces questions seront traitées au cours de quatre chapitres expérimentaux. Les deux premiers chapitres (CHAPITRES 3 et 4) seront consacrés à la manière dont les ex-pressions faciales sont encodées, c’est-à-dire indépendamment ou conjointement aux autres dimensions des visages, chez l’enfant (CHAPITRE 3) et chez le nourrisson (CHAPITRE4). Le rôle de l’expérience des visages sera également abordé. Au cours des deux chapitres suivants (CHAPITRES5et6), seront mises en évidence les trajec-toires développementales chez le nourrisson de deux types d’effets des expressions faciales sur l’attention et la perception, à savoir l’influence des expressions sur la perception du regard (CHAPITRE 5), et l’influence des expressions sur la détection de visages bruités (CHAPITRE6). LeCHAPITRE7 résume et discute l’ensemble des résultats, et propose de nouvelles perspectives de recherche.
Préalablement aux chapitres expérimentaux, le présent chapitre présente briève-ment les différentes méthodes qui seront mises en oeuvre au cours des chapitres suivants, tant chez le nourrisson que chez l’enfant et l’adulte.
2.1 OUTSTANDING QUESTIONS
More than thirty years after the study byT. M. Field et al.(1982) of the perception of emo-tional faces in newborns, and more than a century after the seminal observations ofDarwin (1872), experimental research has clearly demonstrated that the perception of emotional faces emerges during the first year of life but continues to be fine-tuned up to adulthood. Outstand-ing questions concern the interaction of emotional expressions with other facial dimensions in development, mechanisms of development, and a description of developmental trajectories during the first year of life that goes beyond categorization tasks.
2.1.1 Does facial emotional expression processing develop indepen-dently or integrated with the processing of other face dimen-sions?
Neuroimaging studies (EEG, fNIRS) show a modulation of face-related components (e.g.Jessen & Grossmann, 2015) and cortical activity (e.g. Nakato et al.,2011) by facial expressions of emotions in infants and children (SECTION1.2.2), suggestive of an early integration of facial emotions with other dimensions of the face. While the recognition of face identity is generally thought of as an expression-independent process (SECTION1.1.3.2) and classic models stress the functional dissociation between variant and invariant, expression-independent face pro-cessing (e.g. SECTIONS 1.1.1.2and 1.1.1.3), numerous studies in adults suggest that facial emotions may influence, and be influenced by, other dimensions of the face (SECTION1.1.3). Thus, a theoretical question is whether face processing development leads to an increasing in-tegration of cues from initially independent streams, or to an increasing (though incomplete) segregation of cues as encoding gains efficiency. In other words, are interactions between face dimensions a developmental gain-of-function, or a built-in characteristic of face processing? For example, in adults and children, stereotypical beliefs cause facial emotions to directly bias the race categorization of faces (Amodio,2014;Dunham et al.,2013). Can similar effects be described with other face dimensions (e.g. gender, gaze) in infants and children?
2.1.2 Does experience affect how infants perceive emotional expres-sions portrayed by strangers?
Tentative evidence suggests that visual, social experience shapes the perception of facial emo-tions from an early age (SECTION 1.2.2.4) and appears to play a bigger role than cerebral maturation in the emergence of gaze cuing (Peña et al.,2014), another aspects of variant face processing. For example, it has been observed in some studies that infants will discriminate emotional expressions portrayed by their mother more readily than expressions portrayed by their father (Kahana-Kalman & Walker-Andrews,2001;Montague & Walker-Andrews,2002). Would this effect of experience generalize to new faces, for example when perceiving expres-sions or gaze cues in female versus male faces? Surprisingly few studies of emotional face perception in infants have used male faces as stimuli, and even less have used male and female faces stimuli in the same experiment.
2.1.3 Is the development of fear processing continuous or discontin-uous during the first year of life?
Group averages suggest an emergence of fear sensitivity between 5 and 7 months of age ( Lep-pänen & Nelson,2012; SECTION 1.2.2.2), an age at which a categorical perception of some expressions also becomes more evident and robust (Leppänen & Nelson, 2009). However, many studies of fear processing in infants have targeted 6-7 month-olds only, and studies using subtler analysis methods (Yrttiaho et al.,2014), individual developmental trajectories (e.g. Forssman et al., 2014), or specific paradigms (Hoehl, Wiese, & Striano, 2008) suggest a more continuous emergence of fear sensitivity during the first months of life, with some specific processing of fear existing before 6-7 months or even before 5 months. Thus, studies describing a complete developmental trajectory of fear sensitivity before, at and after the piv-otal age of 6-7 months are lacking, and the existing studies typically use the same paradigm (e.g. saccadic latency to a peripheral target, visual preference for a fearful face versus smiling face;SECTION1.2.2.2). How emotional expressions may be perceived before 5-7 months of age remains also poorly understood.
2.2 OBJECTIVES AND OVERVIEW OF THE THESIS
The above questions will be tackled in six studies described in four chapters.
In the first two experimental chapters, we focus on the encoding aspects of emotional faces perception. More specifically, inCHAPTER3we describe an effect of emotional expression on face gender categorization that is present in children as young as 5-6 years. The perceptual determinants of the effect are researched using computational models of face gender catego-rization. InCHAPTER4, over the course of three short studies we concentrate on the visual preference for smiling that is sometimes reported in 3-4 month-old infants and how it is af-fected by experience-sensitive dimensions of the face such as gender and race.
In the following two experimental chapters, we focus on the developmental trajectory, from 3- to 12-months of age, of two different attentional or perceptual aspects of emotional faces perception. InCHAPTER 5, we ask whether gaze-cuing is influenced by positive expressions (smiling) and gender (an experience-sensitive dimension), independently or in interaction. Finally, inCHAPTER6, we research fear sensitivity in an original face-in-noise detection task looking for an effect of emotional saliency on perceptual sensitivity.
Finally, in CHAPTER 7 we summarize the present findings, discuss their relevance and limitations, and outline possible future research directions.
2.3 GENERAL METHODS AND METHODOLOGICAL CONSID -ERATIONS
Before moving on to the experimental contribution we now briefly describe the methods that will be used in preverbal infants, children, and adults, along with a few methodological con-siderations.
2.3.1 Studies in preverbal infants
Preverbal infants can neither follow explicit instructions nor provide verbal or fine motor responses. Thus, behavioral research in infants has developed through the innovative efforts of researchers in designing specific paradigms that draw on infants’ behavioral repertoire (looking, hearing, touching, sucking, grasping) while giving access to the infants’ internal, cognitive world.
2.3.1.1 The preferential looking task
General principle. The preferential looking task was introduced byFantz (1964) and re-mains one of the most used when studying infant visual perception. In this task, the infants are presented with a pair of stimuli on a display, and the amount of time spent looking at either one is measured (FIGURE 2.1). Looking time indexes the relative level of interest for each stimulus in the paired display. In the original procedure, an observer peeps through a small hole in the display, measuring looking time to either stimulus online during presenta-tion. Each infant may be presented with a series of trials featuring different types of stimuli pairings on the display, usually with the left-right side of presentation reversed across tri-als for each stimuli pairing. The experiment may terminate after a fixed number of tritri-als or when the infant becomes fussy. In the first seminal observations it was noted that infants would look longer to patterned than homogeneous surfaces (for example, a black and white grating would be preferred to a gray field), so the technique was heavily used to generate psy-chophysical curves and visual acuity estimates from infants and newborns (Dobson & Teller, 1978). It has been used inCHAPTERS4,5, and6.
The forced-choice preferential looking task was introduced byTeller(1979;1997) as a vari-ant of the original preferential looking task. Interestingly, this varivari-ant does not assume which component of infants’ behavior the observer should measure (first fixation, looking time, facial expression, head turns, number and durations of individual fixations...). Instead, the task of the observer is to directly infer the location of a pattern of interest (for example, a black and
Figure 2.1: Classical example of a preferential looking task. A schematic face is pre-sented paired to a schematic scrambled face during 120 s, and a looking preference for the face stimulus is observed in newborns as well as in infants aged of 1 and 2 months (Fantz, 1961,1964).
white grating versus a gray field) based on the observation of all aspects of infants’ behavior when viewing the stimuli pairing: the ability of an observer to detect whether the infants detected the pattern of interest is used as a proxy for measuring the infants’ detection. The method was calleddouble psychophysicsbecause the observer is also a subject in the sense that he or she is completing a forced-choice task and his or her performance is assessed with regard to an objective standard. A twist of this method has been used inCHAPTER 6as a way to analyze infant looking data.
Comparison with other methods. The reliability of preferential looking to assess visual acuity in infants has been compared to that of other behavioral and electrophysiological meth-ods (Dobson & Teller,1978;Teller,1997). Preferential looking yielded estimates of visual acu-ity that were on par with that obtained with other behavioral methods (direction of first fixa-tion, operant preferential looking, optokinetic nystagmus;Dobson & Teller,1978), but lower than that obtained with electrophysiological methods (VEPs, Visually Evoked Potentials; Nor-cia & Tyler,1985). Assessment of acuity by preferential looking proved to be compatible with psychophysical techniques of threshold estimation as used in adults’ studies, although small adjustments were needed to reduce the number of trials per session (Lewis & Maurer,1986). The discrepancy between the behavioral versus electrophysiological acuity estimations might reflect differences in protocol such as the use of flickering stimuli or signal averaging as well as differences in measurement, the behavioral response of the infants being an indirect con-sequence, or reflection, of perceptual cortical processing (Teller,1997). Overall, preferential looking is considered as a robust, reliable method that has been successfully used to study lower- (Banks & Salapatek,1981;Braddick & Atkinson,2011), mid- (Kellman & Spelke,1983,
in a dis-habituation paradigm) and high-level vision (Macchi Cassia et al., 2004;Pascalis et al., 1995) in infants and newborns. It requires no training of the infant and is also appro-priate for use in infant macaques, allowing cross-species studies of visual development (e.g., Paukner, Huntsberry, & Suomi,2010).
Difficulties and limitations. First, the number of trials that can be obtained from a single infant in a single session of any given task is limited by difficulties in sustaining attention, general fatiguability, and rapid fluctuations in state from sleepy through calm through fussy (Teller, 1979). This limits the possibility to run fully within-subject designs, and renders the careful counterbalancing of conditions all the more important. Psychophysical studies involving the presentation of conditions in which stimuli of interest are barely visible prove especially difficult to run, as infants will quickly loose interest after a few difficult trials and become too fussy for the experiment to continue further (Teller, 1979): in such studies, the ratio of easy to hard trials should be as high as possible. Attrition rates may also be problem-atic.
Second, null results often cannot be interpreted. While this is often the case with null results, it is especially problematic in studies of preferential looking: a robust visual prefer-ence for stimulus A over stimulus B means that infants readily discriminate stimulus A from stimulus B, but the absence of preference doesn’t imply an absence of discrimination (Teller, 1979). Comparisons with different methods, most notably the dishabituation paradigm or Vi-sual Paired Comparison (VPC), help disambiguate the null results of viVi-sual preference studies (e.g.Di Giorgio, Leo, Pascalis, & Simion,2012).
Understanding visual preference. In fact, even positive results may prove difficult to interpret unless a clear direction of preference is expected, such as the strong preference for faces versus objects or scrambled patterns (e.g. as inMacchi Cassia et al.,2004). Specific paradigms such as VPC, dishabituation or violation of expectancy provide stronger hypotheses regarding the direction of the preference that should be observed. But while in most cases in-fants will look longer to the novel, or unexpected, stimuli, suboptimal familiarization can lead to familiarity preferences and the direction of preference may change across age groups (Kidd, Piantadosi, & Aslin,2012;Pascalis & de Haan,2003). The interpretation of spontaneous vi-sual preferences, without prior familiarization or habituation, may prove even more difficult without access to fine-grained data on infants’ visual and social experience: without this data, what is familiar or novel to an individual infant remains unknown. Thus, researchers have started documenting infants’ visual and social experience using infant head-mounted cameras
Figure 2.2: Typical object referencing experiment. Familiarization phase: (A)A cen-tral face with direct gaze and two different objects on the side is presented.(B)The face gazes towards one of the objects. Test phase: (C)The same two objects are presented without the face. A novelty preference for the uncued object is observed. Adapted fromReid and Striano (2005).
(Sugden et al.,2014) or structured parental reports (Rennels & Davis,2008). In some cases, a correlation has been reported between infants’ accumulated visual experience and sponta-neous visual preferences (Liu et al.,2015) or visual categorization performance (Damon et al., manuscript in preparation). This type of paradigm will prove helpful in describing the role of visual and social experience during development.
The particular case of object referencing. Social referencing, the gathering of informa-tion from the expressions and behavior of other people in the environment, typically involves measuring how infants and toddlers may or may not approach unusual toys or situations de-pending on the emotional reactions of their caregiver (Feinman,1982;Walden & Ogan,1988). A simplified version of these paradigms, this time involving visual preference, has been used to study social referencing in infants as young as 3- to 4-months (Hood et al., 1998;Reid & Striano,2005). In this version, a central face is presented and gazes towards one of two objects that are presented on the left and right side of the face. Such a situation elicits a preference for the uncued object when both objects are subsequently presented at test without the central face (FIGURE2.2). This preference for the uncued object may be interpreted as a novelty pref-erence similar to that that are observed in studies of visual memory using familiarization and Visual Paired Comparison (VPC). Here, the novelty preference for the uncued object reflects the encoding, or attentional, bias for the cued object that was elicited during familiarization: due to this bias the cued object is more familiar, and the uncued object more novel, at test. The method has been used inCHAPTER5of the present thesis.
2.3.1.2 Acquisition and analysis of preferential looking data
Experimental setup: stimuli presentation and raw video acquisition. A representa-tion of the experimental setup at the LPNC laboratory in Grenoble is provided below(FIGURE
TVfsetf Video Timer Camera Recorder Tripod Control Screen DisplayfScreen CPU Opaque panels DisplayfPort VGA RCA RCA Seat 60fcm Infant Caregiver A B
Figure 2.3:Experimental setup.(A)Materials and connections. Blue group: video acquisi-tion and video display in real time. Red group: stimuli presentaacquisi-tion. Electrical connecacquisi-tions, keyboard, mouse and speakers are not shown.(B)Example, without panels.
screen with a resolution of 1920 by 1200 pixels and a refresh rate of approximately 60 Hz. This distance is typically employed with infants, as it is estimated that their acuity at this particular distance is optimal. Indeed, visual acuity in infants is severely limited by retinal and convergence immaturity, and only reaches adult levels at the end of the first postnatal year (Norcia & Tyler, 1985). Panels on the right, on the left, and around the display screen limit the infant’s field of vision.
Stimuli presentation is controlled by the experimenter using a control screen with mouse and keyboard. The control screen and the display screen are connected to the same HP Z400 Central Processing Unit (CPU) through a Video Graphics Array (VGA) cable and a Bizlink Display Port cable, respectively.
A Canon XM2 mini-DV camera recorder with 20x optical zoom is mounted on a Hama tripod and records the infant’s gaze from above the display screen. The position and zoom of the camera are controlled by a second experimenter at all times to maximize data quality. During recording, video data is transmitted through an analog RCA connector (yellow RCA) to a For-A VTG-33 video timer and displayed with a timer on a small Nikkai VW58 TV set. The experimenter may use the TV display to monitor the infant’s state, and use the timer to control presentation times depending on the infant’s gaze. A common way to control presentation time is to define trial duration (e.g. 5 or 10 s) from the moment of the infant’s first fixation on the novel display. Such setting allows for individual variations in noticing and orienting to the novel display.
Raw video preprocessing and analysis. Videos are imported from the mini-DV tapes and digitalized. Individual testing sessions are parsed from the video using Windows Movie Maker.
(Department Of Human Communication Science, University College London, 2001). The sam-pling of frames from the video is done with 40 ms precision (25 frames per second) and an initial jitter to randomize sampling distribution. In a typical visual preference paradigm, the experimenter will manually code the infant’s gaze as being either on the left side of the screen, right side (if two stimuli are displayed side by side), or outside of the screen, generating raw looking times data.
Inter-coder reliability. The reliability of manual video analysis is assessed by comparing the looking times data of a subset of the total sample (25%, typically) with the looking times data of that sample as coded by a second, independent coder. One way to estimate inter-coder reliability for numerical data is the correlation coefficient: if both inter-coders agree most of the time, the correlation coefficient between their two data sets will be close to 1. Various correlation coefficients may be used e.g. Kendall’s tau (non-parametric) or Pearson’sr(linear correlation). An inter-coder reliability of or above 0.90 (Pearson’s r) is usually considered acceptable.
The phenomenon of side bias and the question of its handling. A problem that may arise when recording visual preference data from infants in paired stimuli designs is that some infants may fail to look at both stimuli during a trial. The behavior appears to occur more often in infants younger than 4-months, especially when tired. A least two factors sup-port the exclusion of such trials. First, visual preference in paired stimuli designs is supposed to rest upon the comparison of two paired stimuli; obviously such comparison may not occur when one of the two stimuli has not been perceived. Thus, a minimal looking duration towards