Triple redundant signals effect in the visual modality *

r E S u m E n Los tiempos de respuesta en las tareas de reconocimiento visual de objetos disminuye significativamente si los objetivos pueden ser distinguidos por dos atributos redundantes. La ganancia de redundancia para dos atributos se ha encontrado comúnmente, pero la ganancia de redundancia de tres atributos ha sido encontrada solo para estímulos desde tres modalidades diferentes (táctil, auditivo y visual). Este estudio se extiende a aquellos resultados mostrando que el aumento de la redundancia es posible en tres atributos dentro de la misma modalidad visual (color, forma y dirección del movimiento). También se presenta evidencia de que el modelo de activación separada no puede dar cuenta de una ganancia como tal. Palabras clave autores Detección de objetivos, efectos de redundancia; redundancia triple, modelo carrera, modelo coactivación. Palabras clave descriptores Tiempos de respuesta, reconocimiento de objetos, ciencia cognitiva. doi:10.11144/Javeriana.UPSY12-5.trse Para citar este artículo: Engmann, S., & Cousineau, D. (2013). Triple redundant signals effect in the visual modality. Universitas Psychologica, 12(5), 1473-1488. doi:10.11144/Javeriana.UPSY12-5.trse * This research was supported by the Conseil pour la recherche en sciences naturelles et en génie du Canada, the German Academic Exchange Service (DAAD) and the Friedrich-Ebert-Stiftung. ** Université de Montréal *** Université d’Ottawa. For correspondence: Denis Cousineau. École de Psychologie, Université d’Ottawa.E-mail: [email protected] Sonja Engmann, DEniS CouSinEau 1474 Un i v e r s i ta s Ps yc h o l o g i c a V. 12 No. 5 c i e n c i a c o g n i t i va 2013 Introduction The environment surrounding us can be subdivided into distinct sources of information that are used to make a decision about the identity of objects. Sources of information are the different sensory modalities (e.g. auditory, visual) with different types of information within each modality (e.g. color, form, direction of motion within the visual modality). These features are processed and identified separately (although not necessarily independently) by the visual system through specialized processing channels before the object is perceived as a whole (Treisman & Souther, 1985; Kandel, Schwartz & Jessel, 2000; Ungerleider & Mishkin, 1982; Milner & Goodale, 1993). In some cases, a single feature (e.g., the color) is enough to recognize an object. Treisman and Souther (1985) have shown that if a target object differs from several distracters by one distinct feature alone (e.g., a red square among green squares) it can be detected rapidly, accurately and without conscious effort. The detection is also independent of the number of surrounding distracters. This is known as the pop-out effect (Treisman & Souther, 1985). In other cases, a combination of several features is needed for an unambiguous identification. If the joint identification of two or more features is necessary to distinguish a target object from several distracters (e.g., a red square among green squares and red circles), target recognition becomes slower and error-prone. Even if the target itself is unique among the distracters, it shares at least one feature with any one of the distracters, which makes it less easily distinguishable. This task requires central attention and its difficulty increases proportionally with the number of distracters. In yet other cases, target detection or recognition is facilitated by the presence of multiple target attributes. If a target object is defined by several features and the presence of either one of them on its own – as opposed to a combination of all target features – is sufficient to unambiguously recognize the target, target recognition is faster when more than one target feature is present. For example, red squares (i.e. targets with both features) will be detected faster than red circles or green squares (i.e. targets with only one of the two features). This is known as the Redundant Signals Effect (Kinchla, 1974) or the Redundant Target Effect (Miller, 1982). The Redundant Target Effect (RTE) is a phenomenon that has proven to be consistent and stable whenever attention needs to be divided among several modalities, locations or feature dimensions, and when several input channels separately provide the necessary information to perform a task (Miller, 1982; Van der Heijden, La Heij & Boer, 1983; Kinchla & Collyer, 1974; Van der Heijden, 1975). Bimodal and even trimodal detection tasks show facilitation if a stimulus is presented on several different modalities more or less simultaneously (Bimodal: Wundt, 1880; Fidell, 1970; Mulligan & Shaw, 1980; Miller, 1982; Trimodal: Van der Heijden et al., 1983; Diederich & Colonius, 2004, Krummenacher, Müller & Heller, 2001, Miller, 1981, Marzi et al., 1996). The Redundant Target Effect generalizes across target dimensions – form, color, orientation, etc. (Miller, 1982; Mordkoff & Yantis, 1993; Feintuch & Cohen, 2002), as well as letters and words (Morton, 1969) – and modalities (visual, auditory, and tactile; Diederich, 1995). The fact that reaction times profit from redundant signals, at least under most conditions and tasks, provides sound evidence that parallel rather than serial processing of input does happen. The RTE cannot be explained without assuming parallel processing at some stage of the processing pathway (Van der Heijden et al., 1983; Krummenacher et al., 2001; but see Townsend and Nozawa, 1995, who discuss a serial process that violate the Miller bound by assuming that the non-detection of an attribute is faster than its detection; Zehetleitner, Krummenacher, & Müller, 2009, explored this issue). However, several factors influence the size or the appearance of the RTE. Some of these are linked to the experimental design – for example non-target attribute presence or absence. If only one stimulus is present during single target trials, then the RTE is much smaller than if a distracter Triple redundanT signals effecT in The visual modaliTy Un i v e r s i ta s Ps yc h o l o g i c a V. 12 No. 5 c i e n c i a c o g n i t i va 2013 1475 is present on the other channel during single target trials (Miller, 1982; van der Heijden, Schreuder, Maris & Neerincx, 1984). It seems that attention focused on one channel – as opposed to divided attention in cases where two or more channels have to be monitored – is sufficient to reduce, or in some cases, to completely compensate for any redundancy gain (Miller, 1982). The type of task is also important: redundancy gain is typically observed in experimental paradigms of the type Go-NoGo, where a response is required of the participant if, and only if any one of several redundant features is present (Miller, 1982; Mordkoff & Yantis, 1991, 1993; Diederich 1995). Redundancy gain has also been found in a two-alternative-forced-choice paradigm (2AFC; Fidell, 1970), but it is not as large as in Go-NoGo paradigms (Grice & Reed, 1992). Spatial location of two redundant visual targets affects RTE as well: the farther they are apart, the lesser is the redundancy gain (Feintuch & Cohen, 2002; Colonius & Diederich, 2004). However, if targets in spatially different locations are bound together by grouping, the redundancy gain can be increased considerably, as they are then perceived as belonging to the same object (Feintuch & Cohen, 2002). Several types of models have been proposed to explain the redundant target effect: race models, coactivation models and various hybrids of these two. Both race and coactivation models are generally based on the assumption of independent channels that contribute to the accumulation of evidence. It is rather unlikely, however, that this assumption holds in reality. Mordkoff and Yantis (1991) showed that activity on one channel can be influenced by events on another channel, and several authors introduce lateral inhibition between channels to be able to explain their results on various reaction time tasks (Usher & McClelland, 2001; Huber & Cousineau, 2004). The race model was one of the first models proposed to explain the RTE (Raab, 1962). It assumes independent channels separately accumulating evidence in favor of the specific signal or feature to which they are tuned. As soon as one of the channels has accumulated enough evidence to surpass the decision threshold, this channel – the fastest, hence the name of the model – determines the output of the model. The RTE is explained by the notion of statistical facilitation first introduced by Raab (1962). The author showed that when sampling random reaction times across channels, the distribution of the minimal response time of each of these samples will always have a mean lower than any of the response time distributions of the different channels. However, the starting point of the minimal response time distribution cannot be lower than the minimal response time across channels, but the variance of the minimal RT distribution will be smaller than the variance of any of the individual channel distributions. Coactivation models were developed as an alternative to race models for explaining the RTE (Smith, 1968; Miller, 1982; Schwarz, 1989). Coactivation is defined as an activation build-up from different channels to satisfy a single threshold criterion. Coactivation models differ chiefly from race models in that the activation from the different channels is combined, at some point, in the processing of the input. Activation from all channels jointly determines what the response at the next processing level should be. In fact, it is the joint activation of a single threshold criterion from different channels which enables coactivation models to predict a redundancy gain: even if activation on any one channel alone is insufficient to overcome the threshold and make a decision, the pooling of activation from several still weakly activated channels makes it possible to overcome the threshold faster than with any single channel alone. Various authors have compared separate activation and coactivation models (e.g., Mulligan & Shaw, 1980; Fidell, 1970; Kinchla & Collyer, 1974; Eriksen & Schultz, 1979), but the conclusions are not homogeneous and are often contradictory (Mulligan & Shaw, 1980; Fidell, 1970; Eriksen & Schulz, 1979). Several attempts have been made to find a criterion that allows a conclusive distinction between race models and coactivation models. A possible way of excluding separate activation models irrefutably was proposed by Miller (1978). The perforSonja Engmann, DEniS CouSinEau 1476 Un i v e r s i ta s Ps yc h o l o g i c a V. 12 No. 5 c i e n c i a c o g n i t i va 2013 mance of race models on redundant target trials is simply the minimum response time of the different channels that contribute to the redundant signal (Raab, 1962). This allows us to calculate the best possible performance of race models. Two of the most well known methods to calculate the upper limit of race models are the Miller Inequality (Miller, 1978) and the Townsend Bound (Townsend & Nozawa, 1995). If response time distributions on a redundant target task exceed either of these criteria, race models can be refuted as an explanation for redundancy gain; they are not capable of accounting for the amount of gain induced in redundant target trials. The Miller Inequality has been used frequently to refute race models as the sole explanation for the redundancy gain in detection tasks with targets from different dimensions (Krummenacher, Müller & Heller, 2002), different modalities (Diederich & Colonius, 1987), and letter search tasks (Miller, 1982). Even in participants with lateral visual extinction, the RTE induced by a stimulus in the extinct hemisphere was strong enough to violate the Miller Inequality (Marzi et al., 1996). In extending the Miller Inequality to include three redundant targets, Diederich and Colonius (2004) found evidence to refute race models in trimodal detection tasks: the gain observed between double redundant and triple redundant targets alone was too large to be explained solely by separate activation models. Violation of the Miller Inequality, the Townsend Bound (see results) or any other criterion defining the upper limit of race model performance (e.g. Grice, Canham & Gwynn, 1984) is usually interpreted as evidence of coactivation somewhere along the processing pathway. However, Mordkoff and Yantis (1991) suggested an alternative explanation: crosstalk. With the Interactive Race Model, they proposed an extension of separate activation models, which integrates inter-channel crosstalk (positive or negative contingencies between target and target, target and non-target, or two non-targets on two channels) and bias towards one response. Mordkoff and Yantis (1991) explain Miller’s (1982) and similar results in terms of existing contingencies between stimuli on different channels. In a series of experiments with a letter search task, they then show that the “Miller Inequality” is not violated if all contingencies between channels are equated. In a rigorous test of the interactive race model (Schwarz, 1996), most of the results from Mordkoff and Yantis (1991) were replicated. However, under certain conditions (non-simultaneous signal presentation) violation of the “Miller Inequality” was consistently found even when inter-channel contingencies were equated. Miller (1981) also obtained violation of the Inequality in the absence of inter-channel correlations, as did Mordkoff and Yantis (1993). The latter concluded that although inter-channel crosstalk does influence response times, coactivation is at least partly responsible for facilitation in cross-dimensional redundant targets, whereas within a same dimension, separate activation is sufficient to explain facilitation. This study pursues two different goals. First, we wish to investigate if redundancy gain from three redundant target attributes, inside a single modality, is possible. To the best of our knowledge, triple redundancy within any single modality has never been addressed before. One reason for expanding the study of redundancy gain in that direction is that the ecological validity of the target paradigm increases. In a natural context, we rarely see targets, which are defined by only two target attributes. Therefore, increasing the cognitive load and studying triple redundancy gain is likely to reveal interesting insights into processing of visual stimuli. Another reason for choosing a triple redundant paradigm is that it will likely give valuable information about the dynamics of a redundancy gain. It might give us an idea about an upper limit to gain in RTs, limits with respect to the type of target attributes, which can induce triple redundancy gain, and the factors that could hide or inhibit redundancy gain. The second goal of this study is to differentiate between possible causes of redundancy gain. As mentioned above, the literature is not at all unified in attributing redundancy gain to statistical facilitation, to crosstalk or to coactivation. Based on our experimental data, we will exclude all three of these as an isolated explanation of the Triple redundanT signals effecT in The visual modaliTy Un i v e r s i ta s Ps yc h o l o g i c a V. 12 No. 5 c i e n c i a c o g n i t i va 2013 1477 RTE. Since crosstalk is likely to exist, we exclude it by allowing no possible facilitatory contingencies between target attributes on different channels. In a second step, we will reject statistical facilitation by showing that performance of participants on redundant target recognition will be significantly above the Townsend Bound. In a final step, we will also refute coactivation models as an explanation for redundancy gain by comparing the minimal response time, as well as standard deviation and skew of participants’ response time distribution with the coactivation model’s predictions for those response time distribution characteristics.