Why is accurately labelling simple magnitudes so hard? A past, present and future look at simple perceptual judgment

Absolute identification is a deceptively simple task that has been the focus of empirical investigation and theoretical speculation for more than half a century. Observers are shown a set of N stimuli varying on a single dimension (e.g., length or loudness) and each stimulus is given a label (e.g., 1, .., N). They then attempt to identify stimuli presented one at at time by producing the associated label. Since Miller’s (1956) seminal paper the puzzle of why people are severely limited in their capacity to accurately perform absolute identification has endured. Despite the apparent simplicity of absolute identification, many complicated and robust effects are observed in both response latency and accuracy, including capacity limitations, strong sequential effects and effects of the position of a stimulus within the set. Constructing a comprehensive theoretical account of these benchmark effects has proven difficult, and existing accounts all have shortcomings in one way or another. We review classical empirical findings, as well as some newer findings that challenge existing theories. We then discuss a variety of theories, with a focus on the most recent proposals, make some broad conclusions about general classes of models, and discuss the challenges ahead for each class. Absolute or perfect pitch, the ability to identify the notes played on a musical instrument, is a very rare ability, and has been a subject of scientific study since the late 19th century (Ellis, 1876). It is surprising that identifying musical notes is so difficult, since humans routinely identify a huge number of things in day-to-day life: such as faces, voices, and places. Throughout the first half of the 20th century, however, psychology researchers uncovered that the difficulty most people experience in naming notes is representative of a general deficit in identifying simple stimuli that vary on just one dimension. This early research culminated in Miller’s (1956) seminal “7 ± 2” paper, in which he argued that humans were capable of accurately identifying only 5 to 9 stimuli that varied on a single dimension, regardless of the modality of that stimulus. Miller’s work also made prominent the field of study that is the focus of this chapter, absolute identification. In an absolute identification task, a set of N stimuli that vary on a single physical ABSOLUTE IDENTIFICATION 2 dimension are assigned a set of labels (usually the numbers 1 through N). On any given trial, participants are presented with one stimulus and asked to produce the corresponding label. For example, absolute pitch is a version of absolute identification in which the stimuli are tones varying in frequency and the labels are the musical note names A#, C, and so on. Other common versions of absolute identification use lines varying in length, or pure tones varying in loudness. Since Miller’s (1956) work, studies of absolute identification using a variety of stimulus dimensions and perceptual modalities have revealed an intricate pattern of phenomena behind this puzzling limitation in human ability. The complexity of the behavior elicited by such a seemingly simple task has ensured the enduring interest of the area, for example as summarized by Shiffrin and Nosofsky (1994). Absolute identification has also been of interest because of its links to other key areas of psychology, such as categorization – absolute identification is just categorization with one item per category (Nosofsky, 1986, 1997), and magnitude production – the reverse of absolute identification: given a label, participants try to produce the stimulus (De Carlo & Cross, 1990; Zotov, Jones, & Mewhort, 2011). More tantalizing is the suggested link between absolute identification and basic short term memory research. As Miller pointed out, both paradigms focus on memory, and share the same severe performance limit (7 ± 2), suggesting a common, and deep-seated, cognitive mechanism (this is also suggested by recently identified links in the sequential effects between short term memory and absolute identification Malmberg & Annis, 2012). This extensive study of absolute identification has yielded a wide range of robust benchmark phenomena not only in terms of the accuracy of responses but also in dynamic aspects, such as the time to make responses and the effect of previous responses on subsequent responses. Such a richness of data makes absolute identification a difficult challenge for cognitive modeling. We begin this chapter by giving a overview of the benchmark phenomena. These classical benchmarks focus on response accuracy. We then summarize key theoretical approaches that have been applied to absolute identification and describe how the latest models within these approaches have expanded their explanatory reach to response time (RT) as well as accuracy. We finish by discussing some of the recent, and ongoing, issues in the field.