According to one productive and influential approach to cognition, categorization, object recognition, and higher level cognitive processes operate on a set of fixed features, which are the output of lower level perceptual processes. In many situations, however, it is the higher level cognitive process being executed that influences the lower level features that are created. Rather than viewing the repertoire of features as being fixed by low-level processes, we present a theory in which people create features to subserve the representation and categorization of objects. Two types of category learning should be distinguished. Fixed space category learning occurs when new categorizations are representable with the available feature set. Flexible space category learning occurs when new categorizations cannot be represented with the features available. Whether fixed or flexible, learning depends on the featural contrasts and similarities between the new category to be represented and the individual's existing concepts. Fixed feature approaches face one of two problems with tasks that call for new features: If the fixed features are fairly high level and directly useful for categorization, then they will not be flexible enough to represent all objects that might be relevant for a new task. If the fixed features are small, subsymbolic fragments (such as pixels), then regularities at the level of the functional features required to accomplish categorizations will not be captured by these primitives. We present evidence of flexible perceptual changes arising from category learning and theoretical arguments for the importance of this flexibility. We describe conditions that promote feature creation and argue against interpreting them in terms of fixed features. Finally, we discuss the implications of functional features for object categorization, conceptual development, chunking, constructive induction, and formal models of dimensionality reduction.

Eigenfeatures are created by the principal component approach (PCA) used on objects described by a low-level code (i.e., pixels, Gabor jets). We suggest that eigenfeatures act like the flexible features described by Schyns et al. They are particularly suited for face processing and give rise to class-specific effects such as the other-race effect. The PCA approach can be modified to accommodate top-down constraints.

Physiological evidence predicts a model of concept categorisation that evolves through direct interaction with object feature selection. The requirement stated by Schyns et al. for feature plasticity is supported, but important caveats raise a question about the level at which feature identification can occur. Visual attribute selection for feature creation is likely to be directed by top-down and attentional processes.

Schyns et al. argue that flexibility in categorisation implies “feature creation.” We argue that this notion is flawed, that flexibility can be explained by combinations over fixed feature sets, and that feature creation would in any case fail to explain categorisation. We suggest that flexibility in categorisation is due to pragmatic factors influencing feature combination, rendering feature creation unnecessary.

Medical education provides many examples of the development of functional features, but as a response to deliberate instruction. These features require so much specificity and context sensitivity that they seem likely to require the development of new categories of appearances rather than just reweighting old features. A suggested implication is that feature development may help to explain the problematic noticing of features in diagnosis.

Implementational-level evidence of dissociable neural subsystems is a critical element that is missing from the analysis in the Schyns et al. target article. The question of whether fixed or flexible features are used in visual form recognition may have different answers for different subsystems; the evidence that features typically are created during category learning may apply most directly to a specific visual form subsystem.

Schyns, Goldstone & Thibaut's argument is evaluated from a developmental perspective. Theoretically, feature creation is not necessarily problematic; this view derives from the assumption of innate content (primitive feature sets). Alternative assumptions (e.g., Piaget's theory) are possible. Preschool children readily search for novel features in response to task demands. This is compatible with functionalist approaches, but not the rationalist ones criticized by the authors.

The constructivist notion that features are purely functional is incompatible with the classical computational metaphor of mind. I suggest that the discontent expressed by Schyns, Goldstone & Thibaut about fixed-features theories of categorization reflects the growing impact of connectionism. Their perspective is similar to recent research on implicit learning, consciousness, and development. A hard problem remains, however: how to bridge the gap between subsymbolic and symbolic cognition.

Flexible categorization clearly requires an adaptive component, but at what level of representation? We have investigated categorization in sequence learning that requires the extraction of abstract rules, but no modification of sensory primitives. This motivates the need to make explicit the distinction between sensory-level “atomic” features as opposed to concept-level “abstract” features, and the proposal that flexible categorization probably relies on learning at the abstract feature level.

This commentary is an elaboration on Schyns, Goldstone & Thibaut's proposal for flexible features in categorization in the light of three areas not explicitly discussed by the authors: connectionist models of categorization, computational learning theory, and constructivist theories of the mind. In general, the authors' proposal is strongly supported, paving the way for model extensions and for interesting novel cognitive research. Nor is the authors' proposal incompatible with theories positing some fixed set of features.

The learnability of features and their dependence on task and context do not rule out the possibility that primitives used for constructing new features are as small as pixels, nor that they are as large as object parts, or even entire objects. In fact, the simplest approach to feature acquisition may be to treat objects not as if they are composed of unknown primitives according to unknown rules, but rather as if they are what they seem: patterns of atomic features, standing in various similarity relationships to other objects, which serve as holistic features.

The fixed-feature viewpoint Schyns et al. are opposing is not a widely held theoretical position but rather a working assumption of cognitive psychologists – and thus a straw man. We accept their demonstration of new-feature acquisition, but question its ubiquity in category learning. We suggest that new-feature learning (at least in adults) is rarer and more difficult than the authors suggest.

Feature-based theories of concept formation face two dilemmas. First, for many natural concepts, it is hard to see how the concepts of the features could be developmentally more basic. Second, concept formation must be guided by “abstraction heuristics,” but these can be neither universal principles of rational thought nor natural conventions.

Because “people create features to subserve the representation and categorization of objects” (abstract) Schyns et al. “provide an account of feature learning in which the components of a representation have close ties to the categorization history of the organism” (sect. 1.1). This commentary surveys self-organizing neural models that clarify this process. These models suggest how “top-down information should constrain the search for relevant dimensions/features of categorization” (sect. 3.4.2).

The Schyns et al. target article demonstrates that different classifications entail different representations, implying “flexible space learning.” We argue that flexibility is required even at the within-category level.

We address two major limitations of Schyns et al. First, we clarify their concept of “features” by postulating several levels for processing. The composition of the feature set at each level determines the set at the next higher level, following simple structural guidelines. Second, we show that our proposed framework reconciles feature-creation and fixed-feature approaches.

Schyns, Goldstone & Thibaut argue that categorization experience results in the learning of new perceptual features that are not derivable from the learner's existing feature set. We explore the meaning and implications of this “nonderivability” claim and relate it to the question of whether perceptual invariants are learnable, and if so, what might be entailed in learning them.

It is argued that feature creation may not only depend on categorical distinctions that are made during category learning, but also on the choice set during subsequent categorization.

Cognitive theories based on a fixed feature set suffer from frame and symbol grounding problems. Flexible features and other empirically acquired constraints (e.g., analog-to-analog mappings) provide a framework for letting extrinsic relations influence symbol manipulation. By offering a biologically plausible basis for feature learning, nonorthogonal multiresolution analysis and dimensionality reduction, informed by functional constraints, may contribute to a solution to the symbol grounding problem.

Schyns, Goldstone & Thibaut oppose the notion of fixed feature analysis, suggesting the possibility of flexible feature creation in object recognition and categorisation. Such proposals cannot be assessed until clear definitions of the objects in question and their decompositions are formulated. Flexibility may come from the decompositions of objects rather than from feature creation.