Normal and abnormal category-effects in visual object recognition: A legacy of Glyn W. Humphreys

ABSTRACT

Are all categories of objects recognized in the same manner visually? Evidence from neuropsychology suggests they are not, as some brain injured patients are more impaired in recognizing natural objects than artefacts while others show the opposite impairment. In an attempt to explain category-specific deficits for natural objects Glyn Humphreys and colleagues suggested that natural objects are harder to perceptually differentiate than artefacts because natural objects are more structurally similar than artefacts. This explanation was proposed in the context of the Cascade model of visual object naming. While this model has been successful in accounting for a number of observations concerning category-specificity in both patients with brain injury and normal subjects, it has also become clear that there are many important aspects of category-specificity that the model cannot accommodate. These limitations have led to the development of a new model of category-effects at pre-semantic stages in visual object processing, which can be considered a further development of the Cascade model: the Pre-semantic Account of Category-Effects (PACE). Here I give a slightly historical, but primarily integrative, account of this development including recent studies which address important aspects of both the Cascade and the PACE models.

KEYWORDS:

• Category-specificity
• perceptual differentiation
• shape configuration
• visual object recognition

Acknowledgements

I wish to express my gratitude to the Friends of Fakutsi Association (FFA) and Randi Starrfelt for helpful comments.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1. Even though this precaution was voiced approximately 30 years ago it seems relevant today. At the time of writing Multivariate Pattern Analysis (MVPA)—or ‘decoding’—has become a very popular technique for analysing functional neuroimaging data; not least because it seems capable of identifying which brain areas contain information relevant for discriminating between given categories (e.g., faces vs. objects, or natural objects vs. artefacts). However, and as pointed out by Naselaris and Kay (2015), voxels identified with such techniques may in fact represent a feature or variable that is correlated with, but different from, the category-distinction being examined.

2. Seen from an object recognition perspective, there is only one notable difference between the Cascade model and the HIT, and it is that the Cascade model is a purely feed-forward model whereas the HIT invokes the idea of reentrant (top-down) processing. While this difference is important for explaining effects that arise after structural processing it makes no difference for effects that arise prior to semantic processing, which are the focus of the present paper. For this reason the HIT will not be addressed in any detail here.

3. In the HIT model (Humphreys & Forde, 2001) the notion of ‘superordinate structural descriptions’ is no longer mentioned. Rather, the ‘superordinate nodes’ are placed after semantic processing. Hence, in the HIT model the faster superordinate categorization of natural objects is assumed to be a product of mass-activation at the structural level which in turn causes mass activation of semantic representations due to cascade processing.

4. Koen Lamberts’ critique was raised in relation to the HIT model (Humphreys & Forde, 2001), but it applies equally well to the Cascade model on which the HIT is based.

Additional information

Funding

This work was supported by a grant from the Danish Council for Independent Research | Humanities (Grant Number DFF – 4001-00115).