Neuropsychology

Face Recognition Detective

Recognising someone is not one step. Put a difficulty somewhere in the model, see which tasks notice, then work backwards from six fictional profiles.

Simulated - schematic faces generated from seven numbers, fictional cases, no assessment of anybody

Learning objective

By the end you should be able to say which step of person recognition a profile implicates, and why the voice route decides it.

About 30 minutes, in two experiments. Nothing you do here is saved or sent anywhere.

This is not a test, and not a diagnosis

Nothing here assesses the person reading it: no self-test, no score for you, and nothing that can tell anybody whether they have difficulty recognising faces.

All six cases are invented. The faces are drawn from seven numbers by a seeded generator and resemble nobody. No photograph, no published test and no real person appears.

  1. Experiment 1. Put the difficulty at one component and read down the seven task outcomes. Then move it.
  2. Notice which components spare the voice route and which do not.
  3. Experiment 2. Six fictional profiles. Work backwards to the component - and decide when you cannot.

First, a judgement

A fictional person says they no longer recognise colleagues by sight. On testing, they can tell any two unfamiliar faces apart without difficulty, and they know exactly who a colleague is the moment that colleague speaks.

Where is the difficulty most likely to be?

Challenge - a claim to weigh

A fictional report reads: "Performance on a face-matching task was well below the control range, with intact performance on everyday object matching. This confirms a face-specific perceptual impairment."

What is the most important thing missing before that conclusion is safe?

What this demonstrates

Recognising someone is several things in a row

Seeing a face well enough to tell it from another face, finding that face familiar, knowing who the person is, and producing their name are separable steps. They separate because they can be affected independently, and the evidence for that is exactly the pattern you have been building: a difficulty at one step leaves everything before it intact and everything after it stranded.

The second route is what makes the argument

People can also be recognised by voice. A difficulty that ruins recognition from the face while leaving recognition from the voice untouched has to sit on the face route, before the two converge. A difficulty that ruins both sits after they converge, and is therefore not about faces at all. Without the voice tasks, those two very different accounts produce the same face-task profile.

The control task is not optional

Face perception and general visual processing produce identical profiles on every face task. The only thing that separates them is a non-face task - and only if the objects are as hard to tell apart as the faces were. Faces are unusually similar to one another, so an ordinary object task is easier, and an easier task can look intact for reasons that have nothing to do with faces.

Some profiles have no answer

Profile E is below range on all seven tasks, including the voice route and the object task. No single component in this model produces that. Before reaching for a component, the explanations to consider are effort, fatigue, low-level vision such as acuity or contrast, mood, general slowing, unfamiliarity with the specific faces used, and testing in a language or a face population the person has less experience with. Most of these are easier to check than any of the components.

Real difficulties are heterogeneous

Face-recognition difficulties differ in which component is affected, in how severe they are, in whether they were acquired after an injury or present from early life, and in how much a person has learned to work around them using hair, voice, gait, context and clothing. Two people with the same label can have very different profiles, and a profile is a description of tasks rather than a description of a person.