Concepts and knowledge representation - summary of chapter 7 of Cognitive Psychology by Gilhooly, K & Lyddy, F, M

Cognitive Psychology 
Chapter 7
Concepts and knowledge representation

Introduction

Concepts: mental representations of classes of items.
When we treat distinct objects as the same as other distinct objects.
To represent all the distinct object that make up the categories concerned.

Dealing in concepts rather than in distinct individual objects is clearly an efficient way to work and emerges as an inevitable result of who the brain responds to stimulation, in that similar stimuli evoke similar activation patterns and by association will arouse similar memories and action tendencies.

Concepts allow us to organize information in long-term semantic memory very efficiently into hierarchical structures.

Overall, our long-term knowledge about the world is based on concepts and relations among concepts. Also, representations of current situations are in terms of concepts.

All higher-level mental concepts involves imagining possible actions in terms of concepts.

Visual images convey information as to what an object looks like and the image associated with a concept would seem likely to be important in using that concept.

Imagery: the mental representations of sensory properties of objects, experienced as like perceiving the object but with less vividness than in reality.

Theories of conceptual representation

Despite the pervasive role of concepts in cognition, there is no universal agreement on the best way of define concepts in a whole.

Definitional approach

Some concepts are well defined and clear black and white definitions can be given.
Well-defined concepts are the essence of formal subjects such as mathematics and are sought throughout sciences.
Concepts are typically formed from combinations of features that are themselves concepts. Each of these requires its own definition and within a given legal system each would have its own clear criteria.

Many and perhaps most everyday concepts are not so well defined and exhibit a degree of fuzziness. The lack of definitions can have important real life consequences.

Since most concepts that we work with in everyday life are not well defined, a major part of this area of study concerns alternative ways in which ill-defined concepts might be represented and used.

Prototype approaches

Introducing prototypes

Everyday categories have members that vary markedly in how typical they are.
If all concepts were purely definitional and well defined then all examples would be equally representative and decisions about category membership would be clear cut. But over many everyday categories, people reliably judges some examples as more typical of the category than others.

A number of aspects of performance with concepts are affected by typicality (the extend to which an object is representative of a category).

People generally find it quite easy to make typicality judgments.

Some properties are shared more or less widely among group members.
Members of a category shared a family resemblance to each other. ( the tendency for members of a category to be similar to each other but not without have any one characteristic in common to all of them). Members could be given scores for how much they resembled other members of the group.
So, if an item has three attributes and the first was also found in 16 other members, the second in 10 other members and the third in two other members, it was given a family resemblance score of 28.
The more an item has a family resemblance to other items in the category, the more typical it was rated to be.
Typicality judgments could be based on how closely the item resembled all other category members.

The item in a category that has the highest overall family resemblance to the other category members could be said to be a prototype of that category.
Prototype: an ideal example that best represents a category.

Most prototype theories do not propose that the prototype needs to be an actual instance but rather that the prototype is a statistically average member of the category. The prototype does not actually exist.
People can form a prototype without experiencing it directly.
The mental prototype seems to be build up as an average picture of the category members even though the average is never actually experienced.

Levels of categories and prototypes

Categories and concepts typically form into hierarchies.
In a conceptual hierarchy, lower level categories are nested within higher level categories.

When we deal with objects there is often a choice of level of categorization.
If we name a picture as quickly as possible, we will tend to give a

• Highlevel superordinate category level (like tool),
  The superordinate category consists of exemplars that have little in common.
• An intermediate level label (saw)
  Mid-level is the fundamental one and is the basic level of categorization. (categories formed of items that are highly similar and at an intermediate level in a concept hierarchy).
  At this level the members of the categories are highly similar to each other but the category concepts are quite distinct.
• A low-level subordinate category label (cross-cutting handsaw)
  In practice the mid-level concepts seem to be most readily evoked and used.

Another way of thinking about basic levels is to say that a basic level categorization is most informative for communication purposes.

Ad hoc categories: categories formed of items that meet a given goal (like items to take on a picknick).

Overall, prototype approaches have dealt well with some aspects of concept learning and the categorization of new examples, particularly when concrete concepts are involved. Limitations arise in dealing with abstract and ad hoc concepts, knowledge of concepts boundaries and variability and relations among features.

Exemplar-based approaches

Exemplar theories propose that categories are represented purely by stored examples or instances and each example is linked to the category name.
No prototype is assumed.
All examples are store and used. Or only selected, most typical examples are stored and used.

Exemplar models were initially applied to data from experiments in which people learned artificial categories, while prototype models were mainly tested using natural pre-existing concepts and categories.

Theory/knowledge-based approaches

Prototype and exemplar approaches are based on notions of similarity or feature sharing between instances or between instances and prototypes.

Categorization can be driven by knowledge rather than similarity. Causal knowledge plays a clear role in judgments of category membership.
(like: category drunken actions. Bound by knowledge of what people do when they are drunk).
Concepts are thought to include information about their relations to one another and about the relations among the features displayed by their examples.

Essentialism

Many people seem to believe that objects do contain some ‘essence’ which may be hard to define but which makes them what they are.
Essentialism can be seen as a special case of the theory approach and is the view that people tend to believe that category members share some essential properties and that although appearances may be useful guides to category membership, it is the essential properties that are critical.

There are different types of concepts which may have different forms of essential properties.
Three broad types of concepts:

• Nominal
  Clear definitions and so fit the classic definitional approach.
• Natural kind
  Commonly identified as naturally occurring
• Artefact
  Relate to designed and human made objects that are generally defined in terms of their functions (like cars).

Different transformations:

• Functional transformation
• Physical feature transformation
• Molecular transformation

On the essentialist view, experts in different domains, identify the true essences of the concepts they deal with and lay people tend to defer to expert judgments.

Grounded representations versus amodal representations

A more embodied view of concepts as grounded in modality specific systems for:

• Perception
• Action
• Introspection
  No need for amodal abstract symbols.
  Amodal representations: representations that are abstract and do not involve any sensory codes.

The role of simulation in cognition is stressed. Where simulation is the re-enactment of perceptual, motor and introspective states acquired during experience in the world.
Imagery is an example of mental simulation that can play a role in problem solving.

There is evidence for a role in bodily states in conceptual processing.
Like object concepts activate arm movements.

Simulation plays a role in the conceptual task of property verification.
Switching from one modality to another slows verification and property sizes also affect verification.

Lesions to brain areas dealing with different modalities affect different types of conceptual knowledge.

When conceptual knowledge about objects is activated through object names being presented, brain areas that represent the object’s properties in perception become active.
In the property verification task, areas related to the properties being tested become active including brain areas for shape, sound, action and touch

Abstract concepts often involve physical metaphors rooted in sensory experience and motor actions.

Imagery and concepts

Visual images convey information as what an object looks like and the image associated with a concepts would seem likely to be important in using that concept.

Knowledge of concepts is based on re-enactment of previous experiences with category members and so seeing and object would evoke re-enactments of previous ‘object’ related experiences, which would encompass visual experience and possible motor and haptic experience.
These re-enactments or simulations could normally be reported as imagery.
Imagery partially replicates actual experience, but can usually be distinguished as being less vivid and more under the person’s control than actual experiences.

Images may be regarded as representing the appearance of objects such knowledge of what members of common categories look like is an important part of our conceptual knowledge.

Imagery and visuo-spatial processing: overlaps?

Visuo-spatial processing: the mental manipulation of visual or spatial information.
There is interference between imagery tasks and simultaneous visuo-spatial processing, which supports the idea that imagery and perception draw on the same mental and neural resources.

Visual imagery uses the visuo-spatial sketchpad part of working memory while auditory imagery involves the phonological loop component of working memory

Image scanning and comparing

Images are usually generated for a practical purpose.

• Image scanning and comparing

Times to report finding the target part of the image varies in accordance with how far the target was from the starting point of the image. Images are like pictures in the mind.

The symbolic distance effect: the finding that difference judgments between symbolical presented items are made more easily for objects that are indeed widely different in reality.

Images are picture-like representations that are examined by the mind’s eye just as pictures are processed through the brain’s eye.

Critical views of imagery research and theory

• The image scanning results of Kosslyn could reflect participant’s beliefs or tacit knowledge about what should happen in such tasks.
• Pictures can be arbitrarily damaged but images can only be transformed in terms of meaningful components being added or removed. Also, we can perceive pictures without advance warning of their contents but images have to be intentionally constructed and are based on our knowledge of the objects being impaired.

Epiphenomenon: a by-product, it does not contribute to the machine’s functioning.
Imagery experience is a by-product of underlying cognitive processes, but has no actual functional role itself.

Ambiguity of images

Reversible figuers that typically generate alternative and indeed alternating structures.
Like the duck-rabit.

The Gestalt theory of perception:
Ambiguous figures caused unstable representations that resolved themselves into alternating representations.

If images are like precepts, then images should also be ambiguous and reversible.
Images are not exactly like pictures, but rather always have some fixed interpretation on which they are based.
In interpreting a picture and in forming an image of a creature people attend mainly to the face.

Image reversal is sometimes possible, but generally very difficult.

Neuropsychology/ neuroscience of imagery

There is large activation effects in the occipital lobe when people carry out visual imagery tasks as compared to tasks of similar difficulty levels that did int involve visual imagery.

Although similar brain areas were involved in the imagery and perceptual versions of the tasks, the areas most activated in imagery were a reduced set of those activated during perception.

There is involvement of the early visual cortex in imagery tasks.