Schema Theory Act Anderson And Pdp example essay topic

HOW IMPORTANT ARE MENTAL REPRESENTATIONS IN COGNITIVE THEORIES? How the world around us is represented mentally is the corner stone of cognitive architectures. It facilitates understanding of information received and perceived from our environment. The storage and retrieval of knowledge would be impossible without mental representations. Mental representations are the way in which we create 'copies' of the real things around us, which we perceive. A description of a representation is a symbol, sign, image or a depiction that takes the place of a real object in the real world...

Representations were broadly categorized into three. The 'analogue representation' the 'propositional representation' and 'procedural rules'. Analogue representations are those which have an image-like copy quality to them, whereas the propositional representation are based on language-like constructs. Since the arrival of connection ism another representation has been proposed that of sub-symbolic representation. Here mental representations, according to Eysenck and Keane (2002) are "distributed" patterns of activation in a network.

Historically, mental representations have been interpreted by analogy with physical representations, i.e. descriptions and classifications devised for physical representations have been applied to mental representations (Paivio, 1986). Physical representations can be picture-like or language-like (see Table). Physical and mental representations physical representations picture-like language-like examples photographs drawings maps diagrams human-language formal systems: maths, symbolic logic computer programs properties analogue iconic continuous non-analogue non-iconic digital / discrete Table: Types of physical representations (after Paivio, 1986) The representations need then to be categorized for storage in long -term memory. These 'packages' of knowledge are classed as being either procedural knowledge or declarative knowledge. Procedural knowledge is knowing how to do something or precisely what to do. It is sets of rules or procedures and skills like playing the piano.

Declarative knowledge is about facts. Representations allow cognitive models to work as they are the 'substance' the models work on. The models for discussion share common features but are equally differentiated from each other at some level. Before looking at each of the theories mental representations it would be helpful to take a snapshot of the model structures and approaches to learning and processing to gain a fuller understanding of their strengths and weaknesses.

The models compared here are Schema theory (Rummelhart and Norman 1983) ACT Anderson) and PDP. Schema theory is said to offers a unified theory of cognition as it umbrellas all areas of cognition. It is interactive and works on stored knowledge or long-term memory. It does not address any wider structural issues. Schema is about how our learning is influenced by our previous knowledge. Brewer and Tremens (1981) set up an experiment to show that a persons memory for a scene was influenced by the schema for that scene.

They correctly predicted that participants would recall more of the expected items from the room and less of the unexpected items. However a skull which was of low expectancy for the schema was recalled suggesting that recall is not completely schema driven. The mental representations used by schema theory are propositional and symbolic. All information arriving to be processed is interpreted with respect to knowledge in long-term memory and treated accordingly.

It is then assigned slots in an existing schema or a new one is created. Schemas consist of hierarchical organised packages of information with various relationships, variables, slots with values or default settings. Contained within these slots are concepts or sub-schemata. This makes a flexible system. An example of which is a theatre schema: Entertainment Schema Outings THEATRE SLOTS Place (inside) Production Dress code People DEFAULT Everyman Contemporary Smart Friends SPECIFIC EPISODES Bristol Little Hippodrome black Partner dress A schema theory used in language is called scripts, and was proposed by Shank and Abelson (1977) as away of explaining peoples knowledge and expectations for everyday events.

Their well known Restaurant script was designed to test whether people would agree about which events occur in a restaurant. The idea being that we store scripts in memory to allow us to make sense of stories which concern typical events. They found that when scripts written by participants were compared there was general agreement about the main events in that scenario. ACT is very similar to schema theory as the mental representations here are also propositional and symbolic.

Since it is a computer model it can be programmed as a memory system, a language processor or a problem solver (e.g. the Towers Of Hanoi. ). Schema theory focuses only on long term memory, whereas ACT acts on working memory and two kinds of long term memory, declarative and procedural. Of the three models this is the only one to address the overall structure of what is being modelled.

ACT representation is organised similarly to schema theory, in organised packages of information but for declarative memory only and it is not a strictly organised hierarchy but a tangled one. Procedural memory is represented as a production system i.e. the working memory and sets of production condition / action rules. As such it is a well-specified account of all aspects of cognition. Unlike schema theory ACT is modular and processing is strictly serial. Information retrieved from declarative memory is achieved via a spreading activation. The procedural memory has to check conditions (rules) in the long-term memory with the working memory active node patterns to achieve a match.

This gives rise to a bottle-neck in processing as the pattern matching is serial. Goals are achieved by acting on production rules in a prescriptive hierarchy. The last comparison is the PDP model. This model has a network structure whose mental representations are sub symbolic. It has a structure where elementary units (compared to neurons in the brain) are connected together and like schema relates only to long-term memory not addressing overall structure. The nodes themselves do not hold the meaning but meaning is represented in the patterns of activation across the network or a number of localised nodes.

A Multi-layered Connectivist Network Output units Hidden units Input units The difference between the pattern matching here and that seen in ACT is tat PDP is parallel processing and not serial and so is faster at processing. Schema theory gives a good account of top down effects on what we learn or remember. However, the detail in the process and the how in learning are not specified very well. ACT gives a good account of how skill learning takes place by. This is not so well specified and is more creative learning. PDP describes basic associative learning as well.

This may provide some of the explanation as to how schemes are acquired. This models only low-level cognition. Taking each model in turn we can summarise their contribution to learning. Schemas provide the framework and reference on which all of our learning takes place. Anderson and Pichert's 1978 experiment aimed to test whether schemes operate at encoding or retrieval.

They gave participants a story in one schema and then later tested their recall using a different schema. With the premise that if schemes only operate at encoding or retrieval the old schema would not be affected by the new schema. Their results showed that schemes have an effect on both encoding and retrieval. Those new memories are representations built from bottom up sensual information and top down stored schema.

As a result of the schema retrieval our new memories are normalized. Linton's 1975 diary study is a example of this. She remembers the first meeting she attends but subsequent ones blend into one another. The ACT model uses declarative knowledge and via the use of production rules and repetition of a skill, that skill becomes automatic or. PDP models learn by changing the weightings on nodes and the links between them. These models can make generalizations unlike the schema or ACT.

It is a good model of how we forget, known as graceful degradation. It is not an abrupt forgetting as it happens gradually as unused links in the neural net are 'pruned', paralleling human forgetfulness well. Rummelhart's model did not learn. The exceptional feature of the PDP model is if given sufficient examples the model can generalize as if it had learned a prototype.

This is one of the emergent properties of PDP models. An emergent property of a model is one which results from the natural behaviour of the model and that is being explicitly built in. (Cohen 2003). Another emergent property is content address ability, the ability to recall other information about an object when presented with a cue. It is exceptional because the model is essentially a computer program.

There are no emergent properties in Schema theory or ACT. Redundancy of information can be built in however. Generalisation is a key process of Schema theory but has to be programmed into ACT. It is felt that the greatest weakness of schema theory is the lack of process detail.

The slow pattern matching is the greatest weakness for the ACT. PDP has problems with single -track learning. In conclusion schema theory strengths lie in its ability to account for the organisation of our general knowledge and how this influences expectation and interpretation of new events. It works on high level cognitive processes. ACT incorporates a schema-like declarative memory which shares many of the advantages of schema theory. The advantages are in the representation of actions by producing rules and being much more clearly specified.

It can be used in a variety of applications. PDP models come into their own to explain low-level automatic processes, but there is some question as to the ability of the model to explain higher level cognition. The advantage of PDP over ACT is in the natural emergence of key properties, e.g. fast direct pattern matching and content accessibility amongst others. All these models have strengths and weaknesses but their commonality is their need for mental representations of the information being considered.

These representations take different forms i.e. symbols, scripts, rules but ultimately they are mental representations and as such provide the foundation for all three of the theories discussed.

Bibliography

Anderson, J. (1983).
The Architecture of Cognition Harvard University PressAnzai, Y., and Simon, H.A. (1979).
The Theory of Learning by Doing. Psychological Review 86 124-180 Collins, A & Quillin, M.R. (1969) 'Retrieval time from semantic memory' Journal of Verbal Reasoning and Verbal Behaviour, 8,240-7 Eysenck, M.
W., (2000).
Everyday memory. In Eysenck, M.W., & Keane, M.T., Cognitive Psychology: A Student's Handbook 4 then. Hove, UK: Psychology PressRumelhart, D.E. and Norman, D. A (1983) 'Representation in memory in R.
C. Atkinson ed Handbook of Experimental Psychology, Wiley and Sons. Rummelhart, Smolensk y, McClelland & Hinton 1986.