Adaptive Control of Thought- Rational (ACT-R) originally developed by Anderson (1993) is a theory of computational model of human cognitive architecture. As a theory, it proposes a systematic hypothesis on the basic structure of human cognitive system and the functions of these structures in information processing to generate the human cognitive behaviour; as a computational model, to quantitatively simulate and predict human behaviour to a wide range of cognitive tasks. The model has been applied to several modelling domains like Tower of Hanoi, mathematical solving problem in classroom, navigation through a computer maze, computer programming and human memory and has provided good models of human cognition (Andersen et al., 1997).
ACT-R, as its predecessor ACT (Andersen, 1983) involves a distinction between declarative and procedural knowledge. Declarative knowledge corresponds to the things that we are aware we know and can easily describe to others. Examples of declarative knowledge include “George Washington was the first president of the United States” and
“Three plus four is seven”. On the other hand, procedural knowledge is the knowledge that we display in our behaviour but of which we are not conscious of. Andersen (1983) and Andersen et al. (1997) stated that Procedural knowledge basically specifies how to bring declarative knowledge to bear in solving problems.
In ACT-R, declarative knowledge is represented in structures called chunks whereas procedural knowledge is represented as rules called productions. Together, these two components are the building blocks of the ACT-R model.
In ACT-R a chunk is defined by its type and its slots. One can think of chunk types as categories (e.g. birds) and slots as category attributes (e.g. colour and size). They are schema-like structures consisting of an “isa” pointer specifying their category and some number of additional pointers encoding their contents (Andersen et al., 1997). Qin et al.
(2007) further added that a chunk also has a name which can be used to reference it, but the name is not considered to be part of the chunk itself. Below is a chunk encoding the addition fact that 3 + 4 = 7. The name of the chunk is Faet3+4. The isa slot is special and specifies the type of the chunk which is addition-faet in this example.
Faet3+4
isa addition-faet
addend 1 three addend2 four
sum seven
A production rule is a statement of a particular contingency that controls behaviour.
Production rules specify how to retrieve declarative knowledge to solve problems (Andersen et al. 1997; Qin et al. 2007). As an example, consider a child working on the 10s column in the following multicolumn addition problem:
234 + 746 0
An example of a production rule may be:
IF the goal is to add n l and n2 in a column, and nl + n2 = n3
THEN set as a subgoal to write n3 in the eolumn
Qin et al. (2007) noted that the condition of a production rule (the IF part) consists of a specification of the chunks in various buffers. The action of a production rule (the THEN part) consists of modifications of the chunks in the buffers, requests for other chunks to be placed into the buffers, and/or requests for other actions to be taken.
When applied to the preceding problem, this production rule would retrieve the addition fact 3 + 4 = 7 and set the sub-goal to write out 7 in the 10s eolumn. At this point, other productions would apply, which would deal with operations like carrying into or out of the column or writing out the answer. Productions in ACT-R generally have this basic character of responding to the same goal, retrieving information from declarative memory and possibly taking some action or setting a sub goal. In ACT-R, cognition proceeds step by step by the firing of such production rules (Andersen et al., 1997).
3.4.1 ACT-R Theory of Visual Attention
Although ACT-R provided good models of human cognition and well accepted within cognitive fields, however, by the standards of human-eomputer interaction (HCI), it has a serious failing. This is because it ignored many of the details by which the subject interacted with the external environment. For example all the applications of ACT-R involved people reading from a computer screen and using a mouse and keyboard but there was no theory of how this input and output took place. Therefore Andersen et al. (1997) sought to fill this theoretical gap by introducing a theory of how ACT-R interacts with computer applications. In effect, this introduces the theory of visual attention into the original ACT-R theory of higher level cognition. Figure 3.1 provides a basic overview of the system which shows the three related entities; first of which is the ACT-R system itself,
the environment with which the system is interacting (i.e. computer application in this ease), and an iconic memory which is a feature representation of the information on a screen. As can be seen, there is limited number of actions that ACT-R can take- it can issue key strokes and mouse presses to the computer, and it can move its attention around its iconic memory. Wherever it moves its attention, it can synthesise the features located there into declarative chunks that can then be processed by the ACT-R system. The computer program with which it is interacting can issue updates to the screen and then to the iconic memory, either spontaneously or in response to actions of the ACT-R.
Mourn# A ction#
K iy P r # is # s
ACT-R
Attention#!
Shift#
E n v i r o n m o n t (Application) I c o ni c
M emor y
Updat## on W indow#
and thair Contant#
Figure 3.1: relation among ACT-R, the environment and iconic memory (Andersen et ah,
Andersen et al., (1997) concluded that in processing information from a computer screen, users do not have constant access to everything as they often have to search for information. On the other hand, users rarely have to do an exhaustive search of the entire screen to find what we are seeking. ACT-R’s theory of visual attention demonstrated how ACT-R finds and extracts information from the iconic memory. The infonnation in the visual icon consists of features, but ACT-R cannot process visual features directly. It can
only process chunks representing the objects that these features compose. ACT-R can use three basic types of information to guide where attention goes on a screen: Firstly, ACT-R can look in particular locations and directions and secondly look for particular features and lastly request to scan for objects that have not yet been attended. The theory further pointed out that ACT-R can conjoin these in scanning requests, although at one time, it had been argued that attention could be drawn only by single features Tresiman and Gelade (1980).
However, a more current view is that attention can be guided by a conjunction of features, although such conjunction searches tend to be noisier (Wolfe, 1994). An example of conjunction search could be “find the next unattended bar to the left of the current location.”
This theory provides basis for this research through its prescriptions of guidelines for visual attention through order of presentation (i.e. visual arrangement) and information feature or content, as well as the ability of users to conjoin multiple features in searching for information. It is worth mentioning that this theory has also more recently been similarly applied to models of information seeking on the Web such as SNIP-ACT and ACT-IF, which predicts the observed choice of Web links in given tasks by computing the utility of actions based on an analysis of the relationship cues from the user interface to the user’s goals (Pirolli and Card, 1999; Pirolli, 2005; Pirolli and Fu, 2007).