Research Summary : A feature-integration theory of attention Treisman Gelade (1980)

Treisman and Gelade's (1980) feature-integration theory of attention proposes that the entire field of vision is available to be registered by the brain via parallel processing with respect to features (e.g. color, edges, orientation, location, movement, etc.), but the objects are not correctly unified (e.g. the color is not necessarily matched with the shape). The objects are correctly identified only when attention is focused on them which is a serial action. Serial processing combines all the separate features (conjunctions) of a particular object at a particular location together so that the object can be identified (Treisman & Gelade, 1980). If the unattended objects (objects that are not receiving attention) are not confounded with distracters or if the unattended object can be identified with top-down processing due to context or previous knowledge (i.e. the sun is yellow not blue) then a target can be found amidst the background using the faster parallel processing.

 

Treisman and Gelade's (1980) research indicated that individual features can be searched for in parallel but if a conjunction of features must be located then a slower serial search is utilized; A serial search is indicated if the slope (time to find the target) increases in a linear fashion as the number of distracters increases and the negative condition (the target is not on the screen) slope is twice the value of the positive condition; Serial search is utilized even when the there is little similarity between the conjunctive target and the distracters; A parallel search is indicated if the slope of the positive condition is virtually flat as the number of distracters increases; and, Letters are broken into their individual features and then recombined.

 

The remainder of this paper is a sample of the research methodology and results from their experiments.

 

The researchers define the following vocabulary: Dimension is "the complete range of variation" such as color or orientation; Feature "refers to a particular value on a dimension" such as red or vertical (Treisman & Gelade, 1980).

 

In their first experiment the participants scanned for either a feature (blue or S) or a conjunction (a Green T) during different trials. The distracters were variously colored letters. They varied the number of objects on the screen for each test.

 

 

Table # 1 Data from Treisman and Gelade Experiment Number 1
Test condition	Slope positive	Slope negative
Conjunction	28.7	67.1
Feature	3.1	25.1

 

 

They found that the number of objects on the screen had virtually no effect on the feature searches if the target was on the screen which indicates that the entire field of objects were processed in parallel. If the feature target (blue or S) was not on the screen then it took as long to scan as it took to find the conjunctive target (Green T) which indicates that each object was serially scanned looking for the non-existent target. In the conjunctive trials if the target was not on the screen it took twice as long to respond and the slope increased in a linear form as the number of distracters increased indicating a serial and self-terminating process of search. Further trials which allowed for practice did not improve the scan time and it was assumed that practice did not lead to automatic integration or allow for parallel conjunctive searches (Treisman & Gelade, 1980).

 

*** Individual features can be searched for in parallel but if a conjunction of features must be located then a slower serial search is utilized. ***

 

*** A serial search is indicated if the slope (time to find the target) increases in a linear fashion as the number of distracters increases and the negative condition (the target is not on the screen) slope is twice the value of the positive condition***

 

*** A parallel search is indicated if the slope of the positive condition is virtually flat as the number of distracters increases ***

 

The second experiment independent variable was the similarity of the target and distracters, number of objects on the screen, and absence or presence of the target. Low similarity was o/n or red/green. High similarity was t/x or green/blue (where the green and blue inks were very similar). The experiment disproves the supposition that parallel search is used when the conjunction target is easy to find. The conjunction target in the low similarity (easy) test was found more quickly than in the high similarity (hard) test; but become slower (linear positive slope) as the number of objects on the screen increased and in the negative condition the slope was approximately double indicating that serial searching was utilized (Treisman & Gelade, 1980).

 

*** Serial search is utilized even when the there is little similarity between the conjunctive target and the distracters ***

 

The fourth experiment tested the recognition speed for a target distracted by letters that were similar to it and easily confused; or by letters that could be separated into features and then recombined to form an illusionary target (conjunction error). The conditions were presented with various frame sizes (number of distracters) and with presence or absence of the target (positive/negative).

 

C:    Conjunction Condition                R/PQ    T/IZ

S:    Similarity Condition                R/PB    T/IY

c:     Conjunction Condition Control         R/Q        T/Z

s:     Similarity Condition Control            R/B        T/Y

h1:    Heterogeneity Condition            T/PQ

h2:     Heterogeneity Condition Control        T/P

h3:    Heterogeneity Condition Control        T/Q

 

 

 

 

Table # 2 Data from Treisman and Gelade experiment number 4
	Measured Factors
Stub head	Positive Slope	Negative Slope	Ratio of Slopes
Conjunction T/IZ	12.2	34.7	.35
Conjunction R/PQ	27.2	52.1	.52
Average Conjunction	19.7	43.4	.45
Similarity T/IY	5.3	18.1	.29
Similarity R/PB	9.7	40.5	.24
Average Similarity	7.5	29.3	.26
Heterogeneity	4.9	20.5	.24

 

 

 

    The control conditions where a target was to be located within a distracter field of identical letters (e.g. one T in a field of 14 I's) confirmed that letters that are similar in shape are found much more slowly than letters which simply share features (conjunction condition).

    

    The opposite result occurs in the test condition where the target is located either in a distracter field of letters which can create conjunction errors or a field of letters which are simply similar in shape. The time to locate the target in the Similarity Condition and the Heterogeneity Condition were basically the same. This disproves the possibility that the responsible variable was similarity of features and not conjunction of features.

 

    In the positive conjunction condition it took 2 to 3 times longer to find the target than in the condition where the target was hidden among letters which were similar to it. Letters should be so familiar that they are processed as templates but this slower time recognizing letters which share features indicates that the letters are broken into their component features and then a slow serial process must be used to identify each letter without putting it back together incorrectly with its neighbors features.

 

    In the negative condition the Conjunction condition was about twice as long as the positive condition indicating that the search was serial and self-terminating. In the Similarity condition it was about half this time indicating a different method of searching.

 

*** Letters are broken into their individual features and then recombined. ***

 

 

 

 

 

 

 

 

REFERENCES

 

Treisman, A, and Gelade, G. (1980). A feature-integration theory of attention. Cognitive Psychology 12, no. 1: 97-136. MEDLINE, EBSCOhost (accessed July 19, 2009).