How Do We Learn to Categorize Objects   in the World?

Research team:
Researcher: Rubi Hammer
Advisor: Prof. Shaul Hochstein
The Interdisciplinary Center for Neural Computation and the Department of Neurobiology, Hebrew University of Jerusalem

Background
Categorization is a cognitive process that allows us to relate to unfamiliar objects in an appropriate manner based on our familiarity with other objects that are similar to them. To survive in an environment filled with many different objects, people, and animals, we must be able to categorize quickly and in a way that is meaningful. In nature, for example, we can quickly and efficiently differentiate between predators and animals that can be hunted, even though they may be similar in color, form, or size.

Although, as humans, most of us have the ability to categorize objects, we do not fully understand the mechanisms that allow us to do so, although it is a central subject of study for the past several decades. A common belief is that we categorize objects according to their overall similarities. For example, we would expect even a young child who is not familiar with the animals in Picture 1a to be able to pick out the one that does not belong:

Picture a

Picture b

Nevertheless, rough similarities among objects do not always allow us to categorize them correctly. For instance, to be able to differentiate between the seagull and the ducks presented in figure 1b, we need to rely on prior knowledge. Otherwise, we might mistakenly choose animal number 2 as the one that doesn't belong (since it clearly differs from the others in color).

We can acquire such prior knowledge through either direct instruction or by comparing few objects we already know their categorical relation. This type of comparison may help us identify the important characteristics that better define each different category.

The Research Question
Does comparing a pair of objects known to be of the same type vs. comparing a pair of objects known to be of two different types contribute differently for learning new categories?

The Research Process and Methods
The current research has two facets:

1. The theoretical facet, which describes the amount of objective information that can be gained from each of the two comparison processes. We performed the calculations using mathematical and statistical models.
2. The facet of the subjective ability of people, at different stages of development, to make practical use of information available to them from the two comparison processes. To measure this ability, we conducted behavioral experiments to examine the ability of people of various ages to perform under different conditions.

Specifically, we wanted to examine whether the ability to learn principles for categorizing sets of unfamiliar objects by comparing objects of the same type, as opposed to comparing objects of different types, changes with age. The participants in the experiment were asked to determine if the unfamiliar objects appearing together on a computer screen were of the same type or different types. At the pre-learning stage, all the participants could do was guess the right answer. In the second stage, some of the participants learned the rule for categorizing the objects by comparing objects of the same type, while the other participants learned this rule by comparing objects of different types.

Following the learning stage, the participants were once again asked to determine if the pairs of objects appearing on the screen were of the same or different types. By comparing their performance levels at this stage, we were able to estimate the effectiveness of the categorization rule they had learned for both types of comparisons. A schematic diagram of the experiment would look like this:

The principle here is to test the brain as if it was a "black box" that receives sensory input, processes it, and generates appropriate behavior. By recording the way the task is performed by participants in different experimental conditions, we learn about the computational limitations of the human brain.

The Results
We learned from our theoretical analysis that the amount of information that can be gained by comparing objects of the same type will, in most cases, be greater than the amount of information that can be gained by comparing objects of different types. If being informed that the animals labeled 2 and 3 in Picture 1b are of the same type, we can determine, by comparing them, that color is not a relevant characteristic for categorizing animals, since two animals of the same type can have very different colors. Knowing that animal 1 in Picture 1b is not of the same type as the other three will not necessarily teach us the characteristics that differentiate seagulls from ducks. Only in the rare instance when we compare objects of different types that, nevertheless, are very similar in appearance, is there a chance that we will differentiate them based on characteristics that are relevant to their categorization. Knowing that animals 1 and 3, for example, are not of the same type, highlights characteristics such as head and beak shape, which indeed distinguish seagulls from ducks.

The fact that comparing objects of the same type much more often provides the required information points to an evolutionary need for the ability to learn principles for categorizing objects by comparing objects of the same type. On the other hand, comparing objects of different types will be effective in only very rare cases, but nevertheless in those cases this type of comparison can provide further information that can not be gained by comparing objects of the same type.

The question arising from this theoretical analysis is: How do people use these two types of information?
Specifically we tested people capability to learn by either one of the two comparison types when the objective amount of information provided by comparing objects of the same type was identical to that provided by comparing objects of different types. We wanted to test how this objective diference affect the subjective learning strategies people develop.

The research hypothesis is that since comparing objects of the same type is more useful in everyday life scenarios, it will evolve earlier in development.

The findings of our behavioral experiment indicated that when the participants were trained by comparing objects of the same type, young children demonstrated the same learning ability as older people. However, when the participants were trained by comparing objects of different types, the younger children demonstrated less learning ability than older children and adults.

More about...
To confirm or disprove a possible explanation in science, we must isolate the factors relevant to the subject at hand. This principle is valid no matter what type of system our experiment examines-physical, biological, psychological, or social. For example, in conducting a behavioral experiment that examines the influence of human development on the ability to distinguish between different groups of objects, we would choose participants from different age groups (an independent variable) and measure the percentage of success for each group in performing the relevant task (a dependent variable). The results of the experiment provide a quantitative representation of the phenomenon under research. By performing a statistical analysis of the results, we can make inferences about their relevance in more general situations.