Scholar Bria Long on “Drawing as a Window into the Development of Object Representations”

By Kelli Agnich, Teacher

Using a pen or pencil and a piece of paper, draw a phone, a rabbit and a train.

Now, take a look at the distinctions you included to convey the object’s identity: Does your phone have a cord, or does it more closely represent a smartphone? If the latter, what distinguishes your phone from another rectangular screen-based device? Does your rabbit have the telltale long ears and a bushy tail, or might it be mistaken for another furry animal? How about your train? Is it a sleek, modern engine, or does smoke billow from its vintage smokestack?

By the time we’re adults, we can easily think of what an object looks like and produce a drawing of it. We know that trains are large and have wheels used for transportation, while flowers have a stem and an array of petals. But we aren’t born with the knowledge of what different categories look like.  Instead, we must learn which combination of features is associated with each category.

How do we go from knowing nothing about the world to the point at which we simply look around and effortlessly recognize the objects around us? As children develop and learn about the world around them, their growing knowledge likely influences their representation of objects. And this knowledge might be reflected in their drawings: 20 years ago, children’s drawings of a banana and a phone might have been similarly shaped; however, in 2019, illustrations of phones less frequently include a spiral cord or a handset.

Bria Long, a postdoctoral fellow in the Stanford Language and Cognition Lab, presented her research on the development of drawing behaviors in children to the Bing staff on our staff development day last October. She has been collecting drawings of common objects made by students at Bing over this school year in order to study how children learn to draw object categories, and how this is related to what they are learning about these categories.

In previous work, Long collected drawings at a local children’s museum and found a dramatic shift occurs in a child’s ability to produce recognizable drawings between the ages of 4 and 9. To quantify how recognizable children’s drawings are, Long presented these drawings to adults and asked, “What does this look like?” Each adult selected from a predetermined list of objects and Long examined the proportion of adults who recognized a drawing against the age of the child who drew the object. Quite interestingly, the data showed that drawing recognizability plateaued around 6 to 7 years old; however, even very young children’s drawings still contain a rich amount of information, once the observer knows what the child was attempting to draw. For instance, a child’s crescent-shaped drawing might easily be mistaken for a moon—but in fact resembles the banana they were trying to draw.

In an attempt to explain these age-related changes in recognizability, Long examined how the recognizability of children’s drawings is related to children’s age as well as to several other factors: the number of strokes used, the amount of ink used and the time children spent drawing. She found that irrespective of these additional factors, older children’s drawings are, in fact, more recognizable. Long then used deep-neural networks (trained to recognize objects in photographs) to assess the visual similarity between drawings of different categories. This method captures visual similarity relationships between categories: For example, while younger children’s drawings of a train, bus and car might not be recognizable as such, they still share characteristic visual features (e.g., circles designating wheels) that are picked up by this method. Using this technique, Long can then ask how the visual similarity between drawings of objects changes across development. Overall, Long has found that as children get older, their drawings better capture the relevant visual distinctions between categories: They start to include the relevant visual features that distinguish, for example, trains from cars or rabbits from dogs. And one possibility is that this might reflect their increasing knowledge about these different categories.

Using the data from Bing children’s drawings over this year and next, Long will be able to analyze the process by which children begin to draw different object categories. In particular, these studies at Bing present a unique opportunity to track how individual children’s drawings change over development—as many children have a particular way in which they draw certain objects. By characterizing the variation and consistency in children’s drawings of object categories, Long hopes to gain insight into children’s evolving knowledge of the world around them.