Boya Zhou

Final project for the MIT class 4.550/4.570 Computation Design Lab
Development: February-May, 2022
Instructor: Prof. Takehiko Nagakura, Dr. Daniel Tsai & Prof. Guzden Varinlioglu

Overview

Borges was born in Argentina, while his work includes detailed imagination of places all over the world.
His writing map is actually the map of his reading history — his fantasy landscape is based on some geographic references and some fabrication, which like a random mixture.

Project Overview

This study explores the visualization of geological and fantastic landscapes in fictions by machine learning. Can we design a new kind of landscape from the inspiration of text? Can AI go beyond personal bias?

Both landscape and literature are artificial while arranged by human consciousness, which is influenced by personal experience and cultural context. Borges is a writer from Argentina, whose work includes detailed imagination of places all over the world based on his worldwide reading history. Therefore, fantasy landscape in Borges' fictions is a mixture of realistic references and fabrication. With the machine learning of natural language and pix to pix, AI can grab the geographical elements in the novel rapidly and generate a new kind of landscape that goes beyond personal bias. In this way, it is possible to explore the new genre and form of landscape and interactive reader's experiences.

Fig 1. 61 Real Placenames

A dictionary of placenames was input to pick up all the places mentioned in Borges fictions. When locating them on the global map, it is obvious that Borges' stories mostly happen in European and American cities, which echoes his reading history and growing context.

Fig 2. Landscape term dictionary to Lexical Dispersion Plot

A dictionary of landscape terms was input to pick up the landscape words in the text resources.

Fig 3. Image Selection for color coding

Each pair of "place name + landscape term" is searched on the google image to get 10 localized images. These images are color coded according to the landscape terms.

Fig 4. Sketches from different readers for Output Illustration

The machine learning is used to combine the color-coded images from internet resources to a new graphic combination according to the landscape terms appeared in the context. Currently, the combination of fiction landscape is manually color-coded.

Future Development

This experiment tests the relationship between fantasy literature created by human and illustration from internet (big data resources) and machine learning. It shows a new possibility of exploring landscape in a more flexible way. However, the manually color-coding is not precise. A better way to generate it should be explored. Also, it is necessary to figure out how to inspire the design of real landscape with these fantastic and geographic illustrations.