The browser can be accessed
The digital landscape is a result of several abstractions that attempt to map
semantics to space:
- We created a "word embedding model"
(Wikipedia) that represents the relationships between Brandes' words in
a much lower-dimensional space -- in this case, 200 dimensions -- than the
original text, which contains nearly 11,000 individual words.
- Although this in itself is a massive reduction in complexity, we need to further
map these 200 dimensions down to the two dimensions of a computer screen in
order to display it. We accomplish this using t-SNE (Wikipedia), a dimensionality reduction algorithm that
attempts preserve local relationships as well as the global shape.
- Finally, we create an artificial digital landscape of the resulting semantic
space using WebGL, a programming language similar to those used for advanced
In the resulting visualization, you can search for particular terms to locate them on
the digital map. Try words such as:
... pressing the "Søg" [= Search] button to jump to the word's location in the
Although the map appears three-dimensional, in all honesty we need to admit the
height is purely to aid in creating a sense of space. In the future, these mountains
and valleys could represent the word's transformations over time, its rarity, or
other aspects of its use.
- WordVectors R wrapper for word2vec by Ben Schmidt (Assistant professor of
history at Northeastern University and core faculty in the NuLab for Texts,
Maps, and Networks): https://github.com/bmschmidt/wordVectors
- Word-to-Viz visualization library by Doug Duhaime (Digital Humanities Developer,
Yale DHLab). Code forthcoming: https://douglasduhaime.com/
- Vectors: 200
- Window: 30
- Iterations: 30
Further reading on Word Vectors / Word Embedding: