Different regions have distinctive patterns in their place names, even if we limit ourselves to looking at letter or bigram frequencies. This is a set of visualizations that help identify the distinctive ngrams of a few different regions, and a set of scripts which could be a starting point for extending the analysis to other parts of the world.
- Download a list of placenames
- Filter it
- by feature type: For the US, I am only using river names
- by uniqueness: For the US, I only retain words that are unique to a single state
- Bucket the names in a latitude-longitude grid
- For each square in the grid
- count the occurences of each ngram
- and calculate the ratio of this count to the counts across the whole set
- Filter out ngrams
- whose frequency is low
- or whose ratio to the whole set is low
What remains are lat-long buckets of bigrams that met a minimum count threshold, and which are more frequent in a particular bucket than elsewhere.
The pipeline populates the git-ignored cache/ directory,
and copies the final JSON into docs/.
From a fresh checkout, install dependencies for the pipeline:
npm install
Then run the pipeline:
pipeline.sh
Then install dependencies and start jekyll:
cd docs
bundle install
bundle exec jekyll serve