Correlating borrowing events across concepts to derive a data-driven source of evidence for loanword etymologies

(Verena Blaschke & Johannes Dellert)

Given that a target language A borrowed a word for concept C from a donor language B, which other words did A probably borrow from B?

Talk given at the Model and evidence in quantitative comparative linguistics workshop at the 43rd annual conference of the German Linguistic Society (DGfS) (Feb 26, 2021). [Abstract] [Slides]

Running the code

First, generate the bootstrap sample:

python bootstrap.py

To generate the list of concept pairs passing given NPMI and implication strength thresholds (and to generate graphics visualizing grid-searching good thresholds), run:

python inspect_bootstrap_data.py

Create the graph that shows pairs of loanwords that tend to be borrowed together:

python implication_graph.py
dot -Tsvg out\loanwords-80-60.dot -o out\loanwords-80-60.svg

(80% and 60% are the implication strength and NPMI thresholds, respectively. Change the configuration section within the Python script to change the thresholds.)

Other scripts:

• field_size_viz.py visualizes the loanword ratio and number of entries by semantic field
• loanwords.py contains mostly methods called by the other scripts. Running this file produces some files containing an overview of the NPMI values and implication scores in the data.
• semantic_net.py is called by other scripts to calculate the semantic distance between two concepts.
• singleton_implication_viz.py visualizes the number of concept pairs passing the NPMI and implication strength thresholds compared to all other concepts by borrowability.