Note: This is the 2019 EuroVis version of the site, the latest version is here.
This is our attempt at reprocessing the data from our paper 2019 EuroVis paper.
In our paper we made an error in calculating distance (we had measured distance by 10x LAB bins, instead of regular LAB space) so our distance measure was 10x too small, making us underestimate what changes were noticeable.
Additionally, when we try re-running the processing now we get slightly different results, than the paper probably due to library changes and subtle effects on processing (e.g., CSV file edge cases, handling unusual unicode values) though we haven't figured out exactly what the differences are.
This is the reprocessed version of the diagram with correct distance measure. You can also see:
This visualization compares how the translations from online translators compares with our translations based on our data.
Popular terms (bold) in English
and Korean are translated by the probabilistic model (black
arrows) and online translators (gray dashed arrows). The circles
use the average color for the name, and the bars encode the translation
losses. The vertical dashed line denotes approximately 1 just
noticeable difference (JND) in CIELAB space.