-
-
Notifications
You must be signed in to change notification settings - Fork 3
Description
Process Morpheus output to better reflect the order a user would select. For example, δεῖ can technically be imperfect, but it's most likely the present form. The morpheus-perseids-api should present the present conjugation before the imperfect one.
To help develop this, I've created the processing-output branch. It downloads a large number of treebanks, extracts the POSTag, and provides a script for running the code against these examples. (More information here.) We can use this branch to test different ordering methods.
Without any changes, these are the results:
Greek:
Matches: 88079/147382, 59.76%
Any match: 118388/147382, 80.33%
Missing: 6983/147382, 4.74%
Latin:
Matches: 12889/25315, 50.91%
Any match: 21403/25315, 84.55%
Missing: 1853/25315, 7.32%
Total:
Matches: 100968/172697, 58.47%
Any match: 139791/172697, 80.95%
Missing: 8836/172697, 5.12%
(Note: Matches means that the first result is the correct one, Any match means that at least one of the results produced by Morpheus is correct, and Missing means that Morpheus produces no output given the input.)