See ../../README-Natural.md for more info.
These are currently ad hoc and unpolished. It is still not clear how to best evaluate results.
A quick overview:
-
export-dictionary.sh-- Given an AtomSpace containing a learned dictionary (either before or after grammatical clustering), this script will export the dictionary into Link Grammar SQLite3 format. The resulting exported dictionary is directly usable by the Link Grammar parser and generator. -
dict-comp.scm-- Validate a test dictionary against a golden dictionary, using a corpus of sentences. Takes three arguments: the golden dictionary, the test dictionary, and the corpus of sentences that the two should be evaluated on. Example usage:./dict-comp.scm fake-lang learned fake-corpus/corpus-5.txt -
en-dict-comp.scm-- Perform validation of a dictionary against the Link Grammar English dictionary. Takes two arguments: the dictionary name, and a file containing sentences. It reads the sentence file, compares them to parses from the default English dictionary, and tabulates results. Usage:guile -s en-dict-comp.scm <dict-name> <sentence-file-name> -
en-test-data-- Directory containing English-language evaluation corpora. These can be used to measure parser covereage, and to compare the operation of different parsers. This test data was used to examine some of the smaller English-langauge dictionaries, circa 2019, the results of which are given in the diary, part one:learn-lang-diary-part-one.pdf.