Bugfixing the calculation of deletion errors
Deletion errors are internally represented as notes that carry the label of the word that is not present in the result and have the "deletion" flag set. While analyzing the a recognition result (in simonmodeltest) these leaves were inserted at the end of the list of result nodes of the test sample while processing it. This caused the algorithm to count them again - this time as insertion errors - when it encountered the new, trailing nodes. Example: Prompt: Foo Bar Result: Bar Algorithm: 0. Nodes: {Bar} 1. i = 0; Foo is missing, Bar is okay -> Nodes: {Bar[Okay], Foo[Deletion]} 2. i = 1; Foo at the end is too much -> Nodes {Bar[Okay], Foo[Insertion, Deletion]} 2. i = 2; Done After the fix: 0. Nodes: {Bar} 1. i = 0; Foo is missing, Bar is okay -> Nodes: {Foo[Deletion], Bar[Okay]} i = 2 2. i = 2; Done BUG: 310099
parent
2027fbdc
Please register or sign in to comment