"The highlighted tokens are primarily Hindi (and some transliterated or foreign) morphemes and words, often marking diminutives, adjectives, or nouns describing size, quantity, or relation (such as \"छोटा\" for \"small\", \"बहुत\" for \"very\", \"अपनी\" for \"own\", \"लंबा\" for \"long\", etc.), as well as suffixes and inflections that modify meaning or grammatical role. The activations focus on meaningful morphemes that contribute to descriptive or relational semantics in the sentence."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.92 | 0.957 | 0.88 | 0.917 | 0.88 | 0.96 | 0.04 | 0.12 |
fuzz | 0.92 | 0.957 | 0.88 | 0.917 | 0.88 | 0.96 | 0.04 | 0.12 |