"The highlighted tokens are primarily parts of proper nouns, place names, or key content words in multiple languages, often marking named entities, important grammatical particles, or morphemes that contribute to meaning in context. The activations tend to focus on tokens that are semantically or syntactically significant, such as names, locations, and function words that structure information, especially in multilingual or code-switched text."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.47 | 0.483 | 0.86 | 0.619 | 0.86 | 0.08 | 0.92 | 0.14 |
fuzz | 0.53 | 0.516 | 0.96 | 0.671 | 0.96 | 0.1 | 0.9 | 0.04 |