"The highlighted tokens are primarily Japanese content words, such as nouns, verbs, and adjectives, as well as grammatical particles and function words that mark relationships or structure within sentences. These tokens often appear at or near phrase or clause boundaries, and include both standalone words and morphemes that contribute to meaning or grammatical function. The activations also frequently emphasize key information, actions, or entities relevant to the sentence's main topic."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.77 | 0.966 | 0.56 | 0.709 | 0.56 | 0.98 | 0.02 | 0.44 |
fuzz | 0.73 | 0.848 | 0.56 | 0.675 | 0.56 | 0.9 | 0.1 | 0.44 |