"The highlighted tokens are often morphemes, roots, or affixes within words across multiple languages, especially in Portuguese, Spanish, and related Romance languages, as well as some Slavic and Asian languages. These tokens frequently appear in named entities (such as people, places, and organizations), verb conjugations, noun/adjective endings, and common functional words. The activations tend to focus on linguistically meaningful subword units, including those marking tense, plurality, gender, or forming part of proper nouns and technical terms."
Score Type | Accuracy | Precision | Recall | F1 score | TPR | TNR | FPR | FNR |
---|---|---|---|---|---|---|---|---|
detection | 0.52 | 0.51 | 1.0 | 0.676 | 1.0 | 0.04 | 0.96 | 0.0 |
fuzz | 0.51 | 0.505 | 1.0 | 0.671 | 1.0 | 0.02 | 0.98 | 0.0 |