Multilingual Web Register Classifier
The multilingual register classifier was fine-tuned on CORE data in English, Finnish, French, Swedish and Turkish. It predicts 25 hierarchical register labels.
The trained task
The label set includes machine-translated or generated text. The card uses 512-token inputs and gives language-specific probability thresholds. Results in additional languages describe transfer evaluation, not supervised training in every language listed by repository metadata. The actual repository name is web-register-classification-multilingual.Multilingual Web Register Classifier: model card and trained artifactsMultilingual Web Register Classifier: output labels and configuration
Practical assessment
Practical guidance. Choose a threshold for the language and collection being processed. Keep parent and child labels together when interpreting an output. A document's register describes how it is written; it does not establish accuracy or source quality.
Specifications & access
| Detail | Value and conditions |
|---|---|
| Access | Downloadable |
| Model identifiers | web-register-classification-multilingual |
| Base model | XLM-RoBERTa Large.Multilingual Web Register Classifier: model card and trained artifacts |
| Labels | 25 hierarchical register labels.Multilingual Web Register Classifier: output labels and configuration |
| License | Apache 2.0.Multilingual Web Register Classifier: model card and trained artifacts |
Related models
- BGE Multilingual Web Register Classifier
TurkuNLP · Web register classification
This register classifier adapts BGE-M3 RetroMAE and publishes a specific trained cross-validation fold.
Launch date not confirmedDownloadable - English Web Register Classifier
TurkuNLP · Web register classification
The English Web Register Classifier uses a 24-label scheme trained on English CORE data.
Launch date not confirmedDownloadable