Releases: stickeritis/sticker2-models
Small Dutch UD syntax model
Small Dutch UD syntax model:
- Universal tags
- Lemmas
- Morphology
- Universal dependencies
Distilled from finetuned XLM-RoBERTa base, into a ALBERTtransformer with 12 hidden layers, 384 hidden units, 12 attention heads, 128-dimensional piece embeddings, and 4 layer groups.
Small German UD syntax model
Small German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Distilled from finetuned XLM-RoBERTa base, into a ALBERTtransformer with 12 hidden layers, 384 hidden units, 12 attention heads, 128-dimensional piece embeddings, and 4 layer groups.
Medium German UD syntax model
Medium German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Distilled from finetuned XLM-RoBERTa base, into a transformer with 6 hidden layers, 384 hidden units, and 12 attention heads. The vocabulary of the German BERT is used.
Medium Dutch UD syntax model
Medium Dutch UD syntax model:
- Universal tags
- Lemmas
- Morphology
- Universal dependencies
Distilled from finetuned XLM-RoBERTa base, into a transformer with 6 hidden layers, 384 hidden units, and 12 attention heads. The vocabulary of BERTje is used, with some additional word pieces for better coverage of Dutch orthography.
Note: this is the nl-ud-medium-20200430 model updated for sticker 0.4 configuration.
Large Dutch UD syntax model
Medium Dutch UD syntax model:
- Universal tags
- Lemmas
- Morphology
- Universal dependencies
Distilled from finetuned XLM-RoBERTa base, into a transformer with 12 hidden layers, 384 hidden units, and 12 attention heads. The vocabulary of BERTje is used, with some additional word pieces for better coverage of Dutch orthography.
Note: this is the nl-ud-large-20200420 model updated for sticker 0.4 configuration.
Huge Dutch UD syntax model
Huge Dutch UD syntax model:
- Universal tags
- Lemmas
- Morphology
- Universal dependencies
Finetuned XLM-RoBERTa base, 12 hidden layers, 768 hidden units, and 12 attention heads. For production use, one of the smaller models is recommended!
Large German UD syntax model
Large German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Distilled from finetuned XLM-RoBERTa base, into a transformer with 12 hidden layers, 384 hidden units, and 12 attention heads. The vocabulary of the German BERT is used.
Compared to the previous version: use sinusoidal positions, apply TüBa-D/Z-style lemma postprocessing.
Huge German UD syntax model
Huge German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Finetuned XLM-RoBERTa base.
Note: this is the de-ud-large-20200710 model updated for sticker 0.4 configuration.
Large German UD syntax model
Large German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Distilled from finetuned XLM-RoBERTa base, into a transformer with 12 hidden layers, 384 hidden units, and 12 attention heads. The vocabulary of the German BERT is used.
Compared to version 20200523, this model fixes:
- Some instances where nominals were attached to nominals with the
obl
relation. - Some instances where adverbials were attached to verbs with the
amod
relation.
Huge German UD syntax model
Huge German UD syntax model:
- Universal tags
- STTS tags
- Lemmas
- UD morphology
- Morphological tag
- Topological field
- Universal dependencies
- Named entity
Finetuned XLM-RoBERTa base.