Commit
•
74a61aa
1
Parent(s):
fa916ec
Create README.md
Browse files
README.md
ADDED
|
@@ -0,0 +1,104 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
language:
|
| 3 |
+
- multilingual
|
| 4 |
+
- pt
|
| 5 |
+
- en
|
| 6 |
+
tags:
|
| 7 |
+
- bert-base-multilingual-cased
|
| 8 |
+
- semantic role labeling
|
| 9 |
+
- finetuned
|
| 10 |
+
license: Apache 2.0
|
| 11 |
+
datasets:
|
| 12 |
+
- PropBank.Br
|
| 13 |
+
- CoNLL-2012
|
| 14 |
+
metrics:
|
| 15 |
+
- F1 Measure
|
| 16 |
+
---
|
| 17 |
+
|
| 18 |
+
# mBERT fine-tuned on English semantic role labeling
|
| 19 |
+
|
| 20 |
+
## Model description
|
| 21 |
+
|
| 22 |
+
This model is the [`bert-base-multilingual-cased`](https://huggingface.co/bert-base-multilingual-cased) fine-tuned on the English CoNLL formatted OntoNotes v5.0 semantic role labeling data. This is part of a project from which resulted the following models:
|
| 23 |
+
|
| 24 |
+
* [liaad/srl-pt_bertimbau-base](https://huggingface.co/liaad/srl-pt_bertimbau-base)
|
| 25 |
+
* [liaad/srl-pt_bertimbau-large](https://huggingface.co/liaad/srl-pt_bertimbau-large)
|
| 26 |
+
* [liaad/srl-pt_xlmr-base](https://huggingface.co/liaad/srl-pt_xlmr-base)
|
| 27 |
+
* [liaad/srl-pt_xlmr-large](https://huggingface.co/liaad/srl-pt_xlmr-large)
|
| 28 |
+
* [liaad/srl-pt_mbert-base](https://huggingface.co/liaad/srl-pt_mbert-base)
|
| 29 |
+
* [liaad/srl-en_xlmr-base](https://huggingface.co/liaad/srl-en_xlmr-base)
|
| 30 |
+
* [liaad/srl-en_xlmr-large](https://huggingface.co/liaad/srl-en_xlmr-large)
|
| 31 |
+
* [liaad/srl-en_mbert-base](https://huggingface.co/liaad/srl-en_mbert-base)
|
| 32 |
+
* [liaad/srl-enpt_xlmr-base](https://huggingface.co/liaad/srl-enpt_xlmr-base)
|
| 33 |
+
* [liaad/srl-enpt_xlmr-large](https://huggingface.co/liaad/srl-enpt_xlmr-large)
|
| 34 |
+
* [liaad/srl-enpt_mbert-base](https://huggingface.co/liaad/srl-enpt_mbert-base)
|
| 35 |
+
* [liaad/ud_srl-pt_bertimbau-large](https://huggingface.co/liaad/ud_srl-pt_bertimbau-large)
|
| 36 |
+
* [liaad/ud_srl-pt_xlmr-large](https://huggingface.co/liaad/ud_srl-pt_xlmr-large)
|
| 37 |
+
* [liaad/ud_srl-enpt_xlmr-large](https://huggingface.co/liaad/ud_srl-enpt_xlmr-large)
|
| 38 |
+
|
| 39 |
+
For more information, please see the accompanying article (See BibTeX entry and citation info below) and the [project's github](https://github.com/asofiaoliveira/srl_bert_pt).
|
| 40 |
+
|
| 41 |
+
|
| 42 |
+
## Intended uses & limitations
|
| 43 |
+
|
| 44 |
+
#### How to use
|
| 45 |
+
|
| 46 |
+
To use the transformers portion of this model:
|
| 47 |
+
```python
|
| 48 |
+
from transformers import AutoTokenizer, AutoModel
|
| 49 |
+
|
| 50 |
+
tokenizer = AutoTokenizer.from_pretrained("liaad/srl-en_mbert-base")
|
| 51 |
+
model = AutoModel.from_pretrained("liaad/srl-en_mbert-base")
|
| 52 |
+
```
|
| 53 |
+
|
| 54 |
+
To use the full SRL model (transformers portion + a decoding layer), refer to the [project's github](https://github.com/asofiaoliveira/srl_bert_pt).
|
| 55 |
+
|
| 56 |
+
|
| 57 |
+
#### Limitations and bias
|
| 58 |
+
|
| 59 |
+
- The models were trained only for 5 epochs.
|
| 60 |
+
- The English data was preprocessed to match the Portuguese data, so there are some differences in role attributions and some roles were removed from the data.
|
| 61 |
+
|
| 62 |
+
|
| 63 |
+
## Training data
|
| 64 |
+
|
| 65 |
+
Pretrained weights were left identical to the original model [`bert-base-multilingual-cased`](https://huggingface.co/bert-base-multilingual-cased).
|
| 66 |
+
|
| 67 |
+
|
| 68 |
+
## Training procedure
|
| 69 |
+
|
| 70 |
+
The models were trained on the CoNLL-2012 dataset, preprocessed to match the Portuguese PropBank.Br data. They were tested on the PropBank.Br data set as well as on a smaller opinion dataset "Buscapé". For more information, please see the accompanying article (See BibTeX entry and citation info below) and the [project's github](https://github.com/asofiaoliveira/srl_bert_pt).
|
| 71 |
+
|
| 72 |
+
## Eval results
|
| 73 |
+
|
| 74 |
+
|
| 75 |
+
| Model Name | F<sub>1</sub> CV PropBank.Br (in domain) | F<sub>1</sub> Buscapé (out of domain) |
|
| 76 |
+
| --------------- | ------ | ----- |
|
| 77 |
+
| `srl-pt_bertimbau-base` | 76.30 | 73.33 |
|
| 78 |
+
| `srl-pt_bertimbau-large` | 77.42 | 74.85 |
|
| 79 |
+
| `srl-pt_xlmr-base` | 75.22 | 72.82 |
|
| 80 |
+
| `srl-pt_xlmr-large` | 77.59 | 73.84 |
|
| 81 |
+
| `srl-pt_mbert-base` | 72.76 | 66.89 |
|
| 82 |
+
| `srl-en_xlmr-base` | 66.59 | 65.24 |
|
| 83 |
+
| `srl-en_xlmr-large` | 67.60 | 64.94 |
|
| 84 |
+
| `srl-en_mbert-base` | 63.07 | 58.56 |
|
| 85 |
+
| `srl-enpt_xlmr-base` | 76.50 | 73.74 |
|
| 86 |
+
| `srl-enpt_xlmr-large` | **78.22** | 74.55 |
|
| 87 |
+
| `srl-enpt_mbert-base` | 74.88 | 69.19 |
|
| 88 |
+
| `ud_srl-pt_bertimbau-large` | 77.53 | 74.49 |
|
| 89 |
+
| `ud_srl-pt_xlmr-large` | 77.69 | 74.91 |
|
| 90 |
+
| `ud_srl-enpt_xlmr-large` | 77.97 | **75.05** |
|
| 91 |
+
|
| 92 |
+
|
| 93 |
+
### BibTeX entry and citation info
|
| 94 |
+
|
| 95 |
+
```bibtex
|
| 96 |
+
@misc{oliveira2021transformers,
|
| 97 |
+
title={Transformers and Transfer Learning for Improving Portuguese Semantic Role Labeling},
|
| 98 |
+
author={Sofia Oliveira and Daniel Loureiro and Alípio Jorge},
|
| 99 |
+
year={2021},
|
| 100 |
+
eprint={2101.01213},
|
| 101 |
+
archivePrefix={arXiv},
|
| 102 |
+
primaryClass={cs.CL}
|
| 103 |
+
}
|
| 104 |
+
```
|