|
--- |
|
license: apache-2.0 |
|
tags: |
|
- generated_from_trainer |
|
model-index: |
|
- name: bart-base-spelling-nl-1m-3 |
|
results: [] |
|
language: |
|
- nl |
|
metrics: |
|
- bleu |
|
- cer |
|
- wer |
|
- meteor |
|
pipeline_tag: text2text-generation |
|
--- |
|
|
|
# bart-base-spelling-nl |
|
|
|
This model is a Dutch fine-tuned version of |
|
[facebook/bart-base](https://huggingface.co./facebook/bart-base). |
|
|
|
It achieves the following results on an external evaluation set of human-corrected spelling |
|
errors of Dutch snippets of internet text ([errors](https://huggingface.co./antalvdb/bart-base-spelling-nl/blob/main/opentaal-annotaties.txt.errors) |
|
and [corrections](https://huggingface.co./antalvdb/bart-base-spelling-nl/blob/main/opentaal-annotaties.txt.corrections), |
|
run [spell.py](https://huggingface.co./antalvdb/bart-base-spelling-nl/blob/main/spell.py)) |
|
|
|
* CER - 0.024 |
|
* WER - 0.088 |
|
* BLEU - 0.840 |
|
* METEOR - 0.932 |
|
|
|
Note that it is very hard for any spelling corrector to clean more actual spelling errors |
|
than introducing new errors. In other words, most spelling correctors cannot be run |
|
automatically and must be used interactively. |
|
|
|
These are the upper-bound scores when correcting _nothing_. In other words, this is |
|
the actual distance between the errors and their corrections in the evaluation set: |
|
|
|
* CER - 0.010 |
|
* WER - 0.053 |
|
* BLEU - 0.900 |
|
* METEOR - 0.954 |
|
|
|
We are not there yet, clearly. |
|
|
|
## Model description |
|
|
|
This is a fine-tuned version of |
|
[facebook/bart-base](https://huggingface.co./facebook/bart-base) |
|
trained on spelling correction. It leans on the excellent work by |
|
Oliver Guhr ([github](https://github.com/oliverguhr/spelling), |
|
[huggingface](https://huggingface.co./oliverguhr/spelling-correction-english-base)). Training |
|
was performed on an AWS EC2 instance (g5.xlarge) on a single GPU, and |
|
took about two days. |
|
|
|
## Intended uses & limitations |
|
|
|
The intended use for this model is to be a component of the |
|
[Valkuil.net](https://valkuil.net) context-sensitive spelling |
|
checker. |
|
|
|
## Training and evaluation data |
|
|
|
The model was trained on a Dutch dataset composed of 12,351,203 lines |
|
of text, containing a total of 123,131,153 words, from three public Dutch sources, downloaded from the |
|
[Opus corpus](https://opus.nlpl.eu/): |
|
|
|
- nl-europarlv7.txt (2,387,000 lines) |
|
- nl-opensubtitles2016.9m.txt (9,000,000 lines) |
|
- nl-wikipedia.txt (964,203 lines) |
|
|
|
|
|
## Training procedure |
|
|
|
### Training hyperparameters |
|
|
|
The following hyperparameters were used during training: |
|
- learning_rate: 0.0003 |
|
- train_batch_size: 2 |
|
- eval_batch_size: 4 |
|
- seed: 42 |
|
- gradient_accumulation_steps: 16 |
|
- total_train_batch_size: 32 |
|
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08 |
|
- lr_scheduler_type: linear |
|
- num_epochs: 2.0 |
|
|
|
|
|
### Framework versions |
|
|
|
- Transformers 4.27.3 |
|
- Pytorch 2.0.0+cu117 |
|
- Datasets 2.10.1 |
|
- Tokenizers 0.13.2 |
