README.md

---
language: "ca"
tags:
- masked-lm
- BERTa
- catalan
---

# BERTa: RoBERTa-based Catalan language model

## Model description

BERTa is a transformer-based masked language model for the Catalan language. 
It is based on the [RoBERTA](https://github.com/pytorch/fairseq/tree/master/examples/roberta) base model 
and has been trained on a medium-size corpus collected from publicly available corpora and crawlers.

## Training corpora and preprocessing

The training corpus consists of several corpora gathered from web crawling and public corpora.

The publicly available corpora are:

 1. the Catalan part of the [DOGC](http://opus.nlpl.eu/DOGC-v2.php) corpus, a set of documents from the Official Gazette of the Catalan Government
    
 2. the [Catalan Open Subtitles](http://opus.nlpl.eu/download.php?f=OpenSubtitles/v2018/mono/OpenSubtitles.raw.ca.gz), a collection of translated movie subtitles
    
 3. the non-shuffled version of the Catalan part of the [OSCAR](https://traces1.inria.fr/oscar/) corpus \cite{suarez2019asynchronous}, 
    a collection of monolingual corpora, filtered from [Common Crawl](https://commoncrawl.org/about/)
    
 4. The [CaWac](http://nlp.ffzg.hr/resources/corpora/cawac/) corpus, a web corpus of Catalan built from the .cat top-level-domain in late 2013
    the non-deduplicated version

 5. the [Catalan Wikipedia articles](https://ftp.acc.umu.se/mirror/wikimedia.org/dumps/cawiki/20200801/) downloaded on 18-08-2020.

The crawled corpora are:

 6. The Catalan General Crawling, obtained by crawling the 500 most popular .cat and .ad domains
 7. the Catalan Government Crawling, obtained by crawling the .gencat domain and subdomains, belonging to the Catalan Government
    
 8. the ACN corpus with 220k news items from March 2015 until October 2020, crawled from the [Catalan News Agency](https://www.acn.cat/)

To obtain a high-quality training corpus, each corpus have preprocessed with a pipeline of operations, including among the others,
sentence splitting, language detection, filtering of bad-formed sentences and deduplication of repetitive contents.
During the process, we keep document boundaries are kept. 
Finally, the corpora are concatenated and further global deduplication among the corpora is applied.
The final training corpus consists of about 1,8B tokens.


## Tokenization and pretraining 

The training corpus has been tokenized using a byte version of [Byte-Pair Encoding (BPE)](https://github.com/openai/gpt-2)
used in the original [RoBERTA](https://github.com/pytorch/fairseq/tree/master/examples/roberta) model with a vocabulary size of 52,000 tokens. 
The BERTa pretraining consists of a masked language model training that follows the approach employed for the RoBERTa base model
with the same hyperparameters as in the original work.
The training lasted a total of 48 hours with 16 NVIDIA V100 GPUs of 16GB DDRAM.

## Evaluation

## CLUB benchmark

The BERTa model has been fine-tuned on the downstream tasks of the Catalan Language Understanding Evaluation benchmark (CLUB),
that has been created along with the model.

It contains the following tasks and their related datasets:

 1. Part-of-Speech Tagging (POS)
    
    **Catalan-Ancora**: from the [Universal Dependencies treebank](https://github.com/UniversalDependencies/UD_Catalan-AnCora) of the well-known Ancora corpus

 2. Named Entity Recognition (NER)
    
    **[AnCora Catalan 2.0.0](https://zenodo.org/record/4762031#.YKaFjqGxWUk)**: extracted named entities from the original [Ancora](https://doi.org/10.5281/zenodo.4762030) version,
    filtering out some unconventional ones, like book titles, and transcribed them into a standard CONLL-IOB format

 3. Text Classification (TC)
     
    **[TeCla](https://doi.org/10.5281/zenodo.4627197)**: consisting of 137k news pieces from the Catalan News Agency ([ACN](https://www.acn.cat/)) corpus

 4. Semantic Texual Similarity (STS)
    
    **[Catalan semantic texual similarity](https://doi.org/10.5281/zenodo.4529183)**: consting of more than 3000 sentence pairs, annotated with the semantic similarity between them, 
    scraped from the [Catalan Textual Corpus](https://doi.org/10.5281/zenodo.4519349)

 5. Question Answering (QA):
    
    **[ViquiQuAD](https://doi.org/10.5281/zenodo.4562344)**: consisting of more than 15,000 questions outsourced from Catalan Wikipedia randomly chosen from a set of 596 articles which were
                   originally written in Catalan.
    
    **[XQuAD](https://doi.org/10.5281/zenodo.4526223)**: the Catalan translation of XQuAD, a multilingual collection of manual translations of 1,190 
               question-answer pairs from English Wikipedia used only as test set

## Results

Below the evaluation results on the CLUB tasks compared with the multilingual mBERT, XLM-RoBERTa models and 
the Catalan WikiBERT-ca model


| Task        | NER (F1)      | POS (F1)   | STS (Pearson)   | TC (accuracy) | QA (ViquiQuAD) (F1/EM)  | QA (XQuAD) (F1/EM) | 
| ------------|:-------------:| -----:|:------|:-------|:------|:----|
| BERTa       | **88.13** | **98.97** | **79.73** | **74.16** | **86.97/72.29** | **68.89/48.87** |
| mBERT       | 86.38 | 98.82 | 76.34 | 70.56 | 86.97/72.22 | 67.15/46.51 |
| XLM-RoBERTa | 87.66 | 98.89 | 75.40 | 71.68 | 85.50/70.47 | 67.10/46.42 |
| WikiBERT-ca | 77.66 | 97.60 | 77.18 | 73.22 | 85.45/70.75 | 65.21/36.60 |


## Intended uses & limitations
The model is ready-to-use only for masked language modelling to perform the Fill Mask task (try the inference API or read the next section)
However, the is intended to be fine-tuned on non-generative downstream tasks such as Question Answering, Text Classification or Named Entity Recognition.

---

## Using BERTa 
## Load model and tokenizer

``` python
from transformers import AutoTokenizer, AutoModelForMaskedLM
    
tokenizer = AutoTokenizer.from_pretrained("bsc/roberta-base-ca-cased")

model = AutoModelForMaskedLM.from_pretrained("bsc/roberta-base-ca-cased")
```

## Fill Mask task

Below, an example of how to use the masked language modeling task with a pipeline.

```python
>>> from transformers import pipeline
>>> unmasker = pipeline('fill-mask', model='bsc/roberta-base')
>>> unmasker("Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
             "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
             "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
             "i pel nord-oest per la serralada de Collserola "
             "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
             "la línia de costa encaixant la ciutat en un perímetre molt definit.")

[
  {
    "sequence": " Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
                "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
                "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
                "i pel nord-oest per la serralada de Collserola "
                "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
                "la línia de costa encaixant la ciutat en un perímetre molt definit.",
    "score": 0.4177263379096985,
    "token": 734,
    "token_str": " Barcelona"
  },
  {
    "sequence": " Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
                "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
                "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
                "i pel nord-oest per la serralada de Collserola "
                "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
                "la línia de costa encaixant la ciutat en un perímetre molt definit.",
    "score": 0.10696165263652802,
    "token": 3849,
    "token_str": " Badalona"
  },
  {
    "sequence": " Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
                "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
                "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
                "i pel nord-oest per la serralada de Collserola "
                "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
                "la línia de costa encaixant la ciutat en un perímetre molt definit.",
    "score": 0.08135009557008743,
    "token": 19349,
    "token_str": " Collserola"
  },
  {
   "sequence": " Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
                "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
                "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
                "i pel nord-oest per la serralada de Collserola "
                "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
                "la línia de costa encaixant la ciutat en un perímetre molt definit.",
    "score": 0.07330769300460815,
    "token": 4974,
    "token_str": " Terrassa"
  },
  {
    "sequence": " Situada a la costa de la mar Mediterrània, <mask> s'assenta en una plana formada "
                "entre els deltes de les desembocadures dels rius Llobregat, al sud-oest, "
                "i Besòs, al nord-est, i limitada pel sud-est per la línia de costa,"
                "i pel nord-oest per la serralada de Collserola "
                "(amb el cim del Tibidabo, 516,2 m, com a punt més alt) que segueix paral·lela "
                "la línia de costa encaixant la ciutat en un perímetre molt definit.",
    "score": 0.03317456692457199,
    "token": 14333,
    "token_str": " Gavà"
  }
]
```



### BibTeX  citation

```bibtex
Armengol-Estapé J., Carrino CP., Rodriguez-Penagos C., 
de Gibert Bonet O., Armentano-Oller C., Gonzalez-Agirre A., Melero M. 
and Villegas M., "Are Multilingual Models the Best Choice for Moderately 
Under-resourced Languages? A Comprehensive Assessment for Catalan". 
Findings of ACL 2021 (ACL-IJCNLP 2021)
```