Hugging Face's logo

lirondos
/
anglicisms-spanish-flair-cs

Token Classification
Flair
PyTorch
Spanish
anglicisms
loanwords
borrowing
codeswitching
sequence-tagger-model
arxiv:2203.16169
Model card Files Community
2
anglicisms-spanish-flair-cs / README.md
lirondos's picture
lirondos
Update README.md
9946e30
preview code
|
raw
history blame
3.01 kB
metadata
language:
  - es
license: cc-by-4.0
tags:
  - anglicisms
  - loanwords
  - borrowing
  - codeswitching
  - flair
  - token-classification
  - sequence-tagger-model
datasets:
  - coalas
widget:
  - text: >-
      Las fake news sobre la celebrity se reprodujeron por los 'mass media' en
      prime time.
  - text: Me gusta el cine noir y el anime.
  - text: >-
      Benching, estar en el banquillo de tu 'crush' mientras otro juega de
      titular.
  - text: Recetas de noviembre para el batch cooking.
  - text: Utilizaron técnicas de machine learning, big data o blockchain.

anglicisms-spanish-flair-cs

This is a pretrained model for detecting unassimilated English lexical borrowings (a.k.a. anglicisms) on Spanish newswire. This model labels words of foreign origin (fundamentally from English) used in Spanish language, words such as fake news, machine learning, smartwatch, influencer or streaming.

The model is a BiLSTM-CRF model fed with Transformer-based embeddings pretrained on codeswitched data along subword embeddings (BPE and character embeddings). The model was trained on the COALAS corpus for the task of detecting lexical borrowings.

The model considers two labels:

  • ENG: For English lexical borrowings (smartphone, online, podcast)
  • OTHER: For lexical borrowings from any other language (boutique, anime, umami)

The model uses BIO encoding to account for multitoken borrowings.

Metrics (on the test set)

LABEL Precision Recall F1
ALL 90.14 81.79 85.76
ENG 90.16 84.34 87.16
OTHER 85.71 13.04 22.64

There is another mBERT -based model for this same task trained using the Transformers library. That model however produced worse results than this Flair-based model (F1 = 83.55).

Dataset

This model was trained on COALAS, a corpus of Spanish newswire annotated with unassimilated lexical borrowings. The corpus contains 370,000 tokens and includes various written media written in European Spanish. The test set was designed to be as difficult as possible: it covers sources and dates not seen in the training set, includes a high number of OOV words (92% of the borrowings in the test set are OOV) and is very borrowing-dense (20 borrowings per 1,000 tokens).

Set Tokens ENG OTHER Unique
Training 231,126 1,493 28 380
Development 82,578 306 49 316
Test 58,997 1,239 46 987
Total 372,701 3,038 123 1,683