Code-Mixed Probes Show How Pre-Trained Models Generalise On Code-Switched Text

This repository consists of the code and data used for the COLING-LREC 2024 paper titled 'Code-Mixed Probes Show How Pre-Trained Models Generalise On Code-Switched Text' by Frances Laureano De Leon, Harish Tayyar Madabushi, and Mark Lee.

Abstract

Code-switching is a prevalent linguistic phenomenon in which multilingual individuals seamlessly alternate between languages. Despite its widespread use online and recent research trends in this area, research in code-switching presents unique challenges, primarily stemming from the scarcity of labelled data and available resources. In this study we investigate how pre-trained Language Models handle code-switched text in three dimensions: a) the ability of PLMs to detect code-switched text, b) variations in the structural information that PLMs utilise to capture code-switched text, and c) the consistency of semantic information representation in code-switched text. To conduct a systematic and controlled evaluation of the language models in question, we create a novel dataset of well-formed naturalistic code-switched text along with parallel translations into the source languages. Our findings reveal that pre-trained language models are effective in generalising to code-switched text, shedding light on abilities of these models to generalise representations to CS corpora.

PLMs used

1. bert-base-multilingual-uncased
2. xlm-roberta-base
3. xlm-roberta-large

Code

The repository is divided between Semantics, and Layer-wise, and Syntactic folders. Each folder corresponds to the type of experiments performed as presented in the paper. In the syntactic folder we do not include the code for training a structural probe, as we used the code base by Ethan Chi found at https://github.com/ethanachi/multilingual-probing-visualization.

The repository is divided between Semantics, Layer-wise, and Syntactic folders. Each folder corresponds to the type of experiments performed as presented in the paper.

Data

The data folder contains the data used for each of the experiments presented in the paper. The data folder contains subfolders: 'semantics' and 'syntactic'. The data in each subfolder corresponds to the title of each of the experiments mentioned in the paper. There is not a layer-wise folder included because the data used for this set of experiments is publicly available.

warning: some of the data may contain profanity as it was collected from social media sources.

Contact

If you need help or clarification on the data or code, please do not hesitate to contact me: fxl846{at}cs.bham.ac.uk

Citation

If you use the data or code in this repository, please cite:

@inproceedings{laureano-de-leon-etal-2024-code-mixed,
    title = "Code-Mixed Probes Show How Pre-Trained Models Generalise on Code-Switched Text",
    author = "Laureano De Leon, Frances Adriana  and
      Tayyar Madabushi, Harish  and
      Lee, Mark",
    editor = "Calzolari, Nicoletta  and
      Kan, Min-Yen  and
      Hoste, Veronique  and
      Lenci, Alessandro  and
      Sakti, Sakriani  and
      Xue, Nianwen",
    booktitle = "Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)",
    month = may,
    year = "2024",
    address = "Torino, Italia",
    publisher = "ELRA and ICCL",
    url = "https://aclanthology.org/2024.lrec-main.307",
    pages = "3457--3468",
    abstract = "Code-switching is a prevalent linguistic phenomenon in which multilingual individuals seamlessly alternate between languages. Despite its widespread use online and recent research trends in this area, research in code-switching presents unique challenges, primarily stemming from the scarcity of labelled data and available resources. In this study we investigate how pre-trained Language Models handle code-switched text in three dimensions: a) the ability of PLMs to detect code-switched text, b) variations in the structural information that PLMs utilise to capture code-switched text, and c) the consistency of semantic information representation in code-switched text. To conduct a systematic and controlled evaluation of the language models in question, we create a novel dataset of well-formed naturalistic code-switched text along with parallel translations into the source languages. Our findings reveal that pre-trained language models are effective in generalising to code-switched text, shedding light on abilities of these models to generalise representations to CS corpora. We release all our code and data, including the novel corpus, at https://github.com/francesita/code-mixed-probes.",
}