GeoLingIt: Geolocation of Linguistic Variation in Italy

Disclaimer: This dataset is not the official GeoLingIt repository from EVALITA. For the official repository and more information, please visit the EVALITA GeoLingIt page.

Dataset Summary

GeoLingIt is a dataset for studying the geolocation of linguistic variation in Italy using social media posts that exhibit non-standard Italian language. The dataset is part of the EVALITA 2023 evaluation campaign and aims to advance natural language processing (NLP) techniques for non-standard Italian while providing sociolinguistic insights into language variation across Italy.

The dataset includes tweets that feature linguistic patterns from various Italian dialects, regional varieties, and local languages. Social media content, such as tweets, often includes informal language that reflects local dialects and regional varieties. This provides a unique opportunity to explore linguistic variation on a large scale and improve language technologies for Italian and minority languages.

Dataset Structure

Data Format

The dataset supports two main tasks:

• Coarse-grained Geolocation: Predict the region of provenance for each tweet.
• Fine-grained Geolocation: Predict the exact longitude and latitude coordinates for each tweet.

The dataset composed of the following columns:

• id: An anonymized identifier for the tweet.
• text: The content of the tweet with anonymized user mentions, email addresses, URLs, and location mentions.
• region: The region of provenance for the tweet (for coarse-grained geolocation).
• latitude: The latitude coordinate of the tweet (for fine-grained geolocation).
• longitude: The longitude coordinate of the tweet (for fine-grained geolocation).

Source Data

The tweets were collected from Twitter and filtered to include only those that exhibit non-standard Italian language. Each tweet includes geotagging information (latitude, longitude, and place name) within Italy.

Ethical Considerations

The dataset is anonymized to protect user privacy, with user mentions, email addresses, and URLs replaced by placeholders. Latitude and longitude coordinates represent cities as a whole, avoiding specific place identification. The data is intended for aggregate analysis to study diatopic linguistic variation.

Potential Issues

Since the dataset consists of social media posts, it may contain profanities, slurs, and hateful content. Users should be aware of this when working with the data.

Citation

If you use this dataset, please cite the original authors:

@inproceedings{ramponi-casula-2023-diatopit,
    title = "{D}iatop{I}t: A Corpus of Social Media Posts for the Study of Diatopic Language Variation in {I}taly",
    author = "Ramponi, Alan  and
      Casula, Camilla",
    booktitle = "Tenth Workshop on NLP for Similar Languages, Varieties and Dialects (VarDial 2023)",
    month = may,
    year = "2023",
    address = "Dubrovnik, Croatia",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2023.vardial-1.19",
    pages = "187--199",
}