Hugging Face's logo

Datasets:
strickvl
/
afghanwire

Tasks:
Text Classification
Zero-Shot Classification
Summarization
Modalities:
Tabular
Text
Formats:
parquet
Languages:
English
Size:
1K - 10K
DOI:
Libraries:
Datasets
pandas
License:
Dataset card Viewer Files Community
2
afghanwire / README.md
strickvl's picture
strickvl
markdown tags
4201f53 unverified
preview code
|
raw
history blame contribute delete
9.86 kB
metadata
license: cc-by-nc-sa-4.0
configs:
  - config_name: default
    data_files:
      - split: articles
        path: data/articles.parquet
task_categories:
  - text-classification
  - zero-shot-classification
  - summarization
  - feature-extraction
language:
  - en
pretty_name: afghanwire
size_categories:
  - 1K<n<10K

Afghanwire Dataset Description

Dataset Summary

The Afghanwire dataset is a comprehensive collection of translated Afghan media articles from the period of May 2006 to September 2009. It was created by the Afghanwire media agency, founded by Alex Strick van Linschoten and Felix Kuehn. The agency employed a group of Afghan translators who translated articles from Dari and Pashto media sources into English. The dataset includes translated newspaper and magazine articles, as well as summaries of radio and television content. As most of the original media from this period is no longer available online, and certainly not in English, this dataset represents the largest publicly available trove of translated Afghan media for the 2006-2009 period.

The primary purpose of making this dataset available is to serve as a historical artifact. However, it also presents opportunities for various Natural Language Processing (NLP) tasks, such as Named Entity Recognition (NER) for entities that may be underrepresented in standard or smaller models, and potentially sentiment analysis. It is important to note that the dataset is unlabeled and consists solely of translated articles.

Supported Tasks and Leaderboards

There are a variety of potential tasks that could be done on this dataset, including text classification, zero-shot classification, summarization, and feature extraction. There are no leaderboards for this dataset.

Languages

The dataset is only in English, but the original source articles were in Dari and Pashto.

Dataset Structure

Data Instances

Here's a JSON example from the main "articles" table:

{
  "id": "97",
  "paper": "142",
  "title": "Some parliament members left session yesterday",
  "date": "2006-08-13T00:00:00.000Z",
  "author": "-",
  "translator": "19",
  "topic": "0",
  "abstract": "Translated by: H. Stanikzai",
  "comment": "",
  "rating": "0",
  "language": "32",
  "type": "0",
  "article": "Bakhtar, in yesterday session of the parliament, the president of the parliament termed the activities and sessions of the previous week as to be effective, and as an objection on constitutional decree on prisons and confinements the session members left the parliament. The members who left the parliament are mostly from northern Afghanistan and they objected the policies of the government of Afghanistan, they have threatened not participate in the parliament session until the government has changed its policies. The parliament members claims that they are witness for the insecurity in the country, to return the rights and privileges of the military, an in impropriate policy of the government regarding the appointment of cadres the failed reforms in the ministries, bribery in government offices and administrative corruption, the unfair composition the diplomatic representative and the lack of cadre in the central and as well as local government offices. But some other parliament members criticized the action of the members who left the parliament and said that their objection was unlawful and is against the principals of the inner tasks of the parliament.",
  "ok": "0",
  "no_newsletter": "0",
  "eingegeben": 1155479462000,
  "newsletter": "0",
  "free": "0",
  "url": "",
  "top_topic": "0",
  "words": "",
  "translatorcomment": "",
  "datetranslation": "",
  "scan": ""
}

The dataset consists of several supporting tables that are referenced in the main "articles" table, such as papers, article_tags, bib_books, cities, current_events, ethnics, glossary, historical_events, issue, languages, organisations, people, provinces, region_tree, renderbackgrounder, top_topcs, topic, and types.

The dataset as a whole consists of 7990 articles that were translated during the period Afghanwire was open as an organisation.

Data Fields

  • id - basic id for the article
  • paper - id/number for a paper mentioned in papers.parquet
  • title - article title
  • date
  • author (if present)
  • translator (who translated the article)
  • topic (associated with topic table)
  • abstract (sometimes mentions the translator)
  • comment (sometimes also mentions the translator)
  • rating (not always used. was a measure for interest level)
  • language (associated with the separate table)
  • type (associated with the article types table)
  • article - the full translation
  • ok - whether the translation has been edited
  • no_newsletter - a metatag to represent whether the article should be sent out as part of our newsletter or not
  • eingegeben - a unix timestamp for when the article was uploaded to the database
  • newsletter - whether to include the article in our newsletter
  • free - whether to make the article available for free or not
  • url - if available
  • top_topic - what high-level topic the article was associated with
  • words - word count (not always present)
  • translatorcomment - not always present
  • datetranslation - not always present
  • scan - whether there's a scan for the article or not

Data Splits

There are no predefined splits. The dataset is provided as a single large collection.

Dataset Creation

Curation Rationale

The creator of this dataset, Alex Strick van Linschoten, had the database files stored on his hard drive for an extended period. By making this data publicly available, he aims to ensure that it can be utilized by others. The media articles were translated by Afghan translators and represent a snapshot of Afghanistan's media discourse during the 2006-2009 period. As the translations were privately funded and are likely unique, with no other copies existing elsewhere, this dataset is expected to be an extremely valuable resource for scholars and historians.

Source Data

The source data was collected by ordering newspapers and magazines from around Afghanistan to the Afghanwire office on a daily basis. The agency also monitored radio stations. The translators selected articles that they and the agency deemed representative and interesting for readers, and then translated them into English.

It is worth noting that the data was originally used to populate a website and newsletter at afghanwire.com. However, the website is no longer active, and the files only existed in an old MySQL database on the creator's laptop. While the website is partially available on the Internet Archive (snapshot from February 2009), most of the articles were behind a login page, which does not function with the archive snapshots. This dataset aims to make the translated articles accessible to the public.

Annotations

This dataset does not contain any annotations aside from some manual topic classification.

Personal and Sensitive Information

The dataset does not contain any personally identifiable information (PII). All content is sourced from public media outlets and has been translated.

Considerations for Using the Data

Social Impact of Dataset

This dataset is valuable for historians and researchers as it provides insights into the debates and discussions within Afghan society during the 2006-2009 period. It challenges the notion of uniformity in opinions on various issues, such as attitudes towards the Taliban, the Afghan government, and ISAF/NATO forces. By foregrounding Afghan civil society in the discussion of Afghan history, this dataset plays an important role in shifting the focus from foreign powers and military forces to the voices of the Afghan people, which are often overlooked.

The dataset contains information about events, tribes, and ethnic groups from across Afghanistan, including articles about Dai Kundi province, which might have otherwise been lost. Although the Afghanwire office was based in Kabul, efforts were made to obtain newspapers and magazines from the provinces to ensure a representative collection. However, it is acknowledged that there may be a slight bias towards the capital due to the office's location.

Discussion of Biases

The creators of this dataset made a concerted effort to avoid biases in both the selection of articles and the translation process. However, as with any dataset, the potential for biases cannot be entirely eliminated.

Other Known Limitations

Apart from the possibility of a slight overrepresentation of media from Kabul compared to other provinces, there are no other known limitations to this dataset.

Additional Information

Dataset Curators

The dataset was curated by the Afghanwire organization. The translators, Hamid Stanikzai, Atif Mohammadzai, Abdul Hassib Rahimi, and Hamid Safi, selected the articles to be translated and deserve full credit for their work.

Licensing Information

This dataset is licensed under the Creative Commons Attribution-ShareAlike 4.0 International (CC BY-SA 4.0) license. For more information, see https://creativecommons.org/licenses/by-sa/4.0/.

Citation Information

If you use this dataset in your research or project, please cite it as follows:

@misc{afghanwire_2024,
author = {Afghanwire},
title = {Afghanwire Media Database 2006-2009},
year = {2024},
month = {April},
day = {1},
url = {https://huggingface.co/datasets/strickvl/afghanwire}
}