What's in the Box? A Preliminary Analysis of Undesirable Content in the Common Crawl Corpus
arXiv:2105.02732
Abstract
Whereas much of the success of the current generation of neural language models has been driven by increasingly large training corpora, relatively little research has been dedicated to analyzing these massive sources of textual data. In this exploratory analysis, we delve deeper into the Common Crawl, a colossal web corpus that is extensively used for training language models. We find that it contains a significant amount of undesirable content, including hate speech and sexually explicit content, even after filtering procedures. We discuss the potential impacts of this content on language models and conclude with future research directions and a more mindful approach to corpus collection and analysis.
arXiv comment5 pages, 1 figure, 3 tables. Published as a main conference paper at ACL-IJCNLP 2021, submission #87. Code available at https://github.com/josephdviviano/whatsinthebox
References in corpus (26)
- Language Models are Few-Shot Learners
- Semantics derived automatically from language corpora contain human-like biases
- Scaling Laws for Neural Language Models
- Man is to Computer Programmer as Woman is to Homemaker? Debiasing Word Embeddings
- A large annotated corpus for learning natural language inference
- Directions in Abusive Language Training Data: Garbage In, Garbage Out
- Aligning Books and Movies: Towards Story-like Visual Explanations by Watching Movies and Reading Books
- Release Strategies and the Social Impacts of Language Models
- Extracting Training Data from Large Language Models
- Lipstick on a Pig: Debiasing Methods Cover up Systematic Gender Biases in Word Embeddings But do not Remove Them
- Know What You Don't Know: Unanswerable Questions for SQuAD
- Quality at a Glance: An Audit of Web-Crawled Multilingual Datasets
- Data and its (dis)contents: A survey of dataset development and use in machine learning research
- Tackling Online Abuse: A Survey of Automated Abuse Detection Methods
- Deep Learning Models for Multilingual Hate Speech Detection
- Gender Bias in Contextualized Word Embeddings
- Learning Gender-Neutral Word Embeddings
- Examining Gender and Race Bias in Two Hundred Sentiment Analysis Systems
- Hate begets Hate: A Temporal Study of Hate Speech
- Latent Constraints: Learning to Generate Conditionally from Unconditional Generative Models
- Persistent Anti-Muslim Bias in Large Language Models
- Putting Humans in the Natural Language Processing Loop: A Survey
- The Woman Worked as a Babysitter: On Biases in Language Generation
- POINTER: Constrained Progressive Text Generation via Insertion-based Generative Pre-training
- Exploratory Analysis of a Terabyte Scale Web Corpus
- On the Unintended Social Bias of Training Language Generation Models with Data from Local Media