Whose Language Counts as High Quality? Measuring Language Ideologies in Text Data Selection
This work highlights a critical bias in data selection for language models, affecting fairness and representation in AI, and is incremental in exposing specific issues with existing methods.
The study investigated the bias in quality filters used for selecting text data for language models like GPT-3, finding that they favor content from larger, wealthier, and more educated schools, and that these filters do not align with metrics like factuality or literary acclaim.
Language models increasingly rely on massive web dumps for diverse text data. However, these sources are rife with undesirable content. As such, resources like Wikipedia, books, and newswire often serve as anchors for automatically selecting web text most suitable for language modeling, a process typically referred to as quality filtering. Using a new dataset of U.S. high school newspaper articles -- written by students from across the country -- we investigate whose language is preferred by the quality filter used for GPT-3. We find that newspapers from larger schools, located in wealthier, educated, and urban ZIP codes are more likely to be classified as high quality. We then demonstrate that the filter's measurement of quality is unaligned with other sensible metrics, such as factuality or literary acclaim. We argue that privileging any corpus as high quality entails a language ideology, and more care is needed to construct training corpora for language models, with better transparency and justification for the inclusion or exclusion of various texts.