language

Complete dataset of public comments posted to Reddit (http://www.reddit.com) comments from October 2007 to May 2015.

Repository of televised news, including (for many) captions and rough statistics for content

Unstructured dataset of open-source media articles

Dataset of 8 million annotated YouTube videos, including a variety of audio and visual features.

List of datasets useful for deep learning and categorization, including datasets of faces, speech, text, and other images

List of datasets used to study opinion mining, sentiment analysis, and opinion spam detection

Dataset of over 5.8 million Amazon product reviews (including information on product, rating, review text, and more)

Transcripts from British speeches (1895 - 2015), categorized by date, speaker, party, and title

Data from audio recordings of human interaction across various regions of the United States and including a variety of speakers and contexts

Dataset of information derived from and related to one million contemporary songs, with more than 50 variables (including information on track metadata, social networks, and more)