Is there a threshold or a descriptive proportion that is considered not copyright infringement but merely creative and educational use when distributing some copyright protected material along with your own material?

Suppose a research team had a data set, and corresponding to each data point, had a large chunk of copyrighted text. What proportion of this text would they be able to release as preview or demonstration of the material without it being considered copyright infringement?

I would like to cite Google Books, and the popular case of Google Inc. vs. The Authors Guild, where, under Google's book digitization project, they show certain subsections of the book to the general public for no charge, but argue that it is not copyright infringement.