Search

Every word of every paper here, in 4 languages. The index is a file this page downloads the first time you type in the box, and the search runs in your browser. Nothing you type leaves it.

What it looks in

Every block of every page: the paragraphs, the headings, the formulas and the program listings. The prose and the mathematics are indexed apart, so a word you typed finds the paragraph that uses it before the formula that happens to contain the same letters.

A word in more than a third of the blocks of a language is left out of that language's index. It could not narrow a search down and it was most of the file. That is why searching for the returns nothing, in any of the four.

Chinese and Japanese are cut into overlapping pairs of characters rather than into words. It is crude, it finds short words inside longer ones, and it is what a corpus this size can afford without shipping a segmenter to every reader.

Results are per block and not per paper, so a long paper that mentions your word in two places is two lines and each one goes to the section it is in. No paper gets more than three lines, or one paper that keeps saying the word would be the whole page. The last word you type also matches on its beginning, which is why the list keeps moving while you are still typing it.

The two lines under a result are the section it is in and the opening of the block. The opening, and not the sentence that matched: the index carries the first part of every block so that a page of results is not a page of downloads, and in a long paragraph your word may be further in than the part you can see.