THE DOCUMENT OBSERVATORY

A sample is
not the whole story.

Find the main language of a short document. A quoted phrase can pull a language detector off course. Choose a representative window of two or more author sentences, compare the real trigram ranking, then file your conclusion.

Choose your evidence

Select the first and last sentence of one contiguous sample. Stay within the character budget and exclude quoted material. To certify, your sample must contain at least two author sentences, and your conclusion must agree with both the source document and the detector. Model scores rank candidates; they are not probabilities.

Your sample window

Start by comparing a quote-only sample with an author-language sample.