Natural language watermarking: Design, analysis, and a proof-of-concept implementation

Abstract

We describe a scheme for watermarking natural language text by embedding small portions of the watermark bit string in the syntactic structure of a number of selected sentences in the text, with both the selection and embedding keyed (via quadratic residue) to a large prime number. Meaning-preserving transformations of sentences of the text (e.g., translation to another natural language) cannot damage the watermark. Meaning-modifying transformations have a probability, of damaging the watermark, proportional to the watermark length over the number of sentences. Having the key is all that is required for reading the watermark. The approach is best suited for longish meaning-rather than style-oriented "expository" texts (e.g., reports, directives, manuals, etc.), of which governments and industry produce in abundance and which need protection more frequently than fiction or poetry, which are not so tolerant of the small meaning-preserving syntactic changes that the scheme implements.