Text watermarking algorithm based on semantic role labeling

This paper proposes a new approach for text watermarking that uses the semantic roles to embed watermark information. The technology of natural language processing is applied to find and label the three types of semantic roles A0, A1 and ADV in a text. A watermark message is converted into the hexadecimal Unicode and then compressed with the Huffman encoding to form a digit string that consists of the digits 0, 1 and 2. Let the three types of the semantic roles correspond to the three kinds of the digits one to one. The watermark digits are embedded by mapping each digit into the location of a semantic role of the corresponding type. The algorithm does not make any change to the format and content of a text. It has good features of concealment and robustness and can resist various text format transformations and watermark attacks.