Visiting Fellowship - Ozsoy

Text watermarking is an emerging technique in the intersection of natural language processing and the technologies of forensics and security. It works by embedding additional information in the format, structure or content of a text that is transparent with respect to the normal use of the text, but that can be detected when explicitly sought. Such information can be used, inter alia, to trace the origin, provenance, authenticity and use of the text itself. Rather than watermarking by format or layout that may be disrupted by changes in format, the present proposal seeks to manipulate grammatical aspects of the text, such as active/passive voice, that do not change meaning. The proposal seeks to develop such devices for Turkish, a morphologically rich language differing on many points from English, including the semantics of the passive, using grammar based techniques and statistical modeling.