Extracting Shallow Paraphrasing Schemata from Modern Greek Text Using Statistical Significance Testing and Supervised Learning

摘要

Paraphrasing normally involves sophisticated linguistic resources for pre-processing. In the present work Modern Greek paraphrases are automatically generated using statistical significance testing in a novel manner for the extraction of applicable reordering schemata of syntactic constituents. Next, supervised filtering helps remove erroneously generated paraphrases, taking into account the context surrounding the reordering position. The proposed process is knowledge-poor, and thus portable to languages with similar syntax, robust and domain-independent. The intended use of the extracted paraphrases is hiding secret information underneath a cover text.