A Self-Supervised Representation Learning of Sentence Structure for Authorship Attribution

The syntactic structure of sentences in a document substantially informs about its authorial writing style. Sentence representation learning has been widely explored recent years and it shown that improves the generalization different downstream tasks across many domains. Even though utilizing probing methods several studies suggests these learned contextual representations implicitly encode some amount syntax, explicit information further performance deep neural models domain authorship attribution. These observations have motivated us to investigate sentences. In this article, we propose self-supervised framework for structural network contains two components; lexical sub-network which take sequence words their corresponding labels as input, respectively. Due n -to-1 mapping labels, each word will be embedded into vector mainly carries information. We evaluate using tasks, subsequently utilize them attribution task. Our experimental results indicate embeddings significantly improve classification when concatenated with existing pre-trained embeddings.