We propose a new approach for the authorship attribution task that leverages the various linguistic representations learned at different layers of pre-trained transformer-based models.We evaluate our approach on two popular authorship attribution models and three evaluation datasets, in in-domain and out-of-domain scenarios.We found that utilizing various transformer layers improves the robustness of authorship attribution models when tested on outof-domain data, resulting in a much stronger performance.Our analysis gives further insights into how our model's different layers get specialized in representing certain linguistic aspects that we believe benefit the model when tested out of the domain.

The task of keyphrase generation, unlike extraction, aims to generate the phrases which succinctly capture the key information of the source text, that are even absent in the document (i.e., do not match any contiguous sub-sequence of source text). Despite the significant progress achieved by sequence-to-sequence (seq2seq) models in modelling such high entropy task, they are limited by their deterministic modelling capability which limits the generation of a diverse set of keyphrases. To address the above limitation, in this paper, we propose to incorporate Conditional Variational Autoencoder (CoVA) into seq2seq models for its ability to represent a set of keyphrases as a probabilistic distribution which improves the diversity of the generated keyphrases. We model the probabilistic distribution using a hierarchical latent structure where a global latent variable tries to model the diversity among the keyphrases and local latent variables control the generation of each keyphrase to make them coherent. Experimental results on four benchmark datasets of research papers demonstrate the effectiveness of our proposed approach in achieving a large improvement in diversity along with modest gains in quality with respect to previous models.

Discriminating lexical relations among distributionally similar words has always been a challenge for natural language processing (NLP) community. In this paper, we investigate whether the network embedding of distributional thesaurus can be effectively utilized to detect co-hyponymy relations. By extensive experiments over three benchmark datasets, we show that the vector representation obtained by applying node2vec on distributional thesaurus outperforms the state-of-the-art models for binary classification of co-hyponymy vs. hypernymy, as well as co-hyponymy vs. meronymy, by huge margins.