Automatically Ranked Russian Paraphrase Corpus for Text Generation

Abstract

The article is focused on automatic development and ranking of a large corpus for Russian paraphrase generation which proves to be the first corpus of such type in Russian computational linguistics. Existing manually annotated paraphrase datasets for Russian are limited to small-sized ParaPhraser corpus and ParaPlag which are suitable for a set of NLP tasks, such as paraphrase and plagiarism detection, sentence similarity and relatedness estimation, etc. Due to size restrictions, these datasets can hardly be applied in end-to-end text generation solutions. Meanwhile, paraphrase generation requires a large amount of training data. In our study we propose a solution to the problem: we collect, rank and evaluate a new publicly available headline paraphrase corpus (ParaPhraser Plus), and then perform text generation experiments with manual evaluation on automatically ranked corpora using the Universal Transformer architecture.

Text simplification as a controlled text style transfer task

“…We evaluate the golden references (first answer) from the fixed RuSimpleSentEval-2021 test sets (public/private); • Paraphraser. We use a paraphrase model 6 trained on 7000 examples from different sources of various domains: 1) text level: literature domain, prose; back translation (with ru-en translation model 7 ) of the texts from different domains filtered with Bertscore Rouge-L); 2) sentence level: Russian version of Tapaco corpus (Scherrer, 2020) and filtered ParaphraserPlus (Gudkov et al, 2020) corpus. • Fine-tuned paraphraser.…”

Fact-checking benchmark for the Russian Large Language Models

“…The paraphraser 3 was chosen as it's a free model that is provided as an API. The model was trained on 7000 examples from different sources of various domains: 1) text level -texts from different domains filtered with Bertscore (Zhang et al, 2019) and Rouge-L) 2) sentence level -the Russian version of Tapaco corpus (Scherrer, 2020) and filtered Para-phraserPlus (Gudkov et al, 2020) corpus. Russian news dataset for summarization 4 was used as the source data for models generation.…”