Active/passive sentence pairs and their embeddings



This repository contains active/passive voice sentences and their sentence embeddings. The repository also contains the code used to generate these files. The file "active_passive_full_cleaned.tsv" has two columns (for active/passive voice) and 472635 rows, because it contains that many sentence pairs. The corresponding file of sentence embedding vectors is "active_passive_embedding_full.npy". It contains a numpy array of shape (472635, 2, 384). 472635 sentences, two columns for active/passive voice, and the embedding dimension of each sentence is 384. The other files in the folder are used in the generation of the aforementioned two files, but are presumably less interesting. Sentences were extracted from Project Gutenberg without checking for offensive language. There are Project Gutenberg books that contain offensive language. The following is a quote from their Collection Development Policy: "It also means that Project Gutenberg adds eBooks to its collection that contain language or ideas that are outdated, incorrect, offensive, or otherwise inconsistent with today’s societal views, standards or morals." The dataset provided here contains offensive language.
Date made available7 Mar 2023
PublisherKing's College London

Cite this