Text Style Transfer
80 papers with code • 2 benchmarks • 6 datasets
Text Style Transfer is the task of controlling certain attributes of generated text. The state-of-the-art methods can be categorized into two main types which are used on parallel and non-parallel data. Methods on parallel data are typically supervised methods that use a neural sequence-to-sequence model with the encoder-decoder architecture. Methods on non-parallel data are usually unsupervised approaches using Disentanglement, Prototype Editing and Pseudo-Parallel Corpus Construction.
The popular benchmark for this task is the Yelp Review Dataset. Models are typically evaluated with the metrics of Sentiment Accuracy, BLEU, and PPL.
We demonstrate the effectiveness of this cross-alignment method on three tasks: sentiment modification, decipherment of word substitution ciphers, and recovery of word order.
Across all style transfer tasks, our approach yields substantial gains over state-of-the-art non-generative baselines, including the state-of-the-art unsupervised machine translation techniques that our approach generalizes.
Disentangling the content and style in the latent space is prevalent in unpaired text style transfer.
Text attribute transfer aims to automatically rewrite sentences such that they possess certain linguistic attributes, while simultaneously preserving their semantic content.
We prove that this simple modification guides the latent space geometry of the resulting model by encouraging the encoder to map similar texts to similar latent representations.
Beyond Fully-Connected Layers with Quaternions: Parameterization of Hypercomplex Multiplications with $1/n$ Parameters
Recent works have demonstrated reasonable success of representation learning in hypercomplex space.
The T5 model and its unified text-to-text paradigm contributed in advancing the state-of-the-art for many natural language processing tasks.