Educating Text Autoencoders: Latent Representation Guidance via Denoising
This work addresses a key bottleneck for researchers and practitioners in natural language processing seeking controllable text generation, though it is incremental as it builds on existing autoencoder frameworks.
The authors tackled the problem of incoherent latent spaces in generative autoencoders for text, which hinder controllable text generation, by proposing a denoising-augmented adversarial autoencoder (DAAE) that encourages similar texts to map to similar latent representations, resulting in improved generation quality and reconstruction capacity, and enabling zero-shot text style transfer via latent vector arithmetic.
Generative autoencoders offer a promising approach for controllable text generation by leveraging their latent sentence representations. However, current models struggle to maintain coherent latent spaces required to perform meaningful text manipulations via latent vector operations. Specifically, we demonstrate by example that neural encoders do not necessarily map similar sentences to nearby latent vectors. A theoretical explanation for this phenomenon establishes that high capacity autoencoders can learn an arbitrary mapping between sequences and associated latent representations. To remedy this issue, we augment adversarial autoencoders with a denoising objective where original sentences are reconstructed from perturbed versions (referred to as DAAE). 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. In empirical comparisons with various types of autoencoders, our model provides the best trade-off between generation quality and reconstruction capacity. Moreover, the improved geometry of the DAAE latent space enables zero-shot text style transfer via simple latent vector arithmetic.