Autoregressive Text Generation Beyond Feedback Loops
This addresses train-test discrepancies in autoregressive models for text generation, but it is incremental as it builds on existing methods.
The paper tackles the problem of biases from autoregressive feedback in sequential models by combining a latent state space model with a CRF observation model, showing performance improvements in unconditional sentence generation compared to RNN and GAN baselines.
Autoregressive state transitions, where predictions are conditioned on past predictions, are the predominant choice for both deterministic and stochastic sequential models. However, autoregressive feedback exposes the evolution of the hidden state trajectory to potential biases from well-known train-test discrepancies. In this paper, we combine a latent state space model with a CRF observation model. We argue that such autoregressive observation models form an interesting middle ground that expresses local correlations on the word level but keeps the state evolution non-autoregressive. On unconditional sentence generation we show performance improvements compared to RNN and GAN baselines while avoiding some prototypical failure modes of autoregressive models.