Multi-Granularity Representations of Dialog
This work addresses the challenge of improving dialog models for researchers and practitioners by offering an incremental method to enhance representation learning.
The paper tackles the problem of learning generalized latent representations for dialog by introducing a multi-granularity training procedure that modifies negative sampling to control representation granularity, resulting in strong performance gains on next utterance retrieval tasks using MultiWOZ and Ubuntu datasets.
Neural models of dialog rely on generalized latent representations of language. This paper introduces a novel training procedure which explicitly learns multiple representations of language at several levels of granularity. The multi-granularity training algorithm modifies the mechanism by which negative candidate responses are sampled in order to control the granularity of learned latent representations. Strong performance gains are observed on the next utterance retrieval task using both the MultiWOZ dataset and the Ubuntu dialog corpus. Analysis significantly demonstrates that multiple granularities of representation are being learned, and that multi-granularity training facilitates better transfer to downstream tasks.