Variational Learning for Unsupervised Knowledge Grounded Dialogs

Abstract

Recent methods for knowledge grounded dialogs generate responses by incorporating information from an external textual document (Lewis et al. 2020;Guu et al. 2020). These methods do not require the exact document to be known during training and rely on the use of a retrieval system to fetch relevant documents from a large index. The documents used to generate the responses are modeled as latent variables whose prior probabilities need to be estimated. Models such as RAG (Lewis et al. 2020), marginalize the document probabilities over the documents retrieved from the index to define the log likelihood loss function which is optimized end-to-end. In this paper, we develop a variational approach to the above technique wherein, we instead maximize the Evidence Lower bound (ELBO). Using a collection of three publicly available open-conversation datasets, we demonstrate how the posterior distribution, that has information from the ground-truth response, allows for a better approximation of the objective function during training. To overcome the challenges associated with sampling over a large knowledge collection, we develop an efficient approach to approximate the ELBO. To the best of our knowledge we are the first to apply variational training for open-scale unsupervised knowledge grounded dialog systems.1 typical k = 5-10. Note that this an approximation of the document-prior distribution -sampling a document based on the prior distribution, would require retrieval scores for each document in the index collection which can be expensive to compute (Guu et al. 2020;Lewis et al. 2020)

Mix-and-Match: Scalable Dialog Response Retrieval using Gaussian Mixture Embeddings

“…Retrieval based systems for dialog models have been applied in a variety of settings. Existing work has studied the problem of grounding responses in external knowledge such as documents [22,29,35], structured knowledge [30,37], with varying degrees of knowledge-level supervision [36,37]. In such cases, a knowledge instance is first fetched and then a response is generated.…”