Although there has been an abundance of recent work on dialog, there has been relatively little feedback from industry on the technology useful for commercial applications, and the challenges faced in deployment. In this talk, I'll explore (1) Eloquent's human-in-the-loop technology ("On The Job Learning") powering practical real-world dialog, (2) Eloquent's framework for task-oriented dialog enabled by (1), and (3) practical insights "from the trenches" that we hope will be useful to the academic community. We show that with our approach, Eloquent's agent handles a wide range of natural conversations with high accuracy.
Gabor co-founded Eloquent in 2016 after graduating from the Stanford Natural Language Processing Group. Previously he graduated with honors from UC Berkeley in 2010. During and since his Ph.D., he has worked on a wide range of natural language processing tasks, including natural language generation, semantic parsing, relation extraction, natural language inference, and of course recently Conversational AI. He has published 11 papers at top NLP conferences (8 first-author), and is the author of the Stanford CoreNLP's OpenIE system, relation extraction annotator, and new Simple API, as well as co-author of the CoreNLP server. His old academic homepage can be found at: http://cs.stanford.edu/~angeli/.