People intuitively ask natural language dialogue systems for advice on nutrition and dietary guidelines, but systems based on prompted text generation are susceptible to fabricating details, which could be hazardous to non-specialist users. The FoodBench-QA shared task grounds answers in knowledge bases with linked ontologies, in order to evaluate and mitigate fabrication of nutrition information. Our system treats nutrient estimation and entity linking not as a generative problem (predicting numbers from scratch), but as a retrieval problem. We operate on the hypothesis that for structured data like food composition, finding a “real” recipe that is 95% similar is more likely to approximate the correct values than letting the language model fabricate values from sparse context. Our system performed well on food safety labeling from recipe ingredients alone, and it did not benefit from the additional information of recipe titles. In the NER and NEL tasks, our system handled the recipe-focused FCD corpus well, but suffered from poor recall on scientific abstracts and the artificial dataset. These results show the importance of basing information retrieval and question answering in data that is well-matched to the target data.

Large vocabulary domain-agnostic Automatic Speech Recognition (ASR) systems often mistranscribe domain-specific words and phrases. Since these generic ASR systems are the first component of most voice assistants in production, building Natural Language Understanding (NLU) systems that are robust to these errors can be a challenging task. In this paper, we focus on handling ASR errors in named entities, specifically person names, for a voice-based collaboration assistant. We demonstrate an effective method for resolving person names that are mistranscribed by black-box ASR systems, using character and phoneme-based information retrieval techniques and contextual information, which improves accuracy by 40.8% on our production system. We provide a live interactive demo to further illustrate the nuances of this problem and the effectiveness of our solution.

We demonstrate an end-to-end approach for building conversational interfaces from prototype to production that has proven to work well for a number of applications across diverse verticals. Our architecture improves on the standard domain-intent-entity classification hierarchy and dialogue management architecture by leveraging shallow semantic parsing. We observe that NLU systems for industry applications often require more structured representations of entity relations than provided by the standard hierarchy, yet without requiring full semantic parses which are often inaccurate on real-world conversational data. We distinguish two kinds of semantic properties that can be provided through shallow semantic parsing: entity groups and entity roles. We also provide live demos of conversational apps built for two different use cases: food ordering and meeting control.