Exploring Conversational Language Generation for Rich Content about Hotels

Exploring Conversational Language Generation for Rich Content about Hotels Marilyn A. Walker1, Albry Smither2, Shereen Oraby1, Vrindavan Harrison1, Hadar Shemtov3 University of California Santa Cruz1, Google Content Studio2, and Google Research3 Abstract Dialogue systems for hotel and tourist information have typically simplified the richness of the domain, focusing system utterances on only a few selected attributes such as price, location and type of rooms. However, much more content is typically available for hotels, often as many as 50 distinct instantiated attributes for an individual entity. New methods are needed to use this content to generate natural dialogues for hotel information, and in general for any domain with such rich complex content. We describe three experiments aimed at collecting data that can inform an NLG for hotels dialogues, and show, not surprisingly, that the the sentences in the original written hotel descriptions provided on webpages for each hotel are stylistically not a very good match for conversational interaction. We quantify the stylistic features that characterize the differences between the original textual data and the collected dialogic data. We plan to use these in stylistic models for generation, and for scoring retrieved utterances for use in hotel dialogues. KEYWORDS: dialogue, conversation, natural language generation, hotels domain. 1. Introduction Research and advanced development labs in both industry and academia are actively building a new generation of con- versational assistants, to be deployed on mobile devices or on in-home smart speakers, such as Google Home. None of these conversational assistants can currently carry on a co- herent multi-turn conversation in support of a complex de- cision task such as choosing a hotel, where there are many possible options and the user’s choice may involve mak- ing trade-offs among complex personal preferences and the pros and cons of different options. For example, consider the hotel description in the InfoBox in Figure 1, the search result for the typed query “Tell me about Bass Lake Taverne”. These descriptions are written by human writers within Google Content Studio and cover more than 200 thousand hotels worldwide. The descrip- tions are designed to provide travelers with quick, reliable and accurate information that they may need when mak- ing booking decisions, namely a hotel’s amenities, prop- erty, and location. The writers implement many of the de- cisions that a dialogue system would have to make: they make decisions about content selection, content structur- ing, attribute groupings and the final realization of the con- tent (Rambow and Korelsky, 1992). They access multiple sources of information, such as user reviews and the hotels’ own web pages. The descriptions cannot be longer than 650 characters and are optimized for visual scanning. There is currently no method for delivering this content to users via a conversation other than reading the whole InfoBox aloud, or reading individual sections of it. Structured data is also available for each hotel, which in- cludes information about the setting of a hotel and its grounds, the feel of the hotel and its rooms, points of in- terest nearby, …

Content truncated to 500 words