A Case Study: NLG meeting Weather Industry Demand for Quality and Quantity of Textual Weather Forecasts

Abstract

In the highly competitive weather industry, demand for timely, accurate and personalized weather reports is always on the rise. In this paper we present a case study where Arria NLG and the UK national weather agency, the Met Office came together to test the hypothesis that NLG can meet the quality and quantity demands of a real-world use case.

Fuzzy sets across the natural language generation pipeline

“…Within this context, D2T/NLG systems, either alone or as a complementary support to visualization, will allow to improve the understanding of large data sets in many application domains and bring data closer to people. Empiric studies [1] show that visual information alone is not always capable of adequately communicating relevant information about data to users. In this regard NLG and D2T are descriptive approaches that, combined with sound analysis techniques, are starting to prove to be valuable complementary informative tools in the Data Science realm.…”

On the role of linguistic descriptions of data in the building of natural language generation systems

“…An interesting example of both qualitative and quantitative evaluations is described in [42], where Sripada et al detail the extrinsic evaluation process of the weather forecast NLG system by ARRIA NLG after successfully assessing the output quality internally (intrinsically), which consisted in a questionnaire for end-users that use weather information for decision-making. The questionnaire had three questions related to quality assessment (quantitative), but also asked for free text comments (qualitative).…”

“…texts with different detail level and style depending on the final user profile. This approach was further developed and commercialized [40], and is currently used by UK's national weather service, Met Office [41,42] to automatically issue natural language forecasts for every location in the UK.…”

SaferDrive: An NLG-based behaviour change support system for drivers

“…Early systems within this domain already appeared in the 1990s, such as FOG from Goldberg, Driedger, and Kittredge (1994). More recent examples are from Ramos-Soto, Bugarín, Barro, Gallego, Rodríguez, Fraga, and Saunders (2015), who developed a Data-to-Text system for forecasts about air quality, or Sripada, Burnett, Turner, Mastin, and Evans (2014) who developed a Data-to-Text system for the UK national weather service.…”