Using spatial reference frames to generate grounded textual summaries of georeferenced data

Abstract

Summarising georeferenced (can be identified according to it's location) data in natural language is challenging because it requires linking events describing its nongeographic attributes to their underlying geography. This mapping is not straightforward as often the only explicit geographic information such data contains is latitude and longitude. In this paper we present an approach to generating textual summaries of georeferenced data based on spatial reference frames. This approach has been implemented in a data-to-text system we have deployed in the weather forecasting domain.

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

“…RoadSafe [43,44] automatically generates advice for deploying road maintenance vehicles and the deployment routes they must follow, taking into account both meteorological and geographic data (see Fig. 5).…”

Proceedings of the 12th European Workshop on Natural Language Generation - ENLG '09

“…NLG systems can be faced with similar asymmetries, for example when an artificial doctor decides to keep its predictions vague to avoid being contradicted by the facts; a doctor who says "These symptoms will disappear fairly soon" is less likely to get complaints than one who says "These symptoms will have disappeared by midnight next Sunday". Something similar holds for a roadgritting system (like the one in Turner et al 2008), which might easily face lawsuits if it gets things too evidently wrong. Advertisements also come to mind, be-3 Another game with this property was described in De Jaegher (2003), involving a more complex version of the game of the two generals (section 2).…”

“…There can be thousands of dangerous roads on a given night, and it is often impossible to say in a few words exactly which road are dangerous (Turner et al 2008). One summary produced by the generator might approximate the data by saying 'Roads in the Highlands are icey' while another might say 'Roads above 500 metres are icey' (assume this covers a larger set of roads).…”

“…To see how NLG systems could be amenable to a decisiontheoretical analysis, in which the expected payoffs associated with different texts are compared, consider an NLG system that informs roadgrit-ters' decisions about the condition of the roads in Scotland, to help them decide which ones are icy enough to require treatment (e.g. Turner et al 2008).…”

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

“…Adapting techniques used for time series segmentation, this project developed a framework for data summarization in the context of NLG [Sripada et al, 2003b]. This time series summarization framework was later extended to summarizing spatio-temporal data in the ROAD-SAFE system [Turner et al, 2008]. ROADSAFE too was used in an industrial context to produce weather reports (including text in English and a table) for road maintenance in winter months.…”