Nonparametric Learning for Situated Data-to-Text Generation: Helping People to Understand Uncertain Data

Information overload is a pervasive problem in many environments, particularly those in which human decision making is based on extensive data sets. Data-to-text systems have been shown to successfully address this problem by automatically generating textual descriptions of the underlying data. However, when translating (numerical) data into words, an appropriate level of precision needs to be chosen. The following example is from a system which summarises medical time series data for neonatal care: "At 17:24 T1 is 35.7 and T2 is 34.5C" (Gatt et al., 2009). This summary is clearly targeted to experts, such as doctors or nurses, which need precise information for decision making. However, other users, such as visiting parents might be more happy with a description such as "In the evening your baby had normal temperature."

In this project, we will build a data-to-text system that automatically determines the appropriate level of precision for a given context by using statistical machine learning methods. These methods can learn an optimal generation policy from real data and promise to be more robust to new situations than hand-written rules by human experts.
We will also investigate novel feedback-based non-parametric state estimation methods to reduce the data annotation cost for data-to-text systems. Typically, the first step in creating such systems is to manually interpret and align the raw data sources. However, this step is very costly as human experts need to trained for this task. Our new methods promise for data-to-text systems to be rapidly applied to new domains.

The domain we will be targeting for this initial project is pedestrian navigation, where the task is to translate uncertain user positions into walking instructions. The underlying data uncertainty here arises from several sources, such as the user's speech signal, the GPS location, estimated viewshed, walking direction and speed. We will integrate and test our learnt data-to-text generation strategy by integrating it in an existing system and running an evaluation with real users.

One of the outcomes of this project is a data-driven linguistic view on the question of "how to communicate uncertainty", which is an active interdisciplinary research area, including researchers from medicine, law, environmental modelling and climate change.
In future work we will also investigate how the proposed framework transfers to new domains, such as natural language generation from medical data, weather forecasts, or output from complex environmental models.

The overall aim of this research is to provide better interfaces for people to understand "big" data more intuitively. As such, the outcomes of this research have three main impact beneficiaries: (1) academic research investigating how to (automatically) communicate data, (2) informing policy makers how to communicate their findings, and (3) the general public needing to make decisions based on (uncertain) data.

(1) Within the academic community, this proposal aims to bridge the gap between two disciplines, which are both concerned with decision support: data-to-text Natural Language Generation (NLG) systems, and interdisciplinary research working on communicating uncertainty. While other disciplines, such as medicine, environmental modelling, climate change or weather forecasts strongly promote the need for communicating data uncertainty, data-to-text systems still assume that their underlying data is precise and correct. If automatic data-to-text systems are to be widely used within decision support, they must have mechanisms to communicate uncertain data in an effective way. This research will contribute a principled study and data-driven framework for generating descriptions of underlying data uncertainty.

(2) The developed models will not only be beneficial to academics from other disciplines, but also to policy makers, such as the International Panel of Climate Change (IPCC) for example. Currently, the IPCC prescribes a standardised mapping of data uncertainty into words, which is widely recognised and applied beyond climate change research. However, the guidelines by the IPCC are not grounded in linguistic research and have been reported to be problematic in their use. In future, these guidelines could be informed by the outcomes of this research.

(3) Finally, the long-term beneficiary of this research is the general public, who in their daily life have to make decisions based on vast amounts of unstructured information becoming more readily available. For example, the British government recently announced in a white paper that it will be greatly expanding the amount of data which it shares with the rest of us (http://www.guardian.co.uk/politics/2012/jun/27/public-services-data-published-price). However, most people lack the skills and tools to access and interpret large data sets. The overall aim of this research is to provide better direct access to data through user-friendly interfaces, which help people to understand data more intuitively and support decision making.

For a description of how these impact goals will be realised and how their success will be measured, please see "Part III: Pathways to Impact" of this proposal.