Data2Text is actively working on developing solutions to help clients with monitoring and surveillance of complex systems. We cannot publicly describe this work because of confidentiality, but the concepts can be illustrated via research projects carried out at University of Aberdeen.

Smart Award

The Smart award system generates weather forecast texts for multi-point areas, identifying and describing motion of precipitation occurring during the forecast period. This project demonstrates the Data2Text ability to use a combination of different data sources to create concise and clear narrative. The input data includes weather data in the format of Binary GRIB files, which is freely available on the web. The project also uses geospatial data in the format of Shape files. The geospatial data is used for motion detection and for geo-referencing in the text. Example: The following text was generated by the Smart award system using Shape files for the area of the UK, and GRIB binary data for the 8th of February 2011. "Moderate rain in the west of Wales and the south of England will spread north across much of the UK by dawn tomorrow, with a chance of snow over 300m in some areas of Tayside and Grampian towards the early hours."

Pollen Forecast

A simple example of a data-to-text system is the pollen forecast for Scotland generator. This system takes as input 6 numbers which give predicted pollen levels in different parts of Scotland, and from these numbers produces a summary text.

The forecast generator produces texts such as "Grass pollen levels for Tuesday have decreased from the high levels of yesterday." This text shows Data2Text's ability to spot trends in the pollen data over time. The software can also generate texts that summarise the data spatially and produce an easy to read summary. This produces an output such as : "Pollen values of around 4 to 5 across most parts of the country. However, in South Eastern areas, pollen levels will be high with values of 6."

A simple demonstration of this can be found on our demo page.

Sumtime-Turbine

An early project in this area was Sumtime-Turbine, which summarised time-series sensor data from a gas turbine, to help maintenance engineers monitor the turbine; the focus was on supporting long-term surveillance and early problem detection.  This was developed in collaboration with Intelligent Applications (which has since ceased trading).

BabyTalk

A later project was BabyTalk, which developed a suite of systems which generated summaries of clinical data about babies in a neonatal intensive care unit.  This data included time-series sensor data (heart rate, blood pressure, etc); lab results; and records of interventions by medical staff (drugs, surgical procedures, etc). From this rich and heterogenous data set, the Babytalk systems generated:

• short-term summaries (of the past 45 minutes of data), to support real-time decision making about medical interventions 
• summaries of a data over a 12-hour nursing shift, to support handover and ensure staff coming on shift were fully informed about the baby's status 
• non-technical summaries of a day's worth of data for the baby's parents, to keep them informed about their baby's status 

BabyTalk was developed in collaboration with NHS Lothian, and was evaluated in a range of ways, ranging from a controlled experiment which empirically assessed its impact on decision quality, to feedback questionnaires from nurses who used the system on-ward.

The common behind these projects, and the work currently being done by Data2Text for its clients, is that people who monitor complex systems often have tremendous amounts of data available to them, and it is not possible for them to examine all of this data within realistic time constraints. Alarming systems can draw attention to specific problems, but they do not communicate the "big picture" and also may not detect problems in their early stages, when they can be dealt with most cost-effectively.  Summaries produced by Data2Text systems  help the doctor or engineer see the big picture, and detect problems at an early stage.