Using language to save lives by Diana Maynard

How can social media help with natural disasters? Grantham Supervisor Dr Diana Maynard works with a diverse range of people to develop tools which can harness the power of communication to save lives.

From swearing to plastics

Dr Dina Maynard

I have been working in the Natural Language Processing Research Group (NLP) in the Department of Computer Science for over 18 years, developing tools for automatically “understanding” text – for example, what topics people are talking about, what sentiments they are expressing, what kinds of language they are using. To understand text like this means processing potentially millions of documents, something which can’t be done by humans as the volume is too great. NLP can also help us look at things like correlating language with behaviour, for example what kind of messages on Twitter might influence people to change their environmental behaviour?

I work in a team of 10, as a linguist in a group of mostly computer scientists, which means I get to design the tools and oversee a number of projects. While my team take care of the real programming side of things, I get my hands dirty looking at how language is used, which can mean anything from building complex applications for modelling language to compiling lists of swear words used in Twitter. I currently supervise two PhD students (including Grantham Scholar Ye Jiang), who are investigating media reporting of environmental disasters, and social media discussions about Nigerian politics respectively. I am also PI of a European project KNOWMAK, which is developing tools to understand and visualise the state of European research. This includes case studies around the plastic-free ocean and carbon-neutral cities, two of the key areas of focus of the European Commission.

COMRADES and co-creation

One of our other core projects with which I am involved, COMRADES, is about developing tools to aid communication during disasters – something which has very clear practical purposes and can be a literal life saver. A key aspect of dealing with the aftermath of disasters and critical events is to get a quick understanding of the situation. This enables responders to not only assess the gravity of the situation and the needs of the affected community, but also the operational circumstances and the availability of resources.

Crisis situations tend to be chaotic, and it can be hard to get a coherent picture. Even though information about the event is often prolific via social media and mobile communication, it is often tricky to sift through the maelstrom of messages to find what needs are most critical and urgent. This results in potentially vital information quickly becoming unmanageable. In some disasters, up to 80% of messages have turned out to be irrelevant or uninformative. Sometimes they are deliberately misleading, but more often they are either just personal opinions about the event, or they provide incorrect, out-of-date or duplicate information.

COMRADES has worked in Nepal to understand how to give better disaster relief

In the COMRADES project, researchers from 5 different institutions are working together to create an open-source, community resilience platform, designed to help affected citizens reconnect, respond to, and recover from crisis situations. In other words, COMRADES is researching how technologies can help communities to be more resistant to disasters, aiming to optimally share information so that critical help can be provided at the right time and in the right place. By producing a platform to be primarily used by the citizens themselves, a key element of the project is to empower local communities. Furthermore, these communities are actively involved in the design process of these tools and platforms. This user-centered approach is not just talking, but hands-on working together (so-called “joint design” or “co-creation”).

Researchers at the Ushahidi and the Delft University of Technology in the Nertherlands, who are experts in disaster management, have been engaged in field work involving a large and diverse number of local people, ranging from community organisers to even the Prime Minister of Nepal. Over the course of 16 days, they arranged workshops, focus group sessions, and interviews with approximately 150 people in Kathmandu and in remote areas of Nepal, in order to gather input and ideas from people from all walks of life. According to the IFRC World Disaster Report 2013, in the aftermath of crisis events approximately 90 percent of the affected people were saved by the locals themselves. Individual citizens, community groups, and local authorities are the ones who are first on the scene and therefore need to be supported in order to take the correct action. However, it is clear that the role of the governments and the international agencies must equally be recognised and connected to the local communities.

In collaboration with the Cordaid NGO, people from the Laharepauwa community, located in the Rasuwa district (north of Kathmandu), were invited to a workshop, where they mapped out what information was needed and was available to them at various moments following the devastating 2015 earthquake. Using the simple methodology of sticky notes attached to a wall, they discussed with great enthusiasm what communication methods were used (and what could additionally be used) to obtain the various pieces of information that would help them make informed decisions. This helped the research teams determine what information, at what time and through what channels, would best support the community in their decisions.

Fake news

In terms of technology, the project is building a next generation platform to filter messages from citizens as they arrive from social media and SMS messages, identifying those which are most relevant and informative. In the short-term, this enables faster identification of lone messages requesting urgent help; while in the longer term, it also helps to point out unreliable sources of information.

The University of Sheffield team, based in the Computer Science department, has a history of successful work on fake news and rumour detection, as well as developing the open-source GATE toolkit for text processing, which can be used to automatically analyse messages for all kinds of diverse information. This team has been responsible for developing ways to filter and analyse the messages during a crisis and in its aftermath – something that is impossible for humans to do effectively, due to the huge volumes of data. This includes analysing the non-standard language used in social media, detecting important entities such as locations mentioned in the messages and plotting them on a map, and determining the reliability, informativeness and actionability of the information in the messages.

When designing these kinds of technical tools, it is important to make use of the information from the ground, such as understanding what kinds of message or information are most important or urgent at a particular time. These needs might change for different kinds of emergency (e.g. flood, earthquake, fire) and at different times (live people trapped in buildings are extremely urgent to find, but after a few days, the chances of a trapped person still being alive decrease, and so priorities may change).

Community

The real sustainability of this effort, and one of the most exciting things about it, lies in the combination of technology and community engagement. The information processing tools are thus developed alongside input from and interaction with three main communities:

Working directly with these groups not only gives us incredible insight into the conditions and challenges they face, but also affords us the chance to really deliver tools which can change their lives, not just through immediate help during crises, but also long-term by making their communities more resilient.

Life as a very busy academic can be quite challenging at times, especially the frequent travel and long hours, but working on interesting and useful projects definitely makes up for it!