How games are helping computers understand language

We want people to take part in a new online game which is part of major research to help computers improve their understanding of human language.

Perfect for lovers of language and literature, the TileAttack game involves players making annotations to highlight relationships between words and phrases. Working in collaboration with Queen Mary University, the web-based game will collect millions of pieces of data to help computers better understand the complexities of how humans speak and understand phrases.

The need for TileAttack emerged from Phrase Detectives, a web-based game we developed at Essex which has attracted more than 40,000 players. Phrase Detectives is a 'game with a purpose' (GWAP) that uses game elements to encourage players to conduct tasks that result in valuable research resources. Phrase Detectives asks players to provide annotations of natural language text that can then be used to improve automated computer systems better understand human language. 

“By working with real game developers we have taken a whole new approach and developed TileAttack as a more fun game than Phrase Detectives and will attract more players who are also able to compete against each other,” explained Professor Udo Kruschwitz, from our School of Computer Science and Electronic Engineering (CSEE). “This means it will also gather even more valuable data, which will help our research in computational linguistics.”

Some of Europe’s leading computational linguists studied or worked at Essex, which last year celebrated 40 years of computational linguistics research. Essex is also renowned for looking at natural language processing on a much larger scale than other organisations which tend to concentrate on specific areas of language. 

“Our expertise in natural language processing at Essex means we can make sense of the data generated by these games as we know what questions to ask to get the maximum amount of reliable data,” added Professor Kruschwitz. "This work helps push state-of-the-art in artificial intelligence (AI) research by turning tedious annotation tasks into an enjoyable pastime." 

The research will particularly focus on ambiguity in anaphora - the use of a word such as a pronoun referring back to a word used earlier in a text or conversation. The new game will involve players annotating a wide range of copyright-free material including Brothers Grimm fairy tales, Victorian texts and Wikipedia. 

The game was developed by Chris Madge, a PhD student on the Intelligent Games and Game Intelligence Programme (IGGI), a collaborative project between the universities of Essex, Queen Mary, York and Goldsmiths. 

Professor Kruschwitz added: “Human language may seem easy for us to understand, but it is full of complexities when it comes to how phrases, people, places and ideas are connected. Collecting data from people playing TileAttack means we can collect knowledge about natural language to help computers fully understand the rules of language. TileAttack is part of our research direction of 'gamifying' every single step in a long pipeline of natural language processing steps." 

The project is being funded as part of a €2.5 million ERC (European Research Council) Advanced Grant awarded to Essex and Queen Mary led by Professor Massimo Poesio, and also involving Professor Richard Bartle and Dr Jon Chamberlain at Essex. 

The researchers are also working in partnership with the Linguistic Data Consortium, which was formed in 1992 to address the critical data shortage facing language technology research and development.