Towards a Tower of Babel: The Conference on Language Resources and Evaluation

Located at the Bosphorus strait, the metropolis of Istanbul connects Europe and Asia. The history-charged city, in ancient times known as Constantinople, provided the perfect scenery for LREC 2012, the Conference on Language Resources and Evaluation.

Stefan Gindl, researcher and lecturer at the Department of New Media Technology, gave a speech on the usage of games with a purpose in the research field of Sentiment Analysis. Sentiment Analysis is attracting growing interest and has already become a widely demanded technology thanks to its capability of tracking human opinions in very large document collections such as the WWW.

Manual approaches, currently still superior in accuracy thanks to human knowledge of common-sense facts, founder on the sheer amount of available data. Equipping computers with the explicit knowledge and heuristics retrieved by Sentiment Analysis research renders large-scale text analysis feasible in an acceptable amount of time with satisfying accuracy (see webLyzard Media Monitoring and Web Intelligence).

A hurdle, though, is the initial creation of lexical resources used by the algorithms. Expensive expert-knowledge driven approaches providing high quality outpout compete with inexpensive semi-automatic methods of lower quality. These circumstances call for new methods combining both advantages without their downsides. One promising approach is games with a purpose. Creating a setting for human knowledge acquisition in a game-like style motivates people to share their knowledge without providing financial incentives. Moreover, the game-like design further increases the acceptance and concentration span of the players by circumventing frustration by boredom (try it out yourself!). Leveraging the wisdom of the crowds, games with a purpose contribute to the creation of valuable resources for this technology.

Above this, Stefan also presented the outcome of a joint effort of the Interest Group on German Sentiment Analysis (IGGSA). The newly founded cooperation of research departments in Germany, Switzerland and Austria aims to create resources and methods for Sentiment Analysis in German. IGGSA presented the first outcome of their work, a gold standard corpus for the evaluation of approaches to Sentiment Analysis in German, in the fruitful environment LREC 2012.