Theo Meder

TweetGenie: automatic age prediction from tweets by D. Nguyen, R. Gravel, D. Trieschnigg, and T. Meder; with Ching-man Au Yeung as coordinator

A persons language use reveals much about the persons social identity, which is based on the social categories a person belongs to including age and gender. We discuss the development of TweetGenie, a computer program that predicts the age of Twitter users based on their language use. We explore age prediction in three different ways: classifying users into age categories, by life stages, and predicting their exact age. An automatic system achieves better performance than humans on these tasks. Both humans and the automatic systems tend to underpredict the age of older people. We find that most linguistic changes occur when people are young, and that after around 30 years the studied variables show little change, making it difficult to predict the ages of older Twitter users.

MOMFER: A Search Engine of Thompson's Motif-Index of Folk Literature

More than fifty years after the first edition of Thompsons seminal Motif-Indexof Folk Literature, we present an online search engine tailored to fully disclose the index digitally. This search engine, called MOMFER, greatly enhances the searchability of the Motif-Index and provides exciting new ways to explore the collection. This is enabled by our use of modern techniques from both natural language processing and information retrieval. The key feature of the search tool is the way in which it allows users to search the Motif-Index for semantic concepts, such as ‘mythical animals’, ‘mortality’, or ‘emotions’. This paper will explain the motivations for creating the search tool, explicate the production process, and show in a number of case studies how the search tool can be used to explore the index in innovative ways.

In Search of the Dutch Lore of the Land. Old and New Legends throughout the Netherlands

Inspired by The Lore of the Land, published by Jennifer Westwood and Jacqueline Simpson in 2005, a Dutch version of this book, comprising local and regional legends, was published in the Netherlands in 2010 by Willem de Blécourt, Ruben Koman, Jurjen van der Kooi, and Theo Meder. The editors selected over three hundred and fifty legends, representative of the twelve provinces of the Netherlands, for the volume. Some were real ecotypes, having been appropriated by local communities and having led to the erection of associated statues, often for touristic and commercial purposes, as well as for shaping collective identities, while others were almost forgotten tales. The editors also found, somewhat unexpectedly, that quite a number of popular legends had been composed and constructed by romantic writers in the nineteenth century, and that these older legends were much more popular than the recently-invented ones. As the editors also included a number of contemporary legends, they discovered that one such legend had already become popular enough to have a statue erected and a song composed in its honour.

The Apocalypse on Twitter

There was one trending topic on Twitter in December 2012 that we could have seen coming for a few years now: the New Age prophecy of the End of Times on 21 December 2012—all because some Mayan calendar supposedly ended on this date. For 2 weeks long—a week before the Apocalypse and a week after—we monitored Twitter for Dutch words concerning the End of the World. We caught 52,000 tweets in 2 weeks. When did the stream of rumours peek? How many retweets were involved? Was there much micro-variation? What was the overall content of the tweets? What emotions were expressed in the tweets? How did religious people respond? And finally, how many people confessed they were truly scared because of the prophecy? These are intriguing questions that we can answer by using a few basic computational tools. Although the Apocalypse got a lot of attention in the news media, it turned out most Dutch people on Twitter took the End of Days with a grain of salt.

Animacy detection in stories

This paper presents a linguistically uninformed computational model for animacy classification. The model makes use of word n-grams in combination with lower dimensional word embedding representations that are learned from a web-scale corpus. We compare the model to a number of linguistically informed models that use features such as dependency tags and show competitive results. We apply our animacy classifier to a large collection of Dutch folktales to obtain a list of all characters in the stories. We then draw a semantic map of all automatically extracted characters which provides a unique entrance point to the collection.