Shallow parsing for recognizing threats in Dutch tweets

In this paper, we investigate the recognition of threats in Dutch tweets. As tweets often display irregular grammatical form and deviant orthography, analysis by standard means is problematic. Therefore, we have implemented a new shallow parsing mechanism which is driven by handcrafted rules. Experimental results are encouraging, with an F-measure of about 40% on a random sample of Dutch tweets. Moreover, the error analysis shows some clear avenues for further improvement.