SoMeWeTa (short for Social Media and Web Tagger) is a part-of-speech tagger that supports domain adaptation and that can incorporate external sources of information such as Brown clusters and lexica. It is based on the averaged structured perceptron and uses beam search and an early update strategy. It is possible to train and evaluate the tagger on partially annotated data.
SoMeWeTa achieves state-of-the-art results on the German web and social media texts from the EmpiriST 2015 shared task on automatic linguistic annotation of computer-mediated communication / social media. Therefore, SoMeWeTa is particularly well-suited to tag all kinds of written German discourse, for example chats, forums, wiki talk pages, tweets, blog comments, social networks, SMS and WhatsApp dialogues.
In addition, we also provide models trained on German, English and French newspaper texts, as well as models for Bhojpuri and spoken Italian. For all languages, SoMeWeTa achieves highly competitive results close to the current state of the art.
More detailed documentation is available here.