Combining Ontologies and Neural Networks for Analyzing Historical Language Varieties. A Case Study in Middle Low German

Combining Ontologies and Neural Networks for Analyzing Historical Language Varieties. A Case Study in Middle Low German Maria Sukhareva Christian Chiarcos {sukharev|chiarcos}@informatik.uni-frankfurt.de Abstract In this paper, we describe experiments on the morphosyntactic annotation of historical language varieties for the example of Middle Low German (MLG), the official language of the German Hanse during the Middle Ages and a dominant language around the Baltic Sea by the time. To our best knowledge, this is the first experiment in automatically producing morphosyntactic annotations for Middle Low German, and accordingly, no part-of-speech (POS) tagset is currently agreed upon. In our experiment, we illustrate how ontology-based specifications of projected annotations can be employed to circumvent this issue: Instead of training and evaluating against a given tagset, we decomponse it into individual features which are predicted independently by a neural network. By applying consistency constraints (axioms) from an ontology, then, the predicted feature probabilities are decoded into a sound (ontological) representation. Using these representations, we can finally bootstrap a POS tagset capturing only morphosyntactic features which could be reliably predicted. In this way, our approach is capable to optimize precision and recall of morphosyntactic annotations simultaneously with bootstrapping a tagset. Keywords: morphosyntactic annotation, ontology-based annotation, feed-forward neural networks, Middle Low German 1. Background For contemporary languages, most notably those from the Germania (English, German, Dutch, Swedish, etc.), a large amount of linguistic resources has been produced in the last decades, whereas historical and minor Germanic lan- guage varieties attracted attention only recently. With the increased interest in Digital Humanities and computational philology, several annotated corpora were released, e.g., for historical varieties of English (Kroch and Taylor, 2000, PCHE), Icelandic (R¨ognvaldsson et al., 2012), High Ger- man (Linde and Mittmann, 2013; Light, 2013), or the oldest Germanic languages in general (Haug and Jøhndal, 2008, PROIEL). As for continental Germanic, however, it is remarkable that attention focused mostly on the oldest language strata (Old Saxon, Old High German, Gothic), whereas corpora and NLP tools for languages of the High Middle ages (esp., Middle High German and Middle Low German) are not (yet) publicly available.1 In this paper, we specifically focus on the morphosyntac- tic annotation of Middle Low German (MLG), the official language of the German Hanse during the Middle Ages and a dominant language around the Baltic Sea and in Scandi- navia before its decline in the late 16th c. Despite its name, 1Annotated data is to be expected from the German Ref- erence Corpus projects. For Middle High German (http: //referenzkorpus-mhd.uni-bonn.de) and Middle Low German (https://vs1.corpora.uni-hamburg. de/ren), however, no data has been released so far. Also, the Early Modern High German Reference Corpus (http://www.ruhr-uni-bochum.de/wegera/ref)data is not available yet. Independently from these efforts, a Corpus of Historical Low German is currently being created (http://www.chlg.ac.uk), but has not released annotated data either. For Dutch, the situation is slightly better, and a POS-annotated corpus for Middle Dutch is available, albeit limited to 13th century material (http://tst-centrale.org/nl/ producten/corpora/corpus-gysseling/6-59). Low German is not closely related to standard (High) Ger- man, …

Content truncated to 500 words