HiNTS: A Tagset for Middle Low German

HiNTS: A Tagset for Middle Low German Fabian Barteld∗,†,‡, Sarah Ihden∗,†, Katharina Dreessen†, Ingrid Schr¨oder† ∗ these authors contributed equally †Department of Language, Literature and Media, Institute for German Studies, Universit¨at Hamburg, Germany ‡Language Technology Group, Department of Informatics, Universit¨at Hamburg, Germany firstname.lastname@uni-hamburg.de Abstract In this paper, we describe the “Historisches Niederdeutsch Tagset” (HiNTS). This tagset has been developed for annotating parts-of- speech and morphology in Middle Low German texts, a group of historical (1200–1650) dialects of German. A non-standardized language such as Middle Low German has special conditions and requirements which have to be considered when designing a tagset for part of speech and morphology. We explain these requirements, i.e. the need to encode ambiguities while allowing the annotator to be as specific as possible, and our approach for dealing with them in the tagset. We then describe two special features of the tagset. In order to prove the benefit of these tags and corresponding annotation rules, we present example searches and the possible analyses arising from the results of such searches. Besides the usefulness of our tagset, we also considered its reliability in annotation using inter-annotator agreement experiments. The results of these experiments are presented and explained. Keywords: non-standard language, part-of-speech tagset, morphological tagset 1. Introduction In this paper, we describe a tagset called “Historisches Niederdeutsch Tagset” (HiNTS). The tagset was devel- oped at the University of Hamburg in the context of the project ‘Reference Corpus Middle Low German / Low Rhenish (1200–1650)’ (ReN). The aim of the project is the creation of an annotated corpus of Middle Low Ger- man (GML) texts, a group of historical dialects of Ger- man that were used between 1200 and 1650 in the north- ern part of the German language area and in the Baltic re- gions. The corpus will be part of the ‘Corpus of Historical German Texts’, together with the corpora ‘Altdeutsch’ (Old German), ‘Mittelhochdeutsch’ (Middle High German), and ‘Fr¨uhneuhochdeutsch’ (Early New High German).1 ReN provides diplomatically transcribed Middle Low German and Low Rhenish texts from 1200 to 1650 that are lem- matised and annotated with part of speech (POS) and mor- phology using HiNTS, the tagset described in this paper. Since GML is a historical language, which is not well de- scribed and hardly any resources for annotating GML texts exist, there are special requirements for the tagset which we will focus on in this paper. When annotating a historical language, the annotators lack the intuition of a native speaker. Therefore, decisions about a category can only be based on the comparison and statis- tical analysis of texts (Dipper et al., 2013, 2). This however, should be avoided while annotating a text and left for later analysis using the annotated data. We present design deci- sions for our POS tagset that result from this requirement. HiNTS is also used to annotate morphology. Here, many tokens are ambiguous, for example, as regards their inflec- tional ending. To allow the annotators to encode such an ambiguity, HiNTS allows the combination …

Content truncated to 500 words