The SemDaX Corpus ― Sense Annotations with Scalable Sense Inventories

The SemDaX corpus – sense annotations with scalable sense inventories Bolette S. Pedersen1, Anna Braasch1, Anders Johannsen1, Héctor Martínez Alonso1, Sanni Nimb2, Sussi Olsen1, Anders Søgaard1, Nicolai Hartvig Sørensen2 1University of Copenhagen & 2The Society for Danish Language and Literature E-mail: bspedersen@hum.ku.dk, braasch@hum.ku.dk, ajohannsen@hum.ku.dk, alonso@hum.ku.dk, sn@dsl.dk, saolsen@hum.ku.dk nhs@dsl.dk Abstract We launch the SemDaX corpus which is a recently completed Danish human-annotated corpus available through a CLARIN academic license. The corpus includes approx. 90,000 words, comprises six textual domains, and is annotated with sense inventories of different granularity. The aim of the developed corpus is twofold: i) to assess the reliability of the different sense annotation schemes for Danish measured by qualitative analyses and annotation agreement scores, and ii) to serve as training and test data for machine learning algorithms with the practical purpose of developing sense taggers for Danish. To these aims, we take a new approach to human-annotated corpus resources by double annotating a much larger part of the corpus than what is normally seen: for the all-words task we double annotated 60% of the material and for the lexical sample task 100%. We include in the corpus not only the adjucated files, but also the diverging annotations. In other words, we consider not all disagreement to be noise, but rather to contain valuable linguistic information that can help us improve our annotation schemes and our learning algorithms. 1. Annotating with different sense granularity With the general aim of increasing the level of semantic corpus resources for Danish that can be used for training of semantic systems, the SemDaX corpus has been developed as a collaboration between Centre for language Technology, University of Copenhagen, and The Society for Danish Language and Literature. In the project “Semantic Processing across Domains” 2013-2016 granted by the Danish Research Council we build on previous experiences regarding combined lexical-semantic resources, and we experiment with annotation at different levels of sense granularity. Starting in the late 90-ties with the English SemCor corpus based on the Princeton WordNet sense inventory (Landes et al. 1998), sense annotated corpora have been developed for a series of languages. Most of the corpora are annotated with fine-grained sense inventories based on existing wordnets (cf. Petrolito & Bond (2012) for an overview of these), and the resources can roughly be divided into those where all content words are annotated (so-called all-words corpora) and those where only selected content words are annotated (so-called lexical sample corpora). As discussed in Ide & Wilks (2007), Kilgarriff (2007) and others, defining appropriate sense inventories for annotation and word sense disambiguation tasks is however a very hard task. The need for coarser and more manageable sense inventories has emerged, partly driven by poor sense annotator agreement scores in the aforementioned annotations. This has resulted in a series of annotation experiments applying manually and automatically clustered senses, as seen in Agirre & Lacalle (2003), Palmer et al. (2007), Passonneau et al. (2012) Brown et al. (2010), de Melo et al. (2012), and others. The need for “light …

Content truncated to 500 words