SemRelData ― Multilingual Contextual Annotation of Semantic Relations between Nominals: Dataset and Guidelines

SemRelData – Multilingual Contextual Annotation of Semantic Relations between Nominals: Dataset and Guidelines Darina Benikova, Chris Biemann Language Technology Group, Computer Science Department, TU Darmstadt Hochschulstr. 10, 64289 Darmstadt, Germany darina.benikova@gmail.com, biem@cs.tu-darmstadt.de Abstract Semantic relations play an important role in linguistic knowledge representation. Although their role is relevant in the context of written text, there is no approach or dataset that makes use of contextuality of classic semantic relations beyond the boundary of one sentence. We present the SemRelData dataset that contains annotations of semantic relations between nominals in the context of one paragraph. To be able to analyse the universality of this context notion, the annotation was performed on a multi-lingual and multi-genre corpus. To evaluate the dataset, it is compared to large, manually created knowledge resources in the respective languages. The comparison shows that knowledge bases not only have coverage gaps; they also do not account for semantic relations that are manifested in particular contexts only, yet still play an important role for text cohesion. Keywords: semantic relations in context, multilingual, multi-genre, annotated dataset 1. Introduction Although the role of classic semantic relations is consid- ered an important one in context of written text, there was so far no approach or dataset that addresses and quantifies the amount of contextuality of classic semantic relations, in particular beyond the scope of one sentence. We annotated semantic relations between nominals in the context of one paragraph. We only annotated relations that were present in the context, i.e. where the context hints at the existence of such relations. We annotated classical semantic relations, such as syn- onymy, hyperonymy, holonymy and co-hyponymy between nominals. Synonyms are mostly defined as different words with the same meaning, hypernyms are superordinate terms to their subordinate hyponyms, co-hyponyms are words with the same hypernym, and holonyms are terms referring to the whole, which consists of meronyms. The contribution of this paper is threefold: we present Sem- RelData (Semantic Relations Dataset), containing contex- tual semantic relations, its evaluation and an analysis of the impact of contextuality in semantic relations. Our dataset consists of 60 language-parallel documents and ∼60.000 tokens. More specifically, it contains parallel en- cyclopaedic, newspaper and literary texts in English, Ger- man and Russian. The dataset was manually annotated in a double-annotator setting by students with linguistic back- ground and subsequently curated. In the next step, the re- lations were extended using properties of the relations, e.g. transitivity of synonymy. The setting of the multilingual and multi-genre corpus en- abled us to analyse the universality of the context notion. The analysis was performed by comparing our dataset to the largest manually created or revised knowledge bases in the respective languages, i.e. WordNet (Miller and Fellbaum, 1991), GermaNet (Henrich and Hinrichs, 2011) and RuTes (Loukashevich, 2011). Annotated relations not present in these databases were further analysed and man- ually classified according to the reason why they were not present in the respective database. This study highlights the significance of contextuality of se- mantic relations and quantitatively …

Content truncated to 500 words