This article describes the experiment for determining inter-annotator agreement in the semantic annotation of the Latvian WordNet, which was carried out within the project „Latvian Wordnet and Word Sense Disambiguation”. The semantic relations used in the Latvian WordNet are synonymy, near-synonymy, hyponymy, antonymy, meronymy, gradation sets and similarity (defined as „see also”). These semantic relations are determined for four main parts of speech: nouns, verbs, adjectives, and adverbs. During the experiment, 20 words were selected from the online dictionary Tēzaurs.lv, and the team of linguists individually formed semantic links between the senses of these words and other word senses in the dictionary. The obtained data was analysed using both statistical and qualitative analysis methods to determine the level of inter-annotator agreement and potential causes of differing word sense interpretation. The results of the experiment have led to the conclusion that individual annotation yields a medium level of inter-annotator agreement, whereas collective discussions improve this level significantly, even though certain cases remained where the annotators could not reach a consensu
📖 افتح في inklap 🔗 DOI 📮 اطلب بحثاً