Using HurtLex and Best-Worst Scaling to Develop ERIS

ERIS, a lexical resource of Modern Greek for offensive language detection, is the result of cleansing, enriching and assigning graded offensiveness values to the EL branch of HurtLex. ERIS contains 1148 entries and is openly available. Graded values were obtained with the Best-Worst Scaling method that was applied with the Litescale tool. Nouns and adjectives that have humans as a target were found to attract bigger offensiveness values. The classification of the terms in ERIS with the BWS method and a previous classification of a substantial subset of these terms into "offensive (context in/dependent)" with the inter-annotator agreement method are found to stand in a broad correlation, thus validating the methodology that was adopted to produce a more fine-grained and informative affective lexical resource. ERIS contains 1148 terms and their inflectional paradigms. It is openly available under the CC-BY-NC 4.0 license.