Proceedings of the First Workshop on Abusive Language Online

Abstract

ISBN 978-1-945626-66-1 iii IntroductionWe are very pleased to welcome you to the first Workshop on Abusive Language Online (ALW), held at ACL 2017 in Vancouver, Canada. The last few years have seen a surge in abusive behavior online, with governments, social media platforms, and individuals struggling to cope with the consequences and to produce effective methods to combat it. In many cases, online forums, comment sections, and social media interactions have become sites for bullying, scapegoating, and hate speech. These forms of online aggression not only poison the social climate of the online communities that experience it, but can also provoke physical violence and harm.Addressing abusive language necessitates a multidisciplinary approach that requires knowledge from several fields, including, but not limited to: media studies, natural language processing (NLP), psychology, sociology, law, gender studies, communications, and critical race theory. NLP, as a field that directly works with computationally analyzing language, is in a unique position to develop automated methods to analyse, detect, and filter abusive language. By working across disciplinary divides, researchers in all these fields can produce a comprehensive approach to abusive language that blends together computational, social and legal methods.We are therefore very happy to bring researchers of various disciplines together in this one-day workshop to discuss approaches to abusive language. The workshop consists of two invited speaker talks, two panels, and oral and poster presentations. • Carol ToddCarol Todd founded the Amanda Todd Legacy in memory of her daughter Amanda after her death by suicide on October 10, 2012. Amanda's Legacy was created to bring increased awareness and conversations within families and communities about online exploitation, cyberabuse and internet safety. The goal has been to encourage a shift in thinking about bullying type behaviours (both on and offline) to those of REFLECTION and RESPECT as well as to destigmatizing the perceptions related mental health as it can relate to how we treat others.• Brianna Wu Brianna Wu is a 2018 Democratic candidate for U.S. Congress in Massachusetts-District 8. Brianna is also head of development at GSX, a Boston independent videogame studio. Brianna came to national prominence when she and other women working in the tech industry were personally targeted by alt-right hate groups, including one spearheaded by Steve Bannon, now Chief Strategist to Donald Trump. Despite threats on her life and her family, Brianna has never wavered as a voice for the marginalized, including women, people of color and LGBT individuals.We will be hosting the following researchers as our panelists: • Lucas DixonLucas Dixon is Chief Scientist at Jigsaw, an incubator within Alphabet that builds technology to tackle iv some of the toughest global security challenges facing the world today. His work focuses on security, machine intelligence and data visualization.. • Pascale FungPascale Fung is a Professor a...

Studying Generalisability across Abusive Language Detection Datasets

“…The Offensive Language Identification Dataset, OLID (Zampieri et al, 2019a) was used in SemEval-2019 Task 6: 'OffensEval' (Zampieri et al, 2019b). It consists of 14, 100 tweets annotated through a unique hierarchical model whose basic idea was proposed by Waseem et al (2017b). For the shared task, the data was split into (nonstratified) training and test sets containing 13, 240 and 860 tweets, respectively.…”

“…It has been argued by some (Schmidt and Wiegland, 2017;Waseem et al, 2017b) that due to this phenomenon where works tackle restricted subsets of abusive language, it has become difficult to make judgements about whether the features being used can perform well in other subtasks of abusive language detection -as they are often only evaluated on a single dataset, specific to one domain and subtask, and annotated in a specific way. Waseem et al (2017b) proposed that there exists an overlap between these subtasks and subsequently proposed a typology that emphasises identifying the target of abuse and whether the abuse is implicit or explicit. Their typology could potentially be applied to all stages of system development, from data collection to the final model building.…”

“…It also makes some qualitative assessments on why models trained on specific datasets generalise better than others. Additionally, the OLID dataset based on the typology by Waseem et al (2017b) is used to observe whether the hierarchical annotation model is sufficient to synthesise the various subtasks of abusive language detection. To this end, experiments were run using BERT, Bidirectional Encoder Representations from Transformers (Devlin et al, 2018), to compare its performance to other popular models that have been used for abusive language detection (Section 6).…”