Unit Testing for Concepts in Neural Networks

Dealing with Disagreements: Looking Beyond the Majority Vote in Subjective Annotations Aida Mostafazadeh Davani University of Southern California, USA mostafaz@usc.edu Mark D´ıaz Google Research, USA markdiaz@google.com Vinodkumar Prabhakaran Google Research, USA vinodkpg@google.com Abstract Majority voting and averaging are common approaches used to resolve annotator dis- agreements and derive single ground truth labels from multiple annotations. However, annotators may systematically disagree with one another, often reflecting their individ- ual biases and values, especially in the case of subjective tasks such as detecting affect, aggression, and hate speech. Annotator dis- agreements may capture important nuances in such tasks that are often ignored while aggre- gating annotations to a single ground truth. In order to address this, we investigate the effi- cacy of multi-annotator models. In particular, our multi-task based approach treats predict- ing each annotators’ judgements as separate subtasks, while sharing a common learned representation of the task. We show that this approach yields same or better performance than aggregating labels in the data prior to training across seven different binary classifi- cation tasks. Our approach also provides a way to estimate uncertainty in predictions, which we demonstrate better correlate with annota- tion disagreements than traditional methods. Being able to model uncertainty is especially useful in deployment scenarios where knowing when not to make a prediction is important. 1 Introduction Obtaining multiple annotator judgements on the same data instances is a common practice in NLP in order to improve the quality of final labels (Snow et al., 2008; Nowak and R¨uger, 2010). In case of disagreements between annotations, they are often aggregated by majority voting, averaging (Sabou et al., 2014), or adjudicating by an ‘‘expert’’ (Waseem and Hovy, 2016), to derive a singular ground truth or gold label that is later used for training supervised machine learn- ing models. However, in many subjective tasks, there often exists no such single ‘‘right’’ answer (Alm, 2011) and enforcing a single ground truth sacrifices the valuable nuances embedded in an- notator’s assessments of the stimuli and their dis- agreements (Aroyo and Welty, 2013; Cheplygina and Pluim, 2018). Annotators’ sociodemographic factors, moral values, and lived experiences often influence their interpretations, especially in subjective tasks such as identifying political stances (Luo et al., 2020), sentiment (D´ıaz et al., 2018), and online abuse and hate speech (Cowan and Khatchadourian, 2003; Waseem, 2016; Patton et al., 2019). For instance, Waseem (2016) found that feminist and anti-racist activists systematically disagree with crowd work- ers on their hate speech annotations. Similarly, annotators’ political affiliation affects how they annotate the neutrality of political stances (Luo et al., 2020). An adverse effect of majority vote in such cases is limiting representation of mi- nority perspectives in data (Prabhakaran et al., 2021), potentially reinforcing societal disparities and harms. Another consequence of majority voting, when applied to subjective annotations, is that the re- sulting labels may not be internally consistent. For example, consider a scenario where a sentence in a hate-speech dataset is annotated by a set of an- notators, the majority of whom consider …

Content truncated to 500 words