Coreference resolution is an important compo-nent in analyzing narrative text from admin-istrative data (e.g., clinical or police sources).However, existing coreference models trainedon general language corpora suffer from poortransferability due to domain gaps, especiallywhen they are applied to gender-inclusive datawith lesbian, gay, bisexual, and transgender(LGBT) individuals. In this paper, we an-alyzed the challenges of coreference resolu-tion in an exemplary form of administrativetext written in English: violent death nar-ratives from the USA’s Centers for DiseaseControl’s (CDC) National Violent Death Re-porting System. We developed a set of dataaugmentation rules to improve model perfor-mance using a probabilistic data programmingframework. Experiments on narratives froman administrative database, as well as existinggender-inclusive coreference datasets, demon-strate the effectiveness of data augmentationin training coreference models that can betterhandle text data about LGBT individuals.

An abundance of methodological work aims to detect hateful and racist language in text. However, these tools are hampered by problems like low annotator agreement and remain largely disconnected from theoretical work on race and racism in the social sciences. Using annotations of 5188 tweets from 291 annotators, we investigate how annotator perceptions of racism in tweets vary by annotator racial identity and two text features of the tweets: relevant keywords and latent topics identified through structural topic modeling. We provide a descriptive summary of our data and estimate a series of generalized linear models to determine if annotator racial identity and our 12 latent topics, alone or in combination, explain the way racial sentiment was annotated, net of relevant annotator characteristics and tweet features. Our results show that White and non-White annotators exhibit significant differences in ratings when reading tweets with high prevalence of particular, racially-charged topics. We conclude by suggesting how future methodological work can draw on our results and further incorporate social science theory into analyses.

Computational models to detect mental illnesses from text and speech could enhance our understanding of mental health while offering opportunities for early detection and intervention. However, these models are often disconnected from the lived experience of depression and the larger diagnostic debates in mental health. This article investigates these disconnects, primarily focusing on the labels used to diagnose depression, how these labels are computationally represented, and the performance metrics used to evaluate computational models. We also consider how medical instruments used to measure depression, such as the Patient Health Questionnaire (PHQ), contribute to these disconnects. To illustrate our points, we incorporate mixed-methods analyses of 698 interviews on emotional health, which are coupled with self-report PHQ screens for depression. We propose possible strategies to bridge these gaps between modern psychiatric understandings of depression, lay experience of depression, and computational representation.