Interpersonal violence (IPV) is a prominent sociological problem that affects people of all demographic backgrounds. By analyzing how readers interpret, perceive, and react to experiences narrated in social media posts, we explore an understudied source for discourse about abuse. We asked readers to annotate Reddit posts about relationships with vs. without IPV for stakeholder roles and emotion, while measuring their galvanic skin response (GSR), pulse, and facial expression. We map annotations to coreference resolution output to obtain a labeled coreference chain for stakeholders in texts, and apply automated semantic role labeling for analyzing IPV discourse. Findings provide insights into how readers process roles and emotion in narratives. For example, abusers tend to be linked with violent actions and certain affect states. We train classifiers to predict stakeholder categories of coreference chains. We also find that subjects' GSR noticeably changed for IPV texts, suggesting that co-collected measurement-based data about annotators can be used to support text annotation.
more »
« less
Effects of contrastive intonation and grammatical aspect on processing coreference in Mainstream American English
Coreference choices are influenced by multiple factors, including information structural categories such as topic and focus. These information structural categories can be indicated by intonation, yet few studies have investigated how intonation affects subsequent choices for coreference. Using a story continuation experiment with aurally presented stimuli, we show that the location of contrastive focus in Mainstream American English significantly affects the preferred referent for the subject of the next sentence in a short discourse.
more »
« less
- Award ID(s):
- 1251450
- NSF-PAR ID:
- 10028987
- Date Published:
- Journal Name:
- Proceedings of the International Congress of Phonetic Sciences
- ISSN:
- 0301-3162
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Stereotypical character roles-also known as archetypes or dramatis personae-play an important function in narratives: they facilitate efficient communication with bundles of default characteristics and associations and ease understanding of those characters’ roles in the overall narrative. We present a fully unsupervised k-means clustering approach for learning stereotypical roles given only structural plot information. We demonstrate the technique on Vladimir Propp’s structural theory of Russian folktales (captured in the extended ProppLearner corpus, with 46 tales), showing that our approach can induce six out of seven of Propp’s dramatis personae with F1 measures of up to 0.70 (0.58 average), with an additional category for minor characters. We have explored various feature sets and variations of a cluster evaluation method. The best-performing feature set comprises plot functions, unigrams, tf-idf weights, and embeddings over coreference chain heads. Roles that are mentioned more often (Hero, Villain), or have clearly distinct plot patterns (Princess) are more strongly differentiated than less frequent or distinct roles (Dispatcher, Helper, Donor). Detailed error analysis suggests that the quality of the coreference chain and plot functions annotations are critical for this task. We provide all our data and code for reproducibility.more » « less
-
Introductory steel design courses focus on the analysis and design of primary members, which typically include tension members and connections, compression members, flexural members, and beam-columns. Introducing structural steel design concepts to students presents its fair share of challenges. First, it is difficult for students to visualize and accurately predict the potential failure modes of a tension member: yielding of the gross section, rupture of the net section, and block shear. Second, it is also difficult for students to visualize the buckling modes of steel columns, which vary with shape and type of bracing. Students particularly struggle with the determination of buckling modes between strong and weak axes based on effective lengths. Third, flexural failure modes of steel beams are very difficult for students to visualize and understand when each mode controls. The failure modes are complex and fall into three categories for compact shapes: yielding of the cross section, inelastic lateral torsional buckling, and elastic lateral torsional buckling, which is dependent on the unbraced length of the compression flange. Non-compact sections also include local buckling of the flange or web, but identifying the relationship between the unbraced length and beam span and how the unbraced length affects the flexural capacity tends to be the most difficult concept for students to grasp. This paper provides a detailed overview of the design, fabrication, and implementation of three large-scale experiential learning modules for an undergraduate steel design course. The first module focuses on the tension connections by providing physical models of various failure types including yielding of the gross section, rupture of the net section, and block shear; the second module focuses on the capacity of columns with different amounts of lateral bracing about the weak axis; and the third module focuses on the flexural strength of a beam with different unbraced lengths to illustrate the difference between lateral torsional buckling and flange local buckling/yielding of the gross section. The three modules were used throughout the steel design course at Saint Louis University and Rose-Hulman Institute of Technology to illustrate the failure mechanisms associated with the design of steel structures.more » « less
-
Although much literature has established the presence of demographic bias in natural language processing (NLP) models, most work relies on curated bias metrics that may not be reflective of real-world applications. At the same time, practitioners are increasingly using algorithmic tools in high-stakes settings, with particular recent interest in NLP. In this work, we focus on one such setting: child protective services (CPS). CPS workers often write copious free-form text notes about families they are working with, and CPS agencies are actively seeking to deploy NLP models to leverage these data. Given well-established racial bias in this setting, we investigate possible ways deployed NLP is liable to increase racial disparities. We specifically examine word statistics within notes and algorithmic fairness in risk prediction, coreference resolution, and named entity recognition (NER). We document consistent algorithmic unfairness in NER models, possible algorithmic unfairness in coreference resolution models, and little evidence of exacerbated racial bias in risk prediction. While there is existing pronounced criticism of risk prediction, our results expose previously undocumented risks of racial bias in realistic information extraction systems, highlighting potential concerns in deploying them, even though they may appear more benign. Our work serves as a rare realistic examination of NLP algorithmic fairness in a potential deployed setting and a timely investigation of a specific risk associated with deploying NLP in CPS settings.more » « less
-
The food-energy-water (FEW) nexus presents an opportunity to rethink predominant approaches to household behavior change science. We linked emerging FEW nexus research with existing literature examining household consumption and pro-environmental behaviors. While a large body of work examines the environmental impacts of household life and explores pathways to behavior change for sustainability, the literature lacks studies that test interventions in multiple FEW resource categories, leaving researchers unable to identify tensions and tradeoffs in the household system. To guide this developing field and accumulate findings on household behavior across disciplines, we proposed a conceptual typology that synthesizes interdisciplinary analytic traditions to classify behavioral interventions targeting the household FEW nexus. The typology synthesizes behavioral interventions as active, passive, or structural, and household-specific or non-specific, illustrating six distinct categories: information, tailored information, action, gamification, policy/price change, and material/technology provision. A review of 40 studies that guided the typology identifies four significant lessons for future intervention research: household non-specific information and tailored information work better together, feedback is more effective when it is persistent, price-based interventions (information or incentives) are often ineffective, and material/technology provision is very effective but utilized in few household studies. To push forward household resource consumption science, we advocated for a holistic nexus focus that is rooted in interdisciplinarity, coalition building with stakeholders, and data reporting that facilitates knowledge accumulation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
