Abstract
By August 2025, only 36% of U.S. adults reported regularly following the news, down from 51% in 2016. Although older adults are still more likely than younger people to keep up with current events, engagement has fallen across every age group by roughly 15-20% over that period, suggesting a concerning falloff in consistent news consumption. A potential part of this issue is the lack of highly-stimulating engagement in news articles, in comparison to social media apps. To this point, more direct forms of information delivery show potential to bring news to these critically lacking populations. Users aged 18-29 are more likely to get news from social media than any other age group, at 76%. The technical project addresses this opening through an SMS news delivery system, sending short, readable summaries and links to articles with personalization by topic and frequency, as well as AI interaction via SMS to ask questions about the topic. It is important to consider the human and social dimensions of this system because news delivery tools shape what people see, when they see it, and what feels “important”, which can influence civic understanding as much as it improves convenience. Being trusted to deliver news on a broad subject to people means that it is the product’s responsibility to avoid certain types of bias, and such consideration should be put into the research and synthesis algorithms. These choices also affect trust and inclusion, especially when “accessibility” depends on whether information is legible to people across different language backgrounds and levels of institutional access.
The research project analyzes translation technology through the concepts of techno-linguistic bias and epistemic injustice. Techno-linguistic bias refers to the structural tendency of language technologies to favor certain languages and ways of speaking, often because dominant languages are treated as defaults in data collection, modeling, and evaluation. Epistemic injustice occurs when these systems distort, undervalue, or erase the knowledge encoded in less-resourced languages, e.g. by mistranslating culturally central concepts or flattening distinctions that matter within community contexts. In this framing, translation errors can become harms to knowledge when they force minoritized-language meanings into dominant-language categories. To conduct the STS research, I used a literature review of conceptual and case-based work on language technology, data documentation, and revitalization, paired with analysis of case-based research on Indigenous and low-resource language technology projects. Analysis was guided by Helm et al.’s framework for identifying techno-linguistic bias and epistemic injustice, alongside Bird’s decolonial perspective on language technology. The research also drew on work about data statements, language revitalization, actual use patterns in Tetun translation, ethical concerns in Indigenous MT, and ethical competence in translation technology education. The STS research found that the central problem is the conditions under which MT systems for small and endangered languages are built, documented, and governed, since “better performance” can still reproduce linguistic hierarchies and harms when dominant-language defaults are treated as the baseline. The research identified recurring risks around benchmark mismatch, weak documentation, limited community authority, extractive data practices, and tools that are disconnected from revitalization goals. Translation technology for such languages can only be meaningfully “accessible” when its data practices, documentation, and governance are designed to avoid techno-linguistic bias and epistemic injustice by supporting data sovereignty, faithful representation, and revitalization goals. Considered together, the technical and research projects show that increasing news consumption is not only a question of reach, but of how practicality and responsible curation can be balanced as information is delivered to minority audiences. On the technical side, the SMS-based system aims to make regular news engagement more feasible by lowering barriers to access through short, consistent, and direct summaries, addressing the falloff of news consumption in young adults. At the same time, the STS research treats “accessibility” as more than availability, considering how systems quietly privilege dominant languages and make minoritized knowledge less visible or credible through defaults, benchmarks, and extractive data practices. If the capstone system relies on summarization or translation to broaden reach, this research motivates design and governance requirements such as transparent documentation, community-aligned data use, and avoiding one-size-fits-all assumptions about what language is supposed to look like.
