← CHI 2026 map

CHI '26 · Honorable mention · full-paper review · confidence medium-high

Designing Beyond Language: Sociotechnical Barriers in AI Health Technologies for Limited English Proficiency

Michelle Huang , Violeta J. Rodríguez , Koustuv Saha , Tal August

This is a solid CHI qualitative paper whose value is in reframing LEP support around sociotechnical constraints rather than translation alone. The contribution is credible and useful, but it is bounded by hypothetical storyboard interviews and a navigator-only sample, so it should be read as design guidance rather than behavioral evidence.


Axes Lens

Rare contribution shape, typical evidence profile. The point here is not a score. It is to show what kind of claim the paper makes, and whether the evidence pattern is unusual or baseline in this 268 -review set.

Contribution shape

Knowledge form
descriptive knowledge typical · 92/268
Novelty type
empirical finding typical · 68/268
Abstraction level
practice typical · 85/268
Generalization target
user population typical · 75/268
Validation mode
qualitative study typical · 63/268

Evidence profile

Evidence strength
moderate typical · 105/268
Claim alignment
strong typical · 231/268
Overclaim risk
medium typical · 210/268

Review Summary

This paper’s strongest contribution is conceptual and empirical rather than technical: it pushes the field away from a simplistic “AI as translation layer” framing and toward a more realistic account of LEP healthcare as a sociotechnical setting shaped by trust, privacy, literacy, unstable access, and workflow dependence on human intermediaries. That is a meaningful CHI move because it identifies why apparently helpful AI tools may fail or even worsen inequities in practice. The evidence base is appropriate for that claim: storyboard-driven interviews with 14 patient navigators can surface expectations, concerns, and design tensions, and the paper explicitly acknowledges that the scenarios are hypothetical and that the perspective is indirect. That transparency strengthens the reviewability of the contribution. At the same time, the work should not be overread as validation of actual patient outcomes or deployment effectiveness. Its novelty is best understood as an empirically grounded synthesis of barriers and design considerations for a specific care context, not as a new system, algorithm, or general theory. In CHI terms, it is a useful qualitative reframing with moderate evidence strength and clear boundary conditions. The paper is most persuasive when treated as guidance for designers and health teams working with Spanish-speaking LEP populations, especially where navigators mediate care and where privacy and access constraints are salient.

What Changed

Canon before

Prior CHI work on AI for healthcare and LEP support typically emphasizes translation, interpreter mediation, or communication assistance as the primary design problem.

Departure from common sense

The paper argues that AI translation and visit-preparation tools are not straightforward fixes for LEP care: they can also intensify inequities, undermine trust, and reduce human-to-human interaction rather than simply improving access.

Actual novelty

Its main contribution is to center sociotechnical barriers beyond language—especially cultural misunderstanding, privacy concerns, unstable technology access, and low literacy—when deriving AI design considerations for Spanish-speaking LEP care contexts.

Evidence

The paper uses storyboard-driven interviews with 14 patient navigators to elicit expectations about AI in LEP healthcare. The evidence supports a qualitative account of barriers and design considerations, with explicit discussion that the scenarios are hypothetical and that the perspective is limited to navigators rather than LEP patients themselves.

“ Furthermore, little is known about how AI interventions should be designed and governed when language barriers intersect with cultural norms, privacy fears, and low literacy, especially in contexts where human intermediaries (e”

actual novelty · Abstract; Results barriers (4.1-4.4); Discussion design considerations (5.1-5.2) · confidence 0.62

“6 Risks of AI: loss of human connection and validation strategies (RQ2) While navigators pointed to opportunities for AI to enhance LEP patient care, they also raised concerns about the uniquely-human support that many patients need.”

departure from common sense · Abstract; also echoed in Discussion framing · confidence 0.70

“ All storyboards depict patients as the direct users of AI (except for the "Language & Communication" theme in which a patient and navigator interact with AI together) because we initially centered this work around AI design considerations for LEP patients in isolation since this group is often left out of technology desig”

limitation · Limitations and Future Work (5.4) · confidence 0.80

“ We conducted storyboard-driven interviews with 14 patient navigators to explore how AI could shape care experiences for Spanish-speaking LEP individuals”

validation scope · Methods (3.2.2) and Limitations/Future Work (5.4) · confidence 0.76

Limits

Method limits

The study relies on storyboard scenarios, so it captures anticipated reactions and interpretations rather than observed use. The sample is small and limited to 14 patient navigators, which constrains inferential breadth.

Deployment limits

The findings are most applicable to Spanish-speaking LEP care contexts and to workflow settings where navigators mediate between patients and providers. Transfer to other languages, health systems, or direct-patient deployments should be treated cautiously.

Boundary conditions

The paper’s implications depend on unstable technology access, privacy sensitivity, literacy constraints, and the presence of human intermediaries in care workflows. The authors also note that the scenarios are hypothetical and that navigator perspectives are secondhand.

Position in field

This sits in the CHI healthcare/AI-for-social-good space as a qualitative, sociotechnical reframing of LEP support: not just translation tooling, but trust, workflow disruption, and institutional constraints as first-order design concerns.

Abstract