arXiv subjects

Anika Sharma

Publications and source records attributed to Anika Sharma.

4 recordsLinked to original sources

arXiv2026

CARE: A Capability-Based Measurement Framework for Reproductive Equity in Human-AI Interaction

Algorithmic systems mediate sexual and reproductive health (SRH) information seeking. Standard HCI and AI evaluation centers usability, accuracy, and interaction quality, measures designed to assess task performance and interaction quality at the system level. We introduce CARE, the Capability Approach for Reproductive Equity, a measurement framework for human-AI interaction that adds capability outcomes as a unit of evaluation above task performance. CARE functions in two parts. The Normative Design Lens identifies the resources, conversion factors, capabilities, and functionings a system should support. The Evaluation lens assesses how design features, interaction patterns, and social conditions shape capability outcomes, tradeoffs, and lived experiences in use. We apply CARE to SRH-specific chatbots, general-purpose LLMs, and search engine features in a study with 12 participants, demonstrating that it surfaces capability outcomes standard metrics aggregate away. The same design features expanded capabilities for some users while constraining them for others: source-level organization, response format, tone, and SRH-specific features all shaped which capabilities expanded for which users and in which direction. Participants' professional backgrounds, gender identities, and prior AI familiarity further shaped these effects, producing capability outcomes that usability and accuracy metrics, aggregated across users, would not surface. These findings demonstrate capability outcomes as a measurable unit for human-AI interaction evaluation, extending existing metrics with a capability layer above task performance.

cs.HC↗

arXiv2025

Can LLMs Understand What We Cannot Say? Measuring Multilevel Alignment Through Abortion Stigma Across Cognitive, Interpersonal, and Structural Levels

As Large Language Models (LLMs) increasingly mediate stigmatized health decisions, their capacity to understand complex psychological phenomena remains inadequately assessed. Can LLMs understand what we cannot say? We investigate whether LLMs coherently represent abortion stigma across cognitive, interpersonal, and structural levels. We systematically tested 627 demographically diverse personas across five leading LLMs using the validated Individual Level Abortion Stigma Scale (ILAS), examining representation at cognitive (self-judgment), interpersonal (worries about judgment and isolation), and structural (community condemnation and disclosure patterns) levels. Models fail tests of genuine understanding across all dimensions. They underestimate cognitive stigma while overestimating interpersonal stigma, introduce demographic biases assigning higher stigma to younger, less educated, and non-White personas, and treat secrecy as universal despite 36% of humans reporting openness. Most critically, models produce internal contradictions: they overestimate isolation yet predict isolated individuals are less secretive, revealing incoherent representations. These patterns show current alignment approaches ensure appropriate language but not coherent understanding across levels. This work provides empirical evidence that LLMs lack coherent understanding of psychological constructs operating across multiple dimensions. AI safety in high-stakes contexts demands new approaches to design (multilevel coherence), evaluation (continuous auditing), governance and regulation (mandatory audits, accountability, deployment restrictions), and AI literacy in domains where understanding what people cannot say determines whether support helps or harms.

cs.AI↗

arXiv2025

Adversarially Probing Cross-Family Sound Symbolism in 27 Languages

The phenomenon of sound symbolism, the non-arbitrary mapping between word sounds and meanings, has long been demonstrated through anecdotal experiments like Bouba Kiki, but rarely tested at scale. We present the first computational cross-linguistic analysis of sound symbolism in the semantic domain of size. We compile a typologically broad dataset of 810 adjectives (27 languages, 30 words each), each phonemically transcribed and validated with native-speaker audio. Using interpretable classifiers over bag-of-segment features, we find that phonological form predicts size semantics above chance even across unrelated languages, with both vowels and consonants contributing. To probe universality beyond genealogy, we train an adversarial scrubber that suppresses language identity while preserving size signal (also at family granularity). Language prediction averaged across languages and settings falls below chance while size prediction remains significantly above chance, indicating cross-family sound-symbolic bias. We release data, code, and diagnostic tools for future large-scale studies of iconicity.

cs.CL↗

arXiv2017

A Graph Theoretic Analysis of Leverage Centrality

In 2010, Joyce et. al defined the leverage centrality of vertices in a graph as a means to analyze functional connections within the human brain. In this metric a degree of a vertex is compared to the degrees of all it neighbors. We investigate this property from a mathematical perspective. We first outline some of the basic properties and then compute leverage centralities of vertices in different families of graphs. In particular, we show there is a surprising connection between the number of distinct leverage centralities in the Cartesian product of paths and the triangle numbers.

math.CO↗