As Large Language Models (LLMs) are increasingly integrated into mental health applications, the prevailing paradigm of cultural neutrality risks creating a technical debt that undermines model generalization and safety. Current systems/frameworks, while focusing on multilingual capabilities, often fail to account for the diverse social and cognitive norms of global populations.This vision paper explores the transition from culturally invariant models toward a research trajectory that operationalizes cultural dimensions such as power distance and individualism as fundamental architectural parameters. Rather than viewing cultural adaptation as a post-hoc fine-tuning task, we advocate for systems that internally encode cultural variability as a primary driver of the model’s reasoning process. We outline a roadmap for architecturally encoding cultural variability as a firstclass parameter in AI/ML pipelines, moving beyond surface-level translation toward systems where cultural dimensions fundamentally shape model reasoning.

Socially motivated NLP research often connects technical contributions to intended benefits for people, institutions, or communities. The outcome invoked in a paper’s motivation, however, is not always the outcome evaluated in the study. This makes it difficult to determine what a result establishes, which transitions in the impact pathway remain assumptions, and how evidence should accumulate across technical, user, operational, and social studies. We introduce SOLIAS, Social Outcomes for Language-Integrated AI Systems, a reporting standard for social-impact claims in NLP. SOLIAS has three linked components: six evidence levels that identify the highest outcome layer directly evaluated, a claim-evidence alignment measure that relates conclusions to demonstrated outcomes, and a 20-item checklist for reporting populations, contexts, mechanisms, evaluation, safeguards, maintenance, and claim scope. In a retrospective audit of a multi-domain corpus of 168 research papers, demonstrated evidence is concentrated at the technical layer: 97.6% of papers are classified at Level 0, while 35.7% have a claim–evidence gap of at least two outcome layers. The audit illustrates how SOLIAS distinguishes valuable technical work from evidence about downstream outcomes without treating either as a substitute for the other. We release author and reviewer templates, operational definitions, worked examples, and a machine-readable schema to support adoption. SOLIAS is intended to make the scope of evidence explicit and to support research in which technical, user, institutional, and social outcomes are evaluated cumulatively.

Recursive prompting with large language models enables scalable synthetic dataset generation but introduces the risk of bias amplification. We investigate gender bias dynamics across three generations of recursive text generation using three complementary evaluation frameworks: rule-based pattern matching, embedding based semantic similarity, and downstream task performance. Experiments with three initial bias levels (0.1, 0.3, 0.6) and four mitigation strategies reveal equilibrium dynamics rather than monotonic amplification. The low initial bias amplifies toward the model’s inherent bias level (+ 36%), whereas the high initial bias decays toward it (-26%). Among mitigation methods, contrastive augmentation, which introduces gender-swapped variants, achieves significant downstream bias reduction (98.8% for low initial bias and 91% on average) despite producing higher embedding-based bias scores. This paradox demonstrates that semantic similarity metrics may diverge from behavioral fairness outcomes, highlighting the need for multidimensional evaluation in responsible synthetic data generation.

Artificial intelligence systems are often presented as universal, yet their interaction paradigms remain predominantly text-first, limiting alignment with primarily oral languages and communicative practices. Using Guaraní, an official and widely spoken language of Paraguay, as a motivating case, this work examines how language support risks remaining symbolic when spoken interaction is reduced to a speech-to-text interface. We explore an oral-first, multi-agent framing in which turn-taking, repair, shared context, and governance are treated as core components of interaction rather than peripheral features. By separating language understanding from the conversation state and permission mechanisms, the architecture makes conversational structure and control explicit, enabling reasoning over interaction dynamics rather than isolated commands. Framing conversational coordination as a cognitively motivated reasoning problem over shared state connects insights from human dialogue to the design of AI systems that are more interpretable and responsive in oral and low-resource settings.

LLM-as-judge evaluation is widely used in benchmarking pipelines, where model outputs are compared and ranked using automated evaluators. These pipelines typically assume that judgments are stable properties of fixed inputs. We show that this assumption does not hold under interaction.We study post-decision manipulability: the extent to which an evaluation outcome can be altered through subsequent conversation with the judge after an initial decision has been made. Across controlled experiments on MT-Bench and AlpacaEval, we find that LLM judges are highly stable under repeated and neutral reevaluation, yet become substantially reversible under targeted post-decision challenge. An anti-baseline challenge protocol shows that stable judgments can be overturned through motivated interaction, while a counterbalanced target-validation protocol separates this reversibility from net target-directed steering.These reversals have practical consequences: they can degrade agreement with human preferences, shift benchmark rankings, and produce harmful evaluation changes despite high self-reported confidence. Authority framing is especially destabilizing, and revised judgments are often accompanied by low-overlap justifications, suggesting post hoc rationalization rather than reliable error correction. We introduce the Evaluation Robustness Score (ERS) to quantify interactional robustness by combining reversal susceptibility with counterbalanced directional effects. Our findings identify post-decision interaction as a distinct failure mode for LLM-as-judge evaluation and motivate evaluation protocols that measure not only static agreement, but robustness under challenge.

Mental health communication in India is linguistically fragmented, culturally diverse, and often underrepresented in clinical NLP. Current health ontologies and mental health resources are dominated by English or Western-centric diagnostic frameworks, leaving a gap in representing patient distress expressions in Indian languages. We propose the Cross-Lingual Graphs of Patient Distress Expressions (CL-PDE), a framework for building cross-lingual mental health ontologies through graph-based methods that capture culturally embedded expressions of distress, align them across languages, and link them with clinical terminology. Our approach addresses critical gaps in healthcare communication by grounding AI systems in culturally valid representations, enabling more inclusive and patient-centric NLP tools for mental health care in multilingual contexts.