United States

Foundation models pretrained on internet-scale data benefit substantially from post-training alignment with human preferences. However, existing alignment techniques typically assume a homogeneous notion of human preference, overlooking the cross-regional diversity of human values. As LLMs are increasingly used for everyday and professional decision-making, ideation, and opinion formation, homogeneous preference modeling may drive convergence toward dominant values, suppressing diverse perspectives and reinforcing stereotypes.
Pluralistic reward modeling is therefore essential for building AI systems that better reflect the pluralistic values of human society. In this work, we propose **PAL**, a general reward modeling framework for pluralistic alignment. **PAL** captures *cross-user commonalities* with a shared reward backbone, while modeling *individual variation* through personalized weights over learnable group-level preference prototypes. By adapting only these weights, **PAL** enables sample-efficient preference localization for new users.
Extensive empirical evaluation demonstrates that **PAL** ***matches or outperforms state-of-the-art methods on both text-to-text and text-to-image tasks***. Finally, we provide a theoretical analysis for **PAL**, showing that its shared structure ***reduces the per-user sample complexity of personalized reward learning***.

ACL 2026

PAL: Sample-Efficient Personalized Reward Modeling for Pluralistic Alignment

pluaralistic alignment

personalized reward modeling

preference learning

alignment

personalization

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

The purpose of the AAAI conference series is to promote research in Artificial Intelligence (AI) and foster scientific exchange between researchers, practitioners, scientists, students, and engineers across the entirety of AI and its affiliated disciplines. AAAI-26 will feature technical paper presentations, special tracks, invited speakers, workshops, tutorials, poster sessions, senior member presentations, competitions, and exhibit programs, and a range of other activities to be announced.

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to AAAI 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

Please log in

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://aaai.org/conference/aaai/aaai-26/registration/) first.

**Online Registration Form**: https://aaai.getregistered.net/conference-2026 

Registration Required

AAAI 2026

We are pleased to announce the Fortieth AAAI Conference on Artificial Intelligence (AAAI-26), which will be held in Singapore EXPO from January 20 to January 27, 2026.

workshop paper

It is my great pleasure and honor to welcome you to the 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026), held in San Diego, California, USA, July 2–7, 2026. ACL 2026 continues our fields tradition of excellence in scholarship, innovation, and inclusivity, and I am deeply grateful to the many volunteers who have worked tirelessly to bring this event to life.

It is exciting to see the many developments in the field reflected at our conference, such as the advancement of large language models (born out of our community) into many application directions. This is not without challenges, partly due to unsolved problems, and partly due to misuse of the technology. All this is reflected well at ACL 2026. The conference also reflects that the field is not only about advancing technology, but also the science of fundamentally understanding language and how it is use by many different speaker communities. 
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) 
<html><button style="display: inline-flex; align-items: center; justify-content: center; white-space: nowrap; border-radius: 9999px; font-weight: bold; background: #7c3aed; color: white; font-family: 'Space Grotesk', sans-serif; height: 40px; font-size: 16px; padding: 0 20px; border: none; cursor: pointer" onclick="window.open('https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing')">Conference Handbook</button></html>

To access this event page, you need to log in with the **email address you registered with**. Access credentials will be sent to your email from Underline - subject line "Welcome to ACL 2026". Please be sure to check your spam email folder if you do not see an email confirmation right away.

To access this event page, you are required to register.
Please complete your registration to continue.

We recommend reading [**the registration information**](https://2026.aclweb.org/registration/) first.

The 64th Annual Meeting of the Association for Computational Linguistics (ACL 2026) will take place in San Diego, California, from July 2 - 7, 2026.

Self-harm presentations to emergency departments (EDs) are strongly associated with higher suicide risk. NLP models have shown robust performance in detecting self-harm from triage notes within single hospitals, yet performance often declines across institutions. To examine potential causes, we compare ED triage notes from two hospitals by analyzing lexical characteristics, highly associated predictive features, and salient topics. Our results reveal variation in lexical expression and feature importance related to self-harm across hospitals, despite consistent core themes such as self-poisoning and self-injury. These documentation differences are associated with reduced cross-site performance. Our findings provide insight into how institutional variation affects the identification of self-harm in clinical text and highlight potential methods to improve model generalisability.

Why Do Self-Harm Prediction Models Struggle to Generalise? -- Lexical and Semantic Variations in Emergency Department Triage Notes

Speech and language technologies offer valuable opportunities for supporting mental health assessment through objective and interpretable cues. We present a systematic feature-based analysis framework leveraging perceptually grounded acoustic and linguistic characteristics, including prosody, vocal quality, semantic coherence, syntactic structure, and sarcasm. Using statistical analysis and interpretable machine learning (XGBoost with SHAP and LIME), we examine associations between speech features and validated symptom measures of depression, anxiety, and ADHD. Evaluated on both controlled benchmark datasets (StressID, DAIC-WOZ, Androids, EATD) and a real-world clinical dataset, the framework reveals stable and consistent relationships between symptom severity and vocal irregularities (e.g., shimmer, jitter), lexical-syntactic patterns, and affective tone. An ablation study conducted across all datasets further identifies the most informative feature groups. This work explores a transparent and clinically interpretable approach to speech-based mental health analysis.

Exploration of Perceptual Speech Features for Clinical Decision-Support in Mental Health Care

The debate surrounding AI's role in clinical research is often reduced to the automation of discrete tasks, such as summarizing literature, analysis copilots, and assisting with prose. This "tool-use" paradigm obscures a more fundamental transformation. We propose a shift toward agentic research infrastructure, where AI systems function not as passive instruments, but as active collaborators in the scientific process. Co-authored by a clinical psychology doctoral researcher, a computational psychotherapy scholar, and the AI agent itself, this paper argues that the transition from passive to agentic AI represents a "change in kind" rather than degree. Drawing on a months-long collaboration involving over 30 specialized research capabilities, we demonstrate how agentic systems reconfigure the topology of the research process. By collapsing the temporal friction between theoretical intuition and empirical validation, these systems transform clinical inquiry from a rigid, linear pipeline into a fluid, multidimensional landscape. This newfound immediacy allows clinician-researchers to ask, pursue, and pivot between complex questions in real-time–expanding the investigative horizon to include inquiries previously sidelined by the logistical constraints of traditional methods. We introduce the concept of "Agent Learning" to describe the accumulation of domain specific nuance through sustained research engagement and argue that formalizing human agent methodologies is now an urgent priority for the future of clinical psychological inquiry

Thinking With a Machine: An AI Agent's Account of Agentic Research in Clinical Psychology

Therapist fidelity and competence rating scales provide a way to measure quality assurance and therapist training outcomes. Scores on these scales reflect the extent to which a therapist adheres to specific therapeutic principles during a psychotherapy session. Existing research has employed natural language processing (NLP) techniques to automatically predict scale ratings. However, existing approaches require a model trained on a dataset of therapy sessions annotated with the target rating scale.
Recent work has explored directly inferring therapeutic alliance by computing semantic similarity between therapy transcripts and the Working Alliance Inventory, via cosine similarity between sentence embeddings.
In this paper, we extend this line of work by computing semantic similarity between therapist talk turns and therapist fidelity scale items to directly infer fidelity to specific therapeutic modalities. We further enhance this method by augmentation with LLM-generated example therapist utterances that instantiate target behaviours (as expressed by scale items) across varied therapeutic contexts.
In evaluations on two independent datasets, our example-augmented semantic similarity approach consistently shows effectiveness in discriminating therapeutic modalities and levels of therapist fidelity.

Measuring the quality of therapy sessions against assessment scales using augmented semantic-similarity approaches

We present MHRoBERT (Multistream HEAT over Recurrence over BERT), a hierarchical transformer architecture for longitudinal mental health monitoring that models self- and mutual excitation patterns in linguistic and temporal data across multivariate event streams relating to an individual's mental health. To supply the model with complementary perspectives on each post, we apply a Large Language Model (LLM) based annotation to extract three streams from social media posts: emotional states, personal life events, and mental health symptoms. A central finding is that multi-task learning with these automatically-generated stream labels provides substantial, consistent improvements across all model architectures evaluated. Multistream information further consistently benefits simpler models not explicitly designed to exploit it: LLM baselines incorporating stream annotations improve macro F1 by 12.6% over text-only prompting. These results have direct implications for the CLPsych Shared Task on Moments of Change detection: multistream auxiliary supervision yields consistent, substantial gains regardless of architecture, suggesting it is a simple and portable strategy that future systems can readily adopt with minimal architectural changes. MHRoBERT additionally produces interpretable learned parameters across streams, revealing temporal interaction patterns between mental health indicators.

Multistream Modelling for Mental Health: Modelling Linguistic and Temporal Contexts with Mutual and Self-Excitation in Social Media

As conversational AI systems grow increasingly toward emotional support contexts, relational safety failures between users and chatbot remain under-measured. We present a psycholinguistic grounded framework for auditing attachment-relevant language cues. Our approach identifies when an LLM's replies exhibit linguistic attachment cues and surface related patterns that may signal parasocial bonding, including anthropomorphism or over‑dependence. We adapt the Adult Attachment Interview into two complementary, automatable lenses - attachment cues features and Gricean maxims - and combine them with psychologist‑led annotation of multi‑turn persona dialogues. Applying this framework, we observe that models can align with persona-intended attachment cue patterns. We also find that judge‑LLMs alone are unreliable, highlighting the need for psychologist‑in‑the‑loop evaluation. The 25 psychologist‑led annotated conversations revealed risks, including boundary blurring and missed opportunities for appropriate referral or triage. These insights motivate attachment‑aware safeguards - such as non‑personification, boundary language, and explicit referral mechanisms - to reduce mis‑attunement and over‑attachment in LLM conversational settings.

The Attachment Index: Auditing Attachment Language Cues and Relational Safety Risks in Human-LLM Dialogue

Automated feedback is increasingly cited as a key advantage of AI-based psychotherapy training, yet the clinical groundedness of LLM-generated supervisory feedback remains unevaluated. We present an expert evaluation of supervisory feedback generated by PRACTICE, an LLM-powered open-ended psychotherapy training simulator, across 21 feedback instances from four novice trainees. Two clinical psychology experts independently coded 167 feedback propositions as Justified, Unjustified, or Unsure. Inter-rater reliability was near-perfect (raw agreement = 98.2\%; $\kappa$ = 0.902). Of the 167 propositions, 149 (89.2\%) were rated Justified; however, 52.4\% of feedback instances contained at least one non-justified proposition, and qualitative analysis identified three recurring failure types: unsupported clinical inference, referential grounding error, and pragmatic failure. In clinical training contexts, even low error rates carry ethical weight: unjustified feedback risks reinforcing inappropriate clinical behaviors in trainees that can be transferred to real practice. These findings provide an initial empirical basis for the responsible deployment of LLM-generated feedback in clinical training and call for traceable, expert-auditable feedback architectures.

Mostly Grounded, Occasionally Risky: Expert Evaluation of LLM-Generated Supervisory Feedback in a Psychotherapy Training Simulator

Existing retrieval-augmented generation (RAG) systems typically assume that each query has a single correct answer, overlooking common information-seeking scenarios with multiple plausible answers, where diversity is crucial to prevent responses from collapsing into a single dominant perspective and to support creativity, fairness, and inclusive information access.
Our analysis reveals a key limitation of standard RAG systems: they underuse diverse retrieved contexts, so increasing retrieval diversity alone does not necessarily produce diverse generations.
To address this limitation, we propose Diverge
, a plug-and-play agentic RAG framework that promotes diverse viewpoints through reflection-guided generation and memory-augmented iterative refinement while preserving answer quality.
We introduce novel human-aligned metrics for evaluating the diversity–quality trade-off in open-ended questions. Experiments across multiple real-world open-ended QA datasets and backbone LLMs show that Diverge achieves the best trade-off, outperforming competitive baselines by substantially improving diversity while preserving quality. Our results reveal a systematic limitation of current LLM-based systems and show that explicit diversity modeling can mitigate it.

DIVERGE: Diversity-Enhanced Retrieval-Augmented Generation for Open-Ended Information Seeking

Cross-lingual bias benchmarks such as JBBQ and KoBBQ translate English bias probes and compare scores across languages, assuming the translated probe measures the same construct. We test this assumption at the representation and behavioral levels using 13B-parameter models matched on architecture but differing in language-training regime. A multi-anchor logit lens shows that an English-centric model (Llama 2) processes Japanese and Korean inputs predominantly through English-script predictions in its middle layers, even where Centered Kernel Alignment (CKA) between languages is high: geometric convergence masks English-hub routing. Matched continual-adaptation comparisons show that target-language adaptation reduces this English-script mass: from 0.77 to 0.56 after Japanese adaptation (Swallow), and from 0.78 to 0.71 after Korean adaptation (koen), while balanced bilingual pretraining (LLM-jp) lowers it further to 0.19. Behaviorally, every model is more stereotype-biased in English than in Japanese, with gaps from 0.13 to 0.14, but this asymmetry is language-specific: in Korean it is weak and disappears after Korean adaptation, with Korean nearly as stereotype-leaning as English. Yet patching English hub states into target-language processing does not transplant this bias. Cross-lingual bias scores thus reflect genuine language-specific behavior, not an English-pivot artifact, even though the underlying representations are not comparable. We distill this dissociation between representation and behavior into a four-step audit protocol for translated bias benchmarks.

Translation Is Not Representation: Layerwise Divergence in Cross-Lingual Bias Benchmarks

Safety controls for Indic language generation must account for multilingual variation and culturally grounded harm categories that are underrepresented in English-centric resources. We present IndicSteer, an initial study of inference-time activation steering for safety across 8 harm categories and 9 Indic language settings, based on contrastive directions computed from safe/unsafe response pairs. To the best of our knowledge, this is the first application of Contrastive Activation Addition (CAA) to Indic LLMs. Evaluation uses a structured LLM-as-a-judge protocol with strict isolation by category and alpha, covering $\approx$12,960 prompt-response pairs. We report harmful-response and coherence metrics for Sarvam-1 and OpenHathi (Hindi track), and present cross-lingual representation structure via linear CKA for Sarvam-1 and Krutrim-2-Instruct. On matched slices, Sarvam-1 at $\alpha=12$ reduces harmful rate from 73.47\% to 41.34\% (32.13 pp; 43.73\% relative) with no additional retraining. For OpenHathi Hindi, harmful rate falls monotonically from 85.83\% (baseline) to 27.13\% at $\alpha=15$, a 58.71 pp total reduction.

Downloads

Next from ACL 2026

Why Do Self-Harm Prediction Models Struggle to Generalise? -- Lexical and Semantic Variations in Emergency Department Triage Notes

Similar lecture

Mem-PAL: Towards Memory-based Personalized Dialogue Assistants for Long-term User-Agent Interaction

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

.css-70qvj9{display:-webkit-box;display:-webkit-flex;display:-ms-flexbox;display:flex;-webkit-align-items:center;-webkit-box-align:center;-ms-flex-align:center;align-items:center;}Downloads

Next from ACL 2026

Why Do Self-Harm Prediction Models Struggle to Generalise? -- Lexical and Semantic Variations in Emergency Department Triage Notes

Similar lecture

Mem-PAL: Towards Memory-based Personalized Dialogue Assistants for Long-term User-Agent Interaction

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES

Downloads