United States

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&#39;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.

ACL 2026

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

safety evaluation frameworks

safety auditing

attachment theory

chatbot auditing

llm auditing

ai emotional dependency risks

emotional safety in conversational chatbot

stages 

COLING, the International Conference on Computational Linguistics, is one of the premier conferences for natural language processing and computational linguistics. Often grouped within the field of artificial intelligence, but actually pre-dating the development of artificial intelligence, advances in computational linguistics and natural language processing are now some of the major drivers behind the use of artificial intelligence for commercial and social applications – for example, on-line search, machine translation and with voice-assisted conversational devices.

First established in 1965, the biennial COLING conference is held in diverse parts of the globe and attracts participants from both top-ranked research centers and emerging countries. Today, the most important developments in our field are taking place not only in universities and academic research institutes, but also in industrial research departments and in technological startups. COLING conferences provide opportunities for all these communities to showcase their exciting developments.

COLING 2020

COLING, the International Conference on Computational Linguistics, is one of the premier conferences for natural language processing and computational linguistics. Often grouped within the field of artificial intelligence, but actually pre-dating the development of artificial intelligence, advances in computational linguistics and natural language processing are now some of the major drivers behind the use of artificial intelligence for commercial and social applications – for example, on-line search, machine translation and with voice-assisted conversational devices.

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.

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. <br>
[Full Message from General Chair Philipp Koehn](https://drive.google.com/file/d/1QiJFOmMjvDgsDo3i1XzlQnImxe3RU_Cy/view?usp=sharing) <br><br><br>
<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>
<br><br><br><br>

To access this event page, you need to log in with the **email address you registered with**. <br>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.

Registration Required

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

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

AI systems for mental health are developed predominantly using data from Western, Educated, Industrialized, Rich, and Democratic (WEIRD) populations, raising concerns about their validity, fairness, and generalizability across diverse cultural and geographic contexts. This limitation is especially consequential in mental health, where linguistic expression, symptom presentation, help-seeking behavior, and access to care vary substantially across populations. We argue that culture must be treated as a first-class design requirement throughout the AI development lifecycle, not an afterthought applied post-hoc, a principle we term culture by design. Drawing on evidence from NLP, clinical psychology, HCI, and global mental health, we present ten practical recommendations spanning data collection, modeling, evaluation, deployment, and governance, providing researchers and practitioners with a concrete roadmap for building culturally grounded, equitable, and contextually appropriate mental health AI systems.

Culture by Design: A Sociotechnical Framework for Culturally Grounded AI for Mental Health

Globally, the incidence of depression and anxiety continues to rise, and the importance of mental health assessment scales as diagnostic tools has grown accordingly. Researchers are increasingly employing generative AI to produce large volumes of items and entire scales, which in turn elevates the costs of validating their reliability and validity. In this study, we used four open-weight LLMs to complete the GAD-7 and PHQ-9, varying prompts, sampling temperature, and dynamic contextual scenarios to emulate realistic human response patterns. Using multi-group confirmatory factor analysis, differential item functioning analyses, and other psychometric methods, we evaluate the factor structure of LLM-generated responses and assess measurement invariance relative to human responses. Our findings reveal a critical paradox: although open-weight LLMs exhibit exceptionally high internal consistency, they demonstrate severe structural mismatch and fail to achieve scalar measurement invariance against human baselines. Furthermore, pervasive differential item functioning and extreme prompt fragility indicate that these models rely on superficial, stereotype-driven semantic matching rather than simulating stable latent psychological dynamics.

The Reliability Illusion in Synthetic Patients: Psychometric Misalignment of Open-weight LLMs on PHQ-9 and GAD-7

AI and large language models (LLMs) have emerged as promising tools to address global mental health challenges. Despite the global nature of these challenges, there remains a critical shortage of high-quality datasets for training and evaluating such systems. To mitigate this gap, researchers increasingly generate synthetic clinical personas to simulate user data and test digital mental health support systems. However, most validated personas rely on English-centric contexts. This paper investigates whether similar persona-based methods can be used to generate multilingual mental health datasets. We modified nationality and language parameters in personas to generate clinical dialogues in Mandarin, Bengali, and Hindi. We then examined how different LLMs perform when evaluating the depression severity of these generated multilingual datasets against the baseline in English. Our findings indicate that just adding nationality and language parameters in personas might not be adequate, as it can introduce clinical inconsistency across languages. LLM judge models often exhibit inaccuracies in assessing depression severity in non-English texts, with performance varying across different models. This exposes the systemic limitations of applying English-centric personas to multilingual contexts. Ultimately, our work highlights the urgent need for culturally responsive data generation to ensure equitable mental health systems globally.

Creating Multilingual Mental Health Dialogue Datasets: Limits of Persona-Based Localization via Nationality and Language

Cognitive distortions (CDs) are systematically biased patterns of thinking associated with the onset and maintenance of mental health conditions such as depression and anxiety. Computational research on CDs has primarily focused on detection and classification, while the linguistic characterization of distorted language; what psycholinguistic features distinguish distorted from non-distorted text, and whether individual distortion types carry distinct language patterns, remains largely unexplored. Using a Reddit dataset, we apply a Generalized Linear Model (GLM) with bootstrap sampling to LIWC-derived features and find that CD language is psycholinguistically distinct from non-distorted language. We further characterize type-specific psycholinguistic profiles for each CD, and through hierarchical clustering show that CD types are not fully separable, with certain distortions sharing stable linguistic signatures. Together, these findings contribute to the linguistic characterization of CDs, offering an empirically grounded account of the psycholinguistic properties that distinguish distorted language at the level of CDs as a whole and across specific distortion types.

Psycholinguistic Profiles of Cognitive Distortions in Reddit Data

Speech monologues recorded in naturalistic settings provide opportunities to characterize mental illness phenomenology and detect symptom exacerbation. Large language models (LLMs) offer new possibilities for automating this process, as they require annotated data primarily for evaluation rather than training. In this paper, we present a novel automated, multi-agent LLM pipeline for the fine-grained, multi-label extraction of language suggestive of delusional beliefs, associated affective responses, and behavioral responses from transcripts of naturalistic audio diaries collected from people with moderate persecutory ideation. Evaluating an ensemble of three foundation models, we demonstrate that detailed diagnostic prompt instructions successfully reduce false positives for delusional theme classification, but also constrain the interpretation of affective or behavioral responses. Furthermore, comparing multi-agent adjudication frameworks shows that complex conversational debate between agents diminishes accuracy on clinically ambiguous text by inducing premature consensus. Instead, majority voting establishes robust performance (Micro F1 of 0.872 and 0.779 for delusion detection and classification respectively). This work provides a validated and scalable pipeline for the automated detection and characterization of content suggesting delusional beliefs in naturalistic speech.

Automated Detection and Classification of Delusion-related Content in Naturalistic Audio Diaries Using Multi-Agent Language Models

Large language models (LLMs) show promise in generating supportive responses for mental health and counseling applications. However, their responses often lack cultural sensitivity, contextual grounding, and clinically appropriate guidance. This work addresses the gap of how to systematically incorporate domain-specific, clinically validated knowledge into LLMs to improve counseling quality. We utilize and compare two approaches, retrieval-augmented generation (RAG) and a knowledge graph (KG)–based method, designed to support para-counselors. Our KG is constructed manually and clinically validated, capturing causal relationships between stressors, interventions, and outcomes, with contributions from multidisciplinary people. We evaluated multiple LLMs in both settings using BERTScore F1 and SBERT cosine similarity, as well as human evaluation across five metrics, which is designed to directly measure the effectiveness of counseling beyond similarity at the surface level. The results show that KG-based approaches consistently improve contextual relevance, clinical appropriateness, and practical usability compared to RAG alone, demonstrating that structured, expert-validated knowledge plays a critical role in addressing LLMs limitations in counseling tasks.

Enhancing Mental Health Counseling Support in Bangladesh using Culturally-grounded Knowledge

Attention Deficit Hyperactivity Disorder (ADHD) is one of the most common neurodevelopmental disorders in childhood, and its diagnosis relies on assessments combining clinician judgment with standardized rating scales and reports from parents and teachers. While structured instruments such as the Conners' Teacher Rating Scale–Revised Short Form (CTRS-R:S) quantify ADHD-related behaviors, teachers also provide open-ended narratives that may contain complementary signals not captured by structured assessments. However, it remains unclear to what extent teacher narratives encode signals overlooked by rating scales. In this study, we analyze de-identified Turkish teacher evaluation forms collected during clinical ADHD assessments, including both CTRS-R:S scores and open-ended teacher narratives. We compare predictive signals from structured scores and narrative text and identify cases where structured assessments fail to clearly distinguish ADHD from non-ADHD students while narrative-based models capture distinct behavioral patterns. Notably, these cases show minimal overlap with those missed by the narrative model, suggesting that structured and narrative information encode complementary signals. To interpret these differences, we apply a large language model (LLM)-assisted theme discovery pipeline that reveals distinct attention, behavioral, and family-related patterns, highlighting the potential of natural language processing (NLP) to uncover clinically relevant signals from teacher narratives and to complement traditional ADHD screening tools.

When Rating Scales Fall Short: LLM-Assisted Discovery of ADHD Signals in Turkish Teacher Narratives

Social media research on mental health has focused predominantly on detecting and diagnosing conditions at the individual level. In this work, we shift attention to intergroup behavior, examining how two prominent neurodivergent communities, ADHD and autism, adjust their language when engaging with each other on Reddit. Grounded in Communication Accommodation Theory (CAT), we first establish that each community maintains a distinct linguistic profile as measured by the Linguistic Inquiry and Word Count (LIWC) dictionary. We then show that these profiles shift in opposite directions when users cross community boundaries: features that are elevated in one group's home community decrease when its members post in the other group's space, and vice versa, consistent with convergent accommodation. We further demonstrate the practical stakes of this accommodation: classifiers trained to distinguish the two communities from their in-group language lose roughly 30 percentage points of accuracy, falling below chance, when applied to the same users' cross-community posts, confirming that accommodation, not just diagnosis, drives their predictions. Finally, in an exploratory longitudinal analysis around the moment of public diagnosis disclosure, we find that its effects on linguistic style are small and, in some cases, directionally opposite to cross-community accommodation, providing initial evidence that situational audience adaptation and longer-term identity processes may involve
different mechanisms. Our findings contribute to understanding intergroup communication dynamics among neurodivergent populations online and carry implications for community moderation and clinical perspectives on these conditions.

Mirroring Minds: Asymmetric Linguistic Accommodation and Diagnostic Identity in ADHD and Autism Reddit Communities

As Large Language Models (LLMs) demonstrate strong performance on clinical benchmarks, it remains unclear whether this reflects true patient-specific reasoning or reliance on generalized symptom patterns. To address this question, we evaluate LLMs on a counseling competency benchmark to assess their use of patient-specific contextual information. Through controlled experiments with contextual ablations, role framing, Thread-of-Thought (ThoT) prompting, and input perturbations, we find that removing contextual details results in only modest performance drops, and predictions remain relatively stable under input variations. Error analysis further reveals systematic patterns where models favor general clinical associations over context-specific cues, even when such cues are correctly identified during intermediate reasoning. Our findings suggest that achieving passing-level performance may not reflect robust context-sensitive decision-making, revealing a potential gap between benchmark-level clinical competence and reliable utilization of patient-specific contextual information. These results highlight the need for evaluation frameworks that more directly assess context integration in mental health applications.

Downloads

Next from ACL 2026

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

Similar lecture

Parsers Know Best: German PP Attachment Revisited

.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

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

Similar lecture

Parsers Know Best: German PP Attachment Revisited

Downloads