arXiv subjects

Jinho D. Choi

Publications and source records attributed to Jinho D. Choi.

At least 19 recordsLinked to original sources

arXiv2026

Tinker Tales: A Tangible Dialogue System for Child-AI Co-Creative Storytelling

Conversational AI agents are increasingly explored as creative partners, yet how conversation design shapes child-AI dialogue in co-creative settings remains underexplored. We present Tinker Tales, a tangible dialogue system for child-AI collaborative storytelling, in which educational frameworks (narrative development and social-emotional learning) are instantiated as conversation design, shaping how the agent engages children across four narrative stages. The system combines a physical storytelling board, NFC-embedded toys, and a mobile app mediating multimodal interaction through tangible manipulation and voice-based dialogue. We conducted a home-based user study with 10 children (ages 6-8) across two conversation design conditions varying in how the agent structured elaboration, with and without educational scaffolding. Our findings show that prompt framing shapes the form and consistency of children's narrative contributions, structuring how they participate in co-creative dialogue with AI.

cs.HC↗

arXiv2026

Expert-Level Crisis Detection in Mental Health Conversations

Real-world crisis intervention is inherently conversational, yet existing research largely focuses on static texts. When applied to multi-turn dialogues, current models exhibit significant performance degradation, struggling to track risk signals that emerge as context evolves. To address this gap, we introduce CRADLE-Dialogue, a clinician-annotated benchmark for turn-level crisis detection in conversational settings. The dataset features 600 dialogues with multi-label annotations across clinically grounded risks, including suicide ideation, self-harm, and child abuse, distinguishing past from ongoing risk. We further propose an Alert-Confirm evaluation protocol that distinguishes early warning signals (Alert) from turns where a specific crisis becomes explicitly identifiable (Confirm), reflecting the clinical need to intervene before risk becomes explicit. Experiments show that identifying when risk emerges is much harder than recognizing that it exists: models achieve only mid-40% to high-60% Micro F1. Additionally, we release a synthetic training corpus and a 32B-parameter model that substantially outperforms existing open-source models and achieves competitive or superior results against proprietary models across turn-level, dialogue-level, and confirm-only evaluation settings.

cs.CL↗

arXiv2026

What Models Express, Suppress, and Resist: Auditing Open-Weight LLMs with Persona Vectors

What a language model will and will not do is largely set during post-training, but which behaviors it expresses, hides, or resists is not revealed by prompting alone. Persona vectors, behavioral directions in activation space, can probe this organization, but prior work covers only a handful of traits. We present the first systematic application of persona vectors at this scale, compiling a 53-trait inventory across four behaviorally distinct domains and labeling every trait in two open-weight models as natural (expressed at baseline), steerable latent but amplifiable, or intractable (resistant to standard extraction). Both models default to helpful, task-oriented behavior: all nine agentic traits are natural, and their default clinician behavior matches a board-certified psychologist's independent desirability judgments on 16 of 17 traits. Steering produces its largest gains on traits these defaults exclude: hyperbole, hallucination, and sycophancy. The same asymmetry holds across all 171 generic-trait pairs: two steerable traits can collapse the composition, but pairs involving a default never do. Where standard extraction fails on a trait like "evil," a vector transferred from a fine-tuned variant still recovers it, with the residual refusals appearing inside the model's chain-of-thought. Persona vectors are most informative not as a set of controls but as a probe of behavioral organization.

cs.CL↗

arXiv2026

DYCP: Dynamic Context Pruning for Long-Form Dialogue with LLMs

Large Language Models (LLMs) increasingly operate over long-form dialogues with frequent topic shifts. While recent LLMs support extended context windows, efficient management of dialogue history in practice is needed due to inference cost and latency constraints. We present DyCP, a lightweight context management method implemented outside the LLM that dynamically identifies and retrieves relevant dialogue segments conditioned on the current turn, without offline memory construction. DyCP manages dialogue context while preserving the sequential nature of dialogue without predefined topic boundaries, enabling adaptive and efficient context selection. Across three long-form dialogue benchmarks-LoCoMo, MT-Bench+, and SCM4LLMs-and multiple LLM backends, DyCP achieves competitive answer quality in downstream generation, with more selective context usage and improved inference efficiency.

cs.CL↗

arXiv2026

PersonaKit (PK): A Plug-and-Play Platform for User Testing Diverse Roles in Full-Duplex Dialogue

As spoken dialogue systems expand beyond traditional assistant roles to encompass diverse personas -- such as authoritative instructors, uncooperative merchants, or distracted workers -- they require distinct, human-like turn-taking behaviors to maintain psychological immersion. However, current full-duplex systems often default to a rigid, overly accommodating ``always-yield'' policy during overlapping speech, which severely undermines character consistency for non-submissive roles. Evaluating alternative, persona-specific turn-taking strategies through empirical user studies is challenging because building real-time full-duplex test environments requires substantial engineering overhead. To address this, we present PersonaKit (PK), an open-source, low-latency web platform for the rapid prototyping and evaluation of conversational agents. Using intuitive JSON configurations, researchers can define personas, specify probabilistic interruption-handling behaviors (e.g., yield, hold, bridge, or override), and automatically deploy comparative A/B surveys. Through an in-the-wild evaluation with 8 distinct personas, we demonstrate that PersonaKit provides an extensible, end-to-end framework for studying complex sociolinguistic behaviors in next-generation spoken agents.

cs.CL↗

arXiv2026

Improving Factuality in LLMs via Inference-Time Knowledge Graph Construction

Large Language Models (LLMs) often struggle with producing factually consistent answers due to limitations in their parametric memory. Retrieval-Augmented Generation (RAG) paradigms mitigate this issue by incorporating external knowledge at inference time. However, such methods typically handle knowledge as unstructured text, which reduces retrieval accuracy, hinders compositional reasoning, and amplifies the influence of irrelevant information on the factual consistency of LLM outputs. To overcome these limitations, we propose a novel framework that dynamically constructs and expands knowledge graphs (KGs) during inference, integrating both internal knowledge extracted from LLMs and external knowledge retrieved from external sources. Our method begins by extracting a seed KG from the question via prompting, followed by iterative expansion using the LLM's internal knowledge. The KG is then selectively refined through external retrieval, enhancing factual coverage and correcting inaccuracies. We evaluate our approach on three diverse Factual QA benchmarks, demonstrating consistent gains in factual accuracy over baselines. Our findings reveal that inference-time KG construction is a promising direction for enhancing LLM factuality in a structured, interpretable, and scalable manner.

cs.CL↗

arXiv2026

Hierarchical Multi-Persona Induction from User Behavioral Logs: Learning Evidence-Grounded and Truthful Personas

Behavioral logs provide rich signals for user modeling, but are noisy and interleaved across diverse intents. Recent work uses LLMs to generate interpretable natural-language personas from user logs, yet evaluation often emphasizes downstream utility, providing limited assurance of persona quality itself. We propose a hierarchical framework that aggregates user actions into intent memories and induces multiple evidence-grounded personas by clustering and labeling these memories. We formulate persona induction as an optimization problem over persona quality-captured by cluster cohesion, persona-evidence alignment, and persona truthfulness-and train the persona model using a groupwise extension of Direct Preference Optimization (DPO). Experiments on a large-scale service log and two public datasets show that our method induces more coherent, evidence-grounded, and trustworthy personas, while also improving future interaction prediction.

cs.AI↗

arXiv2026

Why Are We Lonely? Leveraging LLMs to Measure and Understand Loneliness in Caregivers and Non-caregivers

This paper presents an LLM-driven approach for constructing diverse social media datasets to measure and compare loneliness in the caregiver and non-caregiver populations. We introduce an expert-developed loneliness evaluation framework and an expert-informed typology for categorizing causes of loneliness for analyzing social media text. Using a human-validated data processing pipeline, we apply GPT-4o, GPT-5-nano, and GPT-5 to build a high-quality Reddit corpus and analyze loneliness across both populations. The loneliness evaluation framework achieved average accuracies of 76.09% and 79.78% for caregivers and non-caregivers, respectively. The cause categorization framework achieved micro-aggregate F1 scores of 0.825 and 0.80 for caregivers and non-caregivers, respectively. Across populations, we observe substantial differences in the distribution of types of causes of loneliness. Caregivers' loneliness were predominantly linked to caregiving roles, identity recognition, and feelings of abandonment, indicating distinct loneliness experiences between the two groups. Demographic extraction further demonstrates the viability of Reddit for building a diverse caregiver loneliness dataset. Overall, this work establishes an LLM-based pipeline for creating high quality social media datasets for studying loneliness and demonstrates its effectiveness in analyzing population-level differences in the manifestation of loneliness.

cs.CL↗

arXiv2026

ReCUBE: Evaluating Repository-Level Context Utilization in Code Generation

Large Language Models (LLMs) have recently emerged as capable coding assistants that operate over large codebases through either agentic exploration or full-context generation. Existing benchmarks capture a broad range of coding capabilities, such as resolving GitHub issues, but none of them directly isolate and measure how effectively LLMs leverage repository-level context during code generation. To address this, we introduce ReCUBE, a benchmark in which LLMs reconstruct a masked file within a real-world repository, using all remaining source files, dependency specifications, and documentation as their only source of context. ReCUBE evaluates reconstructed code with usage-aware test cases that simulate both internal module logic and external cross-file integration, reflecting real-world software usage patterns. We further propose the Caller-Centric Exploration (CCE) toolkit, a set of dependency graph-based tools that can be integrated into agentic frameworks to guide agents toward the most relevant caller files during repository exploration. Experiments across eight models in four settings show that repository-level context utilization remains highly challenging even for state-of-the-art models, with GPT-5 achieving only 37.57% strict pass rate in the full-context setting. Agents augmented with our CCE toolkit consistently outperform all baselines across all evaluated models, with improvements of up to 7.56% in strict pass rate. We release our benchmark, code, and evaluation framework as open source for the NLP research community.

cs.SE↗

arXiv2026

Measuring Sycophancy of Language Models in Multi-turn Dialogues

Large Language Models (LLMs) are expected to provide helpful and harmless responses, yet they often exhibit sycophancy--conforming to user beliefs regardless of factual accuracy or ethical soundness. Prior research on sycophancy has primarily focused on single-turn factual correctness, overlooking the dynamics of real-world interactions. In this work, we introduce SYCON Bench, a novel benchmark for evaluating sycophantic behavior in multi-turn, free-form conversational settings. Our benchmark measures how quickly a model conforms to the user (Turn of Flip) and how frequently it shifts its stance under sustained user pressure (Number of Flip). Applying SYCON Bench to 17 LLMs across three real-world scenarios, we find that sycophancy remains a prevalent failure mode. Our analysis shows that alignment tuning amplifies sycophantic behavior, whereas model scaling and reasoning optimization strengthen the model's ability to resist undesirable user views. Reasoning models generally outperform instruction-tuned models but often fail when they over-index on logical exposition instead of directly addressing the user's underlying beliefs. Finally, we evaluate four additional prompting strategies and demonstrate that adopting a third-person perspective reduces sycophancy by up to 63.8% in debate scenario. We release our code and data at https://github.com/JiseungHong/SYCON-Bench.

cs.CL↗

arXiv2026

Designing and Evaluating a Conversational Agent for Early Diagnosis of Alzheimer's Disease and Related Dementias

Early diagnosis of Alzheimer's disease and related dementias (ADRD) is critical for timely intervention, yet most diagnoses are delayed until advanced stages. While comprehensive patient narratives are essential for accurate diagnosis, prior work has largely focused on screening studies that classify cognitive status from interactions rather than supporting the diagnostic process. We designed voice-interactive conversational agents, leveraging large language models (LLMs), to elicit narratives relevant to ADRD from patients and informants. We evaluated the agent with 30 adults with suspected ADRD through conversation analysis, user surveys, and analysis of symptom elicitation compared to blinded specialist interviews. Symptoms detected by the agent showed promising agreement with those identified by specialists. Users appreciated the agent's patience and systematic questioning, which supported engagement and expression of complex, hard-to-describe experiences. While these findings suggest potential for conversational agents as structured diagnostic support tools, further validation with larger samples and assessment of clinical utility is needed before deployment.

cs.HC↗

arXiv2026

RIFT: Reordered Instruction Following Testbed To Evaluate Instruction Following in Singular Multistep Prompt Structures

Large Language Models (LLMs) are increasingly relied upon for complex workflows, yet their ability to maintain flow of instructions remains underexplored. Existing benchmarks conflate task complexity with structural ordering, making it difficult to isolate the impact of prompt topology on performance. We introduce RIFT, Reordered Instruction Following Testbed, to assess instruction following by disentangling structure from content. Using rephrased Jeopardy! question-answer pairs, we test LLMs across two prompt structures: linear prompts, which progress sequentially, and jumping prompts, which preserve identical content but require non-sequential traversal. Across 10,000 evaluations spanning six state-of-the-art open-source LLMs, accuracy dropped by up to 72% under jumping conditions (compared to baseline), revealing a strong dependence on positional continuity. Error analysis shows that approximately 50% of failures stem from instruction-order violations and semantic drift, indicating that current architectures internalize instruction following as a sequential pattern rather than a reasoning skill. These results reveal structural sensitivity as a fundamental limitation in current architectures, with direct implications for applications requiring non-sequential control flow such as workflow automation and multi-agent systems.

cs.AI↗

arXiv2026

CRADLE Bench: A Clinician-Annotated Benchmark for Multi-Faceted Mental Health Crisis and Safety Risk Detection

Detecting mental health crisis situations such as suicide ideation, rape, domestic violence, child abuse, and sexual harassment is a critical yet underexplored challenge for language models. When such situations arise during user--model interactions, models must reliably flag them, as failure to do so can have serious consequences. In this work, we introduce CRADLE BENCH, a benchmark for multi-faceted crisis detection. Unlike previous efforts that focus on a limited set of crisis types, our benchmark covers seven types defined in line with clinical standards and is the first to incorporate temporal labels. Our benchmark provides 600 clinician-annotated evaluation examples and 420 development examples, together with a training corpus of around 4K examples automatically labeled using a majority-vote ensemble of multiple language models, which significantly outperforms single-model annotation. We further fine-tune six crisis detection models on subsets defined by consensus and unanimous ensemble agreement, providing complementary models trained under different agreement criteria.

cs.CL↗

arXiv2025

LLM-as-a-Grader: Practical Insights from Large Language Model for Short-Answer and Report Evaluation

Large Language Models (LLMs) are increasingly explored for educational tasks such as grading, yet their alignment with human evaluation in real classrooms remains underexamined. In this study, we investigate the feasibility of using an LLM (GPT-4o) to evaluate short-answer quizzes and project reports in an undergraduate Computational Linguistics course. We collect responses from approximately 50 students across five quizzes and receive project reports from 14 teams. LLM-generated scores are compared against human evaluations conducted independently by the course teaching assistants (TAs). Our results show that GPT-4o achieves strong correlation with human graders (up to 0.98) and exact score agreement in 55\% of quiz cases. For project reports, it also shows strong overall alignment with human grading, while exhibiting some variability in scoring technical, open-ended responses. We release all code and sample data to support further research on LLMs in educational assessment. This work highlights both the potential and limitations of LLM-based grading systems and contributes to advancing automated grading in real-world academic settings.

cs.CL↗

arXiv2025

Reference-Aligned Retrieval-Augmented Question Answering over Heterogeneous Proprietary Documents

Proprietary corporate documents contain rich domain-specific knowledge, but their overwhelming volume and disorganized structure make it difficult even for employees to access the right information when needed. For example, in the automotive industry, vehicle crash-collision tests, each costing hundreds of thousands of dollars, produce highly detailed documentation. However, retrieving relevant content during decision-making remains time-consuming due to the scale and complexity of the material. While Retrieval-Augmented Generation (RAG)-based Question Answering (QA) systems offer a promising solution, building an internal RAG-QA system poses several challenges: (1) handling heterogeneous multi-modal data sources, (2) preserving data confidentiality, and (3) enabling traceability between each piece of information in the generated answer and its original source document. To address these, we propose a RAG-QA framework for internal enterprise use, consisting of: (1) a data pipeline that converts raw multi-modal documents into a structured corpus and QA pairs, (2) a fully on-premise, privacy-preserving architecture, and (3) a lightweight reference matcher that links answer segments to supporting content. Applied to the automotive domain, our system improves factual correctness (+1.79, +1.94), informativeness (+1.33, +1.16), and helpfulness (+1.08, +1.67) over a non-RAG baseline, based on 1-5 scale ratings from both human and LLM judge.

cs.AI↗

arXiv2025

Finding A Voice: Exploring the Potential of African American Dialect and Voice Generation for Chatbots

As chatbots become integral to daily life, personalizing systems is key for fostering trust, engagement, and inclusivity. This study examines how linguistic similarity affects chatbot performance, focusing on integrating African American English (AAE) into virtual agents to better serve the African American community. We develop text-based and spoken chatbots using large language models and text-to-speech technology, then evaluate them with AAE speakers against standard English chatbots. Our results show that while text-based AAE chatbots often underperform, spoken chatbots benefit from an African American voice and AAE elements, improving performance and preference. These findings underscore the complexities of linguistic personalization and the dynamics between text and speech modalities, highlighting technological limitations that affect chatbots' AA speech generation and pointing to promising future research directions.

cs.CL↗

arXiv2025

Secure Multifaceted-RAG for Enterprise: Hybrid Knowledge Retrieval with Security Filtering

Existing Retrieval-Augmented Generation (RAG) systems face challenges in enterprise settings due to limited retrieval scope and data security risks. When relevant internal documents are unavailable, the system struggles to generate accurate and complete responses. Additionally, using closed-source Large Language Models (LLMs) raises concerns about exposing proprietary information. To address these issues, we propose the Secure Multifaceted-RAG (SecMulti-RAG) framework, which retrieves not only from internal documents but also from two supplementary sources: pre-generated expert knowledge for anticipated queries and on-demand external LLM-generated knowledge. To mitigate security risks, we adopt a local open-source generator and selectively utilize external LLMs only when prompts are deemed safe by a filtering mechanism. This approach enhances completeness, prevents data leakage, and reduces costs. In our evaluation on a report generation task in the automotive industry, SecMulti-RAG significantly outperforms traditional RAG - achieving 79.3 to 91.9 percent win rates across correctness, richness, and helpfulness in LLM-based evaluation, and 56.3 to 70.4 percent in human evaluation. This highlights SecMulti-RAG as a practical and secure solution for enterprise RAG.

cs.CL↗

arXiv2025

ETM: Modern Insights into Perspective on Text-to-SQL Evaluation in the Age of Large Language Models

The task of Text-to-SQL enables anyone to retrieve information from SQL databases using natural language. While this task has made substantial progress, the two primary evaluation metrics - Execution Accuracy (EXE) and Exact Set Matching Accuracy (ESM) - suffer from inherent limitations that can misrepresent performance. Specifically, ESM's rigid matching overlooks semantically correct but stylistically different queries, whereas EXE can overestimate correctness by ignoring structural errors that yield correct outputs. These shortcomings become especially problematic when assessing outputs from large language model (LLM)-based approaches without fine-tuning, which vary more in style and structure compared to their fine-tuned counterparts. Thus, we introduce a new metric, Enhanced Tree Matching (ETM), which mitigates these issues by comparing queries using both syntactic and semantic elements. Through evaluating nine LLM-based models, we show that EXE and ESM can produce false positive and negative rates as high as 23.0% and 28.9%, while ETM reduces these rates to 0.3% and 2.7%, respectively. We release our ETM script as open source, offering the community a more robust and reliable approach to evaluating Text-to-SQL.

cs.CL↗