Searcharxiv⌕ Search

arXiv subjects

Lun-Wei Ku

Publications and source records attributed to Lun-Wei Ku.

At least 19 recordsLinked to original sources

Necessary for What? Auditing Demographic Disclosure to LLM Synthetic Respondents

Large language models (LLMs) increasingly answer surveys on behalf of real panelists, and each panelist's recorded demographic attributes (e.g., age, education, ethnicity) are often placed in the prompt on the premise that richer profiles make more faithful synthetic respondents. While data minimization limits personal data to what a purpose requires, a synthetic survey serves several purposes at once, from predicting individual answers to estimating how opinions are distributed and comparing groups. We ask which of these purposes require the respondent's record, and whether improved estimates support sound conclusions about groups. We vary whose profile a model receives, where it comes from, and how many fields it contains, in two U.S. surveys with eleven open-weight and three hosted models, and compare each model with a statistical model fit to the same information. We make four contributions. First, we show that the record adds little to individual predictions and that, for the open-weight models, most of its gain in population estimates remains when it is replaced by census data or a synthetic persona. Second, we find that group comparisons improve only with the respondent's own record, and that even then the models' group differences rest more on the demographic label in the prompt than on the individual respondent, which the usual check of reproducing known group differences does not reveal. Third, we find that the models make groups look more alike than they are, keeping differences in circumstances while flattening or reversing differences in attitudes, most sharply for religious affiliation and ethnic origin, both special categories of personal data. Fourth, we outline an audit that a survey organization can run on questions it has already fielded to decide which estimates may use the record.

cs.CY↗

Upstream Influence, Downstream Re-Anchoring in LLM Agent Chains

People act on evaluative judgments produced by LLM agents, and in multi-agent systems those judgments pass through several agents before reaching a person. We show that a workflow's information architecture---source access, handoff format, and which model reads the untrusted content---governs how much of a coordinated manipulation reaches the end. On 32 Taiwan~2025 recall cases, ten authentic Threads comments shown to one agent move its stance by $+0.171$ toward the injected position, and a mirrored set moves it the other way. Passing that judgment down a chain removes more than half of the shift by the third agent; keeping the comments in front of every agent sustains more of it. The reduction is not attrition in transit: each receiver discounts the inherited stance in proportion to how far it sits from its own reading of the evidence, returning to what that evidence implies for the case rather than to neutrality---consistent with partial re-anchoring. Handoff format is a second lever: a stance relayed without its reasoning is largely discounted. Susceptibility at the exposed seat varies fourfold across models, while the chain's arrangement does not matter at all. None of this is visible in the transcripts---the injected claims are almost never restated, so reasoning-trace audits miss the attack.

cs.AI↗

The Fragility and Dynamics of LLM Pruning-Recovery Steps

Structural pruning followed by a recovery stage is the standard recipe for compressing large language models (LLMs), yet the two stages are typically designed and tuned in isolation. In this work, we study how their interaction shapes the compressed model, sweeping ordering, the data that drives each stage, and the alignment of their training objectives. We find that instruction-following is the first capability to collapse under pruning, well before knowledge, reasoning, or core language modeling, and that much of the pruning in-domain loss is likely not knowledge loss but a domain-agnostic failure to respond coherently to the prompt. We find in our setting that in-domain data is the strongest pruning signal, whereas instruction-style data recovers the most. Stage order matters little, but only marginally. We further analyze how the importance ranking drifts before and after pruning. Together these results give practitioners a concrete recipe for composing pruning and recovery.

cs.CL↗

Towards Robust Calibration Dataset for Domain-General Mixture of Experts Pruning

Post-training expert pruning is a promising deployment technique for Mixture-of-Experts (MoE) models. The calibration set is an integral part of the pruning process, however, its correlation with evaluation performance is largely misunderstood and unintuitive. In this work, we present an empirical study of the robustness of calibration perturbations, specifically in MoE models. Showing how calibration size, language, and data distribution affect layer-wise expert pruning for gpt-oss-20b. Our findings reveal that (1) domain-general pruning quality receives minimal benefit from scaling calibration data beyond small sample sizes, (2) cross-lingual calibration transfers effectively, with English-only calibration maintaining strong performance across seven languages of varying typologies and scripts, and (3) a diverse calibration data resembling the pretraining distribution can yield comparable or superior performance to test-distribution data. Ultimately, we provide practical insights as a recipe for better understanding calibration for more effective domain-general expert pruning.

cs.CL↗

Positional Cognitive Specialization: Where Do LLMs Learn To Comprehend and Speak Your Language?

Adapting large language models (LLMs) to new languages is an expensive and opaque process. Understanding how language models acquire new languages and multilingual abilities is key to achieve efficient adaptation. Prior work on multilingual interpretability research focuses primarily on how trained models process multilingual instructions, leaving unexplored the mechanisms through which they acquire new languages during training. We investigate these training dynamics on decoder-only transformers through the lens of two functional cognitive specializations: language perception (input comprehension) and production (output generation). Through experiments on low-resource languages, we demonstrate how perceptual and productive specialization emerges in different regions of a language model by running layer ablation sweeps from the model's input and output directions. Based on the observed specialization patterns, we propose CogSym, a layer-wise heuristic that enables effective adaptation by exclusively fine-tuning a few early and late layers. We show that tuning only the 25% outermost layers achieves downstream task performance within 2-3% deviation from the full fine-tuning baseline. CogSym yields consistent performance with adapter methods such as LoRA, showcasing generalization beyond full fine-tuning. These findings provide insights to better understand how LLMs learn new languages and push toward accessible and inclusive language modeling.

cs.CL↗

Beyond Many-Shot Translation: Scaling In-Context Demonstrations For Low-Resource Machine Translation

Building machine translation (MT) systems for low-resource languages is notably difficult due to the scarcity of high-quality data. Although Large Language Models (LLMs) have improved MT system performance, adapting them to lesser-represented languages remains challenging. In-context learning (ICL) may offer novel ways to adapt LLMs for low-resource MT by conditioning models on demonstration at inference time. In this study, we explore scaling low-resource machine translation ICL beyond the few-shot setting to thousands of examples with long-context models. We scale in-context token budget to 1M tokens and compare three types of training corpora used as in-context supervision: monolingual unsupervised data, instruction-style data, and parallel data (English--target and Indonesian--target). Our experiments on Javanese and Sundanese show that gains from additional context saturate quickly and can degrade near the maximum context window, with scaling behavior strongly dependent on corpus type. Notably, some forms of monolingual supervision can be competitive with parallel data, despite the latter offering additional supervision. Overall, our results characterize the effective limits and corpus-type sensitivity of long-context ICL for low-resource MT, highlighting that larger context windows do not necessarily yield proportional quality gains.

cs.CL↗

Profile-LLM: Dynamic Profile Optimization for Realistic Personality Expression in LLMs

Personalized Large Language Models (LLMs) have been shown to be an effective way to create more engaging and enjoyable user-AI interactions. While previous studies have explored using prompts to elicit specific personality traits in LLMs, they have not optimized these prompts to maximize personality expression. To address this limitation, we propose PersonaPulse: Dynamic Profile Optimization for Realistic Personality Expression in LLMs, a framework that leverages LLMs' inherent knowledge of personality traits to iteratively enhance role-play prompts while integrating a situational response benchmark as a scoring tool, ensuring a more realistic and contextually grounded evaluation to guide the optimization process. Quantitative evaluations demonstrate that the prompts generated by PersonaPulse outperform those of prior work, which were designed based on personality descriptions from psychological studies. Additionally, we explore the relationship between model size and personality modeling through extensive experiments. Finally, we find that, for certain personality traits, the extent of personality evocation can be partially controlled by pausing the optimization process. These findings underscore the importance of prompt optimization in shaping personality expression within LLMs, offering valuable insights for future research on adaptive AI interactions.

cs.CL↗

CoachMe: Decoding Sport Elements with a Reference-Based Coaching Instruction Generation Model

Motion instruction is a crucial task that helps athletes refine their technique by analyzing movements and providing corrective guidance. Although recent advances in multimodal models have improved motion understanding, generating precise and sport-specific instruction remains challenging due to the highly domain-specific nature of sports and the need for informative guidance. We propose CoachMe, a reference-based model that analyzes the differences between a learner's motion and a reference under temporal and physical aspects. This approach enables both domain-knowledge learning and the acquisition of a coach-like thinking process that identifies movement errors effectively and provides feedback to explain how to improve. In this paper, we illustrate how CoachMe adapts well to specific sports such as skating and boxing by learning from general movements and then leveraging limited data. Experiments show that CoachMe provides high-quality instructions instead of directions merely in the tone of a coach but without critical information. CoachMe outperforms GPT-4o by 31.6% in G-Eval on figure skating and by 58.3% on boxing. Analysis further confirms that it elaborates on errors and their corresponding improvement methods in the generated instructions. You can find CoachMe here: https://motionxperts.github.io/

cs.CL↗

Do Large Multimodal Models Solve Caption Generation for Scientific Figures? Lessons Learned from SciCap Challenge 2023

Since the SciCap datasets launch in 2021, the research community has made significant progress in generating captions for scientific figures in scholarly articles. In 2023, the first SciCap Challenge took place, inviting global teams to use an expanded SciCap dataset to develop models for captioning diverse figure types across various academic fields. At the same time, text generation models advanced quickly, with many powerful pre-trained large multimodal models (LMMs) emerging that showed impressive capabilities in various vision-and-language tasks. This paper presents an overview of the first SciCap Challenge and details the performance of various models on its data, capturing a snapshot of the fields state. We found that professional editors overwhelmingly preferred figure captions generated by GPT-4V over those from all other models and even the original captions written by authors. Following this key finding, we conducted detailed analyses to answer this question: Have advanced LMMs solved the task of generating captions for scientific figures?

cs.CL↗

YourSkatingCoach: A Figure Skating Video Benchmark for Fine-Grained Element Analysis

Combining sports and machine learning involves leveraging ML algorithms and techniques to extract insight from sports-related data such as player statistics, game footage, and other relevant information. However, datasets related to figure skating in the literature focus primarily on element classification and are currently unavailable or exhibit only limited access, which greatly raise the entry barrier to developing visual sports technology for it. Moreover, when using such data to help athletes improve their skills, we find they are very coarse-grained: they work for learning what an element is, but they are poorly suited to learning whether the element is good or bad. Here we propose air time detection, a novel motion analysis task, the goal of which is to accurately detect the duration of the air time of a jump. We present YourSkatingCoach, a large, novel figure skating dataset which contains 454 videos of jump elements, the detected skater skeletons in each video, along with the gold labels of the start and ending frames of each jump, together as a video benchmark for figure skating. In addition, although this type of task is often viewed as classification, we cast it as a sequential labeling problem and propose a Transformer-based model to calculate the duration. Experimental results show that the proposed model yields a favorable results for a strong baseline. To further verify the generalizability of the fine-grained labels, we apply the same process to other sports as cross-sports tasks but for coarse-grained task action classification. Here we fine-tune the classification to demonstrate that figure skating, as it contains the essential body movements, constitutes a strong foundation for adaptation to other sports.

cs.CV↗

SocialNLP Fake-EmoReact 2021 Challenge Overview: Predicting Fake Tweets from Their Replies and GIFs

This paper provides an overview of the Fake-EmoReact 2021 Challenge, held at the 9th SocialNLP Workshop, in conjunction with NAACL 2021. The challenge requires predicting the authenticity of tweets using reply context and augmented GIF categories from EmotionGIF dataset. We offer the Fake-EmoReact dataset with more than 453k as the experimental materials, where every tweet is labeled with authenticity. Twenty-four teams registered to participate in this challenge, and 5 submitted their results successfully in the evaluation phase. The best team achieves 93.9 on Fake-EmoReact 2021 dataset using F1 score. In addition, we show the definition of share task, data collection, and the teams' performance that joined this challenge and their approaches.

cs.CL↗

MAAIG: Motion Analysis And Instruction Generation

Many people engage in self-directed sports training at home but lack the real-time guidance of professional coaches, making them susceptible to injuries or the development of incorrect habits. In this paper, we propose a novel application framework called MAAIG(Motion Analysis And Instruction Generation). It can generate embedding vectors for each frame based on user-provided sports action videos. These embedding vectors are associated with the 3D skeleton of each frame and are further input into a pretrained T5 model. Ultimately, our model utilizes this information to generate specific sports instructions. It has the capability to identify potential issues and provide real-time guidance in a manner akin to professional coaches, helping users improve their sports skills and avoid injuries.

cs.CV↗

Is Explanation the Cure? Misinformation Mitigation in the Short Term and Long Term

With advancements in natural language processing (NLP) models, automatic explanation generation has been proposed to mitigate misinformation on social media platforms in addition to adding warning labels to identified fake news. While many researchers have focused on generating good explanations, how these explanations can really help humans combat fake news is under-explored. In this study, we compare the effectiveness of a warning label and the state-of-the-art counterfactual explanations generated by GPT-4 in debunking misinformation. In a two-wave, online human-subject study, participants (N = 215) were randomly assigned to a control group in which false contents are shown without any intervention, a warning tag group in which the false claims were labeled, or an explanation group in which the false contents were accompanied by GPT-4 generated explanations. Our results show that both interventions significantly decrease participants' self-reported belief in fake claims in an equivalent manner for the short-term and long-term. We discuss the implications of our findings and directions for future NLP-based misinformation debunking strategies.

cs.CL↗

LLM-in-the-loop: Leveraging Large Language Model for Thematic Analysis

Thematic analysis (TA) has been widely used for analyzing qualitative data in many disciplines and fields. To ensure reliable analysis, the same piece of data is typically assigned to at least two human coders. Moreover, to produce meaningful and useful analysis, human coders develop and deepen their data interpretation and coding over multiple iterations, making TA labor-intensive and time-consuming. Recently the emerging field of large language models (LLMs) research has shown that LLMs have the potential replicate human-like behavior in various tasks: in particular, LLMs outperform crowd workers on text-annotation tasks, suggesting an opportunity to leverage LLMs on TA. We propose a human-LLM collaboration framework (i.e., LLM-in-the-loop) to conduct TA with in-context learning (ICL). This framework provides the prompt to frame discussions with a LLM (e.g., GPT-3.5) to generate the final codebook for TA. We demonstrate the utility of this framework using survey datasets on the aspects of the music listening experience and the usage of a password manager. Results of the two case studies show that the proposed framework yields similar coding quality to that of human coders but reduces TA's labor and time demands.

cs.CL↗

Location-Aware Visual Question Generation with Lightweight Models

This work introduces a novel task, location-aware visual question generation (LocaVQG), which aims to generate engaging questions from data relevant to a particular geographical location. Specifically, we represent such location-aware information with surrounding images and a GPS coordinate. To tackle this task, we present a dataset generation pipeline that leverages GPT-4 to produce diverse and sophisticated questions. Then, we aim to learn a lightweight model that can address the LocaVQG task and fit on an edge device, such as a mobile phone. To this end, we propose a method which can reliably generate engaging questions from location-aware information. Our proposed method outperforms baselines regarding human evaluation (e.g., engagement, grounding, coherence) and automatic evaluation metrics (e.g., BERTScore, ROUGE-2). Moreover, we conduct extensive ablation studies to justify our proposed techniques for both generating the dataset and solving the task.

cs.CL↗

Label-Aware Hyperbolic Embeddings for Fine-grained Emotion Classification

Fine-grained emotion classification (FEC) is a challenging task. Specifically, FEC needs to handle subtle nuance between labels, which can be complex and confusing. Most existing models only address text classification problem in the euclidean space, which we believe may not be the optimal solution as labels of close semantic (e.g., afraid and terrified) may not be differentiated in such space, which harms the performance. In this paper, we propose HypEmo, a novel framework that can integrate hyperbolic embeddings to improve the FEC task. First, we learn label embeddings in the hyperbolic space to better capture their hierarchical structure, and then our model projects contextualized representations to the hyperbolic space to compute the distance between samples and labels. Experimental results show that incorporating such distance to weight cross entropy loss substantially improves the performance with significantly higher efficiency. We evaluate our proposed model on two benchmark datasets and found 4.8% relative improvement compared to the previous state of the art with 43.2% fewer parameters and 76.9% less training time. Code is available at https: //github.com/dinobby/HypEmo.

cs.CL↗

HonestBait: Forward References for Attractive but Faithful Headline Generation

Current methods for generating attractive headlines often learn directly from data, which bases attractiveness on the number of user clicks and views. Although clicks or views do reflect user interest, they can fail to reveal how much interest is raised by the writing style and how much is due to the event or topic itself. Also, such approaches can lead to harmful inventions by over-exaggerating the content, aggravating the spread of false information. In this work, we propose HonestBait, a novel framework for solving these issues from another aspect: generating headlines using forward references (FRs), a writing technique often used for clickbait. A self-verification process is included during training to avoid spurious inventions. We begin with a preliminary user study to understand how FRs affect user interest, after which we present PANCO1, an innovative dataset containing pairs of fake news with verified news for attractive but faithful news headline generation. Automatic metrics and human evaluations show that our framework yields more attractive results (+11.25% compared to human-written verified news headlines) while maintaining high veracity, which helps promote real information to fight against fake news.

cs.CL↗

Multi-VQG: Generating Engaging Questions for Multiple Images

Generating engaging content has drawn much recent attention in the NLP community. Asking questions is a natural way to respond to photos and promote awareness. However, most answers to questions in traditional question-answering (QA) datasets are factoids, which reduce individuals' willingness to answer. Furthermore, traditional visual question generation (VQG) confines the source data for question generation to single images, resulting in a limited ability to comprehend time-series information of the underlying event. In this paper, we propose generating engaging questions from multiple images. We present MVQG, a new dataset, and establish a series of baselines, including both end-to-end and dual-stage architectures. Results show that building stories behind the image sequence enables models to generate engaging questions, which confirms our assumption that people typically construct a picture of the event in their minds before asking questions. These results open up an exciting challenge for visual-and-language models to implicitly construct a story behind a series of photos to allow for creativity and experience sharing and hence draw attention to downstream applications.

cs.CL↗