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?
Human evaluation is the gold-standard for evaluating text generation models. It is also expensive, and to fit budgetary constraints, a random subset of the test data is often chosen in practice. The randomly selected data may not accurately represent test performance, making this approach economically inefficient for model comparison. Thus, in this work, we develop a suite of selectors to get the most informative datapoints for human evaluation while taking the evaluation costs into account. We show that selectors based on variance in automated metric scores, diversity in model outputs, or Item Response Theory outperform random selection. We further develop an approach to distill these selectors to the scenario where the model outputs are not yet available. In particular, we introduce source-based estimators, which predict item usefulness for human evaluation just based on the source texts. We demonstrate the efficacy of our selectors in two common NLG tasks, machine translation and summarization, and show that up to only ~50% of the test data is needed to produce the same evaluation result as the entire data. Our implementations are published in the subset2evaluate package.
Token representation strategies within large-scale neural architectures often rely on contextually refined embeddings, yet conventional approaches seldom encode structured relationships explicitly within token interactions. Self-attention mechanisms effectively capture dynamic contextual dependencies, but their reliance on learned weight distributions limits the preservation of long-range hierarchical structures in generated sequences. Dependency-aware token encoding introduces a structured approach to embedding initialization, ensuring that relational constraints are embedded within token representations rather than inferred solely through attention dynamics. The proposed encoding mechanism refines token interactions through dependency-weighted attention computations, ensuring that syntactic and semantic dependencies are retained across multiple processing layers. Empirical evaluations indicate reductions in perplexity across diverse linguistic benchmarks, suggesting improvements in contextual coherence and predictive consistency in autoregressive text generation. Computational efficiency assessments reveal a moderate increase in memory consumption and training time, attributed to additional matrix computations within the encoding module, yet scalability remains feasible within conventional transformer architectures. Structured encoding enhances lexical variation and dependency retention, reinforcing linguistic coherence without requiring external syntactic annotations or auxiliary training objectives. Statistical comparisons highlight improvements in dependency alignment, particularly in longer sequences where conventional self-attention models exhibit degradation in hierarchical consistency. Sentence length distributions indicate a reduction in abrupt phrase transitions, further supporting the hypothesis that explicit dependency encoding facilitates more structured phrase generation.
Aligning Large Language Models (LLMs) with human values and away from undesirable behaviors (such as hallucination) has become increasingly important. Recently, steering LLMs towards a desired behavior via activation editing has emerged as an effective method to mitigate harmful generations at inference-time. Activation editing modifies LLM representations by preserving information from positive demonstrations (e.g., truthful) and minimising information from negative demonstrations (e.g., hallucinations). When these demonstrations come from a private dataset, the aligned LLM may leak private information contained in those private samples. In this work, we present the first study of aligning LLM behavior with private datasets. Our work proposes the \textit{\underline{P}rivate \underline{S}teering for LLM \underline{A}lignment (PSA)} algorithm to edit LLM activations with differential privacy (DP) guarantees. We conduct extensive experiments on seven different benchmarks with open-source LLMs of different sizes (0.5B to 7B) and model families (LlaMa, Qwen, Mistral and Gemma). Our results show that PSA achieves DP guarantees for LLM alignment with minimal loss in performance, including alignment metrics, open-ended text generation quality, and general-purpose reasoning. We also develop the first Membership Inference Attack (MIA) for evaluating and auditing the empirical privacy for the problem of LLM steering via activation editing. Our attack is tailored for activation editing and relies solely on the generated texts without their associated probabilities. Our experiments support the theoretical guarantees by showing improved guarantees for our \textit{PSA} algorithm compared to several existing non-private techniques.
Fine-tuning large language models (LLMs) using low-rank adaptation (LoRA) has become a highly efficient approach for downstream tasks, particularly in scenarios with limited computational resources. However, applying LoRA techniques to quantized LLMs poses unique challenges due to the reduced representational precision of quantized weights. In this paper, we introduce CLoQ (Calibrated LoRA initialization for Quantized LLMs), a simplistic initialization strategy designed to overcome these challenges. Our approach focuses on minimizing the layer-wise discrepancy between the original LLM and its quantized counterpart with LoRA components during initialization. By leveraging a small calibration dataset, CLoQ quantizes a pre-trained LLM and determines the optimal LoRA components for each layer, ensuring a strong foundation for subsequent fine-tuning. A key contribution of this work is a novel theoretical result that enables the accurate and closed-form construction of these optimal LoRA components. We validate the efficacy of CLoQ across multiple tasks such as language generation, arithmetic reasoning, and commonsense reasoning, demonstrating that it consistently outperforms existing LoRA fine-tuning methods for quantized LLMs, especially at ultra low-bit widths.
Context-aware processing mechanisms have increasingly become a critical area of exploration for improving the semantic and contextual capabilities of language generation models. The Context-Aware Semantic Recomposition Mechanism (CASRM) was introduced as a novel framework designed to address limitations in coherence, contextual adaptability, and error propagation in large-scale text generation tasks. Through the integration of dynamically generated context vectors and attention modulation layers, CASRM enhances the alignment between token-level representations and broader contextual dependencies. Experimental evaluations demonstrated significant improvements in semantic coherence across multiple domains, including technical, conversational, and narrative text. The ability to adapt to unseen domains and ambiguous inputs was evaluated using a diverse set of test scenarios, highlighting the robustness of the proposed mechanism. A detailed computational analysis revealed that while CASRM introduces additional processing overhead, the gains in linguistic precision and contextual relevance outweigh the marginal increase in complexity. The framework also successfully mitigates error propagation in sequential tasks, improving performance in dialogue continuation and multi-step text synthesis. Additional investigations into token-level attention distribution emphasized the dynamic focus shifts enabled through context-aware enhancements. The findings suggest that CASRM offers a scalable and flexible solution for integrating contextual intelligence into existing language model architectures.
Large language models (LLM) have shown remarkable abilities in text generation, question answering, language translation, reasoning and many other tasks. It continues to advance rapidly and is becoming increasingly influential in various fields, from technology and business to education and entertainment. Despite LLM's success in multiple areas, its ability to play abstract games, such as chess, is underexplored. Chess-playing requires the language models to output legal and reasonable moves from textual inputs. Here, we propose the Large language model ChessLLM to play full chess games. We transform the game into a textual format with the best move represented in the Forsyth-Edwards Notation. We show that by simply supervised fine-tuning, our model has achieved a professional-level Elo rating of 1788 in matches against the standard Elo-rated Stockfish when permitted to sample 10 times. We further show that data quality is important. Long-round data supervision enjoys a 350 Elo rating improvement over short-round data.
Hallucination detection in text generation remains an ongoing struggle for natural language processing (NLP) systems, frequently resulting in unreliable outputs in applications such as machine translation and definition modeling. Existing methods struggle with data scarcity and the limitations of unlabeled datasets, as highlighted by the SHROOM shared task at SemEval-2024. In this work, we propose a novel framework to address these challenges, introducing DeepSeek Few-shot optimization to enhance weak label generation through iterative prompt engineering. We achieved high-quality annotations that considerably enhanced the performance of downstream models by restructuring data to align with instruct generative models. We further fine-tuned the Mistral-7B-Instruct-v0.3 model on these optimized annotations, enabling it to accurately detect hallucinations in resource-limited settings. Combining this fine-tuned model with ensemble learning strategies, our approach achieved 85.5% accuracy on the test set, setting a new benchmark for the SHROOM task. This study demonstrates the effectiveness of data restructuring, few-shot optimization, and fine-tuning in building scalable and robust hallucination detection frameworks for resource-constrained NLP systems.
This paper introduces a novel problem, distributional information embedding, motivated by the practical demands of multi-bit watermarking for large language models (LLMs). Unlike traditional information embedding, which embeds information into a pre-existing host signal, LLM watermarking actively controls the text generation process--adjusting the token distribution--to embed a detectable signal. We develop an information-theoretic framework to analyze this distributional information embedding problem, characterizing the fundamental trade-offs among three critical performance metrics: text quality, detectability, and information rate. In the asymptotic regime, we demonstrate that the maximum achievable rate with vanishing error corresponds to the entropy of the LLM's output distribution and increases with higher allowable distortion. We also characterize the optimal watermarking scheme to achieve this rate. Extending the analysis to the finite-token case, we identify schemes that maximize detection probability while adhering to constraints on false alarm and distortion.
Human motion generation has advanced markedly with the advent of diffusion models. Most recent studies have concentrated on generating motion sequences based on text prompts, commonly referred to as text-to-motion generation. However, the bidirectional generation of motion and text, enabling tasks such as motion-to-text alongside text-to-motion, has been largely unexplored. This capability is essential for aligning diverse modalities and supports unconditional generation. In this paper, we introduce PackDiT, the first diffusion-based generative model capable of performing various tasks simultaneously, including motion generation, motion prediction, text generation, text-to-motion, motion-to-text, and joint motion-text generation. Our core innovation leverages mutual blocks to integrate multiple diffusion transformers (DiTs) across different modalities seamlessly. We train PackDiT on the HumanML3D dataset, achieving state-of-the-art text-to-motion performance with an FID score of 0.106, along with superior results in motion prediction and in-between tasks. Our experiments further demonstrate that diffusion models are effective for motion-to-text generation, achieving performance comparable to that of autoregressive models.
One approach for multilingual data-to-text generation is to translate grammatical configurations upfront from the source language into each target language. These configurations are then used by a surface realizer and in document planning stages to generate output. In this paper, we describe a rule-based NLG implementation of this approach where the configuration is translated by Neural Machine Translation (NMT) combined with a one-time human review, and introduce a cross-language grammar dependency model to create a multilingual NLG system that generates text from the source data, scaling the generation phase without a human in the loop. Additionally, we introduce a method for human post-editing evaluation on the automatically translated text. Our evaluation on the SportSett:Basketball dataset shows that our NLG system performs well, underlining its grammatical correctness in translation tasks.
The rise of large language models (LLMs) has led to more diverse and higher-quality machine-generated text. However, their high expressive power makes it difficult to control outputs based on specific business instructions. In response, benchmarks focusing on the controllability of LLMs have been developed, but several issues remain: (1) They primarily cover major languages like English and Chinese, neglecting low-resource languages like Japanese; (2) Current benchmarks employ task-specific evaluation metrics, lacking a unified framework for selecting models based on controllability across different use cases. To address these challenges, this research introduces LCTG Bench, the first Japanese benchmark for evaluating the controllability of LLMs. LCTG Bench provides a unified framework for assessing control performance, enabling users to select the most suitable model for their use cases based on controllability. By evaluating nine diverse Japanese-specific and multilingual LLMs like GPT-4, we highlight the current state and challenges of controllability in Japanese LLMs and reveal the significant gap between multilingual models and Japanese-specific models.
Large Language Models (LLMs) excel in text generation and understanding, especially in simulating socio-political and economic patterns, serving as an alternative to traditional surveys. However, their global applicability remains questionable due to unexplored biases across socio-demographic and geographic contexts. This study examines how LLMs perform across diverse populations by analyzing public surveys from Chile and the United States, focusing on predictive accuracy and fairness metrics. The results show performance disparities, with LLM consistently outperforming on U.S. datasets. This bias originates from the U.S.-centric training data, remaining evident after accounting for socio-demographic differences. In the U.S., political identity and race significantly influence prediction accuracy, while in Chile, gender, education, and religious affiliation play more pronounced roles. Our study presents a novel framework for measuring socio-demographic biases in LLMs, offering a path toward ensuring fairer and more equitable model performance across diverse socio-cultural contexts.
Improvements in large language models have led to increasing optimism that they can serve as reliable evaluators of natural language generation outputs. In this paper, we challenge this optimism by thoroughly re-evaluating five state-of-the-art factuality metrics on a collection of 11 datasets for summarization, retrieval-augmented generation, and question answering. We find that these evaluators are inconsistent with each other and often misestimate system-level performance, both of which can lead to a variety of pitfalls. We further show that these metrics exhibit biases against highly paraphrased outputs and outputs that draw upon faraway parts of the source documents. We urge users of these factuality metrics to proceed with caution and manually validate the reliability of these metrics in their domain of interest before proceeding.
Large language models (LLMs) exhibit unprecedentedly rich scaling behaviors. In physics, scaling behavior is closely related to phase transitions, critical phenomena, and field theory. To investigate the phase transition phenomena in LLMs, we reformulated the Transformer architecture as an $O(N)$ model. Our study reveals two distinct phase transitions corresponding to the temperature used in text generation and the model's parameter size, respectively. The first phase transition enables us to estimate the internal dimension of the model, while the second phase transition is of \textit{higher-depth} and signals the emergence of new capabilities. As an application, the energy of the $O(N)$ model can be used to evaluate whether an LLM's parameters are sufficient to learn the training data.
Fusing visual understanding into language generation, Multi-modal Large Language Models (MLLMs) are revolutionizing visual-language applications. Yet, these models are often plagued by the hallucination problem, which involves generating inaccurate objects, attributes, and relationships that do not match the visual content. In this work, we delve into the internal attention mechanisms of MLLMs to reveal the underlying causes of hallucination, exposing the inherent vulnerabilities in the instruction-tuning process. We propose a novel hallucination attack against MLLMs that exploits attention sink behaviors to trigger hallucinated content with minimal image-text relevance, posing a significant threat to critical downstream applications. Distinguished from previous adversarial methods that rely on fixed patterns, our approach generates dynamic, effective, and highly transferable visual adversarial inputs, without sacrificing the quality of model responses. Comprehensive experiments on 6 prominent MLLMs demonstrate the efficacy of our attack in compromising black-box MLLMs even with extensive mitigating mechanisms, as well as the promising results against cutting-edge commercial APIs, such as GPT-4o and Gemini 1.5. Our code is available at https://huggingface.co/RachelHGF/Mirage-in-the-Eyes.
Evaluating personalized text generated by large language models (LLMs) is challenging, as only the LLM user, i.e., prompt author, can reliably assess the output, but re-engaging the same individuals across studies is infeasible. This paper addresses the challenge of evaluating personalized text generation by introducing ExPerT, an explainable reference-based evaluation framework. ExPerT leverages an LLM to extract atomic aspects and their evidence from the generated and reference texts, match the aspects, and evaluate their alignment based on content and writing style -- two key attributes in personalized text generation. Additionally, ExPerT generates detailed, fine-grained explanations for every step of the evaluation process, enhancing transparency and interpretability. Our experiments demonstrate that ExPerT achieves a 7.2% relative improvement in alignment with human judgments compared to the state-of-the-art text generation evaluation methods. Furthermore, human evaluators rated the usability of ExPerT's explanations at 4.7 out of 5, highlighting its effectiveness in making evaluation decisions more interpretable.
The Semantic Layered Embedding Diffusion (SLED) mechanism redefines the representation of hierarchical semantics within transformer-based architectures, enabling enhanced contextual consistency across a wide array of linguistic tasks. By introducing a multi-layered diffusion process grounded in spectral analysis, it achieves a complex balance between global and local semantic coherence. Experimental results demonstrate significant improvements in perplexity and BLEU scores, emphasizing the mechanism's ability to adapt effectively across diverse domains, including multilingual and cross-domain text generation. A rigorous mathematical framework underpins the embedding diffusion process, incorporating weighted adjacency matrices, kernel-based refinements, and dynamic layer-wise normalization. Error distribution analysis reveals that SLED addresses challenges in semantic alignment and coherence, outperforming baseline approaches across varied benchmarks. Scalability studies illustrate that its performance gains are maintained consistently across different model sizes, reflecting a practical balance between computational efficiency and linguistic precision. The implementation also achieves energy efficiency, reducing resource consumption during training and inference phases without compromising accuracy. Qualitative case studies further validate its adaptability to extended narratives and context-intensive scenarios, highlighting the mechanism's potential for real-world applications. SLED offers a different perspective on embedding design and its implications for advancing language modeling.
Due to the scarcity of data in low-resourced languages, the development of language models for these languages has been very slow. Currently, pre-trained language models have gained popularity in natural language processing, especially, in developing domain-specific models for low-resourced languages. In this study, we experiment with the impact of using occlusion-based techniques when training a language model for a text generation task. We curate 2 new datasets, the Sepedi monolingual (SepMono) dataset from several South African resources and the Sepedi radio news (SepNews) dataset from the radio news domain. We use the SepMono dataset to pre-train transformer-based models using the occlusion and non-occlusion pre-training techniques and compare performance. The SepNews dataset is specifically used for fine-tuning. Our results show that the non-occlusion models perform better compared to the occlusion-based models when measuring validation loss and perplexity. However, analysis of the generated text using the BLEU score metric, which measures the quality of the generated text, shows a slightly higher BLEU score for the occlusion-based models compared to the non-occlusion models.
Large language models (LLMs) have demonstrated immense potential across various tasks. However, research for exploring and improving the capabilities of LLMs in interpreting graph structures remains limited. To address this gap, we conduct a comprehensive evaluation of prompting current open-source LLMs on graph-to-text generation tasks. Although we explored the optimal prompting strategies and proposed a novel and effective diversity-difficulty-based few-shot sample selection method, we found that the improvements from tuning-free approaches were incremental, as LLMs struggle with planning on complex graphs, particularly those with a larger number of triplets. To further improve LLMs in planning with graph sequences and grounding in truth, we introduce a new graph-to-text dataset, PlanGTG, annotated with two sub-tasks: reordering and attribution. Through extensive automatic and human evaluations, we demonstrate significant improvements in the quality of generated text from both few-shot learning and fine-tuning perspectives using the PlanGTG dataset. Our study paves the way for new research directions in graph-to-text generation. PlanGTG datasets can be found in https://github.com/probe2/kg_text.