Embedding-based similarity metrics between text sequences can be influenced not just by the content dimensions we most care about, but can also be biased by spurious attributes like the text’s source or language. These document confounders cause problems for many applications, but especially those that need to pool texts from different corpora. This paper shows that a debiasing algorithm that removes information about observed confounders from the encoder representations substantially reduces these biases at a minimal computational cost. Document similarity and clustering metrics improve across every embedding variant and task we evaluate—often dramatically. Interestingly, performance on out-of-distribution benchmarks is not impacted, indicating that the embeddings are not otherwise degraded.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Two central capabilities of language models (LMs) are: (i) drawing on prior knowledge about entities, which allows them to answer queries such as What’s the official language of Austria?, and (ii) adapting to new information provided in context, e.g., Pretend the official language of Austria is Tagalog., that is pre-pended to the question. In this article, we introduce targeted persuasion score (TPS), designed to quantify how persuasive a given context is to an LM where persuasion is operationalized as the ability of the context to alter the LM’s answer to the question. In contrast to evaluating persuasiveness only through a model’s most likely answer, TPS provides a more fine-grained view of model behavior. Based on the Wasserstein distance, TPS measures how much a context shifts a model’s original answer distribution towarda target distribution. Empirically, through aseries of experiments, we show that TPS captures a more nuanced notion of persuasiveness than previously proposed metrics.
Large language models increasingly support multiple languages, yet most benchmarks for gender bias remain English-centric. We introduce EuroGEST, a dataset designed to measure gender-stereotypical reasoning in LLMs across English and 29 European languages. EuroGEST builds on an existing expert-informed benchmark covering 16 gender stereotypes, expanded in this work using translation tools, quality estimation metrics, and morphological heuristics. Human evaluations confirm that our data generation method results in high accuracy of both translations and gender labels across languages. We use EuroGEST to evaluate 24 multilingual language models from six model families, demonstrating that the strongest stereotypes in all models across all languages are that women are beautiful, empathetic and neat and men are leaders, strong, tough and professional. We also show that larger models encode gendered stereotypes more strongly and that instruction finetuned models continue to exhibit gendered stereotypes. Our work highlights the need for more multilingual studies of fairness in LLMs and offers scalable methods and resources to audit gender bias across languages.
Humans can quickly learn a new word from a few illustrative examples, and then systematically and flexibly use it in novel contexts. Yet the abilities of current language models for few-shot word learning, and methods for improving these abilities, are underexplored. In this study, we introduce a novel method, Meta-training for IN-context learNing Of Words (Minnow). This method trains language models to generate new examples of a word’s usage given a few in-context examples, using a special placeholder token to represent the new word. This training is repeated on many new words to develop a general word-learning ability. We find that training models from scratch with Minnow on human-scale child-directed language enables strong few-shot word learning, comparable to a large language model (LLM) pre-trained on orders of magnitude more data. Furthermore, through discriminative and generative evaluations, we demonstrate that finetuning pre-trained LLMs with Minnow improves their ability to discriminate between new words, identify syntactic categories of new words, and generate reasonable new usages and definitions for new words, based on one or a few in-context examples. These findings highlight the data efficiency of Minnow and its potential to improve language model performance in word learning tasks.
The integration of external knowledge through Retrieval-Augmented Generation (RAG) has become foundational in enhancing large language models (LLMs) for knowledge-intensive tasks. However, existing RAG paradigms often overlook the cognitive step of applying knowledge, leaving a gap between retrieved facts and task-specific reasoning. In this work, we introduce RAG+, a principled and modular extension that explicitly incorporates application-aware reasoning into the RAG pipeline. RAG+ constructs a dual corpus consisting of knowledge and aligned application examples, created either manually or automatically, and jointly retrieves both during inference. This design enables LLMs not only to access relevant information but also to apply it within structured, goal-oriented reasoning processes. Experiments across mathematical, law, and medical domains, conducted on multiple models, demonstrate that RAG+ consistently outperforms standard RAG variants, achieving average improvements of 3–5%, and peak gains up to 13.5% in complex scenarios. By bridging retrieval with actionable application, RAG+ advances a more cognitively grounded framework for knowledge integration, representing a step toward more interpretable and capable LLMs.
Relational databases are central to modern data management, yet most data exists in unstructured forms like text documents. To bridge this gap, we leverage large language models (LLMs) to automatically synthesize a relational database by generating its schema and populating its tables from raw text. We introduce SQUiD, a novel neurosymbolic framework that decomposes this task into four stages, each with specialized techniques. Our experiments show that SQUiD consistently outperforms baselines across diverse datasets. Our code and datasets are publicly available at: https://github.com/Mushtari-Sadia/SQUiD.
With the development of large language models, their ability to follow simple instructions has significantly improved. However, adhering to complex instructions remains a major challenge. Current approaches to generating complex instructions are often irrelevant to the current instruction requirements or suffer from limited scalability and diversity. Moreover, methods such as back-translation, while effective for simple instruction generation, fail to leverage the rich knowledge and formatting in human written documents. In this paper, we propose a novel **A**utomatic **I**terative **R**efinement (**AIR**) framework to generate complex instructions with constraints, which not only better reflects the requirements of real scenarios but also significantly enhances LLMs’ ability to follow complex instructions. The AIR framework consists of two stages: 1) Generate an initial instruction from a document; 2) Iteratively refine instructions with LLM-as-judge guidance by comparing the model’s output with the document to incorporate valuable constraints. Finally, we construct the AIR-10K dataset with 10K complex instructions and demonstrate that instructions generated with our approach significantly improve the model’s ability to follow complex instructions, outperforming existing methods for instruction generation.
Orthogonal finetuning (OFT) offers highly parameter-efficient adaptation while preventing catastrophic forgetting, but its high runtime and memory demands limit practical deployment. We identify the core computational bottleneck in OFT as its weight-centric implementation, which relies on costly matrix-matrix multiplications with cubic complexity. To overcome this, we propose OFTv2, an input-centric reformulation that instead uses matrix-vector multiplications (i.e., matrix-free computation), reducing the computational cost to quadratic. We further introduce the Cayley–Neumann parameterization, an efficient orthogonal parameterization that approximates the matrix inversion in the Cayley transform via a truncated Neumann series. These modifications allow OFTv2 to achieve up to 10x faster training and 3x lower GPU memory usage without compromising performance. In addition, we extend OFTv2 to support finetuning quantized foundation models and show that it outperforms the popular QLoRA in training stability, efficiency, and memory usage.
Beyond Human Labels: A Multi-Linguistic Auto-Generated Benchmark for Evaluating Large Language Models on Resume Parsing
PDF ↗Efficient resume parsing is critical for global hiring, yet the absence of dedicated benchmarks for evaluating large language models (LLMs) on multilingual, structure-rich resumes hinders progress. To address this, we introduce ResumeBench, the first privacy-compliant benchmark comprising 2,500 synthetic resumes spanning 50 templates, 30 career fields, and 5 languages. These resumes are generated through a human-in-the-loop pipeline that prioritizes realism, diversity, and privacy compliance, which are validated against real-world resumes. This paper evaluates 24 state-of-the-art LLMs on ResumeBench, revealing substantial variations in handling resume complexities. Specifically, top-performing models like GPT-4o exhibit challenges in cross-lingual structural alignment while smaller models show inconsistent scaling effects. Code-specialized LLMs underperform relative to generalists, while JSON outputs enhance schema compliance but fail to address semantic ambiguities. Our findings underscore the necessity for domain-specific optimization and hybrid training strategies to enhance structural and contextual reasoning in LLMs.
Can Large Language Models Outperform Non-Experts in Poetry Evaluation? A Comparative Study Using the Consensual Assessment Technique
PDF ↗This study adapts the Consensual Assessment Technique (CAT) for Large Language Models (LLMs), introducing a novel methodology for poetry evaluation. Using a 90-poem dataset with a ground truth based on publication venue, we demonstrate that this approach allows LLMs to significantly surpass the performance of non-expert human judges. Our method, which leverages forced-choice ranking within small, randomized batches, enabled Claude-3-Opus to achieve a Spearman’s Rank Correlation of 0.87 with the ground truth, dramatically outperforming the best human non-expert evaluation (SRC = 0.38). The LLM assessments also exhibited high inter-rater reliability, underscoring the methodology’s robustness. These findings establish that LLMs, when guided by a comparative framework, can be effective and reliable tools for assessing poetry, paving the way for their broader application in other creative domains.
Enhancing Logical Reasoning in Language Models via Symbolically-Guided Monte Carlo Process Supervision
PDF ↗Large language models (LLMs) have shown strong performance in many reasoning benchmarks. However, recent studies have pointed to memorization, rather than generalization, as one of the leading causes for such performance. LLMs, in fact, are susceptible to content variations, demonstrating a lack of robust planning or symbolic abstractions supporting their reasoning process. To improve reliability, many attempts have been made to combine LLMs with symbolic methods. Nevertheless, existing approaches fail to effectively leverage symbolic representations due to the challenges involved in developing reliable and scalable verification mechanisms. In this paper, we propose to overcome such limitations by synthesizing high-quality symbolic reasoning trajectories with stepwise pseudo-labels at scale via Monte Carlo estimation. A Process Reward Model (PRM) can be efficiently trained based on the synthesized data and then used to select more symbolic trajectories. The trajectories are then employed with Direct Preference Optimization (DPO) and Supervised Fine-Tuning (SFT) to improve logical reasoning and generalization. Our results on benchmarks (i.e., FOLIO and LogicAsker) show the effectiveness of the proposed method with gains on frontier and open-weight models. Moreover, additional experiments on claim verification data reveal that fine-tuning on the generated symbolic reasoning trajectories enhances out-of-domain generalizability, suggesting the potential impact of the proposed method in enhancing planning and logical reasoning.
Reading fluency in any language requires accurate word decoding but also natural prosodic phrasing i.e the grouping of words into rhythmically and syntactically coherent units. This holds for, both, reading aloud and silent reading. While adults pause meaningfully at clause or punctuation boundaries, children aged 8-13 often insert inappropriate pauses due to limited breath control and underdeveloped prosodic awareness. We present a text-based model to predict cognitively appropriate pause locations in children’s reading material. Using a curated dataset of 54 leveled English stories annotated for potential pauses, or prosodic boundaries, by 21 fluent speakers, we find that nearly 30% of pauses occur at non-punctuation locations of the text, highlighting the limitations of using only punctuation-based cues. Our model combines lexical, syntactic, and contextual features with a novel breath duration feature that captures syllable load since the last major boundary. This cognitively motivated approach can model both allowed and “forbidden” pauses. The proposed framework supports applications such as child-directed TTS and oral reading fluency assessment where the proper grouping of words is considered critical to reading comprehension.
Text2Vis: A Challenging and Diverse Benchmark for Generating Multimodal Visualizations from Text
PDF ↗Automated data visualization plays a crucial role in simplifying data interpretation, enhancing decision-making, and improving efficiency. While large language models (LLMs) have shown promise in generating visualizations from natural language, the absence of comprehensive benchmarks limits the rigorous evaluation of their capabilities. We introduce Text2Vis, a benchmark designed to assess text-to-visualization models, covering 20+ chart types and diverse data science queries, including trend analysis, correlation, outlier detection, and predictive analytics. It comprises 1,985 samples, each with a data table, natural language query, short answer, visualization code, and annotated charts. The queries involve complex reasoning, conversational turns, and dynamic data retrieval. We benchmark 11 open-source and closed-source models, revealing significant performance gaps, highlighting key challenges, and offering insights for future advancements. To close this gap, we propose the first cross-modal actor-critic agentic framework that jointly refines the textual answer and visualization code, increasing GPT-4o’s pass rate from 26% to 42% over the direct approach and improving chart quality. We also introduce an automated LLM-based evaluation framework that enables scalable assessment across thousands of samples without human annotation, measuring answer correctness, code execution success, visualization readability, and chart accuracy. We release Text2Vis at <redacted>.
Knowledge Graph Foundation Models (KGFMs) have shown promise in enabling zero-shot reasoning over unseen graphs by learning transferable patterns. However, most existing KGFMs rely solely on graph structure, overlooking the rich semantic signals encoded in textual attributes. We introduce SEMMA, a dual-module KGFM that systematically integrates transferable textual semantics alongside structure. SEMMA leverages Large Language Models (LLMs) to enrich relation identifiers, generating semantic embeddings that subsequently form a textual relation graph, which is fused with the structural component. Across 54 diverse KGs, SEMMA outperforms purely structural baselines like ULTRA in fully inductive link prediction. Crucially, we show that in more challenging generalization settings, where the test-time relation vocabulary is entirely unseen, structural methods collapse while SEMMA is 2x more effective. Our findings demonstrate that textual semantics are critical for generalization in settings where structure alone fails, highlighting the need for foundation models that unify structural and linguistic signals in knowledge reasoning.
Confidence estimation is essential for the reliable deployment of large language models (LLMs). Existing methods are primarily designed for factual QA tasks and often fail to generalize to reasoning tasks. To address this gap, we propose a set of training-free, graph-based confidence estimation methods tailored to reasoning tasks. Our approach models reasoning paths as directed graphs and estimates confidence by exploiting graph properties such as centrality, path convergence, and path weighting. Experiments with two LLMs on three reasoning datasets demonstrate improved confidence estimation and enhanced performance on two downstream tasks.
Automatically assessing handwritten mathematical solutions is an important problem in educational technology with practical applications, but remains a significant challenge due to the diverse formats, unstructured layouts, and symbolic complexity of student work. To address this challenge, we introduce VEHME-a \textbf{V}ision-Language Model for \textbf{E}valuating \textbf{H}andwritten \textbf{M}athematics \textbf{E}xpressions—designed to assess open-form handwritten math responses with high accuracy and interpretable reasoning traces. VEHME integrates a two-phase training pipeline: (i) supervised fine-tuning using structured reasoning data, and (ii) reinforcement learning that aligns model outputs with multi-dimensional grading objectives, including correctness, reasoning depth, and error localization. To enhance spatial understanding, we propose an Expression-Aware Visual Prompting Module, trained on our synthesized multi-line math expressions dataset to robustly guide attention in visually heterogeneous inputs. Evaluated on AIHub and FERMAT datasets, VEHME achieves state-of-the-art performance among open-source models and approaches the accuracy of proprietary systems, demonstrating its potential as a scalable and accessible tool for automated math assessment. Our training and experiment code is publicly available at our GitHub repository.
The Transfer Neurons Hypothesis: An Underlying Mechanism for Language Latent Space Transitions in Multilingual LLMs
PDF ↗Recent studies have suggested a processing framework for multilingual inputs in decoder-based LLMs: early layers convert inputs into English-centric and language-agnostic representations; middle layers perform reasoning within an English-centric latent space; and final layers generate outputs by transforming these representations back into language-specific latent spaces.However, the internal dynamics of such transformation and the underlying mechanism remain underexplored.Towards a deeper understanding of this framework, we propose and empirically validate **The Transfer Neurons Hypothesis**: certain neurons in the MLP module are responsible for transferring representations between language-specific latent spaces and a shared semantic latent space.Furthermore, we show that one function of language-specific neurons, as identified in recent studies, is to facilitate movement between latent spaces.Finally, we show that transfer neurons are critical for reasoning in multilingual LLMs
Label Set Optimization via Activation Distribution Kurtosis for Zero-Shot Classification with Generative Models
PDF ↗In-context learning (ICL) performance is highly sensitive to prompt design, yet the impact of class label options (e.g. lexicon or order) in zero-shot classification remains underexplored. This study proposes LOADS (Label set Optimization via Activation Distribution kurtosiS), a post-hoc method for selecting optimal label sets in zero-shot ICL with large language models (LLMs).LOADS is built upon the observations in our empirical analysis, the first to systematically examine how label option design (i.e., lexical choice, order, and elaboration) impacts classification performance. This analysis shows that the lexical choice of the labels in the prompt (such as agree vs. support in stance classification) plays an important role in both model performance and model’s sensitivity to the label order. A further investigation demonstrates that optimal label words tend to activate fewer outlier neurons in LLMs’ feed-forward networks. LOADS then leverages kurtosis to measure the neuron activation distribution for label selection, requiring only a single forward pass without gradient propagation or labelled data. The LOADS-selected label words consistently demonstrate effectiveness for zero-shot ICL across classification tasks, datasets, models and languages, achieving maximum performance gain from 0.54 to 0.76 compared to the conventional approach of using original dataset label words.
Minimum Bayes Risk (MBR) decoding has seen renewed interest as an alternative to traditional generation strategies. While MBR has proven effective in machine translation, where the variability of a language model’s outcome space is naturally constrained, it may face challenges in more open-ended tasks such as dialogue or instruction-following. We hypothesise that in such settings, applying MBR with standard similarity-based utility functions may result in selecting responses that are broadly representative of the model’s distribution, yet sub-optimal with respect to any particular grouping of generations that share an underlying latent structure. In this work, we introduce three lightweight adaptations to the utility function, designed to make MBR more sensitive to structural variability in the outcome space. To test our hypothesis, we curate a dataset capturing three representative types of latent structure—dialogue act, emotion, and response structure (e.g., a sentence, a paragraph, or a list)—and we propose two metrics to evaluate the structural optimality of MBR. Our analysis demonstrates that common similarity-based utility functions fall short by these metrics. In contrast, our proposed adaptations considerably improve structural optimality. Finally, we evaluate our approaches on real-world instruction-following benchmarks, AlpacaEval and MT-Bench, and show that increased structural sensitivity improves generation quality by up to 13.7 percentage points in win rate.
Extractive Fact Decomposition for Interpretable Natural Language Inference in one Forward Pass
PDF ↗Recent works in Natural Language Inference (NLI) and related tasks, such as automated fact-checking, employ atomic fact decomposition to enhance interpretability and robustness. For this, existing methods rely on resource-intensive generative large language models (LLMs) to perform decomposition. We propose JEDI, an encoder-only architecture that jointly performs extractive atomic fact decomposition and interpretable inference without requiring generative models during inference. To facilitate training, we produce a large corpus of synthetic rationales covering multiple NLI benchmarks. Experimental results demonstrate that JEDI achieves competitive accuracy in distribution and significantly improves robustness out of distribution and in adversarial settings over models based solely on extractive rationale supervision. Our findings show that interpretability and robust generalization in NLI can be realized using encoder-only architectures and synthetic rationales.