Large language models (LLM) have prioritized expanding the context window from which models can incorporate more information. However, training models to handle long contexts presents significant challenges. These include the scarcity of high-quality natural long-context data, the potential for performance degradation on short-context tasks, and the reduced training efficiency associated with attention mechanisms. In this paper, we introduce Untie the Knots (UtK), a novel data augmentation strategy employed during the continue pre-training phase, designed to efficiently enable LLMs to gain long-context capabilities without the need to modify the existing data mixture. In particular, we chunk the documents, shuffle the chunks, and create a complex and knotted structure of long texts; LLMs are then trained to untie these knots and identify relevant segments within seemingly chaotic token sequences. This approach greatly improves the model’s performance by accurately attending to relevant information in long context and the training efficiency is also largely increased. We conduct extensive experiments on models with 7B and 72B parameters, trained on 20 billion tokens, demonstrating that UtK achieves 75% and 84.5% accurracy on RULER at 128K context length, significantly outperforming other long context strategies. The trained models will open-source for further research.

Large language models (LLMs) combined with instruction tuning have shown significant progress in information extraction (IE) tasks, exhibiting strong generalization capabilities to unseen datasets by following annotation guidelines. However, their applicability to low-resource languages remains limited due to lack of both labeled data for fine-tuning, and unlabeled text for pre-training. In this paper, we propose TransFusion, a framework in which models are fine-tuned to use English translations of low-resource language data, enabling more precise predictions through annotation fusion. Based on TransFusion, we introduce GoLLIE-TF, a cross-lingual instruction-tuned LLM for IE tasks, designed to close the performance gap between high and low-resource languages. Our experiments across twelve multilingual IE datasets spanning 50 languages demonstrate that GoLLIE-TF achieves better cross-lingual transfer over the base model. In addition, we show that TransFusion significantly improves low-resource language named entity recognition when applied to proprietary models such as GPT-4 (+5 F1) with a prompting approach, or fine-tuning different language models including decoder-only (+14 F1) and encoder-only (+13 F1) architectures.

In this paper, we introduce AceMath, a suite of frontier math models that excel in solving complex math problems, along with highly effective reward models capable of evaluating generated solutions and reliably identifying the correct ones. To develop the instruction-tuned math models, we propose a supervised fine-tuning (SFT) process that first achieves competitive performance across general domains, followed by targeted fine-tuning for the math domain using a carefully curated set of prompts and synthetically generated responses. The resulting model, AceMath-72B-Instruct greatly outperforms Qwen2.5-Math-72B-Instruct, GPT-4o and Claude-3.5 Sonnet. To develop math-specialized reward model, we first construct AceMath-RewardBench, a comprehensive and robust benchmark for evaluating math reward models across diverse problems and difficulty levels. After that, we present a systematic approach to build our math reward models. The resulting model, AceMath-72B-RM, consistently outperforms state-of-the-art reward models. Furthermore, when combining AceMath-72B-Instruct with AceMath-72B-RM, we achieve the highest average rm@8 score across the math reasoning benchmarks.

Vision Language Models (VLMs) are rapidly advancing in their capability to answer information-seeking questions. As these models are widely deployed in consumer applications, they could lead to new privacy risks due to emergent abilities to identify people in photos, geolocate images, etc. As we demonstrate, somewhat surprisingly, current open-source and proprietary VLMs are very capable image geolocators, making widespread geolocation with VLMs an immediate privacy risk, rather than merely a theoretical future concern. As a first step to address this challenge, we develop a new benchmark, GPTGeoChat, to test the capability of VLMs to moderate geolocation dialogues with users. We collect a set of 1,000 image geolocation conversations between in-house annotators and GPT-4v, which are annotated with the granularity of location information revealed at each turn. Using this new dataset we evaluate the ability of various VLMs to moderate GPT-4v geolocation conversations by determining when too much location information has been revealed. We find that custom fine-tuned models perform on par with prompted API-based models when identifying leaked location information at the country or city level, however fine-tuning on supervised data appears to be needed to accurately moderate finer granularities, such as the name of a restaurant or building.

Although large language models (LLMs) acquire extensive world knowledge and some reasoning abilities, their proficiency in generating humorous sentences remains a challenge. Previous research has demonstrated that the humor generation capabilities of ChatGPT are confined to producing merely 25 unique jokes. In this work, we concentrate on endowing LLMs with the ability of generating puns, a particular category of humor by preference learning method. We propose a multi-stage curriculum preference learning framework to optimize both pun structure preferences and humor preferences. Specifically, we improve the Direct Preference Optimization (DPO) algorithm to address the challenge of multi-objective alignment problem. Besides, to facilitate further advancement in this field, we collect a Chinese Pun (ChinesePun) dataset, containing 2.1k puns and corresponding annotations. Experimental results on both Chinese and English benchmark datasets demonstrate that our method significantly outperforms all the baseline models.

Topic relevance of an essay demands that the composition adheres to a clear theme and aligns well with the essay prompt requirements, a critical aspect of essay quality evaluation. However, existing research of Automatic Essay Scoring (AES) for Chinese essays has overlooked topic relevance and lacks detailed feedback, while Automatic Essay Comment Generation (AECG) faces much complexity and difficulty. Additionally, current Large Language Models, including GPT-4, often make incorrect judgments and provide overly impractical feedback when evaluating topic relevance. This paper introduces TOREE (Topic Relevance Evaluation), a comprehensive dataset developed to assess topic relevance in Chinese primary and middle school students’ essays, which is beneficial for AES, AECG and other applications. Moreover, our proposed two-step method utilizes TOREE through a combination of Supervised Fine-tuning and Preference Learning. Experimental results demonstrate that TOREE is of high quality, and our method significantly enhances models’ performance on two designed tasks for topic relevance evaluation, improving both automatic and human evaluations across four diverse LLMs.

Combining large language models with logical reasoning enhances their capacity to address problems in a robust and reliable manner. Nevertheless, the intricate nature of logical reasoning poses challenges when gathering reliable data from the web to build comprehensive training datasets, subsequently affecting performance on downstream tasks. To address this, we introduce a novel logic-driven data augmentation approach, AMR-LDA. AMR-LDA converts the original text into an Abstract Meaning Representation (AMR) graph, a structured semantic representation that encapsulates the logical structure of the sentence, upon which operations are performed to generate logically modified AMR graphs. The modified AMR graphs are subsequently converted back into text to create augmented data. Notably, our methodology is architecture-agnostic and enhances both generative large language models, such as GPT-3.5 and GPT-4, through prompt augmentation, and discriminative large language models through contrastive learning with logic-driven data augmentation. Empirical evidence underscores the efficacy of our proposed method with improvement in performance across seven downstream tasks, such as reading comprehension requiring logical reasoning, textual entailment, and natural language inference. Furthermore, our method leads on the ReClor leaderboard at https://eval.ai/web/challenges/challenge-page/503/leaderboard/1347. The source code and data are publicly available at https://github.com/Strong-AI-Lab/Logical-Equivalence-driven-AMR-Data-Augmentation-for-Representation-Learning.

We present a human-in-the-loop evaluation framework for fact-checking novel misinformation claims and identifying social media messages that support them. Our approach extracts check-worthy claims, which are aggregated and ranked for review. Stance classifiers are then used to identify tweets supporting novel misinformation claims, which are further reviewed to determine whether they violate relevant policies. To demonstrate the feasibility of our approach, we develop a baseline system based on modern NLP methods for human-in-the-loop fact-checking in the domain of COVID-19 treatments. We make our data and detailed annotation guidelines available to support the evaluation of human-in-the-loop systems that identify novel misinformation directly from raw user-generated content.

Training machine learning models to successfully perform scientific fact-checking tasks is challenging due to the expertise bottleneck that limits the availability of appropriate training datasets. In this task, models use textual evidence to confirm scientific claims, which requires data that contains extensive domain-expert annotation. Consequently, the number of existing scientific-fact-checking datasets and the sizes of those datasets are limited. However, these limitations do not apply to multiple-choice question datasets because of the necessity of domain exams in the modern education system. As one of the first steps towards addressing the fact-checking dataset scarcity problem in scientific domains, we propose a pipeline for automatically converting multiple-choice questions into fact-checking data, which we call Multi2Claim. By applying the proposed pipeline, we generated two large-scale datasets for scientific-fact-checking tasks: Med-Fact and Gsci-Fact for the medical and general science domains, respectively. These two datasets are among the first examples of large-scale scientific-fact-checking datasets. We developed baseline models for the verdict prediction task using each dataset. Additionally, we demonstrated that the datasets could be used to improve performance with respect to the F 1 weighted metric on existing fact-checking datasets such as SciFact, HEALTHVER, COVID-Fact, and CLIMATE-FEVER. In some cases, the improvement in performance was up to a 26% increase.

Pre-trained vision and language models have demonstrated state-of-the-art capabilities over existing tasks involving images and texts, including visual question answering. However, it remains unclear whether these models possess the capability to answer questions that are not only querying visual content but knowledge-intensive and information-seeking. In this study, we introduce InfoSeek, a visual question answering dataset tailored for information-seeking questions that cannot be answered with only common sense knowledge. Using InfoSeek, we analyze various pre-trained visual question answering models and gain insights into their characteristics. Our findings reveal that state-of-the-art pre-trained multi-modal models (e.g., PaLI-X, BLIP2, InstructBLIP) face challenges in answering visual information-seeking questions, but fine-tuning on the InfoSeek dataset elicits models to use fine-grained knowledge that was learned during pre-training. Furthermore, we show that accurate visual entity recognition can be used to improve performance on InfoSeek by retrieving relevant documents, showing a significant space for improvement.

Translating training data into many languages has emerged as a practical solution for improving cross-lingual transfer. For tasks that involve span-level annotations, such as information extraction or question answering, an additional label projection step is required to map annotated spans onto the translated texts. Recently, a few efforts have utilized a simple mark-then-translate method to jointly perform translation and projection by inserting special markers around the labeled spans in the original sentence. However, as far as we are aware, no empirical analysis has been conducted on how this approach compares to traditional annotation projection based on word alignment. In this paper, we present an extensive empirical study across 57 languages and three tasks (QA, NER, and Event Extraction) to evaluate the effectiveness and limitations of both methods, filling an important gap in the literature. Experimental results show that our optimized version of mark-then-translate, which we call EasyProject, is easily applied to many languages and works surprisingly well, outperforming the more complex word alignment-based methods. We analyze several key factors that affect the end-task performance, and show EasyProject works well because it can accurately preserve label span boundaries after translation. We will publicly release all our code and data.

Attention describes cognitive processes that are important to many human phenomena including reading. The term is also used to describe the way in which transformer neural networks perform natural language processing. While attention appears to be very different under these two contexts, this paper presents an analysis of the correlations between transformer attention and overt human attention during reading tasks. An extensive analysis of human eye tracking datasets showed that the dwell times of human eye movements were strongly correlated with the attention patterns occurring in the early layers of pre-trained transformers such as BERT. Additionally, the strength of a correlation was not related to the number of parameters within a transformer. This suggests that something about the transformers’ architecture determined how closely the two measures were correlated.

Multi-hop question answering (QA) requires reasoning over multiple documents to answer a complex question and provide interpretable supporting evidence. However, providing supporting evidence is not enough to demonstrate that a model has performed the desired reasoning to reach the correct answer. Most existing multi-hop QA methods fail to answer a large fraction of sub-questions, even if their parent questions are answered correctly. In this paper, we propose the Prompt-based Conservation Learning (PCL) framework for multi-hop QA, which acquires new knowledge from multi-hop QA tasks while conserving old knowledge learned on single-hop QA tasks, mitigating forgetting. Specifically, we first train a model on existing single-hop QA tasks, and then freeze this model and expand it by allocating additional sub-networks for the multi-hop QA task. Moreover, to condition pre-trained language models to stimulate the kind of reasoning required for specific multi-hop questions, we learn soft prompts for the novel sub-networks to perform type-specific reasoning. Experimental results on the HotpotQA benchmark show that PCL is competitive for multi-hop QA and retains good performance on the corresponding single-hop sub-questions, demonstrating the efficacy of PCL in mitigating knowledge loss by forgetting.

Transformers that are pre-trained on multilingual corpora, such as, mBERT and XLM-RoBERTa, have achieved impressive cross-lingual transfer capabilities. In the zero-shot transfer setting, only English training data is used, and the fine-tuned model is evaluated on another target language. While this works surprisingly well, substantial variance has been observed in target language performance between different fine-tuning runs, and in the zero-shot setup, no target-language development data is available to select among multiple fine-tuned models. Prior work has relied on English dev data to select among models that are fine-tuned with different learning rates, number of steps and other hyperparameters, often resulting in suboptimal choices. In this paper, we show that it is possible to select consistently better models when small amounts of annotated data are available in auxiliary pivot languages. We propose a machine learning approach to model selection that uses the fine-tuned model’s own internal representations to predict its cross-lingual capabilities. In extensive experiments we find that this method consistently selects better models than English validation data across twenty five languages (including eight low-resource languages), and often achieves results that are comparable to model selection using target language development data.

Multilingual pre-trained Transformers, such as mBERT (Devlin et al., 2019) and XLM-RoBERTa (Conneau et al., 2020a), have been shown to enable effective cross-lingual zero-shot transfer. However, their performance on Arabic information extraction (IE) tasks is not very well studied. In this paper, we pre-train a customized bilingual BERT, dubbed GigaBERT, that is designed specifically for Arabic NLP and English-to-Arabic zero-shot transfer learning. We study GigaBERT’s effectiveness on zero-short transfer across four IE tasks: named entity recognition, part-of-speech tagging, argument role labeling, and relation extraction. Our best model significantly outperforms mBERT, XLM-RoBERTa, and AraBERT (Antoun et al., 2020) in both the supervised and zero-shot transfer settings. We have made our pre-trained models publicly available at: https://github.com/lanwuwei/GigaBERT.

Rich entity representations are useful for a wide class of problems involving entities. Despite their importance, there is no standardized benchmark that evaluates the overall quality of entity representations. In this work, we propose EntEval: a test suite of diverse tasks that require nontrivial understanding of entities including entity typing, entity similarity, entity relation prediction, and entity disambiguation. In addition, we develop training techniques for learning better entity representations by using natural hyperlink annotations in Wikipedia. We identify effective objectives for incorporating the contextual information in hyperlinks into state-of-the-art pretrained language models (Peters et al., 2018) and show that they improve strong baselines on multiple EntEval tasks.

We introduce entity post-modifier generation as an instance of a collaborative writing task. Given a sentence about a target entity, the task is to automatically generate a post-modifier phrase that provides contextually relevant information about the entity. For example, for the sentence, “Barack Obama, _______, supported the #MeToo movement.”, the phrase “a father of two girls” is a contextually relevant post-modifier. To this end, we build PoMo, a post-modifier dataset created automatically from news articles reflecting a journalistic need for incorporating entity information that is relevant to a particular news event. PoMo consists of more than 231K sentences with post-modifiers and associated facts extracted from Wikidata for around 57K unique entities. We use crowdsourcing to show that modeling contextual relevance is necessary for accurate post-modifier generation. We adapt a number of existing generation approaches as baselines for this dataset. Our results show there is large room for improvement in terms of both identifying relevant facts to include (knowing which claims are relevant gives a >20% improvement in BLEU score), and generating appropriate post-modifier text for the context (providing relevant claims is not sufficient for accurate generation). We conduct an error analysis that suggests promising directions for future research.