EMNLP 2024 Papers

While recent preference alignment algorithms for language models have demonstrated promising results, supervised fine-tuning (SFT) remains imperative for achieving successful convergence. In this paper, we revisit SFT in the context of preference alignment, emphasizing that a minor penalty for the disfavored style is sufficient for preference alignment. Building on this foundation, we introduce a straightforward reference model-free monolithic odds ratio preference optimization algorithm, ORPO, eliminating the need for an additional preference alignment phase. We demonstrate, both empirically and theoretically, that the odds ratio is a sensible choice for contrasting favored and disfavored styles during SFT across diverse sizes from 125M to 7B. Specifically, fine-tuning Phi-2 (2.7B), Llama-2 (7B), and Mistral (7B) with ORPO on the UltraFeedback alone surpasses the performance of state-of-the-art language models including Llama-2 Chat and Zephyr with more than 7B and 13B parameters: achieving up to 12.20% on AlpacaEval 2.0 (Figure 1), and 7.32 in MT-Bench (Table 2). We release code and model checkpoints for Mistral-ORPO-𝛼 (7B) and Mistral-ORPO-𝛽 (7B).

We introduce OMG-QA, a new resource for question answering that is designed to evaluate the effectiveness of question answering systems that perform retrieval augmented generation (RAG) in scenarios that demand reasoning on multi-modal, multi-document contexts. These systems, given a user query, must retrieve relevant contexts from the web, which may include non-textual information, and then reason and synthesize these contents to generate a detailed, coherent answer. Unlike existing open-domain QA datasets, OMG-QA requires systems to navigate and integrate diverse modalities and a broad pool of information sources, making it uniquely challenging. We conduct a thorough evaluation and analysis of a diverse set of QA systems, featuring various retrieval frameworks, document retrievers, document indexing approaches, evidence retrieval methods, and LLMs tasked with both information retrieval and generation. Our findings reveal significant limitations in existing approaches using RAG or LLM agents to address open questions that require long-form answers supported by multi-modal evidence. We believe that OMG-QA will be a valuable resource for developing QA systems that are better equipped to handle open-domain, multi-modal information-seeking tasks.

To meet application needs, event extraction has shifted from simple entities to unconventional entities serving as event arguments. However, current corpora with unconventional entities as event arguments are limited in event types and lack rich multi-events and shared arguments. Financial commentary not only describes the basic elements of an event but also states the background, scope, manner, condition, result, and tool used for the event, as well as the tense, intensity, and emotions of actions or state changes. Therefore, it is not suitable to develop event types that include only a few specific roles, as these cannot comprehensively capture the event’s semantics. Also, there are affluent complex entities serving as event arguments, multiple events, and shared event arguments. To advance the practicality of event extraction technology, this paper first develops a general open event template from the perspective of understanding the meaning of events, aiming to comprehensively reveal useful information about events. This template includes 21 event argument roles, divided into three categories: core event roles, situational event roles, and adverbial roles. Then, based on the constructed event template, Chinese financial commentaries are collected and manually annotated to create a corpus OEE-CFC supporting open event extraction. This corpus includes 17,469 events, 44,221 arguments, 3,644 complex arguments, and 5,898 shared arguments. Finally, based on the characteristics of OEE-CFC, we design four types of prompts, and two models for event argument extraction are developed, with experiments conducted on the prompts.

Warning: Contents of this paper may be upsetting.Public attitudes towards key societal issues, expressed on online media, are of immense value in policy and reform efforts, yet challenging to understand at scale. We study one such social issue: homelessness in the U.S., by leveraging the remarkable capabilities of large language models to assist social work experts in analyzing millions of posts from Twitter. We introduce a framing typology: Online Attitudes Towards Homelessness (OATH) Frames: nine hierarchical frames capturing critiques, responses and perceptions. We release annotations with varying degrees of assistance from language models, with immense benefits in scaling: 6.5× speedup in annotation time while only incurring a 3 point F1 reduction in performance with respect to the domain experts. Our experiments demonstrate the value of modeling OATH-Frames over existing sentiment and toxicity classifiers. Our large-scale analysis with predicted OATH-Frames on 2.4M posts on homelessness reveal key trends in attitudes across states, time periods and vulnerable populations, enabling new insights on the issue. Our work provides a general framework to understand nuanced public attitudes at scale, on issues beyond homelessness.

Language models struggle with handling numerical data and performing arithmetic operations. We hypothesize that this limitation can be partially attributed to non-intuitive textual numbers representation. When a digit is read or generated by a causal language model it does not know its place value (e.g. thousands vs. hundreds) until the entire number is processed. To address this issue, we propose a simple adjustment to how numbers are represented by including the count of digits before each number. For instance, instead of “42”, we suggest using “2:42” as the new format. This approach, which we term NumeroLogic, offers an added advantage in number generation by serving as a Chain of Thought (CoT). By requiring the model to consider the number of digits first, it enhances the reasoning process before generating the actual number. We use arithmetic tasks to demonstrate the effectiveness of the NumeroLogic formatting. We further demonstrate NumeroLogic applicability to general natural language modeling, improving language understanding performance in the MMLU benchmark.

Quantitative information plays a crucial role in understanding and interpreting the content of documents. Many user queries contain quantities and cannot be resolved without understanding their semantics, e.g., “car that costs less than $10k”. Yet, modern search engines apply the same ranking mechanisms for both words and quantities, overlooking magnitude and unit information. In this paper, we introduce two quantity-aware ranking techniques designed to rank both the quantity and textual content either jointly or independently. These techniques incorporate quantity information in available retrieval systems and can address queries with numerical conditions equal, greater than, and less than. To evaluate the effectiveness of our proposed models, we introduce two novel quantity-aware benchmark datasets in the domains of finance and medicine and compare our method against various lexical and neural models. The code and data are available under https://github.com/satya77/QuantityAwareRankers.

This paper presents a series of investigations into an interesting phenomenon where we observe performance increases in large language models (LLMs) when providing a prompt that causes and exploits hallucination. We propose null-shot prompting, a counter-intuitive approach where we intentionally instruct LLMs to look at and utilize information from a null section. We investigate null-shot prompting on a wide range of tasks, including arithmetic reasoning, commonsense reasoning, and reading comprehension. We observe a substantial increase in performance in arithmetic reasoning tasks for various models, with up to a 44.62% increase compared to a baseline in one model. Therefore, we investigate deeper into this task by utilizing a more challenging mathematics problem-solving benchmark. We observe that LLMs benefit from hallucination in null-shot prompting in this task and discuss the mathematical topics that benefit the most from introducing hallucination in the prompt. We continue our investigation by evaluating hallucination detection abilities of the LLMs when using null-shot prompting. We find surprising results where hallucination in prompts can improve hallucination detection abilities of many LLMs. We also examine the effects of introducing both reasoning, which is known to mitigate hallucination, and hallucination simultaneously in the prompt and observe another surprising turn for the mathematics problem-solving benchmark with many performance improvements. We hope this paper will spark more interest, investigations, and discussions on how hallucination in prompts LLMs and even bolsters them in certain cases.

Large Language Models (LLMs) have shown impressive abilities in data annotation, opening the way for new approaches to solve classic NLP problems. In this paper, we show how to use LLMs to create NuNER, a compact language representation model specialized in the Named Entity Recognition (NER) task. NuNER can be fine-tuned to solve downstream NER problems in a data-efficient way, outperforming similar-sized foundation models in the few-shot regime and competing with much larger LLMs. We find that the size and entity-type diversity of the pre-training dataset are key to achieving good performance. We view NuNER as a member of the broader family of task-specific foundation models, recently unlocked by LLMs. NuNER and NuNER’s dataset are open-sourced with MIT License.

Large language models (LLMs) are still struggling in aligning with human preference in complex tasks and scenarios. They are prone to overfit into the unexpected patterns or superficial styles in the training data. We conduct an empirical study that only selects the top-10% most updated parameters in LLMs for alignment training, and see improvements in the convergence process and final performance. It indicates the existence of redundant neurons in LLMs for alignment training. To reduce its influence, we propose a low-redundant alignment method named **ALLO**, focusing on optimizing the most related neurons with the most useful supervised signals. Concretely, we first identify the neurons that are related to the human preference data by a gradient-based strategy, then identify the alignment-related key tokens by reward models for computing loss. Besides, we also decompose the alignment process into the forgetting and learning stages, where we first forget the tokens with unaligned knowledge and then learn aligned knowledge, by updating different ratios of neurons, respectively. Experimental results on 10 datasets have shown the effectiveness of ALLO. Our code and data will be publicly released.

Iterative preference learning, though yielding superior performances, requires online annotated preference labels. In this work, we study strategies to save annotation budgets while achieving competitive or even better performances for iterative preference learning. Built on intuitions from active learning, we empirically show that annotating those response pairs with small margins is generally better than large or random. Besides, experiments under the multi-iteration scenario suggest allocating more annotation budgets in the earlier iterations rather than later ones.

The rapid development of large language models has led to the widespread adoption of Retrieval-Augmented Generation (RAG), which integrates external knowledge to alleviate knowledge bottlenecks and mitigate hallucinations. However, the existing RAG paradigm inevitably suffers from the impact of flawed information introduced during the retrieval phrase, thereby diminishing the reliability and correctness of the generated outcomes. In this paper, we propose Credibility-aware Generation (CAG), a universally applicable framework designed to mitigate the impact of flawed information in RAG. At its core, CAG aims to equip models with the ability to discern and process information based on its credibility. To this end, we propose an innovative data transformation framework that generates data based on credibility, thereby effectively endowing models with the capability of CAG. Furthermore, to accurately evaluate the models’ capabilities of CAG, we construct a comprehensive benchmark covering three critical real-world scenarios. Experimental results demonstrate that our model can effectively understand and employ credibility for generation, significantly outperform other models with retrieval augmentation, and exhibit robustness despite the increasing noise in the context.

Visual storytelling consists in generating a natural language story given a temporally ordered sequence of images. This task is not only challenging for models, but also very difficult to evaluate with automatic metrics since there is no consensus about what makes a story ‘good’. In this paper, we introduce a novel method that measures story quality in terms of human likeness regarding three key aspects highlighted in previous work: visual grounding, coherence, and repetitiveness. We then use this method to evaluate the stories generated by several models, showing that the foundation model LLaVA obtains the best result, but only slightly so compared to TAPM, a 50-times smaller visual storytelling model. Upgrading the visual and language components of TAPM results in a model that yields competitive performance with a relatively low number of parameters. Finally, we carry out a human evaluation study, whose results suggest that a ‘good’ story may require more than a human-like level of visual grounding, coherence, and repetition.

With the size and cost of large transformer-based language models growing, recently, there has been interest in shortcut casting of early transformer hidden-representations to final-representations for cheaper model inference. In particular, shortcutting pre-trained transformers with linear transformations over early layers has been shown to improve precision in early inference. However, for large language models, even this becomes computationally expensive. In this work, we propose Narrow Jump to Conclusions (NJTC) and Normalized Narrow Jump to Conclusions (N-NJTC) - parameter efficient alternatives to standard linear shortcutting that reduces shortcut parameter count by over 97%. We show that N-NJTC reliably outperforms Identity shortcuts at early stages and offers stable precision from all transformer block levels for GPT-2-XL, Phi3-Mini and Llama2-7B transformer models, demonstrating the viability of more parameter efficient short-cutting approaches.

In recent years, Large Language Models (LLMs) have demonstrated remarkable capabilities in parsing textual data and generating code. However, their performance in tasks involving tabular data, especially those requiring symbolic reasoning, faces challenges due to the structural variance and inconsistency in table cell values often found in web tables. In this paper, we introduce NormTab, a novel framework aimed at enhancing the symbolic reasoning performance of LLMs by normalizing web tables. We study table normalization as a stand-alone, one-time preprocessing step using LLMs to support symbolic reasoning on tabular data. Our experimental evaluation, conducted on challenging web table datasets such as WikiTableQuestion and TabFact, demonstrates that leveraging NormTab significantly improves symbolic reasoning performance, showcasing the importance and effectiveness of web table normalization for enhancing LLM-based symbolic reasoning tasks.

Available training data for named entity recognition (NER) often contains a significant percentage of incorrect labels for entity types and entity boundaries. Such label noise poses challenges for supervised learning and may significantly deteriorate model quality. To address this, prior work proposed various noise-robust learning approaches capable of learning from data with partially incorrect labels. These approaches are typically evaluated using simulated noise where the labels in a clean dataset are automatically corrupted. However, as we show in this paper, this leads to unrealistic noise that is far easier to handle than real noise caused by human error or semi-automatic annotation. To enable the study of the impact of various types of real noise, we introduce NoiseBench, an NER benchmark consisting of clean training data corrupted with 6 types of real noise, including expert errors, crowdsourcing errors, automatic annotation errors and LLM errors. We present an analysis that shows that real noise is significantly more challenging than simulated noise, and show that current state-of-the-art models for noise-robust learning fall far short of their achievable upper bound. We release NoiseBench for both English and German to the research community.

We present a framework for detecting and categorizing noise in literary texts, demonstrated through its application to Danish and Norwegian literature from the late 19-th century. Noise, understood as “aberrant sonic behaviour,” is not only an auditory phenomenon but also a cultural construct tied to the processes of civilization and urbanization.We begin by utilizing topic modeling techniques to identify noise-related documents, followed by fine-tuning BERT-based language models trained on Danish and Norwegian texts to analyze a corpus of over 800 novels.We identify and track the prevalence of noise in these texts, offering insights into the literary perceptions of noise during the Scandinavian “Modern Breakthrough” period (1870-1899). Our contributions include the development of a comprehensive dataset annotated for noise-related segments and their categorization into human-made, non-human-made, and musical noises. This study illustrates the framework’s potential for enhancing the understanding of the relationship between noise and its literary representations, providing a deeper appreciation of the auditory elements in literary works, including as sources for cultural history.

Research in vision and language has made considerable progress thanks to benchmarks such as COCO. COCO captions focused on unambiguous facts in English; ArtEmis introduced subjective emotions and ArtELingo introduced some multilinguality (Chinese and Arabic). However we believe there should be more multilinguality. Hence, we present ArtELingo-28, a vision-language benchmark that spans 28 languages and encompasses approximately 200,000 annotations (140 annotations per image). Traditionally, vision research focused on unambiguous class labels, whereas ArtELingo-28 emphasizes diversity of opinions over languages and cultures. The challenge is to build machine learning systems that assign emotional captions to images. Baseline results will be presented for three novel conditions: Zero-Shot, Few-Shot and One-vs-All Zero-Shot. We find that cross-lingual transfer is more successful for culturally-related languages. Data and code will be made publicly available.

Credit risk monitoring is an essential process for financial institutions to evaluate the creditworthiness of borrowing entities and minimize potential losses. Traditionally, this involves the periodic assessment of news regarding client companies to identify events which can impact their financial standing. This process can prove arduous and delay a timely response to credit impacting events. The News Risk Alerting System (NRAS) proactively identifies credit-relevant news related to clients and alerts the relevant Credit Officer (CO). This production system has been deployed for nearly three years and has alerted COs to over 2700 credit-relevant events with an estimated precision of 77%.

Identifying important neurons for final predictions is essential for understanding the mechanisms of large language models. Due to computational constraints, current attribution techniques struggle to operate at neuron level. In this paper, we propose a static method for pinpointing significant neurons. Compared to seven other methods, our approach demonstrates superior performance across three metrics. Additionally, since most static methods typically only identify “value neurons” directly contributing to the final prediction, we propose a method for identifying “query neurons” which activate these “value neurons”. Finally, we apply our methods to analyze six types of knowledge across both attention and feed-forward network (FFN) layers. Our method and analysis are helpful for understanding the mechanisms of knowledge storage and set the stage for future research in knowledge editing. The code is available on https://github.com/zepingyu0512/neuron-attribution.