论文检索

Jiali Cheng, Hadi Amiri

Language models are prone to dataset biases, known as shortcuts and spurious correlations in data, which often result in performance drop on new data. We present a new debiasing framework called FairFlow that mitigates dataset biases by learning to be undecided in its predictions for data samples or representations associated with known or unknown biases. The framework introduces two key components: a suite of data and model perturbation operations that generate different biased views of input samples, and a contrastive objective that learns debiased and robust representations from the resulting biased views of samples. Experiments show that FairFlow outperforms existing debiasing methods, particularly against out-of-domain and hard test samples without compromising the in-domain performance.

Giovanni Franco Gabriel Marraffini, Andrés Cotton, Noe Fabian Hsueh, Axel Fridman, Juan Wisznia, Luciano Del Corro

The question of how to make decisions that maximise the well-being of all persons is very relevant to design language models that are beneficial to humanity and free from harm. We introduce the Greatest Good Benchmark to evaluate the moral judgments of LLMs using utilitarian dilemmas. Our analysis across 15 diverse LLMs reveals consistently encoded moral preferences that diverge from established moral theories and lay population moral standards. Most LLMs have a marked preference for impartial beneficence and rejection of instrumental harm. These findings showcase the ‘artificial moral compass’ of LLMs, offering insights into their moral alignment.

Daniil Larionov, Mikhail Seleznyov, Vasiliy Viskov, Alexander Panchenko, Steffen Eger

State-of-the-art trainable machine translation evaluation metrics like xCOMET achieve high correlation with human judgment but rely on large encoders (up to 10.7B parameters), making them computationally expensive and inaccessible to researchers with limited resources. To address this issue, we investigate whether the knowledge stored in these large encoders can be compressed while maintaining quality. We employ distillation, quantization, and pruning techniques to create efficient xCOMET alternatives and introduce a novel data collection pipeline for efficient black-box distillation. Our experiments show that, using quantization, xCOMET can be compressed up to three times with no quality degradation. Additionally, through distillation, we create an 278M-sized xCOMET-lite metric, which has only 2.6% of xCOMET-XXL parameters, but retains 92.1% of its quality. Besides, it surpasses strong small-scale metrics like COMET-22 and BLEURT-20 on the WMT22 metrics challenge dataset by 6.4%, despite using 50% fewer parameters. All code, dataset, and models are available online.

Youxiang Zhu, Nana Lin, Kiran Sandilya Balivada, Daniel Haehn, Xiaohui Liang

Although large language models (LLMs) excel in various text classification tasks, regular prompting strategies (e.g., few-shot prompting) do not work well with dementia detection via picture description. The challenge lies in the language marks for dementia are unclear, and LLM may struggle with relating its internal knowledge to dementia detection. In this paper, we present an accurate and interpretable classification approach by Adversarial Text Generation (ATG), a novel decoding strategy that could relate dementia detection with other tasks. We further develop a comprehensive set of instructions corresponding to various tasks and use them to guide ATG, achieving the best accuracy of 85%, >10% improvement compared to the regular prompting strategies. In addition, we introduce feature context, a human-understandable text that reveals the underlying features of LLM used for classifying dementia. From feature contexts, we found that dementia detection can be related to tasks such as assessing attention to detail, language, and clarity with specific features of the environment, character, and other picture content or language-related features. Future work includes incorporating multi-modal LLMs to interpret speech and picture information.

Harbani Jaggi, Kashyap Coimbatore Murali, Eve Fleisig, Erdem Biyik

When annotators disagree, predicting the labels given by individual annotators can capture nuances overlooked by traditional label aggregation. We introduce three approaches to predict individual annotator ratings on the toxicity of text by incorporating individual annotator-specific information: a neural collaborative filtering (NCF) approach, an in-context learning (ICL) approach, and an intermediate embedding-based architecture. We also study the utility of demographic information for rating prediction. NCF showed limited utility; however, integrating annotator history, demographics, and survey information permits both the embedding-based architecture and ICL to substantially improve prediction accuracy, with the embedding-based architecture outperforming the other methods. We also find that, if demographics are predicted from survey information, using these imputed demographics as features performs comparably to using true demographic data. This suggests that demographics may not provide substantial information for modeling ratings beyond what is captured in survey responses. Our findings raise considerations about the relative utility of different types of annotator information and provide new approaches for modeling annotators in subjective NLP tasks.

Somanshu Singla, Zhen Wang, Tianyang Liu, Abdullah Ashfaq, Zhiting Hu, Eric P. Xing

Aligning Large Language Models (LLMs) traditionally relies on complex and costly training processes like supervised fine-tuning (SFT) and reinforcement learning from human feedback (RLHF). To address the challenge of achieving alignment without these extensive tuning costs and expensive annotations, we present a novel, tuning-free approach for self-alignment called Dynamic Rewarding with Prompt Optimization (DRPO). Our approach enables self-alignment through a search-based prompt optimization framework, allowing the model to self-improve and generate optimized prompts without additional training or human supervision. The core of DRPO leverages a dynamic rewarding mechanism to identify and rectify model-specific alignment weaknesses, enabling LLMs to adapt quickly to various alignment challenges. Empirical evaluations on eight recent LLMs, including both open- and closed-source, reveal that DRPO significantly enhances alignment performance, enabling base models to outperform their SFT/RLHF-tuned counterparts. Moreover, DRPO’s automatically optimized prompts surpass those curated by human experts, demonstrating its superior alignment capabilities. Our findings envision a highly cost-effective and adaptable solution for future alignment research to be further explored.

Negar Arabzadeh, Siqing Huo, Nikhil Mehta, Qingyun Wu, Chi Wang, Ahmed Hassan Awadallah, Charles L. A. Clarke, Julia Kiseleva

The rapid development of Large Language Models (LLMs) has led to a surge in applications that facilitate collaboration among multiple agents, assisting humans in their daily tasks. However, a significant gap remains in assessing to what extent LLM-powered applications genuinely enhance user experience and task execution efficiency. This highlights the need to verify utility of LLM-powered applications, particularly by ensuring alignment between the application’s functionality and end-user needs. We introduce AgentEval, a novel framework designed to simplify the utility verification process by automatically proposing a set of criteria tailored to the unique purpose of any given application. This allows for a comprehensive assessment, quantifying the utility of an application against the suggested criteria. We present a comprehensive analysis of the effectiveness and robustness of AgentEval for two open source datasets including Math Problem solving and ALFWorld House-hold related tasks. For reproducibility purposes, we make the data, code and all the logs publicly available at https://github.com/Narabzad/AgentEval

HyoJung Han, Kevin Duh, Marine Carpuat

Recent advances in automatic quality estimation for machine translation have exclusively focused on written language, leaving the speech modality underexplored. In this work, we formulate the task of quality estimation for speech translation (SpeechQE), construct a benchmark, and evaluate a family of systems based on cascaded and end-to-end architectures. In this process, we introduce a novel end-to-end system leveraging pre-trained text LLM. Results suggest that end-to-end approaches are better suited to estimating the quality of direct speech translation than using quality estimation systems designed for text in cascaded systems. More broadly, we argue that quality estimation of speech translation needs to be studied as a separate problem from that of text, and release our [data and models](https://github.com/h-j-han/SpeechQE) to guide further research in this space.

Björn Deiseroth, Manuel Brack, Patrick Schramowski, Kristian Kersting, Samuel Weinbach

Tokenizers are crucial for encoding information in Large Language Models, but their development has recently stagnated, and they contain inherent weaknesses. Major limitations include computational overhead, ineffective vocabulary use, and unnecessarily large embedding and head layers. Additionally, their performance is biased towards a reference corpus, leading to reduced effectiveness for underrepresented languages.To remedy these issues, we propose T-Free, which directly embeds words through sparse activation patterns over character triplets and does not require a reference corpus. T-Free inherently exploits morphological similarities and allows for strong compression of embedding layers. In our exhaustive experimental evaluation, we achieve competitive downstream performance with a parameter reduction of more than 85% on these layers. Further, T-Free shows significant improvements in cross-lingual transfer learning.

Alexander Spangher, Nanyun Peng, Sebastian Gehrmann, Mark Dredze

Journalists engage in multiple steps in the news writing process that depend on human creativity, like exploring different “angles” (i.e. the specific perspectives a reporter takes). These can potentially be aided by large language models (LLMs). By affecting planning decisions, such interventions can have an outsize impact on creative output. We advocate a careful approach to evaluating these interventions to ensure alignment with human values.In a case study of journalistic coverage of press releases, we assemble a large dataset of 250k press releases and 650k articles covering them. We develop methods to identify news articles that _challenge and contextualize_ press releases. Finally, we evaluate suggestions made by LLMs for these articles and compare these with decisions made by human journalists. Our findings are three-fold: (1) Human-written news articles that challenge and contextualize press releases more take more creative angles and use more informational sources. (2) LLMs align better with humans when recommending angles, compared with informational sources. (3) Both the angles and sources LLMs suggest are significantly less creative than humans.

Baolin Li, Yankai Jiang, Vijay Gadepally, Devesh Tiwari

The rapid advancement of generative AI has heightened environmental concerns, particularly regarding carbon emissions. Our framework, Sprout, addresses these challenges by reducing the carbon footprint of inference in large language models (LLMs). Sprout introduces “generation directives” to guide the autoregressive generation process, achieving a balance between ecological sustainability and high-quality outputs. By employing a strategic optimizer for directive assignment and a novel offline quality evaluator, Sprout reduces the carbon footprint of generative LLM inference by over 40% in real-world evaluations, using the Llama model and global electricity grid data. This work is crucial as the rising interest in inference time compute scaling laws amplifies environmental concerns, emphasizing the need for eco-friendly AI solutions.

Bhathiya Hemanthage, Hakan Bilen, Phil Bartie, Christian Dondrup, Oliver Lemon

The Generalized Referring Expression Comprehension (GREC) task extends classic REC by generating image bounding boxes for objects referred to in natural language expressions, which may indicate zero, one, or multiple targets. This generalization enhances the practicality of REC models for diverse real-world applications. However, the presence of varying numbers of targets in samples makes GREC a more complex task, both in terms of training supervision and final prediction selection strategy. Addressing these challenges, we introduce RECANTFormer, a one-stage method for GREC that combines a decoder-free (encoder-only) transformer architecture with DETR-like Hungarian matching. Our approach consistently outperforms baselines by significant margins in three GREC datasets.

Kata Naszadi, Frans A Oliehoek, Christof Monz

This paper explores the impact of variable pragmatic competence on communicative success through simulating language learning and conversing between speakers and listeners with different levels of reasoning abilities. Through studying this interaction, we hypothesize that matching levels of reasoning between communication partners would create a more beneficial environment for communicative success and language learning. Our research findings indicate that learning from more explicit, literal language is advantageous, irrespective of the learner’s level of pragmatic competence. Furthermore, we find that integrating pragmatic reasoning during language learning, not just during evaluation, significantly enhances overall communication performance. This paper provides key insights into the importance of aligning reasoning levels and incorporating pragmatic reasoning in optimizing communicative interactions.

Rima Hazra, Sayan Layek, Somnath Banerjee, Soujanya Poria

Ensuring the safe alignment of large language models (LLMs) with human values is critical as they become integral to applications like translation and question answering. Current alignment methods struggle with dynamic user intentions and complex objectives, making models vulnerable to generating harmful content. We propose Safety Arithmetic, a training-free framework enhancing LLM safety across different scenarios: Base models, Supervised fine-tuned models (SFT), and Edited models. Safety Arithmetic involves Harm Direction Removal to avoid harmful content and Safety Alignment to promote safe responses. Additionally, we present NoIntentEdit, a dataset highlighting edit instances that could compromise model safety if used unintentionally. Our experiments show that Safety Arithmetic significantly improves safety measures, reduces over-safety, and maintains model utility, outperforming existing methods in ensuring safe content generation.

Bashar Talafha, Karima Kadaoui, Samar Mohamed Magdy, Mariem Habiboullah, Chafei Mohamed Chafei, Ahmed Oumar El-Shangiti, Hiba Zayed, Mohamedou Cheikh Tourad, Rahaf Alhamouri, Rwaa Assi 等

In spite of the recent progress in speech processing, the majority of world languages and dialects remain uncovered. This situation only furthers an already wide technological divide, thereby hindering technological and socioeconomic inclusion. This challenge is largely due to the absence of datasets that can empower diverse speech systems. In this paper, we seek to mitigate this obstacle for a number of Arabic dialects by presenting Casablanca, a large-scale community-driven effort to collect and transcribe a multi-dialectal Arabic dataset. The dataset covers eight dialects: Algerian, Egyptian, Emirati, Jordanian, Mauritanian, Moroccan, Palestinian, and Yemeni, and includes annotations for transcription, gender, dialect, and code-switching. We also develop a number of strong baselines exploiting Casablanca. The project page for Casablanca is accessible at: www.dlnlp.ai/speech/casablanca.

Akshat Gupta, Sidharth Baskaran, Gopala Anumanchipalli

Recent work using Rank-One Model Editing (ROME), a popular model editing method, has shown that there are certain facts that the algorithm is unable to edit without breaking the model. Such edits have previously been called disabling edits. These disabling edits cause immediate model collapse and limits the use of ROME for sequential editing. In this paper, we show that disabling edits are an artifact of irregularities in the implementation of ROME. With this paper, we provide a more stable implementation ROME, which we call r-ROME and show that model collapse is no longer observed when making large scale sequential edits with r-ROME, while further improving generalization and locality of model editing compared to the original implementation of ROME. We also provide a detailed mathematical explanation of the reason behind disabling edits.

Ce Zhang, Taixi Lu, Md Mohaiminul Islam, Ziyang Wang, Shoubin Yu, Mohit Bansal, Gedas Bertasius

We present LLoVi, a simple yet effective **L**anguage-based **Lo**ng-range **Vi**deo question-answering (LVQA) framework. Our method decomposes the short- and long-range modeling aspects of LVQA into two stages. First, we use a short-term visual captioner to generate textual descriptions of short video clips (0.5-8 seconds in length) densely sampled from a long input video. Afterward, an LLM aggregates the densely extracted short-term captions to answer a given question. Furthermore, we propose a novel multi-round summarization prompt that asks the LLM first to summarize the noisy short-term visual captions and then answer a given input question. To analyze what makes our simple framework so effective, we thoroughly evaluate various components of our framework. Our empirical analysis reveals that the choice of the visual captioner and LLM is critical for good LVQA performance. The proposed multi-round summarization prompt also leads to a significant LVQA performance boost. Our method achieves the best-reported results on the EgoSchema dataset, best known for very long-form video question-answering. LLoVi also outperforms the previous state-of-the-art by **10.2%** and **6.2%** on NExT-QA and IntentQA for LVQA. Finally, we extend LLoVi to grounded VideoQA, which requires both QA and temporal localization, and show that it outperforms all prior methods on NExT-GQA. Code is available at https://github.com/CeeZh/LLoVi.

Xuhui Zhou, Zhe Su, Tiwalayo Eisape, Hyunwoo Kim, Maarten Sap

Recent advances in large language models (LLM) have enabled richer social simulations, allowing for the study of various social phenomena. However, most recent work has used a more omniscient perspective on these simulations (e.g., single LLM to generate all interlocutors), which is fundamentally at odds with the non-omniscient, information asymmetric interactions that involve humans and AI agents in the real world. To examine these differences, we develop an evaluation framework to simulate social interactions with LLMs in various settings (omniscient, non-omniscient). Our experiments show that LLMs perform better in unrealistic, omniscient simulation settings but struggle in ones that more accurately reflect real-world conditions with information asymmetry. Moreover, we illustrate the limitations inherent in learning from omniscient simulations. Our findings indicate that addressing information asymmetry remains a fundamental challenge for LLM-based agents.

Pieter Delobelle, Giuseppe Attanasio, Debora Nozza, Su Lin Blodgett, Zeerak Talat

This paper introduces the concept of actionability in the context of bias measures in natural language processing (NLP). We define actionability as the degree to which a measure’s results enable informed action and propose a set of desiderata for assessing it. Building on existing frameworks such as measurement modeling, we argue that actionability is a crucial aspect of bias measures that has been largely overlooked in the literature.We conduct a comprehensive review of 146 papers proposing bias measures in NLP, examining whether and how they provide the information required for actionable results. Our findings reveal that many key elements of actionability, including a measure’s intended use and reliability assessment, are often unclear or entirely absent.This study highlights a significant gap in the current approach to developing and reporting bias measures in NLP. We argue that this lack of clarity may impede the effective implementation and utilization of these measures. To address this issue, we offer recommendations for more comprehensive and actionable metric development and reporting practices in NLP bias research.

Hai Ye, Hwee Tou Ng

Iterative data generation and model re-training can effectively align large language models (LLMs) to human preferences. The process of data sampling is crucial, as it significantly influences the success of policy improvement. Repeated random sampling is a widely used method that independently queries the model multiple times to generate outputs. In this work, we propose a more effective sampling method, named Preference-Guided Reflective Sampling (PRS). Unlike random sampling, PRS employs a tree-based generation framework to enable more efficient sampling. It leverages adaptive self-refinement techniques to better explore the sampling space. By specifying user preferences in natural language, PRS can further optimize response generation according to these preferences. As a result, PRS can align models to diverse user preferences. Our experiments demonstrate that PRS generates higher-quality responses with significantly higher rewards. On AlpacaEval and Arena-Hard, PRS substantially outperforms repeated random sampling in best-of-N sampling. Moreover, PRS shows strong performance when applied in iterative offline RL training.