Authors: Tomasz Zi\k{e}tkiewicz
Abstract: This paper presents a new approach to the problem of correcting speech recognition errors by means of post-editing. It consists of using a neural sequence tagger that learns how to correct an ASR (Automatic Speech Recognition) hypothesis word by word and a corrector module that applies corrections returned by the tagger. The proposed solution is applicable to any ASR system, regardless of its architecture, and provides high-precision control over errors being corrected. This is especially crucial in production environments, where avoiding the introduction of new mistakes by the error correction model may be more important than the net gain in overall results. The results show that the performance of the proposed error correction models is comparable with previous approaches while requiring much smaller resources to train, which makes it suitable for industrial applications, where both inference latency and training times are critical factors that limit the use of other techniques.
Authors: Leonardo Cotta, Chris J. Maddison
Abstract: Frontier Large Language Models (LLMs) are increasingly being deployed for high-stakes decision-making. On the other hand, these models are still consistently making predictions that contradict users' or society's expectations, e.g., hallucinating, or discriminating. Thus, it is important that we develop test-time strategies to improve their trustworthiness. Inspired by prior work, we leverage causality as a tool to formally encode two aspects of trustworthiness in LLMs: fairness and robustness. Under this perspective, existing test-time solutions explicitly instructing the model to be fair or robust implicitly depend on the LLM's causal reasoning capabilities. In this work, we explore the opposite approach. Instead of explicitly asking the LLM for trustworthiness, we design prompts to encode the underlying causal inference algorithm that will, by construction, result in more trustworthy predictions. Concretely, we propose out-of-context prompting as a test-time solution to encourage fairness and robustness in LLMs. Out-of-context prompting leverages the user's prior knowledge of the task's causal model to apply (random) counterfactual transformations and improve the model's trustworthiness. Empirically, we show that out-of-context prompting consistently improves the fairness and robustness of frontier LLMs across five different benchmark datasets without requiring additional data, finetuning or pre-training.
Authors: Sari Masri, Yaqeen Raddad, Fidaa Khandaqji, Huthaifa I. Ashqar, Mohammed Elhenawy
Abstract: Recently, with the rapid development in the fields of technology and the increasing amount of text t available on the internet, it has become urgent to develop effective tools for processing and understanding texts in a way that summaries the content without losing the fundamental essence of the information. Given this challenge, we have developed an advanced text summarization system targeting Arabic textbooks. Relying on modern natu-ral language processing models such as MT5, AraBART, AraT5, and mBART50, this system evaluates and extracts the most important sentences found in biology textbooks for the 11th and 12th grades in the Palestinian curriculum, which enables students and teachers to obtain accurate and useful summaries that help them easily understand the content. We utilized the Rouge metric to evaluate the performance of the trained models. Moreover, experts in education Edu textbook authoring assess the output of the trained models. This approach aims to identify the best solutions and clarify areas needing improvement. This research provides a solution for summarizing Arabic text. It enriches the field by offering results that can open new horizons for research and development in the technologies for understanding and generating the Arabic language. Additionally, it contributes to the field with Arabic texts through creating and compiling schoolbook texts and building a dataset.
Authors: Luis Lugo, Valentin Vielzeuf
Abstract: Sustainable artificial intelligence focuses on data, hardware, and algorithms to make machine learning models more environmentally responsible. In particular, machine learning models for speech representations are computationally expensive, generating environmental concerns because of their high energy consumption. Thus, we propose a sustainable self-supervised model to learn speech representation, combining optimizations in neural layers and training to reduce computing costs. The proposed model improves over a resource-efficient baseline, reducing both memory usage and computing cost estimations. It pretrains using a single GPU in less than a day. On top of that, it improves the error rate performance of the baseline in downstream task evaluations. When comparing it to large speech representation approaches, there is an order of magnitude reduction in memory usage, while computing cost reductions represent almost three orders of magnitude improvement.
Authors: Haw-Shiuan Chang, Nanyun Peng, Mohit Bansal, Anil Ramakrishna, Tagyoung Chung
Abstract: Decoding methods for large language models (LLMs) usually struggle with the tradeoff between ensuring factuality and maintaining diversity. For example, a higher p threshold in the nucleus (top-p) sampling increases the diversity but decreases the factuality, and vice versa. In this paper, we propose REAL (Residual Entropy from Asymptotic Line) sampling, a decoding method that achieves improved factuality and diversity over nucleus sampling by predicting an adaptive threshold of $p$. Specifically, REAL sampling predicts the step-wise likelihood of an LLM to hallucinate, and lowers the p threshold when an LLM is likely to hallucinate. Otherwise, REAL sampling increases the p threshold to boost the diversity. To predict the step-wise hallucination likelihood without supervision, we construct a Token-level Hallucination Forecasting (THF) model to predict the asymptotic entropy (i.e., inherent uncertainty) of the next token by extrapolating the next-token entropies from a series of LLMs with different sizes. If a LLM's entropy is higher than the asymptotic entropy (i.e., the LLM is more uncertain than it should be), the THF model predicts a high hallucination hazard, which leads to a lower p threshold in REAL sampling. In the FactualityPrompts benchmark, we demonstrate that REAL sampling based on a 70M THF model can substantially improve the factuality and diversity of 7B LLMs simultaneously, judged by both retrieval-based metrics and human evaluation. After combined with contrastive decoding, REAL sampling outperforms 9 sampling methods, and generates texts that are more factual than the greedy sampling and more diverse than the nucleus sampling with $p=0.5$. Furthermore, the predicted asymptotic entropy is also a useful unsupervised signal for hallucination detection tasks.
Authors: Dojun Park, Jiwoo Lee, Seohyun Park, Hyeyun Jeong, Youngeun Koo, Soonha Hwang, Seonwoo Park, Sungeun Lee
Abstract: As the capabilities of LLMs expand, it becomes increasingly important to evaluate them beyond basic knowledge assessment, focusing on higher-level language understanding. This study introduces MultiPragEval, a robust test suite designed for the multilingual pragmatic evaluation of LLMs across English, German, Korean, and Chinese. Comprising 1200 question units categorized according to Grice's Cooperative Principle and its four conversational maxims, MultiPragEval enables an in-depth assessment of LLMs' contextual awareness and their ability to infer implied meanings. Our findings demonstrate that Claude3-Opus significantly outperforms other models in all tested languages, establishing a state-of-the-art in the field. Among open-source models, Solar-10.7B and Qwen1.5-14B emerge as strong competitors. This study not only leads the way in the multilingual evaluation of LLMs in pragmatic inference but also provides valuable insights into the nuanced capabilities necessary for advanced language comprehension in AI systems.
Authors: Jason Wu, Eldon Schoop, Alan Leung, Titus Barik, Jeffrey P. Bigham, Jeffrey Nichols
Abstract: Large language models (LLMs) struggle to consistently generate UI code that compiles and produces visually relevant designs. Existing approaches to improve generation rely on expensive human feedback or distilling a proprietary model. In this paper, we explore the use of automated feedback (compilers and multi-modal models) to guide LLMs to generate high-quality UI code. Our method starts with an existing LLM and iteratively produces improved models by self-generating a large synthetic dataset using an original model, applying automated tools to aggressively filter, score, and de-duplicate the data into a refined higher quality dataset. The original LLM is improved by finetuning on this refined dataset. We applied our approach to several open-source LLMs and compared the resulting performance to baseline models with both automated metrics and human preferences. Our evaluation shows the resulting models outperform all other downloadable baselines and approach the performance of larger proprietary models.
Authors: Dasun Athukoralage, Thushari Atapattu, Menasha Thilakaratne, Katrina Falkner
Abstract: This paper presents our approaches for the SMM4H24 Shared Task 5 on the binary classification of English tweets reporting children's medical disorders. Our first approach involves fine-tuning a single RoBERTa-large model, while the second approach entails ensembling the results of three fine-tuned BERTweet-large models. We demonstrate that although both approaches exhibit identical performance on validation data, the BERTweet-large ensemble excels on test data. Our best-performing system achieves an F1-score of 0.938 on test data, outperforming the benchmark classifier by 1.18%.
Authors: Lin Shi, Weicheng Ma, Soroush Vosoughi
Abstract: LLM-as-a-Judge offers a promising alternative to human judges across various tasks, yet inherent biases, particularly position bias - a systematic preference for answers based on their position in the prompt - compromise its effectiveness. Our study investigates this issue by developing a framework to systematically study and quantify position bias using metrics such as repetitional consistency, positional consistency, and positional fairness. We conduct experiments with 9 judge models across 22 tasks from the MTBench and DevBench benchmarks and nearly 40 answer-generating models, generating approximately 80,000 evaluation instances. This comprehensive assessment reveals significant variations in bias across judges and tasks. Although GPT-4 often excels in positional consistency and fairness, some more cost-effective models perform comparably or even better in specific tasks, highlighting essential trade-offs between consistency, fairness, and cost. Our results also demonstrate high consistency of judgment across repetitions, confirming that position bias is not due to random variations. This research significantly contributes to the field by introducing new concepts for understanding position bias and providing a multi-dimensional framework for evaluation. These insights guide the selection of optimal judge models, enhance benchmark design, and lay the foundation for future research into effective debiasing strategies, ultimately enhancing the reliability of LLM evaluators.
Authors: Amogh Mannekote, Jinseok Nam, Ziming Li, Jian Gao, Kristy Elizabeth Boyer, Bonnie J. Dorr
Abstract: Existing benchmark corpora of task-oriented dialogue are collected either using a "machines talking to machines" approach or by giving template-based goal descriptions to crowdworkers. These methods, however, often produce utterances that are markedly different from natural human conversations in which people often convey their preferences in indirect ways, such as through small talk. We term such utterances as Indirect User Requests (IURs). Understanding such utterances demands considerable world knowledge and reasoning capabilities on the listener's part. Our study introduces an LLM-based pipeline to automatically generate realistic, high-quality IURs for a given domain, with the ultimate goal of supporting research in natural language understanding (NLU) and dialogue state tracking (DST) for task-oriented dialogue systems. Our findings show that while large LLMs such as GPT-3.5 and GPT-4 generate high-quality IURs, achieving similar quality with smaller models is more challenging. We release IndirectRequests, a dataset of IURs that advances beyond the initial Schema-Guided Dialog (SGD) dataset in that it provides a challenging testbed for testing the "in the wild" performance of NLU and DST models.
Authors: Runyan Yang, Huibao Yang, Xiqing Zhang, Tiantian Ye, Ying Liu, Yingying Gao, Shilei Zhang, Chao Deng, Junlan Feng
Abstract: Recently, there have been attempts to integrate various speech processing tasks into a unified model. However, few previous works directly demonstrated that joint optimization of diverse tasks in multitask speech models has positive influence on the performance of individual tasks. In this paper we present a multitask speech model -- PolySpeech, which supports speech recognition, speech synthesis, and two speech classification tasks. PolySpeech takes multi-modal language model as its core structure and uses semantic representations as speech inputs. We introduce semantic speech embedding tokenization and speech reconstruction methods to PolySpeech, enabling efficient generation of high-quality speech for any given speaker. PolySpeech shows competitiveness across various tasks compared to single-task models. In our experiments, multitask optimization achieves performance comparable to single-task optimization and is especially beneficial for specific tasks.
Authors: Yiran Wang, Masao Utiyama
Abstract: Recently, binary representation has been proposed as a novel representation that lies between continuous and discrete representations. It exhibits considerable information-preserving capability when being used to replace continuous input vectors. In this paper, we investigate the feasibility of further introducing it to the output side, aiming to allow models to output binary labels instead. To preserve the structural information on the output side along with label information, we extend the previous contrastive hashing method as structured contrastive hashing. More specifically, we upgrade CKY from label-level to bit-level, define a new similarity function with span marginal probabilities, and introduce a novel contrastive loss function with a carefully designed instance selection strategy. Our model achieves competitive performance on various structured prediction tasks, and demonstrates that binary representation can be considered a novel representation that further bridges the gap between the continuous nature of deep learning and the discrete intrinsic property of natural languages.
Authors: Yizhang Zhu, Shiyin Du, Boyan Li, Yuyu Luo, Nan Tang
Abstract: Large Language Models (LLMs) have demonstrated impressive capabilities across a range of scientific tasks including mathematics, physics, and chemistry. Despite their successes, the effectiveness of LLMs in handling complex statistical tasks remains systematically under-explored. To bridge this gap, we introduce StatQA, a new benchmark designed for statistical analysis tasks. StatQA comprises 11,623 examples tailored to evaluate LLMs' proficiency in specialized statistical tasks and their applicability assessment capabilities, particularly for hypothesis testing methods. We systematically experiment with representative LLMs using various prompting strategies and show that even state-of-the-art models such as GPT-4o achieve a best performance of only 64.83%, indicating significant room for improvement. Notably, while open-source LLMs (e.g. LLaMA-3) show limited capability, those fine-tuned ones exhibit marked improvements, outperforming all in-context learning-based methods (e.g. GPT-4o). Moreover, our comparative human experiments highlight a striking contrast in error types between LLMs and humans: LLMs primarily make applicability errors, whereas humans mostly make statistical task confusion errors. This divergence highlights distinct areas of proficiency and deficiency, suggesting that combining LLM and human expertise could lead to complementary strengths, inviting further investigation into their collaborative potential.
Authors: Trang Le, Daniel Lazar, Suyoun Kim, Shan Jiang, Duc Le, Adithya Sagar, Aleksandr Livshits, Ahmed Aly, Akshat Shrivastava
Abstract: Spoken Language Understanding (SLU) is a critical component of voice assistants; it consists of converting speech to semantic parses for task execution. Previous works have explored end-to-end models to improve the quality and robustness of SLU models with Deliberation, however these models have remained autoregressive, resulting in higher latencies. In this work we introduce PRoDeliberation, a novel method leveraging a Connectionist Temporal Classification-based decoding strategy as well as a denoising objective to train robust non-autoregressive deliberation models. We show that PRoDeliberation achieves the latency reduction of parallel decoding (2-10x improvement over autoregressive models) while retaining the ability to correct Automatic Speech Recognition (ASR) mistranscriptions of autoregressive deliberation systems. We further show that the design of the denoising training allows PRoDeliberation to overcome the limitations of small ASR devices, and we provide analysis on the necessity of each component of the system.
Authors: David Wadden, Kejian Shi, Jacob Morrison, Aakanksha Naik, Shruti Singh, Nitzan Barzilay, Kyle Lo, Tom Hope, Luca Soldaini, Shannon Zejiang Shen, Doug Downey, Hannaneh Hajishirzi, Arman Cohan
Abstract: We present SciRIFF (Scientific Resource for Instruction-Following and Finetuning), a dataset of 137K instruction-following demonstrations for 54 tasks covering five essential scientific literature understanding capabilities: information extraction, summarization, question answering, claim verification, and classification. SciRIFF demonstrations are notable for their long input contexts, detailed task specifications, and complex structured outputs. While instruction-following resources are available in specific domains such as clinical medicine and chemistry, SciRIFF is the first dataset focused on extracting and synthesizing information from research literature across a wide range of scientific fields. To demonstrate the utility of SciRIFF, we develop a sample-efficient strategy to adapt a general instruction-following model for science by performing additional finetuning on a mix of general-domain and SciRIFF demonstrations. In evaluations on nine held-out scientific tasks, our model -- called SciTulu -- improves over a strong LLM baseline by 28.1% and 6.5% at the 7B and 70B scales respectively, while maintaining general instruction-following performance within 2% of the baseline. We are optimistic that SciRIFF will facilitate the development and evaluation of LLMs to help researchers navigate the ever-growing body of scientific literature. We release our dataset, model checkpoints, and data processing and evaluation code to enable further research.
Authors: Yiwei Li, Fei Mi, Yitong Li, Yasheng Wang, Bin Sun, Shaoxiong Feng, Kan Li
Abstract: Stochastic sampling strategies such as top-k and top-p have been widely used in dialogue generation task. However, as an open-domain chatting system, there will be two different conversation scenarios, i.e. chit-chat and knowledge-based question answering. In the former situation, responses diversity is essential due to the one-to-many nature in dialogue. The latter, on the other hand, requires less randomness given that stochastic decoding strategy entails the risk of generating incorrect information. As a result, an adaptive and flexible decoding strategy is needed to cope with these two scenarios simultaneously. To this end, we propose the dynamic decoding strategy (DDS), which can adjust the decoding space w.r.t. different contexts. In DDS, both sequence-level and token-level adaptive search can be achieved to adjust the decoding process in a unified framework. Besides, our adaptive algorithm can not only be used during model inference, but it can also be applied during the model training stage to further enhance the performance. Comprehensive experiments indicate that the proposed decoding strategy can consistently improve the performance of pre-trained dialogue models when coupled with four well-used stochastic decoding algorithms.
Authors: Bing Han, Long Zhou, Shujie Liu, Sanyuan Chen, Lingwei Meng, Yanming Qian, Yanqing Liu, Sheng Zhao, Jinyu Li, Furu Wei
Abstract: With the help of discrete neural audio codecs, large language models (LLM) have increasingly been recognized as a promising methodology for zero-shot Text-to-Speech (TTS) synthesis. However, sampling based decoding strategies bring astonishing diversity to generation, but also pose robustness issues such as typos, omissions and repetition. In addition, the high sampling rate of audio also brings huge computational overhead to the inference process of autoregression. To address these issues, we propose VALL-E R, a robust and efficient zero-shot TTS system, building upon the foundation of VALL-E. Specifically, we introduce a phoneme monotonic alignment strategy to strengthen the connection between phonemes and acoustic sequence, ensuring a more precise alignment by constraining the acoustic tokens to match their associated phonemes. Furthermore, we employ a codec-merging approach to downsample the discrete codes in shallow quantization layer, thereby accelerating the decoding speed while preserving the high quality of speech output. Benefiting from these strategies, VALL-E R obtains controllablity over phonemes and demonstrates its strong robustness by approaching the WER of ground truth. In addition, it requires fewer autoregressive steps, with over 60% time reduction during inference. This research has the potential to be applied to meaningful projects, including the creation of speech for those affected by aphasia. Audio samples will be available at: https://aka.ms/valler.
URLs: https://aka.ms/valler.
Authors: Rahul Kumar, Amar Raja Dibbu, Shrutendra Harsola, Vignesh Subrahmaniam, Ashutosh Modi
Abstract: Several large-scale datasets (e.g., WikiSQL, Spider) for developing natural language interfaces to databases have recently been proposed. These datasets cover a wide breadth of domains but fall short on some essential domains, such as finance and accounting. Given that accounting databases are used worldwide, particularly by non-technical people, there is an imminent need to develop models that could help extract information from accounting databases via natural language queries. In this resource paper, we aim to fill this gap by proposing a new large-scale Text-to-SQL dataset for the accounting and financial domain: BookSQL. The dataset consists of 100k natural language queries-SQL pairs, and accounting databases of 1 million records. We experiment with and analyze existing state-of-the-art models (including GPT-4) for the Text-to-SQL task on BookSQL. We find significant performance gaps, thus pointing towards developing more focused models for this domain.
Authors: Yida Chen, Aoyu Wu, Trevor DePodesta, Catherine Yeh, Kenneth Li, Nicholas Castillo Marin, Oam Patel, Jan Riecke, Shivam Raval, Olivia Seow, Martin Wattenberg, Fernanda Vi\'egas
Abstract: Conversational LLMs function as black box systems, leaving users guessing about why they see the output they do. This lack of transparency is potentially problematic, especially given concerns around bias and truthfulness. To address this issue, we present an end-to-end prototype-connecting interpretability techniques with user experience design-that seeks to make chatbots more transparent. We begin by showing evidence that a prominent open-source LLM has a "user model": examining the internal state of the system, we can extract data related to a user's age, gender, educational level, and socioeconomic status. Next, we describe the design of a dashboard that accompanies the chatbot interface, displaying this user model in real time. The dashboard can also be used to control the user model and the system's behavior. Finally, we discuss a study in which users conversed with the instrumented system. Our results suggest that users appreciate seeing internal states, which helped them expose biased behavior and increased their sense of control. Participants also made valuable suggestions that point to future directions for both design and machine learning research. The project page and video demo of our TalkTuner system are available at https://bit.ly/talktuner-project-page
Authors: Jaehoon Kim, Seungwan Jin, Sohyun Park, Someen Park, Kyungsik Han
Abstract: Detecting implicit hate speech that is not directly hateful remains a challenge. Recent research has attempted to detect implicit hate speech by applying contrastive learning to pre-trained language models such as BERT and RoBERTa, but the proposed models still do not have a significant advantage over cross-entropy loss-based learning. We found that contrastive learning based on randomly sampled batch data does not encourage the model to learn hard negative samples. In this work, we propose Label-aware Hard Negative sampling strategies (LAHN) that encourage the model to learn detailed features from hard negative samples, instead of naive negative samples in random batch, using momentum-integrated contrastive learning. LAHN outperforms the existing models for implicit hate speech detection both in- and cross-datasets. The code is available at https://github.com/Hanyang-HCC-Lab/LAHN
Authors: Bulat Khaertdinov, Pedro Jeuris, Annanda Sousa, Enrique Hortal
Abstract: Recent advancements in Deep and Self-Supervised Learning (SSL) have led to substantial improvements in Speech Emotion Recognition (SER) performance, reaching unprecedented levels. However, obtaining sufficient amounts of accurately labeled data for training or fine-tuning the models remains a costly and challenging task. In this paper, we propose a multi-view SSL pre-training technique that can be applied to various representations of speech, including the ones generated by large speech models, to improve SER performance in scenarios where annotations are limited. Our experiments, based on wav2vec 2.0, spectral and paralinguistic features, demonstrate that the proposed framework boosts the SER performance, by up to 10% in Unweighted Average Recall, in settings with extremely sparse data annotations.
Authors: Yuxi Feng, Raymond Li, Zhenan Fan, Giuseppe Carenini, Mohammadreza Pourreza, Weiwei Zhang, Yong Zhang
Abstract: While in-context Learning (ICL) has proven to be an effective technique to improve the performance of Large Language Models (LLMs) in a variety of complex tasks, notably in translating natural language questions into Structured Query Language (NL2SQL), the question of how to select the most beneficial demonstration examples remains an open research problem. While prior works often adapted off-the-shelf encoders to retrieve examples dynamically, an inherent discrepancy exists in the representational capacities between the external retrievers and the LLMs. Further, optimizing the selection of examples is a non-trivial task, since there are no straightforward methods to assess the relative benefits of examples without performing pairwise inference. To address these shortcomings, we propose DeTriever, a novel demonstration retrieval framework that learns a weighted combination of LLM hidden states, where rich semantic information is encoded. To train the model, we propose a proxy score that estimates the relative benefits of examples based on the similarities between output queries. Experiments on two popular NL2SQL benchmarks demonstrate that our method significantly outperforms the state-of-the-art baselines on one-shot NL2SQL tasks.
Authors: Dongsuk Jang, Hyeryun Park, Jiye Son, Hyeonuk Hwang, Sujin Kim, Jinwook Choi
Abstract: In the rapidly evolving field of healthcare, the integration of artificial intelligence (AI) has become a pivotal component in the automation of clinical workflows, ushering in a new era of efficiency and accuracy. This study focuses on the transformative capabilities of the fine-tuned KoELECTRA model in comparison to the GPT-4 model, aiming to facilitate automated information extraction from thyroid operation narratives. The current research landscape is dominated by traditional methods heavily reliant on regular expressions, which often face challenges in processing free-style text formats containing critical details of operation records, including frozen biopsy reports. Addressing this, the study leverages advanced natural language processing (NLP) techniques to foster a paradigm shift towards more sophisticated data processing systems. Through this comparative study, we aspire to unveil a more streamlined, precise, and efficient approach to document processing in the healthcare domain, potentially revolutionizing the way medical data is handled and analyzed.
Authors: Chris Yuhao Liu, Yaxuan Wang, Jeffrey Flanigan, Yang Liu
Abstract: Large language models (LLMs) have advanced to encompass extensive knowledge across diverse domains. Yet controlling what a large language model should not know is important for ensuring alignment and thus safe use. However, accurately and efficiently unlearning knowledge from an LLM remains challenging due to the potential collateral damage caused by the fuzzy boundary between retention and forgetting, and the large computational requirements for optimization across state-of-the-art models with hundreds of billions of parameters. In this work, we present Embedding-COrrupted (ECO) Prompts, a lightweight unlearning framework for large language models to address both the challenges of knowledge entanglement and unlearning efficiency. Instead of relying on the LLM itself to unlearn, we enforce an unlearned state during inference by employing a prompt classifier to identify and safeguard prompts to forget. We learn corruptions added to prompt embeddings via zeroth order optimization toward the unlearning objective offline and corrupt prompts flagged by the classifier during inference. We find that these embedding-corrupted prompts not only lead to desirable outputs that satisfy the unlearning objective but also closely approximate the output from a model that has never been trained on the data intended for forgetting. Through extensive experiments on unlearning, we demonstrate the superiority of our method in achieving promising unlearning at nearly zero side effects in general domains and domains closely related to the unlearned ones. Additionally, we highlight the scalability of our method to 100 LLMs, ranging from 0.5B to 236B parameters, incurring no additional cost as the number of parameters increases.
Authors: Jie Ruan, Wenqing Wang, Xiaojun Wan
Abstract: Human evaluation serves as the gold standard for assessing the quality of Natural Language Generation (NLG) systems. Nevertheless, the evaluation guideline, as a pivotal element ensuring reliable and reproducible human assessment, has received limited attention.Our investigation revealed that only 29.84% of recent papers involving human evaluation at top conferences release their evaluation guidelines, with vulnerabilities identified in 77.09% of these guidelines. Unreliable evaluation guidelines can yield inaccurate assessment outcomes, potentially impeding the advancement of NLG in the right direction. To address these challenges, we take an initial step towards reliable evaluation guidelines and propose the first human evaluation guideline dataset by collecting annotations of guidelines extracted from existing papers as well as generated via Large Language Models (LLMs). We then introduce a taxonomy of eight vulnerabilities and formulate a principle for composing evaluation guidelines. Furthermore, a method for detecting guideline vulnerabilities has been explored using LLMs, and we offer a set of recommendations to enhance reliability in human evaluation. The annotated human evaluation guideline dataset and code for the vulnerability detection method are publicly available online.
Authors: Jie Ruan, Xiao Pu, Mingqi Gao, Xiaojun Wan, Yuesheng Zhu
Abstract: Human evaluation is viewed as a reliable evaluation method for NLG which is expensive and time-consuming. To save labor and costs, researchers usually perform human evaluation on a small subset of data sampled from the whole dataset in practice. However, different selection subsets will lead to different rankings of the systems. To give a more correct inter-system ranking and make the gold standard human evaluation more reliable, we propose a Constrained Active Sampling Framework (CASF) for reliable human judgment. CASF operates through a Learner, a Systematic Sampler and a Constrained Controller to select representative samples for getting a more correct inter-system ranking.Experiment results on 137 real NLG evaluation setups with 44 human evaluation metrics across 16 datasets and 5 NLG tasks demonstrate CASF receives 93.18% top-ranked system recognition accuracy and ranks first or ranks second on 90.91% of the human metrics with 0.83 overall inter-system ranking Kendall correlation.Code and data are publicly available online.
Authors: Javad Pourmostafa Roshan Sharami, Dimitar Shterionov, Pieter Spronck
Abstract: The quality of output from large language models (LLMs), particularly in machine translation (MT), is closely tied to the quality of in-context examples (ICEs) provided along with the query, i.e., the text to translate. The effectiveness of these ICEs is influenced by various factors, such as the domain of the source text, the order in which the ICEs are presented, the number of these examples, and the prompt templates used. Naturally, selecting the most impactful ICEs depends on understanding how these affect the resulting translation quality, which ultimately relies on translation references or human judgment. This paper presents a novel methodology for in-context learning (ICL) that relies on a search algorithm guided by domain-specific quality estimation (QE). Leveraging the XGLM model, our methodology estimates the resulting translation quality without the need for translation references, selecting effective ICEs for MT to maximize translation quality. Our results demonstrate significant improvements over existing ICL methods and higher translation performance compared to fine-tuning a pre-trained language model (PLM), specifically mBART-50.
Authors: Taiming Lu, Lingfeng Shen, Xinyu Yang, Weiting Tan, Beidi Chen, Huaxiu Yao
Abstract: Reinforcement Learning from Human Feedback (RLHF) involves training policy models (PMs) and reward models (RMs) to align language models with human preferences. Instead of focusing solely on PMs and RMs independently, we propose to examine their interactions during fine-tuning, introducing the concept of seamlessness. Our study starts with observing the saturation phenomenon, where continual improvements in RM and PM do not translate into RLHF progress. Our analysis shows that RMs fail to assign proper scores to PM responses, resulting in a 35% mismatch rate with human preferences, highlighting a significant discrepancy between PM and RM. To measure seamlessness between PM and RM without human effort, we propose an automatic metric, SEAM. SEAM quantifies the discrepancies between PM and RM judgments induced by data samples. We validate the effectiveness of SEAM in data selection and model augmentation. Our experiments demonstrate that (1) using SEAM-filtered data for RL training improves RLHF performance by 4.5%, and (2) SEAM-guided model augmentation results in a 4% performance improvement over standard augmentation methods.
Authors: Jo\~ao Vitorino, Eva Maia, Isabel Pra\c{c}a
Abstract: Large Language Models (LLMs) are valuable for text classification, but their vulnerabilities must not be disregarded. They lack robustness against adversarial examples, so it is pertinent to understand the impacts of different types of perturbations, and assess if those attacks could be replicated by common users with a small amount of perturbations and a small number of queries to a deployed LLM. This work presents an analysis of the effectiveness, efficiency, and practicality of three different types of adversarial attacks against five different LLMs in a sentiment classification task. The obtained results demonstrated the very distinct impacts of the word-level and character-level attacks. The word attacks were more effective, but the character and more constrained attacks were more practical and required a reduced number of perturbations and queries. These differences need to be considered during the development of adversarial defense strategies to train more robust LLMs for intelligent text classification applications.
Authors: Napat Laosaengpha, Thanit Tativannarat, Chawan Piansaddhayanon, Attapol Rutherford, Ekapol Chuangsuwanich
Abstract: Learning job title representation is a vital process for developing automatic human resource tools. To do so, existing methods primarily rely on learning the title representation through skills extracted from the job description, neglecting the rich and diverse content within. Thus, we propose an alternative framework for learning job titles through their respective job description (JD) and utilize a Job Description Aggregator component to handle the lengthy description and bidirectional contrastive loss to account for the bidirectional relationship between the job title and its description. We evaluated the performance of our method on both in-domain and out-of-domain settings, achieving a superior performance over the skill-based approach.
Authors: Hao Yang, Yanyan Zhao, Yang Wu, Shilong Wang, Tian Zheng, Hongbo Zhang, Wanxiang Che, Bing Qin
Abstract: Compared to traditional sentiment analysis, which only considers text, multimodal sentiment analysis needs to consider emotional signals from multimodal sources simultaneously and is therefore more consistent with the way how humans process sentiment in real-world scenarios. It involves processing emotional information from various sources such as natural language, images, videos, audio, physiological signals, etc. However, although other modalities also contain diverse emotional cues, natural language usually contains richer contextual information and therefore always occupies a crucial position in multimodal sentiment analysis. The emergence of ChatGPT has opened up immense potential for applying large language models (LLMs) to text-centric multimodal tasks. However, it is still unclear how existing LLMs can adapt better to text-centric multimodal sentiment analysis tasks. This survey aims to (1) present a comprehensive review of recent research in text-centric multimodal sentiment analysis tasks, (2) examine the potential of LLMs for text-centric multimodal sentiment analysis, outlining their approaches, advantages, and limitations, (3) summarize the application scenarios of LLM-based multimodal sentiment analysis technology, and (4) explore the challenges and potential research directions for multimodal sentiment analysis in the future.
Authors: Pia Pachinger, Janis Goldzycher, Anna Maria Planitzer, Wojciech Kusa, Allan Hanbury, Julia Neidhardt
Abstract: Model interpretability in toxicity detection greatly profits from token-level annotations. However, currently such annotations are only available in English. We introduce a dataset annotated for offensive language detection sourced from a news forum, notable for its incorporation of the Austrian German dialect, comprising 4,562 user comments. In addition to binary offensiveness classification, we identify spans within each comment constituting vulgar language or representing targets of offensive statements. We evaluate fine-tuned language models as well as large language models in a zero- and few-shot fashion. The results indicate that while fine-tuned models excel in detecting linguistic peculiarities such as vulgar dialect, large language models demonstrate superior performance in detecting offensiveness in AustroTox. We publish the data and code.
Authors: Zhi Qu, Chenchen Ding, Taro Watanabe
Abstract: Understanding representation transfer in multilingual neural machine translation can reveal the representational issue causing the zero-shot translation deficiency. In this work, we introduce the identity pair, a sentence translated into itself, to address the lack of the base measure in multilingual investigations, as the identity pair represents the optimal state of representation among any language transfers. In our analysis, we demonstrate that the encoder transfers the source language to the representational subspace of the target language instead of the language-agnostic state. Thus, the zero-shot translation deficiency arises because representations are entangled with other languages and are not transferred effectively to the target language. Based on our findings, we propose two methods: 1) low-rank language-specific embedding at the encoder, and 2) language-specific contrastive learning of the representation at the decoder. The experimental results on Europarl-15, TED-19, and OPUS-100 datasets show that our methods substantially enhance the performance of zero-shot translations by improving language transfer capacity, thereby providing practical evidence to support our conclusions.
Authors: Mingyu Zheng, Xinwei Feng, Qingyi Si, Qiaoqiao She, Zheng Lin, Wenbin Jiang, Weiping Wang
Abstract: Although great progress has been made by previous table understanding methods including recent approaches based on large language models (LLMs), they rely heavily on the premise that given tables must be converted into a certain text sequence (such as Markdown or HTML) to serve as model input. However, it is difficult to access such high-quality textual table representations in some real-world scenarios, and table images are much more accessible. Therefore, how to directly understand tables using intuitive visual information is a crucial and urgent challenge for developing more practical applications. In this paper, we propose a new problem, multimodal table understanding, where the model needs to generate correct responses to various table-related requests based on the given table image. To facilitate both the model training and evaluation, we construct a large-scale dataset named MMTab, which covers a wide spectrum of table images, instructions and tasks. On this basis, we develop Table-LLaVA, a generalist tabular multimodal large language model (MLLM), which significantly outperforms recent open-source MLLM baselines on 23 benchmarks under held-in and held-out settings. The code and data is available at this https://github.com/SpursGoZmy/Table-LLaVA
Authors: Qianli Wang, Tatiana Anikina, Nils Feldhus, Simon Ostermann, Sebastian M\"oller
Abstract: Conversational explainable artificial intelligence (ConvXAI) systems based on large language models (LLMs) have garnered significant interest from the research community in natural language processing (NLP) and human-computer interaction (HCI). Such systems can provide answers to user questions about explanations, have the potential to enhance users' comprehension and offer more information about the decision-making and generation processes of LLMs. Currently available ConvXAI systems are based on intent recognition rather than free chat. Thus, reliably grasping users' intentions in ConvXAI systems still presents a challenge, because there is a broad range of XAI methods to map requests onto and each of them can have multiple slots to take care of. In order to bridge this gap, we present CoXQL, the first dataset for user intent recognition in ConvXAI, covering 31 intents, seven of which require filling additional slots. Subsequently, we enhance an existing parsing approach by incorporating template validations, and conduct an evaluation of several LLMs on CoXQL using different parsing strategies. We conclude that the improved parsing approach (MP+) surpasses the performance of previous approaches. We also discover that intents with multiple slots remain highly challenging for LLMs.
Authors: Zile Qiao, Wei Ye, Yong Jiang, Tong Mo, Pengjun Xie, Weiping Li, Fei Huang, Shikun Zhang
Abstract: Retrieval-augmented language models (RALMs) have recently shown great potential in mitigating the limitations of implicit knowledge in LLMs, such as untimely updating of the latest expertise and unreliable retention of long-tail knowledge. However, since the external knowledge base, as well as the retriever, can not guarantee reliability, potentially leading to the knowledge retrieved not being helpful or even misleading for LLM generation. In this paper, we introduce Supportiveness-based Knowledge Rewriting (SKR), a robust and pluggable knowledge rewriter inherently optimized for LLM generation. Specifically, we introduce the novel concept of "supportiveness"--which represents how effectively a knowledge piece facilitates downstream tasks--by considering the perplexity impact of augmented knowledge on the response text of a white-box LLM. Based on knowledge supportiveness, we first design a training data curation strategy for our rewriter model, effectively identifying and filtering out poor or irrelevant rewrites (e.g., with low supportiveness scores) to improve data efficacy. We then introduce the direct preference optimization (DPO) algorithm to align the generated rewrites to optimal supportiveness, guiding the rewriter model to summarize augmented content that better improves the final response. Comprehensive evaluations across six popular knowledge-intensive tasks and four LLMs have demonstrated the effectiveness and superiority of SKR. With only 7B parameters, SKR has shown better knowledge rewriting capability over GPT-4, the current state-of-the-art general-purpose LLM.
Authors: Duanyu Feng, Bowen Qin, Chen Huang, Youcheng Huang, Zheng Zhang, Wenqiang Lei
Abstract: The success of the reward model in distinguishing between responses with subtle safety differences depends critically on the high-quality preference dataset, which should capture the fine-grained nuances of harmful and harmless responses. This motivates the need to develop a dataset involving preference margins, which accurately quantify how harmless one response is compared to another. In this paper, we take the first step to propose an effective and cost-efficient framework to promote the margin-enhanced preference dataset development. Our framework, Legend, Leverages representation engineering to annotate preference datasets. It constructs the specific direction within the LLM's embedding space that represents safety. By leveraging this safety direction, Legend can then leverage the semantic distances of paired responses along this direction to annotate margins automatically. We experimentally demonstrate our effectiveness in both reward modeling and harmless alignment for LLMs. Legend also stands out for its efficiency, requiring only the inference time rather than additional training. This efficiency allows for easier implementation and scalability, making Legend particularly valuable for practical applications in aligning LLMs with safe conversations.
Authors: Huijie Yao, Wengang Zhou, Hao Zhou, Houqiang Li
Abstract: Spoken language glossification (SLG) aims to translate the spoken language text into the sign language gloss, i.e., a written record of sign language. In this work, we present a framework named $S$emi-$S$upervised $S$poken $L$anguage $G$lossification ($S^3$LG) for SLG. To tackle the bottleneck of limited parallel data in SLG, our $S^3$LG incorporates large-scale monolingual spoken language text into SLG training. The proposed framework follows the self-training structure that iteratively annotates and learns from pseudo labels. Considering the lexical similarity and syntactic difference between sign language and spoken language, our $S^3$LG adopts both the rule-based heuristic and model-based approach for auto-annotation. During training, we randomly mix these complementary synthetic datasets and mark their differences with a special token. As the synthetic data may be less quality, the $S^3$LG further leverages consistency regularization to reduce the negative impact of noise in the synthetic data. Extensive experiments are conducted on public benchmarks to demonstrate the effectiveness of the $S^3$LG. Our code is available at \url{https://github.com/yaohj11/S3LG}.
Authors: Micol Spitale, Jiaee Cheong, Hatice Gunes
Abstract: Recent studies show bias in many machine learning models for depression detection, but bias in LLMs for this task remains unexplored. This work presents the first attempt to investigate the degree of gender bias present in existing LLMs (ChatGPT, LLaMA 2, and Bard) using both quantitative and qualitative approaches. From our quantitative evaluation, we found that ChatGPT performs the best across various performance metrics and LLaMA 2 outperforms other LLMs in terms of group fairness metrics. As qualitative fairness evaluation remains an open research question we propose several strategies (e.g., word count, thematic analysis) to investigate whether and how a qualitative evaluation can provide valuable insights for bias analysis beyond what is possible with quantitative evaluation. We found that ChatGPT consistently provides a more comprehensive, well-reasoned explanation for its prediction compared to LLaMA 2. We have also identified several themes adopted by LLMs to qualitatively evaluate gender fairness. We hope our results can be used as a stepping stone towards future attempts at improving qualitative evaluation of fairness for LLMs especially for high-stakes tasks such as depression detection.
Authors: Isidora Jekni\'c, David Schlangen, Alexander Koller
Abstract: Collaboration is an integral part of human dialogue. Typical task-oriented dialogue games assign asymmetric roles to the participants, which limits their ability to elicit naturalistic role-taking in collaboration and its negotiation. We present a novel and simple online setup that favors balanced collaboration: a two-player 2D object placement game in which the players must negotiate the goal state themselves. We show empirically that human players exhibit a variety of role distributions, and that balanced collaboration improves task performance. We also present an LLM-based baseline agent which demonstrates that automatic playing of our game is an interesting challenge for artificial systems.
Authors: Tikhonov Pavel, Anastasiya Ianina, Valentin Malykh
Abstract: Extractive summarization is a task of highlighting the most important parts of the text. We introduce a new approach to extractive summarization task using hidden clustering structure of the text. Experimental results on CNN/DailyMail demonstrate that our approach generates more accurate summaries than both extractive and abstractive methods, achieving state-of-the-art results in terms of ROUGE-2 metric exceeding the previous approaches by 10%. Additionally, we show that hidden structure of the text could be interpreted as aspects.
Authors: Gregorios A Katsios, Ning Sa, Tomek Strzalkowski
Abstract: The identification of Figurative Language (FL) features in text is crucial for various Natural Language Processing (NLP) tasks, where understanding of the author's intended meaning and its nuances is key for successful communication. At the same time, the use of a specific blend of various FL forms most accurately reflects a writer's style, rather than the use of any single construct, such as just metaphors or irony. Thus, we postulate that FL features could play an important role in Authorship Attribution (AA) tasks. We believe that our is the first computational study of AA based on FL use. Accordingly, we propose a Multi-task Figurative Language Model (MFLM) that learns to detect multiple FL features in text at once. We demonstrate, through detailed evaluation across multiple test sets, that the our model tends to perform equally or outperform specialized binary models in FL detection. Subsequently, we evaluate the predictive capability of joint FL features towards the AA task on three datasets, observing improved AA performance through the integration of MFLM embeddings.
Authors: Aman Ahluwalia, Suhrud Wani
Abstract: Large Language Models (LLMs) demonstrate remarkable capabilities in replicating human tasks and boosting productivity. However, their direct application for data extraction presents limitations due to a prioritisation of fluency over factual accuracy and a restricted ability to manipulate specific information. Therefore to overcome these limitations, this research leverages the knowledge representation power of pre-trained LLMs and the targeted information access enabled by RAG models, this research investigates a general-purpose accurate data scraping recipe for RAG models designed for language generation. To capture knowledge in a more modular and interpretable way, we use pre trained language models with a latent knowledge retriever, which allows the model to retrieve and attend over documents from a large corpus. We utilised RAG model architecture and did an in-depth analysis of their capabilities under three tasks: (i) Semantic Classification of HTML elements, (ii) Chunking HTML text for effective understanding, and (iii) comparing results from different LLMs and ranking algorithms. While previous work has developed dedicated architectures and training procedures for HTML understanding and extraction, we show that LLMs pre-trained on standard natural language with an addition of effective chunking, searching and ranking algorithms, can prove to be efficient data scraping tool to extract complex data from unstructured text. Future research directions include addressing the challenges of provenance tracking and dynamic knowledge updates within the proposed RAG-based data extraction framework. By overcoming these limitations, this approach holds the potential to revolutionise data extraction from vast repositories of textual information.
Authors: Benjamin Hsu, Xiaoyu Liu, Huayang Li, Yoshinari Fujinuma, Maria Nadejde, Xing Niu, Yair Kittenplon, Ron Litman, Raghavendra Pappagari
Abstract: Document translation poses a challenge for Neural Machine Translation (NMT) systems. Most document-level NMT systems rely on meticulously curated sentence-level parallel data, assuming flawless extraction of text from documents along with their precise reading order. These systems also tend to disregard additional visual cues such as the document layout, deeming it irrelevant. However, real-world documents often possess intricate text layouts that defy these assumptions. Extracting information from Optical Character Recognition (OCR) or heuristic rules can result in errors, and the layout (e.g., paragraphs, headers) may convey relationships between distant sections of text. This complexity is particularly evident in widely used PDF documents, which represent information visually. This paper addresses this gap by introducing M3T, a novel benchmark dataset tailored to evaluate NMT systems on the comprehensive task of translating semi-structured documents. This dataset aims to bridge the evaluation gap in document-level NMT systems, acknowledging the challenges posed by rich text layouts in real-world applications.
Authors: Wen-Ding Li, Kevin Ellis
Abstract: Programming-by-Examples (PBE) aims to generate an algorithm from input-output examples. Such systems are practically and theoretically important: from an end-user perspective, they are deployed to millions of people, and from an AI perspective, PBE corresponds to a very general form of few-shot inductive inference. Given the success of Large Language Models (LLMs) in code-generation tasks, we investigate here the extent to which LLMs can be said to have `solved' PBE. We experiment on classic domains such as lists and strings, and an uncommon graphics programming domain not well represented in typical pretraining data. We find that pretrained models are not effective at PBE, but that they can be fine-tuned for much higher performance, provided the test problems are in-distribution. We analyze empirically what causes these models to succeed and fail, and take steps toward understanding how to achieve better out-of-distribution generalization. Collectively these results suggest that LLMs make strong progress toward solving the typical suite of PBE tasks, potentially increasing the flexibility and applicability of PBE systems, while also identifying ways in which LLMs still fall short.
Authors: Junrui Ni, Liming Wang, Yang Zhang, Kaizhi Qian, Heting Gao, Mark Hasegawa-Johnson, Chang D. Yoo
Abstract: Recent advancements in supervised automatic speech recognition (ASR) have achieved remarkable performance, largely due to the growing availability of large transcribed speech corpora. However, most languages lack sufficient paired speech and text data to effectively train these systems. In this article, we tackle the challenge of developing ASR systems without paired speech and text corpora by proposing the removal of reliance on a phoneme lexicon. We explore a new research direction: word-level unsupervised ASR. Using a curated speech corpus containing only high-frequency English words, our system achieves a word error rate of nearly 20% without parallel transcripts or oracle word boundaries. Furthermore, we experimentally demonstrate that an unsupervised speech recognizer can emerge from joint speech-to-speech and text-to-text masked token-infilling. This innovative model surpasses the performance of previous unsupervised ASR models trained with direct distribution matching.
Authors: Anirudh Sundar, Jin Xu, William Gay, Christopher Richardson, Larry Heck
Abstract: An emerging area of research in situated and multimodal interactive conversations (SIMMC) includes interactions in scientific papers. Since scientific papers are primarily composed of text, equations, figures, and tables, SIMMC methods must be developed specifically for each component to support the depth of inquiry and interactions required by research scientists. This work introduces Conversational Papers (cPAPERS), a dataset of conversational question-answer pairs from reviews of academic papers grounded in these paper components and their associated references from scientific documents available on arXiv. We present a data collection strategy to collect these question-answer pairs from OpenReview and associate them with contextual information from LaTeX source files. Additionally, we present a series of baseline approaches utilizing Large Language Models (LLMs) in both zero-shot and fine-tuned configurations to address the cPAPERS dataset.
Authors: Xinyan Zhao, Yuan Sun, Wenlin Liu, Chau-Wai Wong
Abstract: This study is among the first to develop different prototypes of generative AI (GenAI) chatbots powered by GPT 4 to communicate hurricane preparedness information to diverse residents. Drawing from the Computers Are Social Actors (CASA) paradigm and the literature on disaster vulnerability and cultural tailoring, this study conducted a between-subjects experiment with 441 Black, Hispanic, and Caucasian residents of Florida. A computational analysis of chat logs (N = 7,848) shows that anthropomorphism and personalization are key communication topics in GenAI chatbot-user interactions. SEM results (N = 441) suggest that GenAI chatbots varying in tone formality and cultural tailoring significantly predict bot perceptions and, subsequently, hurricane preparedness outcomes. These results highlight the potential of using GenAI chatbots to improve diverse communities' disaster preparedness.
Authors: Zijin Hong, Zheng Yuan, Qinggang Zhang, Hao Chen, Junnan Dong, Feiran Huang, Xiao Huang
Abstract: Generating accurate SQL according to natural language questions (text-to-SQL) is a long-standing problem since it is challenging in user question understanding, database schema comprehension, and SQL generation. Conventional text-to-SQL systems include human engineering and deep neural networks. Subsequently, pre-trained language models (PLMs) have been developed and utilized for text-to-SQL tasks, achieving promising performance. As modern databases become more complex and corresponding user questions more challenging, PLMs with limited comprehension capabilities can lead to incorrect SQL generation. This necessitates more sophisticated and tailored optimization methods, which, in turn, restricts the applications of PLM-based systems. Most recently, large language models (LLMs) have demonstrated significant abilities in natural language understanding as the model scale remains increasing. Therefore, integrating the LLM-based implementation can bring unique opportunities, challenges, and solutions to text-to-SQL research. In this survey, we present a comprehensive review of LLM-based text-to-SQL. Specifically, we propose a brief overview of the current challenges and the evolutionary process of text-to-SQL. Then, we provide a detailed introduction to the datasets and metrics designed to evaluate text-to-SQL systems. After that, we present a systematic analysis of recent advances in LLM-based text-to-SQL. Finally, we discuss the remaining challenges in this field and propose expectations for future directions.
Authors: Yutong Wang, Jiali Zeng, Xuebo Liu, Fandong Meng, Jie Zhou, Min Zhang
Abstract: Large language models (LLMs) have exhibited remarkable performance in various natural language processing tasks. Techniques like instruction tuning have effectively enhanced the proficiency of LLMs in the downstream task of machine translation. However, the existing approaches fail to yield satisfactory translation outputs that match the quality of supervised neural machine translation (NMT) systems. One plausible explanation for this discrepancy is that the straightforward prompts employed in these methodologies are unable to fully exploit the acquired instruction-following capabilities. To this end, we propose the TasTe framework, which stands for translating through self-reflection. The self-reflection process includes two stages of inference. In the first stage, LLMs are instructed to generate preliminary translations and conduct self-assessments on these translations simultaneously. In the second stage, LLMs are tasked to refine these preliminary translations according to the evaluation results. The evaluation results in four language directions on the WMT22 benchmark reveal the effectiveness of our approach compared to existing methods. Our work presents a promising approach to unleash the potential of LLMs and enhance their capabilities in MT. The codes and datasets are open-sourced at https://github.com/YutongWang1216/ReflectionLLMMT.
Authors: Yuling Gu, Oyvind Tafjord, Bailey Kuehl, Dany Haddad, Jesse Dodge, Hannaneh Hajishirzi
Abstract: Progress in AI is often demonstrated by new models claiming improved performance on tasks measuring model capabilities. Evaluating language models in particular is challenging, as small changes to how a model is evaluated on a task can lead to large changes in measured performance. There is no common standard setup, so different models are evaluated on the same tasks in different ways, leading to claims about which models perform best not being reproducible. We propose OLMES, a completely documented, practical, open standard for reproducible LLM evaluations. In developing this standard, we identify and review the varying factors in evaluation practices adopted by the community - such as details of prompt formatting, choice of in-context examples, probability normalizations, and task formulation. In particular, OLMES supports meaningful comparisons between smaller base models that require the unnatural "cloze" formulation of multiple-choice questions against larger models that can utilize the original formulation. OLMES includes well-considered recommendations guided by results from existing literature as well as new experiments investigating open questions.
Authors: Zhangchen Xu, Fengqing Jiang, Luyao Niu, Yuntian Deng, Radha Poovendran, Yejin Choi, Bill Yuchen Lin
Abstract: High-quality instruction data is critical for aligning large language models (LLMs). Although some models, such as Llama-3-Instruct, have open weights, their alignment data remain private, which hinders the democratization of AI. High human labor costs and a limited, predefined scope for prompting prevent existing open-source data creation methods from scaling effectively, potentially limiting the diversity and quality of public alignment datasets. Is it possible to synthesize high-quality instruction data at scale by extracting it directly from an aligned LLM? We present a self-synthesis method for generating large-scale alignment data named Magpie. Our key observation is that aligned LLMs like Llama-3-Instruct can generate a user query when we input only the left-side templates up to the position reserved for user messages, thanks to their auto-regressive nature. We use this method to prompt Llama-3-Instruct and generate 4 million instructions along with their corresponding responses. We perform a comprehensive analysis of the extracted data and select 300K high-quality instances. To compare Magpie data with other public instruction datasets, we fine-tune Llama-3-8B-Base with each dataset and evaluate the performance of the fine-tuned models. Our results indicate that in some tasks, models fine-tuned with Magpie perform comparably to the official Llama-3-8B-Instruct, despite the latter being enhanced with 10 million data points through supervised fine-tuning (SFT) and subsequent feedback learning. We also show that using Magpie solely for SFT can surpass the performance of previous public datasets utilized for both SFT and preference optimization, such as direct preference optimization with UltraFeedback. This advantage is evident on alignment benchmarks such as AlpacaEval, ArenaHard, and WildBench.
Authors: Ditto PS, Jithin VG, Adarsh MS
Abstract: In recent years, large language models have demonstrated remarkable performance across various natural language processing (NLP) tasks. However, deploying these models for real-world applications often requires efficient inference solutions to handle the computational demands. In this paper, we explore the utilization of CPUs for accelerating the inference of large language models. Specifically, we introduce a parallelized approach to enhance throughput by 1) Exploiting the parallel processing capabilities of modern CPU architectures, 2) Batching the inference request. Our evaluation shows the accelerated inference engine gives an 18-22x improvement in the generated token per sec. The improvement is more with longer sequence and larger models. In addition to this, we can also run multiple workers in the same machine with NUMA node isolation to further improvement in tokens/s. Table 2, we have received 4x additional improvement with 4 workers. This would also make Gen-AI based products and companies environment friendly, our estimates shows that CPU usage for Inference could reduce the power consumption of LLMs by 48.9% while providing production ready throughput and latency.
Authors: Wanaiu Huang
Abstract: Semantic information is vital for human interaction, and decoding it from brain activity enables non-invasive clinical augmentative and alternative communication. While there has been significant progress in reconstructing visual images, few studies have focused on the language aspect. To address this gap, leveraging the powerful capabilities of the decoder-based vision-language pretrained model CoCa, this paper proposes BrainChat, a simple yet effective generative framework aimed at rapidly accomplishing semantic information decoding tasks from brain activity, including fMRI question answering and fMRI captioning. BrainChat employs the self-supervised approach of Masked Brain Modeling to encode sparse fMRI data, obtaining a more compact embedding representation in the latent space. Subsequently, BrainChat bridges the gap between modalities by applying contrastive loss, resulting in aligned representations of fMRI, image, and text embeddings. Furthermore, the fMRI embeddings are mapped to the generative Brain Decoder via cross-attention layers, where they guide the generation of textual content about fMRI in a regressive manner by minimizing caption loss. Empirically, BrainChat exceeds the performance of existing state-of-the-art methods in the fMRI captioning task and, for the first time, implements fMRI question answering. Additionally, BrainChat is highly flexible and can achieve high performance without image data, making it better suited for real-world scenarios with limited data.
Authors: Jun Gao, Qian Qiao, Ziqiang Cao, Zili Wang, Wenjie Li
Abstract: In-context learning (ICL) facilitates Large Language Models (LLMs) exhibiting emergent ability on downstream tasks without updating billions of parameters. However, in the area of multi-modal Large Language Models (MLLMs), two problems hinder the application of multi-modal ICL: (1) Most primary MLLMs are only trained on single-image datasets, making them unable to read multi-modal demonstrations. (2) With the demonstrations increasing, thousands of visual tokens highly challenge hardware and degrade ICL performance. During preliminary explorations, we discovered that the inner LLM tends to focus more on the linguistic modality within multi-modal demonstrations to generate responses. Therefore, we propose a general and light-weighted framework \textbf{AIM} to tackle the mentioned problems through \textbf{A}ggregating \textbf{I}mage information of \textbf{M}ultimodal demonstrations to the dense latent space of the corresponding linguistic part. Specifically, AIM first uses the frozen backbone MLLM to read each image-text demonstration and extracts the vector representations on top of the text. These vectors naturally fuse the information of the image-text pair, and AIM transforms them into fused virtual tokens acceptable for the inner LLM via a trainable projection layer. Ultimately, these fused tokens function as variants of multi-modal demonstrations, fed into the MLLM to direct its response to the current query as usual. Because these fused tokens stem from the textual component of the image-text pair, a multi-modal demonstration is nearly reduced to a pure textual demonstration, thus seamlessly applying to any MLLMs. With its de facto MLLM frozen, AIM is parameter-efficient and we train it on public multi-modal web corpora which have nothing to do with downstream test tasks.
Authors: Lichang Chen, Jiuhai Chen, Chenxi Liu, John Kirchenbauer, Davit Soselia, Chen Zhu, Tom Goldstein, Tianyi Zhou, Heng Huang
Abstract: Reinforcement learning with human feedback~(RLHF) is critical for aligning Large Language Models (LLMs) with human preference. Compared to the widely studied offline version of RLHF, \emph{e.g.} direct preference optimization (DPO), recent works have shown that the online variants achieve even better alignment. However, online alignment requires on-the-fly generation of new training data, which is costly, hard to parallelize, and suffers from varying quality and utility. In this paper, we propose a more efficient data exploration strategy for online preference tuning (OPTune), which does not rely on human-curated or pre-collected teacher responses but dynamically samples informative responses for on-policy preference alignment. During data generation, OPTune only selects prompts whose (re)generated responses can potentially provide more informative and higher-quality training signals than the existing responses. In the training objective, OPTune reweights each generated response (pair) by its utility in improving the alignment so that learning can be focused on the most helpful samples. Throughout our evaluations, OPTune'd LLMs maintain the instruction-following benefits provided by standard preference tuning whilst enjoying 1.27-1.56x faster training speed due to the efficient data exploration strategy.
Authors: Nirmalya Thakur, Vanessa Su, Mingchen Shao, Kesha A. Patel, Hongseok Jeong, Victoria Knieling, Andrew Brian
Abstract: The work of this paper presents a dataset that contains the data of 4011 videos about the ongoing outbreak of measles published on 264 websites on the internet between January 1, 2024, and May 31, 2024. The dataset is available at https://dx.doi.org/10.21227/40s8-xf63. These websites primarily include YouTube and TikTok, which account for 48.6% and 15.2% of the videos, respectively. The remainder of the websites include Instagram and Facebook as well as the websites of various global and local news organizations. For each of these videos, the URL of the video, title of the post, description of the post, and the date of publication of the video are presented as separate attributes in the dataset. After developing this dataset, sentiment analysis (using VADER), subjectivity analysis (using TextBlob), and fine-grain sentiment analysis (using DistilRoBERTa-base) of the video titles and video descriptions were performed. This included classifying each video title and video description into (i) one of the sentiment classes i.e. positive, negative, or neutral, (ii) one of the subjectivity classes i.e. highly opinionated, neutral opinionated, or least opinionated, and (iii) one of the fine-grain sentiment classes i.e. fear, surprise, joy, sadness, anger, disgust, or neutral. These results are presented as separate attributes in the dataset for the training and testing of machine learning algorithms for performing sentiment analysis or subjectivity analysis in this field as well as for other applications. Finally, this paper also presents a list of open research questions that may be investigated using this dataset.
Authors: Shahin Amiriparian, Lukas Christ, Alexander Kathan, Maurice Gerczuk, Niklas M\"uller, Steffen Klug, Lukas Stappen, Andreas K\"onig, Erik Cambria, Bj\"orn Schuller, Simone Eulitz
Abstract: The Multimodal Sentiment Analysis Challenge (MuSe) 2024 addresses two contemporary multimodal affect and sentiment analysis problems: In the Social Perception Sub-Challenge (MuSe-Perception), participants will predict 16 different social attributes of individuals such as assertiveness, dominance, likability, and sincerity based on the provided audio-visual data. The Cross-Cultural Humor Detection Sub-Challenge (MuSe-Humor) dataset expands upon the Passau Spontaneous Football Coach Humor (Passau-SFCH) dataset, focusing on the detection of spontaneous humor in a cross-lingual and cross-cultural setting. The main objective of MuSe 2024 is to unite a broad audience from various research domains, including multimodal sentiment analysis, audio-visual affective computing, continuous signal processing, and natural language processing. By fostering collaboration and exchange among experts in these fields, the MuSe 2024 endeavors to advance the understanding and application of sentiment analysis and affective computing across multiple modalities. This baseline paper provides details on each sub-challenge and its corresponding dataset, extracted features from each data modality, and discusses challenge baselines. For our baseline system, we make use of a range of Transformers and expert-designed features and train Gated Recurrent Unit (GRU)-Recurrent Neural Network (RNN) models on them, resulting in a competitive baseline system. On the unseen test datasets of the respective sub-challenges, it achieves a mean Pearson's Correlation Coefficient ($\rho$) of 0.3573 for MuSe-Perception and an Area Under the Curve (AUC) value of 0.8682 for MuSe-Humor.
Authors: Jayaram Raghuram, George Kesidis, David J. Miller
Abstract: A Trojan in a language model can be inserted when the model is refined for a particular application such as determining the sentiment of product reviews. In this paper, we clarify and empirically explore variations of the data-poisoning threat model. We then empirically assess two simple defenses each for a different defense scenario. Finally, we provide a brief survey of related attacks and defenses.
Authors: Ahmad Rashid, Ruotian Wu, Julia Grosse, Agustinus Kristiadi, Pascal Poupart
Abstract: Large language models (LLMs) can significantly be improved by aligning to human preferences -- the so-called reinforcement learning from human feedback (RLHF). However, the cost of fine-tuning an LLM is prohibitive for many users. Due to their ability to bypass LLM finetuning, tokenwise reward-guided text generation (RGTG) methods have recently been proposed. They use a reward model trained on full sequences to score partial sequences during a tokenwise decoding, in a bid to steer the generation towards sequences with high rewards. However, these methods have so far been only heuristically motivated and poorly analyzed. In this work, we show that reward models trained on full sequences are not compatible with scoring partial sequences. To alleviate this issue, we propose to explicitly train a Bradley-Terry reward model on partial sequences, and autoregressively sample from the implied tokenwise policy during decoding time. We study the property of this reward model and the implied policy. In particular, we show that this policy is proportional to the ratio of two distinct RLHF policies. We show that our simple approach outperforms previous RGTG methods and achieves similar performance as strong offline baselines but without large-scale LLM finetuning.
Authors: Saffron Huang, Divya Siddarth, Liane Lovitt, Thomas I. Liao, Esin Durmus, Alex Tamkin, Deep Ganguli
Abstract: There is growing consensus that language model (LM) developers should not be the sole deciders of LM behavior, creating a need for methods that enable the broader public to collectively shape the behavior of LM systems that affect them. To address this need, we present Collective Constitutional AI (CCAI): a multi-stage process for sourcing and integrating public input into LMs-from identifying a target population to sourcing principles to training and evaluating a model. We demonstrate the real-world practicality of this approach by creating what is, to our knowledge, the first LM fine-tuned with collectively sourced public input and evaluating this model against a baseline model trained with established principles from a LM developer. Our quantitative evaluations demonstrate several benefits of our approach: the CCAI-trained model shows lower bias across nine social dimensions compared to the baseline model, while maintaining equivalent performance on language, math, and helpful-harmless evaluations. Qualitative comparisons of the models suggest that the models differ on the basis of their respective constitutions, e.g., when prompted with contentious topics, the CCAI-trained model tends to generate responses that reframe the matter positively instead of a refusal. These results demonstrate a promising, tractable pathway toward publicly informed development of language models.
Authors: Lin Zhang, Xin Wang, Erica Cooper, Mireia Diez, Federico Landini, Nicholas Evans, Junichi Yamagishi
Abstract: This paper defines Spoof Diarization as a novel task in the Partial Spoof (PS) scenario. It aims to determine what spoofed when, which includes not only locating spoof regions but also clustering them according to different spoofing methods. As a pioneering study in spoof diarization, we focus on defining the task, establishing evaluation metrics, and proposing a benchmark model, namely the Countermeasure-Condition Clustering (3C) model. Utilizing this model, we first explore how to effectively train countermeasures to support spoof diarization using three labeling schemes. We then utilize spoof localization predictions to enhance the diarization performance. This first study reveals the high complexity of the task, even in restricted scenarios where only a single speaker per audio file and an oracle number of spoofing methods are considered. Our code is available at https://github.com/nii-yamagishilab/PartialSpoof.
Authors: Andrea Burns, Kate Saenko, Bryan A. Plummer
Abstract: Mobile app user interfaces (UIs) are rich with action, text, structure, and image content that can be utilized to learn generic UI representations for tasks like automating user commands, summarizing content, and evaluating the accessibility of user interfaces. Prior work has learned strong visual representations with local or global captioning losses, but fails to retain both granularities. To combat this, we propose Textual Foresight, a novel pretraining objective for learning UI screen representations. Textual Foresight generates global text descriptions of future UI states given a current UI and local action taken. Our approach requires joint reasoning over elements and entire screens, resulting in improved UI features: on generation tasks, UI agents trained with Textual Foresight outperform state-of-the-art by 2% with 28x fewer images. We train with our newly constructed mobile app dataset, OpenApp, which results in the first public dataset for app UI representation learning. OpenApp enables new baselines, and we find Textual Foresight improves average task performance over them by 5.7% while having access to 2x less data.
Authors: Elaheh Baharlouei, Mahsa Shafaei, Yigeng Zhang, Hugo Jair Escalante, Thamar Solorio
Abstract: We address the challenge of detecting questionable content in online media, specifically the subcategory of comic mischief. This type of content combines elements such as violence, adult content, or sarcasm with humor, making it difficult to detect. Employing a multimodal approach is vital to capture the subtle details inherent in comic mischief content. To tackle this problem, we propose a novel end-to-end multimodal system for the task of comic mischief detection. As part of this contribution, we release a novel dataset for the targeted task consisting of three modalities: video, text (video captions and subtitles), and audio. We also design a HIerarchical Cross-attention model with CAPtions (HICCAP) to capture the intricate relationships among these modalities. The results show that the proposed approach makes a significant improvement over robust baselines and state-of-the-art models for comic mischief detection and its type classification. This emphasizes the potential of our system to empower users, to make informed decisions about the online content they choose to see. In addition, we conduct experiments on the UCF101, HMDB51, and XD-Violence datasets, comparing our model against other state-of-the-art approaches showcasing the outstanding performance of our proposed model in various scenarios.
Authors: Yerbolat Khassanov, Zhipeng Chen, Tianfeng Chen, Tze Yuang Chong, Wei Li, Jun Zhang, Lu Lu, Yuxuan Wang
Abstract: This paper addresses challenges in integrating new languages into a pre-trained multilingual automatic speech recognition (mASR) system, particularly in scenarios where training data for existing languages is limited or unavailable. The proposed method employs a dual-pipeline with low-rank adaptation (LoRA). It maintains two data flow pipelines-one for existing languages and another for new languages. The primary pipeline follows the standard flow through the pre-trained parameters of mASR, while the secondary pipeline additionally utilizes language-specific parameters represented by LoRA and a separate output decoder module. Importantly, the proposed approach minimizes the performance degradation of existing languages and enables a language-agnostic operation mode, facilitated by a decoder selection strategy. We validate the effectiveness of the proposed method by extending the pre-trained Whisper model to 19 new languages from the FLEURS dataset
Authors: Roger Waleffe, Wonmin Byeon, Duncan Riach, Brandon Norick, Vijay Korthikanti, Tri Dao, Albert Gu, Ali Hatamizadeh, Sudhakar Singh, Deepak Narayanan, Garvit Kulshreshtha, Vartika Singh, Jared Casper, Jan Kautz, Mohammad Shoeybi, Bryan Catanzaro
Abstract: Selective state-space models (SSMs) like Mamba overcome some of the shortcomings of Transformers, such as quadratic computational complexity with sequence length and large inference-time memory requirements from the key-value cache. Moreover, recent studies have shown that SSMs can match or exceed the language modeling capabilities of Transformers, making them an attractive alternative. In a controlled setting (e.g., same data), however, studies so far have only presented small scale experiments comparing SSMs to Transformers. To understand the strengths and weaknesses of these architectures at larger scales, we present a direct comparison between 8B-parameter Mamba, Mamba-2, and Transformer models trained on the same datasets of up to 3.5T tokens. We also compare these models to a hybrid architecture consisting of 43% Mamba-2, 7% attention, and 50% MLP layers (Mamba-2-Hybrid). Using a diverse set of tasks, we answer the question of whether Mamba models can match Transformers at larger training budgets. Our results show that while pure SSMs match or exceed Transformers on many tasks, they lag behind Transformers on tasks which require strong copying or in-context learning abilities (e.g., 5-shot MMLU, Phonebook) or long-context reasoning. In contrast, we find that the 8B Mamba-2-Hybrid exceeds the 8B Transformer on all 12 standard tasks we evaluated (+2.65 points on average) and is predicted to be up to 8x faster when generating tokens at inference time. To validate long-context capabilities, we provide additional experiments evaluating variants of the Mamba-2-Hybrid and Transformer extended to support 16K, 32K, and 128K sequences. On an additional 23 long-context tasks, the hybrid model continues to closely match or exceed the Transformer on average. To enable further study, we release the checkpoints as well as the code used to train our models as part of NVIDIA's Megatron-LM project.
Authors: Anfeng Xu, Kevin Huang, Tiantian Feng, Lue Shen, Helen Tager-Flusberg, Shrikanth Narayanan
Abstract: Speech foundation models, trained on vast datasets, have opened unique opportunities in addressing challenging low-resource speech understanding, such as child speech. In this work, we explore the capabilities of speech foundation models on child-adult speaker diarization. We show that exemplary foundation models can achieve 39.5% and 62.3% relative reductions in Diarization Error Rate and Speaker Confusion Rate, respectively, compared to previous speaker diarization methods. In addition, we benchmark and evaluate the speaker diarization results of the speech foundation models with varying the input audio window size, speaker demographics, and training data ratio. Our results highlight promising pathways for understanding and adopting speech foundation models to facilitate child speech understanding.
Authors: Eungbeom Kim, Hantae Kim, Kyogu Lee
Abstract: Transformer encoder with connectionist temporal classification (CTC) framework is widely used for automatic speech recognition (ASR). However, knowledge distillation (KD) for ASR displays a problem of disagreement between teacher-student models in frame-level alignment which ultimately hinders it from improving the student model's performance. In order to resolve this problem, this paper introduces a self-knowledge distillation (SKD) method that guides the frame-level alignment during the training time. In contrast to the conventional method using separate teacher and student models, this study introduces a simple and effective method sharing encoder layers and applying the sub-model as the student model. Overall, our approach is effective in improving both the resource efficiency as well as performance. We also conducted an experimental analysis of the spike timings to illustrate that the proposed method improves performance by reducing the alignment disagreement.
Authors: Luis Miguel Vieira da Silva, Aljosha K\"ocher, Felix Gehlhoff, Alexander Fay
Abstract: To achieve a flexible and adaptable system, capability ontologies are increasingly leveraged to describe functions in a machine-interpretable way. However, modeling such complex ontological descriptions is still a manual and error-prone task that requires a significant amount of effort and ontology expertise. This contribution presents an innovative method to automate capability ontology modeling using Large Language Models (LLMs), which have proven to be well suited for such tasks. Our approach requires only a natural language description of a capability, which is then automatically inserted into a predefined prompt using a few-shot prompting technique. After prompting an LLM, the resulting capability ontology is automatically verified through various steps in a loop with the LLM to check the overall correctness of the capability ontology. First, a syntax check is performed, then a check for contradictions, and finally a check for hallucinations and missing ontology elements. Our method greatly reduces manual effort, as only the initial natural language description and a final human review and possible correction are necessary, thereby streamlining the capability ontology generation process.
Authors: Joseba Fernandez de Landa, Rodrigo Agerri
Abstract: Social media users express their political preferences via interaction with other users, by spontaneous declarations or by participation in communities within the network. This makes a social network such as Twitter a valuable data source to study computational science approaches to political learning inference. In this work we focus on three diverse regions in Spain (Basque Country, Catalonia and Galicia) to explore various methods for multi-party categorization, required to analyze evolving and complex political landscapes, and compare it with binary left-right approaches. We use a two-step method involving unsupervised user representations obtained from the retweets and their subsequent use for political leaning detection. Comprehensive experimentation on a newly collected and curated dataset comprising labeled users and their interactions demonstrate the effectiveness of using Relational Embeddings as representation method for political ideology detection in both binary and multi-party frameworks, even with limited training data. Finally, data visualization illustrates the ability of the Relational Embeddings to capture intricate intra-group and inter-group political affinities.
Authors: Masaya Kawamura, Ryuichi Yamamoto, Yuma Shirahata, Takuya Hasumi, Kentaro Tachibana
Abstract: We introduce LibriTTS-P, a new corpus based on LibriTTS-R that includes utterance-level descriptions (i.e., prompts) of speaking style and speaker-level prompts of speaker characteristics. We employ a hybrid approach to construct prompt annotations: (1) manual annotations that capture human perceptions of speaker characteristics and (2) synthetic annotations on speaking style. Compared to existing English prompt datasets, our corpus provides more diverse prompt annotations for all speakers of LibriTTS-R. Experimental results for prompt-based controllable TTS demonstrate that the TTS model trained with LibriTTS-P achieves higher naturalness than the model using the conventional dataset. Furthermore, the results for style captioning tasks show that the model utilizing LibriTTS-P generates 2.5 times more accurate words than the model using a conventional dataset. Our corpus, LibriTTS-P, is available at https://github.com/line/LibriTTS-P.
Authors: Thomas Roland Barillot, Alex De Castro
Abstract: This works reports evidence for the topological signatures of ambiguity in sentence embeddings that could be leveraged for ranking and/or explanation purposes in the context of vector search and Retrieval Augmented Generation (RAG) systems. We proposed a working definition of ambiguity and designed an experiment where we have broken down a proprietary dataset into collections of chunks of varying size - 3, 5, and 10 lines and used the different collections successively as queries and answers sets. It allowed us to test the signatures of ambiguity with removal of confounding factors. Our results show that proxy ambiguous queries (size 10 queries against size 3 documents) display different distributions of homologies 0 and 1 based features than proxy clear queries (size 5 queries against size 10 documents). We then discuss those results in terms increased manifold complexity and/or approximately discontinuous embedding submanifolds. Finally we propose a strategy to leverage those findings as a new scoring strategy of semantic similarities.
Authors: Jayneel Parekh, Pegah Khayatan, Mustafa Shukor, Alasdair Newson, Matthieu Cord
Abstract: Large multimodal models (LMMs) combine unimodal encoders and large language models (LLMs) to perform multimodal tasks. Despite recent advancements towards the interpretability of these models, understanding internal representations of LMMs remains largely a mystery. In this paper, we present a novel framework for the interpretation of LMMs. We propose a dictionary learning based approach, applied to the representation of tokens. The elements of the learned dictionary correspond to our proposed concepts. We show that these concepts are well semantically grounded in both vision and text. Thus we refer to these as "multi-modal concepts". We qualitatively and quantitatively evaluate the results of the learnt concepts. We show that the extracted multimodal concepts are useful to interpret representations of test samples. Finally, we evaluate the disentanglement between different concepts and the quality of grounding concepts visually and textually. We will publicly release our code.
Authors: Pingzhi Li, Xiaolong Jin, Yu Cheng, Tianlong Chen
Abstract: Large Language Models~(LLMs) have become foundational in the realm of natural language processing, demonstrating performance improvements as model sizes increase. The Mixture-of-Experts~(MoE) approach offers a promising way to scale LLMs more efficiently by using fewer computational FLOPs through sparse activation. However, it suffers from significant memory overheads, necessitating model compression techniques. Post-training quantization, a popular method for model compression, proves less effective when directly applied to MoE models due to MoE's overlooked inherent sparsity. This paper explores several MoE structure-aware quantization heuristics, ranging from coarse to fine granularity, from MoE block to individual linear weight. Our investigations reveal critical principles: different MoE structures (i.e., blocks, experts, linear layers) require varying numbers of weight bits for effective and efficient quantization. Conclusions are supported by extensive benchmarking across two representative MoE models and six tasks. We further introduce novel enhancements to more accurately identify the most critical weights in MoE quantization that necessitate higher bit allocations, including the linear weight outlier scorer and MoE block scorer. Additionally, subsequent experiments validate our findings in the context of both weight and activation quantization.
Authors: Iwen E. Kang, Christophe Van Gysel, Man-Hung Siu
Abstract: Voice assistants increasingly use on-device Automatic Speech Recognition (ASR) to ensure speed and privacy. However, due to resource constraints on the device, queries pertaining to complex information domains often require further processing by a search engine. For such applications, we propose a novel Transformer based model capable of rescoring and rewriting, by exploring full context of the N-best hypotheses in parallel. We also propose a new discriminative sequence training objective that can work well for both rescore and rewrite tasks. We show that our Rescore+Rewrite model outperforms the Rescore-only baseline, and achieves up to an average 8.6% relative Word Error Rate (WER) reduction over the ASR system by itself.
Authors: Hanieh Khorashadizadeh, Fatima Zahra Amara, Morteza Ezzabady, Fr\'ed\'eric Ieng, Sanju Tiwari, Nandana Mihindukulasooriya, Jinghua Groppe, Soror Sahri, Farah Benamara, Sven Groppe
Abstract: This survey investigates the synergistic relationship between Large Language Models (LLMs) and Knowledge Graphs (KGs), which is crucial for advancing AI's capabilities in understanding, reasoning, and language processing. It aims to address gaps in current research by exploring areas such as KG Question Answering, ontology generation, KG validation, and the enhancement of KG accuracy and consistency through LLMs. The paper further examines the roles of LLMs in generating descriptive texts and natural language queries for KGs. Through a structured analysis that includes categorizing LLM-KG interactions, examining methodologies, and investigating collaborative uses and potential biases, this study seeks to provide new insights into the combined potential of LLMs and KGs. It highlights the importance of their interaction for improving AI applications and outlines future research directions.
Authors: Yuanchao Li, Peter Bell, Catherine Lai
Abstract: Text data is commonly utilized as a primary input to enhance Speech Emotion Recognition (SER) performance and reliability. However, the reliance on human-transcribed text in most studies impedes the development of practical SER systems, creating a gap between in-lab research and real-world scenarios where Automatic Speech Recognition (ASR) serves as the text source. Hence, this study benchmarks SER performance using ASR transcripts with varying Word Error Rates (WERs) on well-known corpora: IEMOCAP, CMU-MOSI, and MSP-Podcast. Our evaluation includes text-only and bimodal SER with diverse fusion techniques, aiming for a comprehensive analysis that uncovers novel findings and challenges faced by current SER research. Additionally, we propose a unified ASR error-robust framework integrating ASR error correction and modality-gated fusion, achieving lower WER and higher SER results compared to the best-performing ASR transcript. This research is expected to provide insights into SER with ASR assistance, especially for real-world applications.
Authors: Sanyam Kapoor, Nate Gruver, Manley Roberts, Katherine Collins, Arka Pal, Umang Bhatt, Adrian Weller, Samuel Dooley, Micah Goldblum, Andrew Gordon Wilson
Abstract: When using large language models (LLMs) in high-stakes applications, we need to know when we can trust their predictions. Some works argue that prompting high-performance LLMs is sufficient to produce calibrated uncertainties, while others introduce sampling methods that can be prohibitively expensive. In this work, we first argue that prompting on its own is insufficient to achieve good calibration and then show that fine-tuning on a small dataset of correct and incorrect answers can create an uncertainty estimate with good generalization and small computational overhead. We show that a thousand graded examples are sufficient to outperform baseline methods and that training through the features of a model is necessary for good performance and tractable for large open-source models when using LoRA. We also investigate the mechanisms that enable reliable LLM uncertainty estimation, finding that many models can be used as general-purpose uncertainty estimators, applicable not just to their own uncertainties but also the uncertainty of other models. Lastly, we show that uncertainty estimates inform human use of LLMs in human-AI collaborative settings through a user study.
Authors: Chun-Yi Kuan, Wei-Ping Huang, Hung-yi Lee
Abstract: Large audio-language models (LALMs) enhance traditional large language models by integrating audio perception capabilities, allowing them to tackle audio-related tasks. Previous research has primarily focused on assessing the performance of LALMs across various tasks, yet overlooking their reliability, particularly concerning issues like object hallucination. In our study, we introduce methods to assess the extent of object hallucination of publicly available LALMs. Our findings reveal that LALMs are comparable to specialized audio captioning models in their understanding of audio content, but struggle to answer discriminative questions, specifically those requiring the identification of the presence of particular object sounds within an audio clip. This limitation highlights a critical weakness in current LALMs: their inadequate understanding of discriminative queries. Moreover, we explore the potential of prompt engineering to enhance LALMs' performance on discriminative questions.
Authors: Xuehai He, Weixi Feng, Kaizhi Zheng, Yujie Lu, Wanrong Zhu, Jiachen Li, Yue Fan, Jianfeng Wang, Linjie Li, Zhengyuan Yang, Kevin Lin, William Yang Wang, Lijuan Wang, Xin Eric Wang
Abstract: Multimodal Language Language Models (MLLMs) demonstrate the emerging abilities of "world models" -- interpreting and reasoning about complex real-world dynamics. To assess these abilities, we posit videos are the ideal medium, as they encapsulate rich representations of real-world dynamics and causalities. To this end, we introduce MMWorld, a new benchmark for multi-discipline, multi-faceted multimodal video understanding. MMWorld distinguishes itself from previous video understanding benchmarks with two unique advantages: (1) multi-discipline, covering various disciplines that often require domain expertise for comprehensive understanding; (2) multi-faceted reasoning, including explanation, counterfactual thinking, future prediction, etc. MMWorld consists of a human-annotated dataset to evaluate MLLMs with questions about the whole videos and a synthetic dataset to analyze MLLMs within a single modality of perception. Together, MMWorld encompasses 1,910 videos across seven broad disciplines and 69 subdisciplines, complete with 6,627 question-answer pairs and associated captions. The evaluation includes 2 proprietary and 10 open-source MLLMs, which struggle on MMWorld (e.g., GPT-4V performs the best with only 52.3\% accuracy), showing large room for improvement. Further ablation studies reveal other interesting findings such as models' different skill sets from humans. We hope MMWorld can serve as an essential step towards world model evaluation in videos.
Authors: Soufiane Hayou, Nikhil Ghosh, Bin Yu
Abstract: In this paper, we study the role of initialization in Low Rank Adaptation (LoRA) as originally introduced in Hu et al. (2021). Essentially, to start from the pretrained model as initialization for finetuning, one can either initialize B to zero and A to random (default initialization in PEFT package), or vice-versa. In both cases, the product BA is equal to zero at initialization, which makes finetuning starts from the pretrained model. These two initialization schemes are seemingly similar. They should in-principle yield the same performance and share the same optimal learning rate. We demonstrate that this is an incorrect intuition and that the first scheme (initializing B to zero and A to random) on average yields better performance compared to the other scheme. Our theoretical analysis shows that the reason behind this might be that the first initialization allows the use of larger learning rates (without causing output instability) compared to the second initialization, resulting in more efficient learning of the first scheme. We validate our results with extensive experiments on LLMs.
Authors: Xianhang Li, Haoqin Tu, Mude Hui, Zeyu Wang, Bingchen Zhao, Junfei Xiao, Sucheng Ren, Jieru Mei, Qing Liu, Huangjie Zheng, Yuyin Zhou, Cihang Xie
Abstract: Web-crawled image-text pairs are inherently noisy. Prior studies demonstrate that semantically aligning and enriching textual descriptions of these pairs can significantly enhance model training across various vision-language tasks, particularly text-to-image generation. However, large-scale investigations in this area remain predominantly closed-source. Our paper aims to bridge this community effort, leveraging the powerful and \textit{open-sourced} LLaMA-3, a GPT-4 level LLM. Our recaptioning pipeline is simple: first, we fine-tune a LLaMA-3-8B powered LLaVA-1.5 and then employ it to recaption 1.3 billion images from the DataComp-1B dataset. Our empirical results confirm that this enhanced dataset, Recap-DataComp-1B, offers substantial benefits in training advanced vision-language models. For discriminative models like CLIP, we observe enhanced zero-shot performance in cross-modal retrieval tasks. For generative models like text-to-image Diffusion Transformers, the generated images exhibit a significant improvement in alignment with users' text instructions, especially in following complex queries. Our project page is https://www.haqtu.me/Recap-Datacomp-1B/
Authors: Raphael Tang, Xinyu Zhang, Lixinyu Xu, Yao Lu, Wenyan Li, Pontus Stenetorp, Jimmy Lin, Ferhan Ture
Abstract: Diffusion models are the state of the art in text-to-image generation, but their perceptual variability remains understudied. In this paper, we examine how prompts affect image variability in black-box diffusion-based models. We propose W1KP, a human-calibrated measure of variability in a set of images, bootstrapped from existing image-pair perceptual distances. Current datasets do not cover recent diffusion models, thus we curate three test sets for evaluation. Our best perceptual distance outperforms nine baselines by up to 18 points in accuracy, and our calibration matches graded human judgements 78% of the time. Using W1KP, we study prompt reusability and show that Imagen prompts can be reused for 10-50 random seeds before new images become too similar to already generated images, while Stable Diffusion XL and DALL-E 3 can be reused 50-200 times. Lastly, we analyze 56 linguistic features of real prompts, finding that the prompt's length, CLIP embedding norm, concreteness, and word senses influence variability most. As far as we are aware, we are the first to analyze diffusion variability from a visuolinguistic perspective. Our project page is at http://w1kp.com
URLs: http://w1kp.com
Authors: Angana Borah, Manash Pratim Barman, Amit Awekar
Abstract: A representation learning method is considered stable if it consistently generates similar representation of the given data across multiple runs. Word Embedding Methods (WEMs) are a class of representation learning methods that generate dense vector representation for each word in the given text data. The central idea of this paper is to explore the stability measurement of WEMs using intrinsic evaluation based on word similarity. We experiment with three popular WEMs: Word2Vec, GloVe, and fastText. For stability measurement, we investigate the effect of five parameters involved in training these models. We perform experiments using four real-world datasets from different domains: Wikipedia, News, Song lyrics, and European parliament proceedings. We also observe the effect of WEM stability on three downstream tasks: Clustering, POS tagging, and Fairness evaluation. Our experiments indicate that amongst the three WEMs, fastText is the most stable, followed by GloVe and Word2Vec.
Authors: Xiao Yu, Yuang Qi, Kejiang Chen, Guoqiang Chen, Xi Yang, Pengyuan Zhu, Xiuwei Shang, Weiming Zhang, Nenghai Yu
Abstract: Large language models (LLMs) have the potential to generate texts that pose risks of misuse, such as plagiarism, planting fake reviews on e-commerce platforms, or creating inflammatory false tweets. Consequently, detecting whether a text is generated by LLMs has become increasingly important. Existing high-quality detection methods usually require access to the interior of the model to extract the intrinsic characteristics. However, since we do not have access to the interior of the black-box model, we must resort to surrogate models, which impacts detection quality. In order to achieve high-quality detection of black-box models, we would like to extract deep intrinsic characteristics of the black-box model generated texts. We view the generation process as a coupled process of prompt and intrinsic characteristics of the generative model. Based on this insight, we propose to decouple prompt and intrinsic characteristics (DPIC) for LLM-generated text detection method. Specifically, given a candidate text, DPIC employs an auxiliary LLM to reconstruct the prompt corresponding to the candidate text, then uses the prompt to regenerate text by the auxiliary LLM, which makes the candidate text and the regenerated text align with their prompts, respectively. Then, the similarity between the candidate text and the regenerated text is used as a detection feature, thus eliminating the prompt in the detection process, which allows the detector to focus on the intrinsic characteristics of the generative model. Compared to the baselines, DPIC has achieved an average improvement of 6.76\% and 2.91\% in detecting texts from different domains generated by GPT4 and Claude3, respectively.
Authors: Kung-Hsiang Huang, Hou Pong Chan, Kathleen McKeown, Heng Ji
Abstract: Considerable advancements have been made to tackle the misrepresentation of information derived from reference articles in the domains of fact-checking and faithful summarization. However, an unaddressed aspect remains - the identification of social media posts that manipulate information within associated news articles. This task presents a significant challenge, primarily due to the prevalence of personal opinions in such posts. We present a novel task, identifying manipulation of news on social media, which aims to detect manipulation in social media posts and identify manipulated or inserted information. To study this task, we have proposed a data collection schema and curated a dataset called ManiTweet, consisting of 3.6K pairs of tweets and corresponding articles. Our analysis demonstrates that this task is highly challenging, with large language models (LLMs) yielding unsatisfactory performance. Additionally, we have developed a simple yet effective basic model that outperforms LLMs significantly on the ManiTweet dataset. Finally, we have conducted an exploratory analysis of human-written tweets, unveiling intriguing connections between manipulation and the domain and factuality of news articles, as well as revealing that manipulated sentences are more likely to encapsulate the main story or consequences of a news outlet.
Authors: Gregor Geigle, Radu Timofte, Goran Glava\v{s}
Abstract: Vision-and-language (VL) models with separate encoders for each modality (e.g., CLIP) have become the go-to models for zero-shot image classification and image-text retrieval. They are, however, mostly evaluated in English as multilingual benchmarks are limited in availability. We introduce Babel-ImageNet, a massively multilingual benchmark that offers (partial) translations of ImageNet labels to 100 languages, built without machine translation or manual annotation. We instead automatically obtain reliable translations by linking them -- via shared WordNet synsets -- to BabelNet, a massively multilingual lexico-semantic network. We evaluate 11 public multilingual CLIP models on zero-shot image classification (ZS-IC) on our benchmark, demonstrating a significant gap between English ImageNet performance and that of high-resource languages (e.g., German or Chinese), and an even bigger gap for low-resource languages (e.g., Sinhala or Lao). Crucially, we show that the models' ZS-IC performance highly correlates with their performance in image-text retrieval, validating the use of Babel-ImageNet to evaluate multilingual models for the vast majority of languages without gold image-text data. Finally, we show that the performance of multilingual CLIP can be drastically improved for low-resource languages with parameter-efficient language-specific training. We make our code and data publicly available: \url{https://github.com/gregor-ge/Babel-ImageNet}
Authors: Angana Borah, Daria Pylypenko, Cristina Espana-Bonet, Josef van Genabith
Abstract: Recent work has shown evidence of 'Clever Hans' behavior in high-performance neural translationese classifiers, where BERT-based classifiers capitalize on spurious correlations, in particular topic information, between data and target classification labels, rather than genuine translationese signals. Translationese signals are subtle (especially for professional translation) and compete with many other signals in the data such as genre, style, author, and, in particular, topic. This raises the general question of how much of the performance of a classifier is really due to spurious correlations in the data versus the signals actually targeted for by the classifier, especially for subtle target signals and in challenging (low resource) data settings. We focus on topic-based spurious correlation and approach the question from two directions: (i) where we have no knowledge about spurious topic information and its distribution in the data, (ii) where we have some indication about the nature of spurious topic correlations. For (i) we develop a measure from first principles capturing alignment of unsupervised topics with target classification labels as an indication of spurious topic information in the data. We show that our measure is the same as purity in clustering and propose a 'topic floor' (as in a 'noise floor') for classification. For (ii) we investigate masking of known spurious topic carriers in classification. Both (i) and (ii) contribute to quantifying and (ii) to mitigating spurious correlations.
Authors: Zonglin Yang, Xinya Du, Junxian Li, Jie Zheng, Soujanya Poria, Erik Cambria
Abstract: Hypothetical induction is recognized as the main reasoning type when scientists make observations about the world and try to propose hypotheses to explain those observations. Past research on hypothetical induction is under a constrained setting: (1) the observation annotations in the dataset are carefully manually handpicked sentences (resulting in a close-domain setting); and (2) the ground truth hypotheses are mostly commonsense knowledge, making the task less challenging. In this work, we tackle these problems by proposing the first dataset for social science academic hypotheses discovery, with the final goal to create systems that automatically generate valid, novel, and helpful scientific hypotheses, given only a pile of raw web corpus. Unlike previous settings, the new dataset requires (1) using open-domain data (raw web corpus) as observations; and (2) proposing hypotheses even new to humanity. A multi-module framework is developed for the task, including three different feedback mechanisms to boost performance, which exhibits superior performance in terms of both GPT-4 based and expert-based evaluation. To the best of our knowledge, this is the first work showing that LLMs are able to generate novel (''not existing in literature'') and valid (''reflecting reality'') scientific hypotheses.
Authors: Bochuan Cao, Yuanpu Cao, Lu Lin, Jinghui Chen
Abstract: Recently, Large Language Models (LLMs) have made significant advancements and are now widely used across various domains. Unfortunately, there has been a rising concern that LLMs can be misused to generate harmful or malicious content. Though a line of research has focused on aligning LLMs with human values and preventing them from producing inappropriate content, such alignments are usually vulnerable and can be bypassed by alignment-breaking attacks via adversarially optimized or handcrafted jailbreaking prompts. In this work, we introduce a Robustly Aligned LLM (RA-LLM) to defend against potential alignment-breaking attacks. RA-LLM can be directly constructed upon an existing aligned LLM with a robust alignment checking function, without requiring any expensive retraining or fine-tuning process of the original LLM. Furthermore, we also provide a theoretical analysis for RA-LLM to verify its effectiveness in defending against alignment-breaking attacks. Through real-world experiments on open-source large language models, we demonstrate that RA-LLM can successfully defend against both state-of-the-art adversarial prompts and popular handcrafted jailbreaking prompts by reducing their attack success rates from nearly 100% to around 10% or less.
Authors: Saurabh Srivastava, Chengyue Huang, Weiguo Fan, Ziyu Yao
Abstract: Large language models (LLMs) have revolutionized zero-shot task performance, mitigating the need for task-specific annotations while enhancing task generalizability. Despite its advancements, current methods using trigger phrases such as "Let's think step by step" remain limited. This study introduces PRomPTed, an approach that optimizes the zero-shot prompts for individual task instances following an innovative manner of "LLMs in the loop". Our comprehensive evaluation across 13 datasets and 10 task types based on GPT-4 reveals that PRomPTed significantly outperforms both the naive zero-shot approaches and a strong baseline (i.e., "Output Refinement") which refines the task output instead of the input prompt. Our experimental results also confirmed the generalization of this advantage to the relatively weaker GPT-3.5. Even more intriguingly, we found that leveraging GPT-3.5 to rewrite prompts for the stronger GPT-4 not only matches but occasionally exceeds the efficacy of using GPT-4 as the prompt rewriter. Our research thus presents a huge value in not only enhancing zero-shot LLM performance but also potentially enabling supervising LLMs with their weaker counterparts, a capability attracting much interest recently. Finally, our additional experiments confirm the generalization of the advantages to open-source LLMs such as Mistral 7B and Mixtral 8x7B.
Authors: David Arps, Laura Kallmeyer, Younes Samih, Hassan Sajjad
Abstract: We introduce SPUD (Semantically Perturbed Universal Dependencies), a framework for creating nonce treebanks for the multilingual Universal Dependencies (UD) corpora. SPUD data satisfies syntactic argument structure, provides syntactic annotations, and ensures grammaticality via language-specific rules. We create nonce data in Arabic, English, French, German, and Russian, and demonstrate two use cases of SPUD treebanks. First, we investigate the effect of nonce data on word co-occurrence statistics, as measured by perplexity scores of autoregressive (ALM) and masked language models (MLM). We find that ALM scores are significantly more affected by nonce data than MLM scores. Second, we show how nonce data affects the performance of syntactic dependency probes. We replicate the findings of M\"uller-Eberstein et al. (2022) on nonce test data and show that the performance declines on both MLMs and ALMs wrt. original test data. However, a majority of the performance is kept, suggesting that the probe indeed learns syntax independently from semantics.
Authors: Chong Li, Shaonan Wang, Jiajun Zhang, Chengqing Zong
Abstract: Multilingual generative models obtain remarkable cross-lingual in-context learning capabilities through pre-training on large-scale corpora. However, they still exhibit a performance bias toward high-resource languages and learn isolated distributions of multilingual sentence representations, which may hinder knowledge transfer across languages. To bridge this gap, we propose a simple yet effective cross-lingual alignment framework exploiting pairs of translation sentences. It aligns the internal sentence representations across different languages via multilingual contrastive learning and aligns outputs by following cross-lingual instructions in the target language. Experimental results show that even with less than 0.1 {\textperthousand} of pre-training tokens, our alignment framework significantly boosts the cross-lingual abilities of generative language models and mitigates the performance gap. Further analyses reveal that it results in a better internal multilingual representation distribution of multilingual models.
Authors: Ryuto Koike, Masahiro Kaneko, Naoaki Okazaki
Abstract: To combat the misuse of Large Language Models (LLMs), many recent studies have presented LLM-generated-text detectors with promising performance. When users instruct LLMs to generate texts, the instruction can include different constraints depending on the user's need. However, most recent studies do not cover such diverse instruction patterns when creating datasets for LLM detection. In this paper, we reveal that even task-oriented constraints -- constraints that would naturally be included in an instruction and are not related to detection-evasion -- cause existing powerful detectors to have a large variance in detection performance. We focus on student essay writing as a realistic domain and manually create task-oriented constraints based on several factors for essay quality. Our experiments show that the standard deviation (SD) of current detector performance on texts generated by an instruction with such a constraint is significantly larger (up to an SD of 14.4 F1-score) than that by generating texts multiple times or paraphrasing the instruction. We also observe an overall trend where the constraints can make LLM detection more challenging than without them. Finally, our analysis indicates that the high instruction-following ability of LLMs fosters the large impact of such constraints on detection performance.
Authors: Zhexin Zhang, Junxiao Yang, Pei Ke, Fei Mi, Hongning Wang, Minlie Huang
Abstract: While significant attention has been dedicated to exploiting weaknesses in LLMs through jailbreaking attacks, there remains a paucity of effort in defending against these attacks. We point out a pivotal factor contributing to the success of jailbreaks: the intrinsic conflict between the goals of being helpful and ensuring safety. Accordingly, we propose to integrate goal prioritization at both training and inference stages to counteract. Implementing goal prioritization during inference substantially diminishes the Attack Success Rate (ASR) of jailbreaking from 66.4% to 3.6% for ChatGPT. And integrating goal prioritization into model training reduces the ASR from 71.0% to 6.6% for Llama2-13B. Remarkably, even in scenarios where no jailbreaking samples are included during training, our approach slashes the ASR by half. Additionally, our findings reveal that while stronger LLMs face greater safety risks, they also possess a greater capacity to be steered towards defending against such attacks, both because of their stronger ability in instruction following. Our work thus contributes to the comprehension of jailbreaking attacks and defenses, and sheds light on the relationship between LLMs' capability and safety. Our code is available at \url{https://github.com/thu-coai/JailbreakDefense_GoalPriority}.
URLs: https://github.com/thu-coai/JailbreakDefense_GoalPriority
Authors: Xiangru Tang, Yuliang Liu, Zefan Cai, Yanjun Shao, Junjie Lu, Yichi Zhang, Zexuan Deng, Helan Hu, Kaikai An, Ruijun Huang, Shuzheng Si, Sheng Chen, Haozhe Zhao, Liang Chen, Yan Wang, Tianyu Liu, Zhiwei Jiang, Baobao Chang, Yin Fang, Yujia Qin, Wangchunshu Zhou, Yilun Zhao, Arman Cohan, Mark Gerstein
Abstract: Despite Large Language Models (LLMs) like GPT-4 achieving impressive results in function-level code generation, they struggle with repository-scale code understanding (e.g., coming up with the right arguments for calling routines), requiring a deeper comprehension of complex file interactions. Also, recently, people have developed LLM agents that attempt to interact with repository code (e.g., compiling and evaluating its execution), prompting the need to evaluate their performance. These gaps have motivated our development of ML-Bench, a benchmark rooted in real-world programming applications that leverage existing code repositories to perform tasks. Addressing the need for LLMs to interpret long code contexts and translate instructions into precise, executable scripts, ML-Bench encompasses annotated 9,641 examples across 18 GitHub repositories, challenging LLMs to accommodate user-specified arguments and documentation intricacies effectively. To evaluate both LLMs and AI agents, two setups are employed: ML-LLM-Bench for assessing LLMs' text-to-code conversion within a predefined deployment environment, and ML-Agent-Bench for testing autonomous agents in an end-to-end task execution within a Linux sandbox environment. Our findings indicate that while GPT-4o leads with a Pass@5 rate surpassing 50%, there remains significant scope for improvement, highlighted by issues such as hallucinated outputs and difficulties with bash script generation. Notably, in the more demanding ML-Agent-Bench, GPT-4o achieves a 76.47% success rate, reflecting the efficacy of iterative action and feedback in complex task resolution.
Authors: Ehsan Latif, Luyang Fang, Ping Ma, Xiaoming Zhai
Abstract: This study proposes a method for knowledge distillation (KD) of fine-tuned Large Language Models (LLMs) into smaller, more efficient, and accurate neural networks. We specifically target the challenge of deploying these models on resource-constrained devices. Our methodology involves training the smaller student model (Neural Network) using the prediction probabilities (as soft labels) of the LLM, which serves as a teacher model. This is achieved through a specialized loss function tailored to learn from the LLM's output probabilities, ensuring that the student model closely mimics the teacher's performance. To validate the performance of the KD approach, we utilized a large dataset, 7T, containing 6,684 student-written responses to science questions and three mathematical reasoning datasets with student-written responses graded by human experts. We compared accuracy with state-of-the-art (SOTA) distilled models, TinyBERT, and artificial neural network (ANN) models. Results have shown that the KD approach has 3% and 2% higher scoring accuracy than ANN and TinyBERT, respectively, and comparable accuracy to the teacher model. Furthermore, the student model size is 0.03M, 4,000 times smaller in parameters and x10 faster in inferencing than the teacher model and TinyBERT, respectively. The significance of this research lies in its potential to make advanced AI technologies accessible in typical educational settings, particularly for automatic scoring.
Authors: Junjie Wang, Dan Yang, Binbin Hu, Yue Shen, Wen Zhang, Jinjie Gu
Abstract: In this paper, we explore a new way for user targeting, where non-expert marketers could select their target users solely given demands in natural language form. The key to this issue is how to transform natural languages into practical structured logical languages, i.e., the structured understanding of marketer demands. In practical scenarios, the demands of non-expert marketers are often abstract and diverse. Considering the impressive natural language processing ability of large language models (LLMs), we try to leverage LLMs to solve this issue. To stimulate the LLMs' reasoning ability, the chain-of-thought (CoT) prompting method is widely used, but existing methods still have some limitations in our scenario: (1) Previous methods either use simple "Let's think step by step" spells or provide fixed examples in demonstrations without considering compatibility between prompts and concrete questions, making LLMs ineffective when the marketers' demands are abstract and diverse. (2) Previous methods are often implemented in closed-source models or excessively large models, which is not suitable in industrial practical scenarios. Based on these, we propose ARALLM (i.e., Analogical Reasoning Augmented Large Language Models) consisting of two modules: Analogical Reasoning based Prompting and Reasoning-Augmented Multi-Task Model Distillation. Part of our data and code can be found at https://github.com/alipay/Analogic-Reasoning-Augmented-Large-Language-Model.
URLs: https://github.com/alipay/Analogic-Reasoning-Augmented-Large-Language-Model.
Authors: Yuu Jinnai, Ukyo Honda, Tetsuro Morimura, Peinan Zhang
Abstract: One of the most important challenges in text generation systems is to produce outputs that are not only correct but also diverse. Recently, Minimum Bayes-Risk (MBR) decoding has gained prominence for generating sentences of the highest quality among the decoding algorithms. However, existing algorithms proposed for generating diverse outputs are predominantly based on beam search or random sampling, thus their output quality is capped by these underlying methods. In this paper, we investigate an alternative approach -- we develop diversity-promoting decoding algorithms by enforcing diversity objectives to MBR decoding. We propose two variants of MBR, Diverse MBR (DMBR) and $k$-medoids MBR (KMBR), methods to generate a set of sentences with high quality and diversity. We evaluate DMBR and KMBR on a variety of directed text generation tasks using encoder-decoder models and a large language model with prompting. The experimental results show that the proposed method achieves a better trade-off than the diverse beam search and sampling algorithms.
Authors: Zhen Li, Xiaohan Xu, Tao Shen, Can Xu, Jia-Chen Gu, Yuxuan Lai, Chongyang Tao, Shuai Ma
Abstract: In the rapidly evolving domain of Natural Language Generation (NLG) evaluation, introducing Large Language Models (LLMs) has opened new avenues for assessing generated content quality, e.g., coherence, creativity, and context relevance. This paper aims to provide a thorough overview of leveraging LLMs for NLG evaluation, a burgeoning area that lacks a systematic analysis. We propose a coherent taxonomy for organizing existing LLM-based evaluation metrics, offering a structured framework to understand and compare these methods. Our detailed exploration includes critically assessing various LLM-based methodologies, as well as comparing their strengths and limitations in evaluating NLG outputs. By discussing unresolved challenges, including bias, robustness, domain-specificity, and unified evaluation, this paper seeks to offer insights to researchers and advocate for fairer and more advanced NLG evaluation techniques.
Authors: Yihao Li, Ru Zhang, Jianyi Liu
Abstract: While Large Language Models (LLMs) demonstrate exceptional performance in a multitude of Natural Language Processing (NLP) tasks, they encounter challenges in practical applications, including issues with hallucinations, inadequate knowledge updating, and limited transparency in the reasoning process. To overcome these limitations, this study innovatively proposes a collaborative training-free reasoning scheme involving tight cooperation between Knowledge Graph (KG) and LLMs. This scheme first involves using LLMs to iteratively explore KG, selectively retrieving a task-relevant knowledge subgraph to support reasoning. The LLMs are then guided to further combine inherent implicit knowledge to reason on the subgraph while explicitly elucidating the reasoning process. Through such a cooperative approach, our scheme achieves more reliable knowledge-based reasoning and facilitates the tracing of the reasoning results. Experimental results show that our scheme significantly progressed across multiple datasets, notably achieving over a 10% improvement on the QALD10 dataset compared to the best baseline and the fine-tuned state-of-the-art (SOTA) work. Building on this success, this study hopes to offer a valuable reference for future research in the fusion of KG and LLMs, thereby enhancing LLMs' proficiency in solving complex issues.
Authors: Zhibo Chu, Shiwen Ni, Zichong Wang, Xi Feng, Min Yang, Wenbin Zhang
Abstract: Language models serve as a cornerstone in natural language processing (NLP), utilizing mathematical methods to generalize language laws and knowledge for prediction and generation. Over extensive research spanning decades, language modeling has progressed from initial statistical language models (SLMs) to the contemporary landscape of large language models (LLMs). Notably, the swift evolution of LLMs has reached the ability to process, understand, and generate human-level text. Nevertheless, despite the significant advantages that LLMs offer in improving both work and personal lives, the limited understanding among general practitioners about the background and principles of these models hampers their full potential. Notably, most LLMs reviews focus on specific aspects and utilize specialized language, posing a challenge for practitioners lacking relevant background knowledge. In light of this, this survey aims to present a comprehensible overview of LLMs to assist a broader audience. It strives to facilitate a comprehensive understanding by exploring the historical background of language models and tracing their evolution over time. The survey further investigates the factors influencing the development of LLMs, emphasizing key contributions. Additionally, it concentrates on elucidating the underlying principles of LLMs, equipping audiences with essential theoretical knowledge. The survey also highlights the limitations of existing work and points out promising future directions.
Authors: Jiwon Song, Kyungseok Oh, Taesu Kim, Hyungjun Kim, Yulhwa Kim, Jae-Joon Kim
Abstract: Large language models (LLMs) have proven to be highly effective across various natural language processing tasks. However, their large number of parameters poses significant challenges for practical deployment. Pruning, a technique aimed at reducing the size and complexity of LLMs, offers a potential solution by removing redundant components from the network. Despite the promise of pruning, existing methods often struggle to achieve substantial end-to-end LLM inference speedup. In this paper, we introduce SLEB, a novel approach designed to streamline LLMs by eliminating redundant transformer blocks. We choose the transformer block as the fundamental unit for pruning, because LLMs exhibit block-level redundancy with high similarity between the outputs of neighboring blocks. This choice allows us to effectively enhance the processing speed of LLMs. Our experimental results demonstrate that SLEB outperforms previous LLM pruning methods in accelerating LLM inference while also maintaining superior perplexity and accuracy, making SLEB as a promising technique for enhancing the efficiency of LLMs. The code is available at: https://github.com/jiwonsong-dev/SLEB.
Authors: Wei Zhai, Hongzhi Qi, Qing Zhao, Jianqiang Li, Ziqi Wang, Han Wang, Bing Xiang Yang, Guanghui Fu
Abstract: In the current environment, psychological issues are prevalent and widespread, with social media serving as a key outlet for individuals to share their feelings. This results in the generation of vast quantities of data daily, where negative emotions have the potential to precipitate crisis situations. There is a recognized need for models capable of efficient analysis. While pre-trained language models have demonstrated their effectiveness broadly, there's a noticeable gap in pre-trained models tailored for specialized domains like psychology. To address this, we have collected a huge dataset from Chinese social media platforms and enriched it with publicly available datasets to create a comprehensive database encompassing 3.36 million text entries. To enhance the model's applicability to psychological text analysis, we integrated psychological lexicons into the pre-training masking mechanism. Building on an existing Chinese language model, we performed adaptive training to develop a model specialized for the psychological domain. We evaluated our model's performance across six public datasets, where it demonstrated improvements compared to eight other models. Additionally, in the qualitative comparison experiment, our model provided psychologically relevant predictions given the masked sentences. Due to concerns regarding data privacy, the dataset will not be made publicly available. However, we have made the pre-trained models and codes publicly accessible to the community via: https://github.com/zwzzzQAQ/Chinese-MentalBERT.
Authors: Weixiang Zhao, Zhuojun Li, Shilong Wang, Yang Wang, Yulin Hu, Yanyan Zhao, Chen Wei, Bing Qin
Abstract: Emotional Intelligence (EI), consisting of emotion perception, emotion cognition and emotion expression, plays the critical roles in improving user interaction experience for the current large language model (LLM) based conversational general AI assistants. Previous works mainly focus on raising the emotion perception ability of them via naive fine-tuning on EI-related classification or regression tasks. However, this leads to the incomplete enhancement of EI and catastrophic forgetting of the general intelligence (GI). To this end, we first introduce \textsc{EiBench}, a large-scale collection of EI-related tasks in the text-to-text formation with task instructions that covers all three aspects of EI, which lays a solid foundation for the comprehensive EI enhancement of LLMs. Then a novel \underline{\textbf{Mo}}dular \underline{\textbf{E}}motional \underline{\textbf{I}}ntelligence enhancement method (\textbf{MoEI}), consisting of Modular Parameter Expansion and intra-inter modulation, is proposed to comprehensively enhance the EI of LLMs without compromise their GI. Extensive experiments on two representative LLM-based assistants, Flan-T5 and LLaMA-2-Chat, demonstrate the effectiveness of MoEI to improving EI while maintain GI.
Authors: Frank Wildenburg, Michael Hanna, Sandro Pezzelle
Abstract: In everyday language use, speakers frequently utter and interpret sentences that are semantically underspecified, namely, whose content is insufficient to fully convey their message or interpret them univocally. For example, to interpret the underspecified sentence "Don't spend too much", which leaves implicit what (not) to spend, additional linguistic context or outside knowledge is needed. In this work, we propose a novel Dataset of semantically Underspecified Sentences grouped by Type (DUST) and use it to study whether pre-trained language models (LMs) correctly identify and interpret underspecified sentences. We find that newer LMs are reasonably able to identify underspecified sentences when explicitly prompted. However, interpreting them correctly is much harder for any LMs. Our experiments show that when interpreting underspecified sentences, LMs exhibit little uncertainty, contrary to what theoretical accounts of underspecification would predict. Overall, our study reveals limitations in current models' processing of sentence semantics and highlights the importance of using naturalistic data and communicative scenarios when evaluating LMs' language capabilities.
Authors: Xiao Li, Bolin Zhu, Sichen Liu, Yin Zhu, Yiwei Liu, Gong Cheng
Abstract: The application of formulas is a fundamental ability of humans when addressing numerical reasoning problems. However, existing numerical reasoning datasets seldom explicitly indicate the formulas employed during the reasoning steps. To bridge this gap, we construct a dataset for formula-based numerical reasoning called FormulaReasoning, which consists of 5,420 reasoning-based questions. We employ it to conduct evaluations of LLMs with size ranging from 7B to over 100B parameters utilizing zero-shot and few-shot chain-of-thought methods, and we further explore using retrieval-augmented LLMs provided with an external formula database associated with our dataset. We also experiment with supervised methods where we divide the reasoning process into formula generation, parameter extraction, and numerical calculation, and perform data augmentation. Our empirical findings underscore the significant potential for improvement in existing models when applied to our complex, formula-driven FormulaReasoning.
Authors: Houcemeddine Turki, Kawthar Ellouze, Hager Ben Ammar, Mohamed Ali Hadj Taieb, Imed Adel, Mohamed Ben Aouicha, Pier Luigi Farri, Abderrezak Bennour
Abstract: Tunisian Arabic (ISO 693-3: aeb) isa distinct variety native to Tunisia, derived from Arabic and enriched by various historical influences. This research introduces the "Normalized Orthography for Tunisian Arabic" (NOTA), an adaptation of CODA* guidelines for transcribing Tunisian Arabic using Arabic script. The aim is to enhance language resource development by ensuring user-friendliness and consistency. The updated standard addresses challenges in accurately representing Tunisian phonology and morphology, correcting issues from transcriptions based on Modern Standard Arabic.
Authors: Howard Yen, Tianyu Gao, Danqi Chen
Abstract: Extending large language models (LLMs) to process longer inputs is crucial for a wide range of applications. However, the substantial computational cost of transformers and limited generalization of positional encoding restrict the size of their context window. We introduce Context Expansion with Parallel Encoding (CEPE), a framework that can be applied to any existing decoder-only LLMs to extend their context window. CEPE employs a small encoder to process long inputs chunk by chunk, enabling the frozen decoder to utilize additional contexts via cross-attention. CEPE is efficient, generalizable, and versatile: trained with 8K-token documents, it extends the context window of LLAMA-2 to 128K tokens, offering 10x the throughput with only 1/6 of the memory. CEPE yields strong performance on language modeling and in-context learning. CEPE also excels in retrieval-augmented applications, while existing long-context models degenerate with retrieved contexts. We further introduce a CEPE variant that can extend the context window of instruction-tuned models using only unlabeled data, and showcase its effectiveness on LLAMA-2-CHAT, leading to a strong instruction-following model that can leverage very long contexts on downstream tasks.
Authors: Shicheng Xu, Liang Pang, Mo Yu, Fandong Meng, Huawei Shen, Xueqi Cheng, Jie Zhou
Abstract: Retrieval-augmented generation (RAG) enhances large language models (LLMs) by incorporating additional information from retrieval. However, studies have shown that LLMs still face challenges in effectively using the retrieved information, even ignoring it or being misled by it. The key reason is that the training of LLMs does not clearly make LLMs learn how to utilize input retrieved texts with varied quality. In this paper, we propose a novel perspective that considers the role of LLMs in RAG as ``Information Refiner'', which means that regardless of correctness, completeness, or usefulness of retrieved texts, LLMs can consistently integrate knowledge within the retrieved texts and model parameters to generate the texts that are more concise, accurate, and complete than the retrieved texts. To this end, we propose an information refinement training method named InFO-RAG that optimizes LLMs for RAG in an unsupervised manner. InFO-RAG is low-cost and general across various tasks. Extensive experiments on zero-shot prediction of 11 datasets in diverse tasks including Question Answering, Slot-Filling, Language Modeling, Dialogue, and Code Generation show that InFO-RAG improves the performance of LLaMA2 by an average of 9.39\% relative points. InFO-RAG also shows advantages in in-context learning and robustness of RAG.
Authors: Philip Feldman, James R. Foulds, Shimei Pan
Abstract: Large language models (LLMs) like ChatGPT demonstrate the remarkable progress of artificial intelligence. However, their tendency to hallucinate -- generate plausible but false information -- poses a significant challenge. This issue is critical, as seen in recent court cases where ChatGPT's use led to citations of non-existent legal rulings. This paper explores how Retrieval-Augmented Generation (RAG) can counter hallucinations by integrating external knowledge with prompts. We empirically evaluate RAG against standard LLMs using prompts designed to induce hallucinations. Our results show that RAG increases accuracy in some cases, but can still be misled when prompts directly contradict the model's pre-trained understanding. These findings highlight the complex nature of hallucinations and the need for more robust solutions to ensure LLM reliability in real-world applications. We offer practical recommendations for RAG deployment and discuss implications for the development of more trustworthy LLMs.
Authors: Bryan Li, Tamer Alkhouli, Daniele Bonadiman, Nikolaos Pappas, Saab Mansour
Abstract: The development of large language models (LLM) has shown progress on reasoning, though studies have largely considered either English or simple reasoning tasks. To address this, we introduce a multilingual structured reasoning and explanation dataset, termed xSTREET, that covers four tasks across six languages. xSTREET exposes a gap in base LLM performance between English and non-English reasoning tasks. We then propose two methods to remedy this gap, building on the insight that LLMs trained on code are better reasoners. First, at training time, we augment a code dataset with multilingual comments using machine translation while keeping program code as-is. Second, at inference time, we bridge the gap between training and inference by employing a prompt structure that incorporates step-by-step code primitives to derive new facts and find a solution. Our methods show improved multilingual performance on xSTREET, most notably on the scientific commonsense reasoning subtask. Furthermore, the models show no regression on non-reasoning tasks, thus demonstrating our techniques maintain general-purpose abilities.
Authors: Qian Chen, Dongyang Li, Xiaofeng He, Hongzhao Li, Hongyu Yi
Abstract: The black-box nature of deep learning models in NLP hinders their widespread application. The research focus has shifted to Hierarchical Attribution (HA) for its ability to model feature interactions. Recent works model non-contiguous combinations with a time-costly greedy search in Eculidean spaces, neglecting underlying linguistic information in feature representations. In this work, we introduce a novel method, namely Poincare Explanation (PE), for modeling feature interactions with hyperbolic spaces in a time efficient manner. Specifically, we take building text hierarchies as finding spanning trees in hyperbolic spaces. First we project the embeddings into hyperbolic spaces to elicit inherit semantic and syntax hierarchical structures. Then we propose a simple yet effective strategy to calculate Shapley score. Finally we build the the hierarchy with proving the constructing process in the projected space could be viewed as building a minimum spanning tree and introduce a time efficient building algorithm. Experimental results demonstrate the effectiveness of our approach.
Authors: Gaurish Thakkar, Sherzod Hakimov, Marko Tadi\'c
Abstract: In recent years, multimodal natural language processing, aimed at learning from diverse data types, has garnered significant attention. However, there needs to be more clarity when it comes to analysing multimodal tasks in multi-lingual contexts. While prior studies on sentiment analysis of tweets have predominantly focused on the English language, this paper addresses this gap by transforming an existing textual Twitter sentiment dataset into a multimodal format through a straightforward curation process. Our work opens up new avenues for sentiment-related research within the research community. Additionally, we conduct baseline experiments utilising this augmented dataset and report the findings. Notably, our evaluations reveal that when comparing unimodal and multimodal configurations, using a sentiment-tuned large language model as a text encoder performs exceptionally well.
Authors: Tianle Li, Ge Zhang, Quy Duc Do, Xiang Yue, Wenhu Chen
Abstract: Large Language Models (LLMs) have made significant strides in handling long sequences. Some models like Gemini could even to be capable of dealing with millions of tokens. However, their performance evaluation has largely been confined to metrics like perplexity and synthetic tasks, which may not fully capture their true abilities in more challenging, real-world scenarios. We introduce a benchmark (LongICLBench) for long in-context learning in extreme-label classification using six datasets with 28 to 174 classes and input lengths from 2K to 50K tokens. Our benchmark requires LLMs to comprehend the entire input to recognize the massive label spaces to make correct predictions. We evaluate on 15 long-context LLMs and find that they perform well on less challenging classification tasks with smaller label space and shorter demonstrations. However, they struggle with more challenging task like Discovery with 174 labels, suggesting a gap in their ability to process long, context-rich sequences. Further analysis reveals a bias towards labels presented later in the sequence and a need for improved reasoning over multiple pieces of information. Our study reveals that long context understanding and reasoning is still a challenging task for the existing LLMs. We believe LongICLBench could serve as a more realistic evaluation for the future long-context LLMs.
Authors: Amirhossein Abaskohi, Amirhossein Dabiriaghdam, Lele Wang, Giuseppe Carenini
Abstract: Memes, combining text and images, frequently use metaphors to convey persuasive messages, shaping public opinion. Motivated by this, our team engaged in SemEval-2024 Task 4, a hierarchical multi-label classification task designed to identify rhetorical and psychological persuasion techniques embedded within memes. To tackle this problem, we introduced a caption generation step to assess the modality gap and the impact of additional semantic information from images, which improved our result. Our best model utilizes GPT-4 generated captions alongside meme text to fine-tune RoBERTa as the text encoder and CLIP as the image encoder. It outperforms the baseline by a large margin in all 12 subtasks. In particular, it ranked in top-3 across all languages in Subtask 2a, and top-4 in Subtask 2b, demonstrating quantitatively strong performance. The improvement achieved by the introduced intermediate step is likely attributable to the metaphorical essence of images that challenges visual encoders. This highlights the potential for improving abstract visual semantics encoding.
Authors: Seyed Mahed Mousavi, Simone Alghisi, Giuseppe Riccardi
Abstract: LLMs acquire knowledge from massive data snapshots collected at different timestamps. Their knowledge is then commonly evaluated using static benchmarks. However, factual knowledge is generally subject to time-sensitive changes, and static benchmarks cannot address those cases. We present an approach to dynamically evaluate the knowledge in LLMs and their time-sensitiveness against Wikidata, a publicly available up-to-date knowledge graph. We evaluate the time-sensitive knowledge in twenty-four private and open-source LLMs, as well as the effectiveness of four editing methods in updating the outdated facts. Our results show that 1) outdatedness is a critical problem across state-of-the-art LLMs; 2) LLMs output inconsistent answers when prompted with slight variations of the question prompt; and 3) the performance of the state-of-the-art knowledge editing algorithms is very limited, as they can not reduce the cases of outdatedness and output inconsistency.
Authors: Geyu Lin, Bin Wang, Zhengyuan Liu, Nancy F. Chen
Abstract: Multilingual proficiency presents a significant challenge for large language models (LLMs). English-centric models are usually suboptimal in other languages, particularly those that are linguistically distant from English. This performance discrepancy mainly stems from the imbalanced distribution of training data across languages during pre-training and instruction tuning stages. To address this problem, we propose a novel approach called CrossIn, which utilizes a mixed composition of cross-lingual instruction tuning data. Our method leverages the compressed representation shared by various languages to efficiently enhance the model's task-solving capabilities and multilingual proficiency within a single process. In addition, we introduce a multi-task and multi-faceted benchmark to evaluate the effectiveness of CrossIn. Experimental results demonstrate that our method substantially improves performance across tasks and languages, and we provide extensive insights into the impact of cross-lingual data volume and the integration of translation data on enhancing multilingual consistency and accuracy.
Authors: Wei Xu, Jianlong Chen, Zhicheng Ding, Jinyin Wang
Abstract: This paper explores the importance of text sentiment analysis and classification in the field of natural language processing, and proposes a new approach to sentiment analysis and classification based on the bidirectional gated recurrent units (GRUs) model. The study firstly analyses the word cloud model of the text with six sentiment labels, and then carries out data preprocessing, including the steps of removing special symbols, punctuation marks, numbers, stop words and non-alphabetic parts. Subsequently, the data set is divided into training set and test set, and through model training and testing, it is found that the accuracy of the validation set is increased from 85% to 93% with training, which is an increase of 8%; at the same time, the loss value of the validation set decreases from 0.7 to 0.1 and tends to be stable, and the model is gradually close to the actual value, which can effectively classify the text emotions. The confusion matrix shows that the accuracy of the model on the test set reaches 94.8%, the precision is 95.9%, the recall is 99.1%, and the F1 score is 97.4%, which proves that the model has good generalisation ability and classification effect. Overall, the study demonstrated an effective method for text sentiment analysis and classification with satisfactory results.
Authors: Lin Ai, Zheng Hui, Zizhou Liu, Julia Hirschberg
Abstract: Machine Reading Comprehension (MRC) poses a significant challenge in the field of Natural Language Processing (NLP). While mainstream MRC methods predominantly leverage extractive strategies using encoder-only models such as BERT, generative approaches face the issue of out-of-control generation -- a critical problem where answers generated are often incorrect, irrelevant, or unfaithful to the source text. To address these limitations in generative models for MRC, we introduce the Question-Attended Span Extraction (QASE) module. Integrated during the fine-tuning phase of pre-trained generative language models (PLMs), QASE significantly enhances their performance, allowing them to surpass the extractive capabilities of advanced Large Language Models (LLMs) such as GPT-4 in few-shot settings. Notably, these gains in performance do not come with an increase in computational demands. The efficacy of the QASE module has been rigorously tested across various datasets, consistently achieving or even surpassing state-of-the-art (SOTA) results, thereby bridging the gap between generative and extractive models in extractive MRC tasks.
Authors: Anwoy Chatterjee, Eshaan Tanwar, Subhabrata Dutta, Tanmoy Chakraborty
Abstract: Large Language Models (LLMs) have transformed NLP with their remarkable In-context Learning (ICL) capabilities. Automated assistants based on LLMs are gaining popularity; however, adapting them to novel tasks is still challenging. While colossal models excel in zero-shot performance, their computational demands limit widespread use, and smaller language models struggle without context. This paper investigates whether LLMs can generalize from labeled examples of predefined tasks to novel tasks. Drawing inspiration from biological neurons and the mechanistic interpretation of the Transformer architecture, we explore the potential for information sharing across tasks. We design a cross-task prompting setup with three LLMs and show that LLMs achieve significant performance improvements despite no examples from the target task in the context. Cross-task prompting leads to a remarkable performance boost of 107% for LLaMA-2 7B, 18.6% for LLaMA-2 13B, and 3.2% for GPT 3.5 on average over zero-shot prompting, and performs comparable to standard in-context learning. The effectiveness of generating pseudo-labels for in-task examples is demonstrated, and our analyses reveal a strong correlation between the effect of cross-task examples and model activation similarities in source and target input tokens. This paper offers a first-of-its-kind exploration of LLMs' ability to solve novel tasks based on contextual signals from different task examples.
Authors: Ziyin Zhang, Lizhen Xu, Zhaokun Jiang, Hongkun Hao, Rui Wang
Abstract: We present GSM-MC and MATH-MC, two multiple-choice (MC) datasets constructed by collecting answers and incorrect predictions on GSM8K and MATH from 60 open-source models. Through extensive experiments, we show that LLMs' performance on the MC versions of these two popular benchmarks is strongly correlated with their performance on the original versions and is quite robust to distractor choices and option orders, while the evaluation time is reduced by a factor of up to 30. Following a similar procedure, we introduce PythonIO, a new program output prediction MC dataset constructed from two other popular LLM evaluation benchmarks, HumanEval and MBPP. Our data and code are available at https://github.com/Geralt-Targaryen/MC-Evaluation.
Authors: Ulugbek Salaev
Abstract: As Uzbek language is agglutinative, has many morphological features which words formed by combining root and affixes. Affixes play an important role in the morphological analysis of words, by adding additional meanings and grammatical functions to words. Inflectional endings are utilized to express various morphological features within the language. This feature introduces numerous possibilities for word endings, thereby significantly expanding the word vocabulary and exacerbating issues related to data sparsity in statistical models. This paper present modeling of the morphological analysis of Uzbek words, including stemming, lemmatizing, and the extraction of morphological information while considering morpho-phonetic exceptions. Main steps of the model involve developing a complete set of word-ending with assigned morphological information, and additional datasets for morphological analysis. The proposed model was evaluated using a curated test set comprising 5.3K words. Through manual verification of stemming, lemmatizing, and morphological feature corrections carried out by linguistic specialists, it obtained a word-level accuracy of over 91%. The developed tool based on the proposed model is available as a web-based application and an open-source Python library.
Authors: Ruochen Zhao, Wenxuan Zhang, Yew Ken Chia, Deli Zhao, Lidong Bing
Abstract: As LLMs evolve on a daily basis, there is an urgent need for a trustworthy evaluation method that can provide robust evaluation results in a timely fashion. Currently, as static benchmarks are prone to contamination concerns, users tend to trust human voting platforms, such as Chatbot Arena. However, human annotations require extensive manual efforts. To provide an automatic, robust, and trustworthy evaluation framework, we innovatively propose the Auto-Arena of LLMs, which automates the entire evaluation process with LLM agents. Firstly, an examiner LLM devises queries. Then, a pair of candidate LLMs engage in a multi-round peer-battle around the query, during which the LLM's true performance gaps become visible. Finally, a committee of LLM judges collectively discuss and determine the winner, which alleviates bias and promotes fairness. In our extensive experiment on the 17 newest LLMs, Auto-Arena shows the highest correlation with human preferences, providing a promising alternative to human evaluation platforms.
Authors: Yichuan Wang, Yang Liu, Yu Yan, Qun Wang, Shulei Wu, Xucheng Huang, Ling Jiang
Abstract: We introduce Xmodel-LM, a compact and efficient 1.1B language model pre-trained on around 2 trillion tokens. Trained on our self-built dataset (Xdata), which balances Chinese and English corpora based on downstream task optimization, Xmodel-LM exhibits remarkable performance despite its smaller size. It notably surpasses existing open-source language models of similar scale. Our model checkpoints and code are publicly accessible on GitHub at https://github.com/XiaoduoAILab/XmodelLM.
Authors: Zhenyun Deng, Michael Schlichtkrull, Andreas Vlachos
Abstract: Selecting which claims to check is a time-consuming task for human fact-checkers, especially from documents consisting of multiple sentences and containing multiple claims. However, existing claim extraction approaches focus more on identifying and extracting claims from individual sentences, e.g., identifying whether a sentence contains a claim or the exact boundaries of the claim within a sentence. In this paper, we propose a method for document-level claim extraction for fact-checking, which aims to extract check-worthy claims from documents and decontextualise them so that they can be understood out of context. Specifically, we first recast claim extraction as extractive summarization in order to identify central sentences from documents, then rewrite them to include necessary context from the originating document through sentence decontextualisation. Evaluation with both automatic metrics and a fact-checking professional shows that our method is able to extract check-worthy claims from documents more accurately than previous work, while also improving evidence retrieval.
Authors: Thomas Bott, Florian Lux, Ngoc Thang Vu
Abstract: In recent years, prompting has quickly become one of the standard ways of steering the outputs of generative machine learning models, due to its intuitive use of natural language. In this work, we propose a system conditioned on embeddings derived from an emotionally rich text that serves as prompt. Thereby, a joint representation of speaker and prompt embeddings is integrated at several points within a transformer-based architecture. Our approach is trained on merged emotional speech and text datasets and varies prompts in each training iteration to increase the generalization capabilities of the model. Objective and subjective evaluation results demonstrate the ability of the conditioned synthesis system to accurately transfer the emotions present in a prompt to speech. At the same time, precise tractability of speaker identities as well as overall high speech quality and intelligibility are maintained.
Authors: Reid McIlroy-Young, Katrina Brown, Conlan Olson, Linjun Zhang, Cynthia Dwork
Abstract: The development of generative language models that can create long and coherent textual outputs via autoregression has lead to a proliferation of uses and a corresponding sweep of analyses as researches work to determine the limitations of this new paradigm. Unlike humans, these 'Large Language Models' (LLMs) are highly sensitive to small changes in their inputs, leading to unwanted inconsistency in their behavior. One problematic inconsistency when LLMs are used to answer multiple-choice questions or analyze multiple inputs is order dependency: the output of an LLM can (and often does) change significantly when sub-sequences are swapped, despite both orderings being semantically identical. In this paper we present Set-Based Prompting, a technique that guarantees the output of an LLM will not have order dependence on a specified set of sub-sequences. We show that this method provably eliminates order dependency, and that it can be applied to any transformer-based LLM to enable text generation that is unaffected by re-orderings. Delving into the implications of our method, we show that, despite our inputs being out of distribution, the impact on expected accuracy is small, where the expectation is over the order of uniformly chosen shuffling of the candidate responses, and usually significantly less in practice. Thus, Set-Based Prompting can be used as a 'dropped-in' method on fully trained models. Finally, we discuss how our method's success suggests that other strong guarantees can be obtained on LLM performance via modifying the input representations.
Authors: Alina Leidinger, Robert van Rooij, Ekaterina Shutova
Abstract: Recent scholarship on reasoning in LLMs has supplied evidence of impressive performance and flexible adaptation to machine generated or human feedback. Nonmonotonic reasoning, crucial to human cognition for navigating the real world, remains a challenging, yet understudied task. In this work, we study nonmonotonic reasoning capabilities of seven state-of-the-art LLMs in one abstract and one commonsense reasoning task featuring generics, such as 'Birds fly', and exceptions, 'Penguins don't fly' (see Fig. 1). While LLMs exhibit reasoning patterns in accordance with human nonmonotonic reasoning abilities, they fail to maintain stable beliefs on truth conditions of generics at the addition of supporting examples ('Owls fly') or unrelated information ('Lions have manes'). Our findings highlight pitfalls in attributing human reasoning behaviours to LLMs, as well as assessing general capabilities, while consistent reasoning remains elusive.
Authors: Hidetoshi Matsuo, Mizuho Nishio, Takaaki Matsunaga, Koji Fujimoto, Takamichi Murakami
Abstract: Background: Structured radiology reports remains underdeveloped due to labor-intensive structuring and narrative-style reporting. Deep learning, particularly large language models (LLMs) like GPT-3.5, offers promise in automating the structuring of radiology reports in natural languages. However, although it has been reported that LLMs are less effective in languages other than English, their radiological performance has not been extensively studied. Purpose: This study aimed to investigate the accuracy of TNM classification based on radiology reports using GPT3.5-turbo (GPT3.5) and the utility of multilingual LLMs in both Japanese and English. Material and Methods: Utilizing GPT3.5, we developed a system to automatically generate TNM classifications from chest CT reports for lung cancer and evaluate its performance. We statistically analyzed the impact of providing full or partial TNM definitions in both languages using a Generalized Linear Mixed Model. Results: Highest accuracy was attained with full TNM definitions and radiology reports in English (M = 94%, N = 80%, T = 47%, and ALL = 36%). Providing definitions for each of the T, N, and M factors statistically improved their respective accuracies (T: odds ratio (OR) = 2.35, p < 0.001; N: OR = 1.94, p < 0.01; M: OR = 2.50, p < 0.001). Japanese reports exhibited decreased N and M accuracies (N accuracy: OR = 0.74 and M accuracy: OR = 0.21). Conclusion: This study underscores the potential of multilingual LLMs for automatic TNM classification in radiology reports. Even without additional model training, performance improvements were evident with the provided TNM definitions, indicating LLMs' relevance in radiology contexts.
Authors: Shoutao Guo, Shaolei Zhang, Zhengrui Ma, Min Zhang, Yang Feng
Abstract: Simultaneous Machine Translation (SiMT) generates target translations while reading the source sentence. It relies on a policy to determine the optimal timing for reading sentences and generating translations. Existing SiMT methods generally adopt the traditional Transformer architecture, which concurrently determines the policy and generates translations. While they excel at determining policies, their translation performance is suboptimal. Conversely, Large Language Models (LLMs), trained on extensive corpora, possess superior generation capabilities, but it is difficult for them to acquire translation policy through the training methods of SiMT. Therefore, we introduce Agent-SiMT, a framework combining the strengths of LLMs and traditional SiMT methods. Agent-SiMT contains the policy-decision agent and the translation agent. The policy-decision agent is managed by a SiMT model, which determines the translation policy using partial source sentence and translation. The translation agent, leveraging an LLM, generates translation based on the partial source sentence. The two agents collaborate to accomplish SiMT. Experiments demonstrate that Agent-SiMT attains state-of-the-art performance.
Authors: Frederic Kirstein, Jan Philip Wahle, Bela Gipp, Terry Ruas
Abstract: Abstractive dialogue summarization is the task of distilling conversations into informative and concise summaries. Although reviews have been conducted on this topic, there is a lack of comprehensive work detailing the challenges of dialogue summarization, unifying the differing understanding of the task, and aligning proposed techniques, datasets, and evaluation metrics with the challenges. This article summarizes the research on Transformer-based abstractive summarization for English dialogues by systematically reviewing 1262 unique research papers published between 2019 and 2024, relying on the Semantic Scholar and DBLP databases. We cover the main challenges present in dialog summarization (i.e., language, structure, comprehension, speaker, salience, and factuality) and link them to corresponding techniques such as graph-based approaches, additional training tasks, and planning strategies, which typically overly rely on BART-based encoder-decoder models. We find that while some challenges, like language, have seen considerable progress, mainly due to training methods, others, such as comprehension, factuality, and salience, remain difficult and hold significant research opportunities. We investigate how these approaches are typically assessed, covering the datasets for the subdomains of dialogue (e.g., meeting, medical), the established automatic metrics and human evaluation approaches for assessing scores and annotator agreement. We observe that only a few datasets span across all subdomains. The ROUGE metric is the most used, while human evaluation is frequently reported without sufficient detail on inner-annotator agreement and annotation guidelines. Additionally, we discuss the possible implications of the recently explored large language models and conclude that despite a potential shift in relevance and difficulty, our described challenge taxonomy remains relevant.
Authors: Maciej P. Polak, Shrey Modi, Anna Latosinska, Jinming Zhang, Ching-Wen Wang, Shaonan Wang, Ayan Deep Hazra, Dane Morgan
Abstract: Accurate and comprehensive material databases extracted from research papers are crucial for materials science and engineering, but their development requires significant human effort. With large language models (LLMs) transforming the way humans interact with text, LLMs provide an opportunity to revolutionize data extraction. In this study, we demonstrate a simple and efficient method for extracting materials data from full-text research papers leveraging the capabilities of LLMs combined with human supervision. This approach is particularly suitable for mid-sized databases and requires minimal to no coding or prior knowledge about the extracted property. It offers high recall and nearly perfect precision in the resulting database. The method is easily adaptable to new and superior language models, ensuring continued utility. We show this by evaluating and comparing its performance on GPT-3 and GPT-3.5/4 (which underlie ChatGPT), as well as free alternatives such as BART and DeBERTaV3. We provide a detailed analysis of the method's performance in extracting sentences containing bulk modulus data, achieving up to 90% precision at 96% recall, depending on the amount of human effort involved. We further demonstrate the method's broader effectiveness by developing a database of critical cooling rates for metallic glasses over twice the size of previous human curated databases.
Authors: Yuu Jinnai, Tetsuro Morimura, Ukyo Honda, Kaito Ariu, Kenshi Abe
Abstract: Minimum Bayes Risk (MBR) decoding has been shown to be a powerful alternative to beam search decoding in a variety of text generation tasks. MBR decoding selects a hypothesis from a pool of hypotheses that has the least expected risk under a probability model according to a given utility function. Since it is impractical to compute the expected risk exactly over all possible hypotheses, two approximations are commonly used in MBR. First, it integrates over a sampled set of hypotheses rather than over all possible hypotheses. Second, it estimates the probability of each hypothesis using a Monte Carlo estimator. While the first approximation is necessary to make it computationally feasible, the second is not essential since we typically have access to the model probability at inference time. We propose Model-Based MBR (MBMBR), a variant of MBR that uses the model probability itself as the estimate of the probability distribution instead of the Monte Carlo estimate. We show analytically and empirically that the model-based estimate is more promising than the Monte Carlo estimate in text generation tasks. Our experiments show that MBMBR outperforms MBR in several text generation tasks, both with encoder-decoder models and with large language models.
Authors: Francesco Paissan, Elisabetta Farella
Abstract: Contrastive Language-Audio Pretraining (CLAP) became of crucial importance in the field of audio and speech processing. Its employment ranges from sound event detection to text-to-audio generation. However, one of the main limitations is the considerable amount of data required in the training process and the overall computational complexity during inference. This paper investigates how we can reduce the complexity of contrastive language-audio pre-trained models, yielding an efficient model that we call tinyCLAP. We derive an unimodal distillation loss from first principles and explore how the dimensionality of the shared, multimodal latent space can be reduced via pruning. TinyCLAP uses only 6% of the original Microsoft CLAP parameters with a minimal reduction (less than 5%) in zero-shot classification performance across the three sound event detection datasets on which it was tested
Authors: Andr\'es Villa, Juan Carlos Le\'on Alc\'azar, Alvaro Soto, Bernard Ghanem
Abstract: Large Vision and Language Models have enabled significant advances in fully supervised and zero-shot visual tasks. These large architectures serve as the baseline to what is currently known as Instruction Tuning Large Vision and Language models (IT-LVLMs). IT-LVLMs are general-purpose multi-modal assistants whose responses are modulated by natural language instructions and visual data. Despite this versatility, IT-LVLM effectiveness in fundamental computer vision problems remains unclear, primarily due to the absence of a standardized evaluation benchmark. This paper introduces a Multi-modal Evaluation Benchmark named MERLIM, a scalable test-bed to assess the capabilities of IT-LVLMs on fundamental computer vision tasks. MERLIM contains over 300K image-question pairs and has a strong focus on detecting cross-modal "hallucination" events in IT-LVLMs. Our results bring important insights on the performance of state-of-the-art IT-LVMLs including limitations at identifying fine-grained visual concepts, object hallucinations across tasks, and biases towards the language query. Our findings also suggest that these models have weak visual grounding, but manage to make adequate guesses from global visual patterns or language biases contained in the LLM component.
Authors: Yuu Jinnai, Kaito Ariu
Abstract: Minimum Bayes-Risk (MBR) decoding is shown to be a powerful alternative to beam search decoding for a wide range of text generation tasks. However, MBR requires a huge amount of time for inference to compute the MBR objective, which makes the method infeasible in many situations where response time is critical. Confidence-based pruning (CBP) (Cheng and Vlachos, 2023) has recently been proposed to reduce the inference time in machine translation tasks. Although it is shown to significantly reduce the amount of computation, it requires hyperparameter tuning using a development set to be effective. To this end, we propose Approximate Minimum Bayes-Risk (AMBR) decoding, a hyperparameter-free method to run MBR decoding approximately. AMBR is derived from the observation that the problem of computing the sample-based MBR objective is the medoid identification problem. AMBR uses the Correlated Sequential Halving (CSH) algorithm (Baharav and Tse, 2019), the best approximation algorithm to date for the medoid identification problem, to compute the sample-based MBR objective. We evaluate AMBR on machine translation, text summarization, and image captioning tasks. The results show that AMBR achieves on par with CBP, with CBP selecting hyperparameters through an Oracle for each given computation budget.
Authors: Emanuele La Malfa, Christoph Weinhuber, Orazio Torre, Fangru Lin, Samuele Marro, Anthony Cohn, Nigel Shadbolt, Michael Wooldridge
Abstract: Many reasoning, planning, and problem-solving tasks share an intrinsic algorithmic nature: correctly simulating each step is a sufficient condition to solve them correctly. This work studies to what extent Large Language Models (LLMs) can simulate coding and algorithmic tasks to provide insights into general capabilities in such algorithmic reasoning tasks. We introduce benchmarks for straight-line programs, code that contains critical paths, and approximate and redundant instructions. We further assess the simulation capabilities of LLMs with sorting algorithms and nested loops and show that a routine's computational complexity directly affects an LLM's ability to simulate its execution. While the most powerful LLMs exhibit relatively strong simulation capabilities, the process is fragile, seems to rely heavily on pattern recognition, and is affected by memorisation. We propose a novel off-the-shelf prompting method, Chain of Simulation (CoSm), which instructs LLMs to simulate code execution line by line/follow the computation pattern of compilers. CoSm efficiently helps LLMs reduce memorisation and shallow pattern recognition while improving simulation performance. We consider the success of CoSm in code simulation to be inspirational for other general routine simulation reasoning tasks.
Authors: Jianrui Zhang, Mu Cai, Tengyang Xie, Yong Jae Lee
Abstract: We propose CounterCurate, a framework to comprehensively improve the visio-linguistic compositional reasoning capability for both contrastive and generative multimodal models. In particular, we identify two critical under-explored problems: the neglect of the physically grounded reasoning (counting and position understanding) and the potential of using highly capable text and image generation models for semantic counterfactual fine-tuning. Our work pioneers an approach that addresses these gaps. We first spotlight the near-chance performance of multimodal models like CLIP and LLaVA in physically grounded compositional reasoning. We then apply simple data augmentation using grounded image generation model GLIGEN to generate fine-tuning data, resulting in significant performance improvements: +33% and +37% for CLIP and LLaVA, respectively, on our newly curated Flickr30k-Positions benchmark. Moreover, we exploit the capabilities of high-performing text generation and image generation models, specifically GPT-4V and DALLE-3, to curate challenging semantic counterfactuals, thereby further enhancing compositional reasoning capabilities on benchmarks such as SugarCrepe, where CounterCurate outperforms GPT-4V. To facilitate future research, we release our code, dataset, benchmark, and checkpoints at https://countercurate.github.io.
Authors: Natalia Tomashenko, Xiaoxiao Miao, Pierre Champion, Sarina Meyer, Xin Wang, Emmanuel Vincent, Michele Panariello, Nicholas Evans, Junichi Yamagishi, Massimiliano Todisco
Abstract: The task of the challenge is to develop a voice anonymization system for speech data which conceals the speaker's voice identity while protecting linguistic content and emotional states. The organizers provide development and evaluation datasets and evaluation scripts, as well as baseline anonymization systems and a list of training resources formed on the basis of the participants' requests. Participants apply their developed anonymization systems, run evaluation scripts and submit evaluation results and anonymized speech data to the organizers. Results will be presented at a workshop held in conjunction with Interspeech 2024 to which all participants are invited to present their challenge systems and to submit additional workshop papers.
Authors: Hanze Dong, Wei Xiong, Bo Pang, Haoxiang Wang, Han Zhao, Yingbo Zhou, Nan Jiang, Doyen Sahoo, Caiming Xiong, Tong Zhang
Abstract: We present the workflow of Online Iterative Reinforcement Learning from Human Feedback (RLHF) in this technical report, which is widely reported to outperform its offline counterpart by a large margin in the recent large language model (LLM) literature. However, existing open-source RLHF projects are still largely confined to the offline learning setting. In this technical report, we aim to fill in this gap and provide a detailed recipe that is easy to reproduce for online iterative RLHF. In particular, since online human feedback is usually infeasible for open-source communities with limited resources, we start by constructing preference models using a diverse set of open-source datasets and use the constructed proxy preference model to approximate human feedback. Then, we discuss the theoretical insights and algorithmic principles behind online iterative RLHF, followed by a detailed practical implementation. Our trained LLM, LLaMA-3-8B-SFR-Iterative-DPO-R, achieves impressive performance on LLM chatbot benchmarks, including AlpacaEval-2, Arena-Hard, and MT-Bench, as well as other academic benchmarks such as HumanEval and TruthfulQA. We have shown that supervised fine-tuning (SFT) and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets. Further, we have made our models, curated datasets, and comprehensive step-by-step code guidebooks publicly available. Please refer to https://github.com/RLHFlow/RLHF-Reward-Modeling and https://github.com/RLHFlow/Online-RLHF for more detailed information.
URLs: https://github.com/RLHFlow/RLHF-Reward-Modeling, https://github.com/RLHFlow/Online-RLHF
Authors: Kun Zhou, Shengkui Zhao, Yukun Ma, Chong Zhang, Hao Wang, Dianwen Ng, Chongjia Ni, Nguyen Trung Hieu, Jia Qi Yip, Bin Ma
Abstract: Recent language model-based text-to-speech (TTS) frameworks demonstrate scalability and in-context learning capabilities. However, they suffer from robustness issues due to the accumulation of errors in speech unit predictions during autoregressive language modeling. In this paper, we propose a phonetic enhanced language modeling method to improve the performance of TTS models. We leverage self-supervised representations that are phonetically rich as the training target for the autoregressive language model. Subsequently, a non-autoregressive model is employed to predict discrete acoustic codecs that contain fine-grained acoustic details. The TTS model focuses solely on linguistic modeling during autoregressive training, thereby reducing the error propagation that occurs in non-autoregressive training. Both objective and subjective evaluations validate the effectiveness of our proposed method.
Authors: Jianing Yang, Xuweiyi Chen, Nikhil Madaan, Madhavan Iyengar, Shengyi Qian, David F. Fouhey, Joyce Chai
Abstract: The integration of language and 3D perception is crucial for developing embodied agents and robots that comprehend and interact with the physical world. While large language models (LLMs) have demonstrated impressive language understanding and generation capabilities, their adaptation to 3D environments (3D-LLMs) remains in its early stages. A primary challenge is the absence of large-scale datasets that provide dense grounding between language and 3D scenes. In this paper, we introduce 3D-GRAND, a pioneering large-scale dataset comprising 40,087 household scenes paired with 6.2 million densely-grounded scene-language instructions. Our results show that instruction tuning with 3D-GRAND significantly enhances grounding capabilities and reduces hallucinations in 3D-LLMs. As part of our contributions, we propose a comprehensive benchmark 3D-POPE to systematically evaluate hallucination in 3D-LLMs, enabling fair comparisons among future models. Our experiments highlight a scaling effect between dataset size and 3D-LLM performance, emphasizing the critical role of large-scale 3D-text datasets in advancing embodied AI research. Notably, our results demonstrate early signals for effective sim-to-real transfer, indicating that models trained on large synthetic data can perform well on real-world 3D scans. Through 3D-GRAND and 3D-POPE, we aim to equip the embodied AI community with essential resources and insights, setting the stage for more reliable and better-grounded 3D-LLMs. Project website: https://3d-grand.github.io
Authors: Zijian Hei, Weiling Liu, Wenjie Ou, Juyi Qiao, Junming Jiao, Zhiqing Zhu, Guowen Song
Abstract: Retrieval-Augmented Generation (RAG) has significantly demonstrated the performance of Large Language Models (LLMs) in the knowledge-intensive tasks, such as Question-Answering (QA). RAG expands the query context by incorporating external knowledge bases to enhance the response accuracy. However, it would be inefficient to access LLMs multiple times for each query and unreliable to retrieve all the relevant documents by a single query. We find that even though there is low relevance between some critical documents and query, it is possible to retrieve the remaining documents by combining parts of the documents with the query. To mine the relevance, a two-stage retrieval framework called Dynamic-Relevant Retrieval-Augmented Generation (DR-RAG) is proposed to improve document retrieval recall and the accuracy of answers while maintaining efficiency. Also, a small classifier is applied to two different selection strategies to determine the contribution of the retrieved documents to answering the query and retrieve the relatively relevant documents. Meanwhile, DR-RAG call the LLMs only once, which significantly improves the efficiency of the experiment. The experimental results on multi-hop QA datasets show that DR-RAG can significantly improve the accuracy of the answers and achieve new progress in QA systems.
Authors: Teun van der Weij, Felix Hofst\"atter, Ollie Jaffe, Samuel F. Brown, Francis Rhys Ward
Abstract: Trustworthy capability evaluations are crucial for ensuring the safety of AI systems, and are becoming a key component of AI regulation. However, the developers of an AI system, or the AI system itself, may have incentives for evaluations to understate the AI's actual capability. These conflicting interests lead to the problem of sandbagging $\unicode{x2013}$ which we define as "strategic underperformance on an evaluation". In this paper we assess sandbagging capabilities in contemporary language models (LMs). We prompt frontier LMs, like GPT-4 and Claude 3 Opus, to selectively underperform on dangerous capability evaluations, while maintaining performance on general (harmless) capability evaluations. Moreover, we find that models can be fine-tuned, on a synthetic dataset, to hide specific capabilities unless given a password. This behaviour generalizes to high-quality, held-out benchmarks such as WMDP. In addition, we show that both frontier and smaller models can be prompted, or password-locked, to target specific scores on a capability evaluation. Even more, we found that a capable password-locked model (Llama 3 70b) is reasonably able to emulate a less capable model (Llama 2 7b). Overall, our results suggest that capability evaluations are vulnerable to sandbagging. This vulnerability decreases the trustworthiness of evaluations, and thereby undermines important safety decisions regarding the development and deployment of advanced AI systems.