Authors: Qianxi Li, Yingyue Cao, Jikun Kang, Tianpei Yang, Xi Chen, Jun Jin, Matthew E. Taylor
Fine-tuning Large Language Models (LLMs) adapts a trained model to specific downstream tasks, significantly improving task-specific performance. Supervised Fine-Tuning (SFT) is a common approach, where an LLM is trained to produce desired answers. However, LLMs trained with SFT sometimes make simple mistakes and result in hallucinations on reasoning tasks such as question-answering. Without external feedback, it is difficult for SFT to learn a good mapping between the question and the desired answer, especially with a small dataset. This paper introduces an alternative to SFT called Natural Language Feedback for Finetuning LLMs (LaFFi). LaFFi has LLMs directly predict the feedback they will receive from an annotator. We find that requiring such reflection can significantly improve the accuracy in in-domain question-answering tasks, providing a promising direction for the application of natural language feedback in the realm of SFT LLMs. Additional ablation studies show that the portion of human-annotated data in the annotated datasets affects the fine-tuning performance.
Authors: Dongsheng Wang, Natraj Raman, Mathieu Sibue, Zhiqiang Ma, Petr Babkin, Simerjot Kaur, Yulong Pei, Armineh Nourbakhsh, Xiaomo Liu
Enterprise documents such as forms, invoices, receipts, reports, contracts, and other similar records, often carry rich semantics at the intersection of textual and spatial modalities. The visual cues offered by their complex layouts play a crucial role in comprehending these documents effectively. In this paper, we present DocLLM, a lightweight extension to traditional large language models (LLMs) for reasoning over visual documents, taking into account both textual semantics and spatial layout. Our model differs from existing multimodal LLMs by avoiding expensive image encoders and focuses exclusively on bounding box information to incorporate the spatial layout structure. Specifically, the cross-alignment between text and spatial modalities is captured by decomposing the attention mechanism in classical transformers to a set of disentangled matrices. Furthermore, we devise a pre-training objective that learns to infill text segments. This approach allows us to address irregular layouts and heterogeneous content frequently encountered in visual documents. The pre-trained model is fine-tuned using a large-scale instruction dataset, covering four core document intelligence tasks. We demonstrate that our solution outperforms SotA LLMs on 14 out of 16 datasets across all tasks, and generalizes well to 4 out of 5 previously unseen datasets.
Authors: Jinlong Xue, Yayue Deng, Yingming Gao, Ya Li
Recent advancements in diffusion models and large language models (LLMs) have significantly propelled the field of AIGC. Text-to-Audio (TTA), a burgeoning AIGC application designed to generate audio from natural language prompts, is attracting increasing attention. However, existing TTA studies often struggle with generation quality and text-audio alignment, especially for complex textual inputs. Drawing inspiration from state-of-the-art Text-to-Image (T2I) diffusion models, we introduce Auffusion, a TTA system adapting T2I model frameworks to TTA task, by effectively leveraging their inherent generative strengths and precise cross-modal alignment. Our objective and subjective evaluations demonstrate that Auffusion surpasses previous TTA approaches using limited data and computational resource. Furthermore, previous studies in T2I recognizes the significant impact of encoder choice on cross-modal alignment, like fine-grained details and object bindings, while similar evaluation is lacking in prior TTA works. Through comprehensive ablation studies and innovative cross-attention map visualizations, we provide insightful assessments of text-audio alignment in TTA. Our findings reveal Auffusion's superior capability in generating audios that accurately match textual descriptions, which further demonstrated in several related tasks, such as audio style transfer, inpainting and other manipulations. Our implementation and demos are available at https://auffusion.github.io.
Authors: Ife Adebara, AbdelRahim Elmadany, Muhammad Abdul-Mageed
Low-resource African languages pose unique challenges for natural language processing (NLP) tasks, including natural language generation (NLG). In this paper, we develop Cheetah, a massively multilingual NLG language model for African languages. Cheetah supports 517 African languages and language varieties, allowing us to address the scarcity of NLG resources and provide a solution to foster linguistic diversity. We demonstrate the effectiveness of Cheetah through comprehensive evaluations across seven generation downstream tasks. In five of the seven tasks, Cheetah significantly outperforms other models, showcasing its remarkable performance for generating coherent and contextually appropriate text in a wide range of African languages. We additionally conduct a detailed human evaluation to delve deeper into the linguistic capabilities of Cheetah. The introduction of Cheetah has far-reaching benefits for linguistic diversity. By leveraging pretrained models and adapting them to specific languages, our approach facilitates the development of practical NLG applications for African communities. The findings of this study contribute to advancing NLP research in low-resource settings, enabling greater accessibility and inclusion for African languages in a rapidly expanding digital landscape. We will publicly release our models for research.
Authors: Jun Zhao, Zhihao Zhang, Qi Zhang, Tao Gui, Xuanjing Huang
In recent times, substantial advancements have been witnessed in large language models (LLMs), exemplified by ChatGPT, showcasing remarkable proficiency across a range of complex tasks. However, many mainstream LLMs (e.g. LLaMA) are pretrained on English-dominant corpus, which limits their performance in other non-English languages. In this paper, we focus on how to effectively transfer the capabilities of language generation and following instructions to a non-English language. To answer this question, we conduct an extensive empirical investigation based on LLaMA, accumulating over 1440 GPU hours. We analyze the impact of key factors such as vocabulary extension, further pretraining, and instruction tuning on transfer. To accurately assess the model's level of knowledge, we employ four widely used standardized testing benchmarks: C-Eval, MMLU, AGI-Eval, and GAOKAO-Bench. Furthermore, a comprehensive evaluation of the model's response quality is conducted, considering aspects such as accuracy, fluency, informativeness, logical coherence, and harmlessness, based on LLM-Eval, a benchmarks consisting instruction tasks from 17 diverse categories. Our evaluation results demonstrate that comparable performance to state-of-the-art transfer models can be achieved with less than 1% of the pretraining data, both in terms of knowledge alignment and response quality. Furthermore, the experimental outcomes across the thirteen low-resource languages also exhibit similar trends. We anticipate that the conclusions revealed by the experiments will aid the community in developing non-English LLMs.
Authors: Jerrold Soh
We propose and evaluate an automated pipeline for discovering significant topics from legal decision texts by passing features synthesized with topic models through penalised regressions and post-selection significance tests. The method identifies case topics significantly correlated with outcomes, topic-word distributions which can be manually-interpreted to gain insights about significant topics, and case-topic weights which can be used to identify representative cases for each topic. We demonstrate the method on a new dataset of domain name disputes and a canonical dataset of European Court of Human Rights violation cases. Topic models based on latent semantic analysis as well as language model embeddings are evaluated. We show that topics derived by the pipeline are consistent with legal doctrines in both areas and can be useful in other related legal analysis tasks.
Authors: Zhichao Yin, Binyuan Hui, Min Yang, Fei Huang, Yongbin Li
Recently, substantial advancements in pre-trained vision-language models have greatly enhanced the capabilities of multi-modal dialog systems. These models have demonstrated significant improvements by fine-tuning on downstream tasks. However, the existing pre-trained models primarily focus on effectively capturing the alignment between vision and language modalities, often ignoring the intricate nature of dialog context. In this paper, we propose a parameter-efficient prompt-tuning method named DialCLIP for multi-modal dialog retrieval. Specifically, our approach introduces a multi-modal context prompt generator to learn context features which are subsequently distilled into prompts within the pre-trained vision-language model CLIP. Besides, we introduce domain prompt to mitigate the disc repancy from the downstream dialog data. To facilitate various types of retrieval, we also design multiple experts to learn mappings from CLIP outputs to multi-modal representation space, with each expert being responsible to one specific retrieval type. Extensive experiments show that DialCLIP achieves state-of-the-art performance on two widely recognized benchmark datasets (i.e., PhotoChat and MMDialog) by tuning a mere 0.04% of the total parameters. These results highlight the efficacy and efficiency of our proposed approach, underscoring its potential to advance the field of multi-modal dialog retrieval.
Authors: Triet Huynh Minh, Quan Le Bao
Poetry generation has been a challenging task in the field of Natural Language Processing, as it requires the model to understand the nuances of language, sentiment, and style. In this paper, we propose using Large Language Models to generate Vietnamese poems from natural language prompts, thereby facilitating an intuitive process with enhanced content control. Our most efficacious model, the GPT-3 Babbage variant, achieves a custom evaluation score of 0.8, specifically tailored to the "luc bat" genre of Vietnamese poetry. Furthermore, we also explore the idea of paraphrasing poems into normal text prompts and yield a relatively high score of 0.718 in the "luc bat" genre. This experiment presents the potential for cross-Language poem-to-poem translation with translated poems as the inputs while concurrently maintaining complete control over the generated content.
Authors: Xianjun Yang, Stephen D. Wilson, Linda Petzold
This paper presents the development of a specialized chatbot for materials science, leveraging the Llama-2 language model, and continuing pre-training on the expansive research articles in the materials science domain from the S2ORC dataset. The methodology involves an initial pretraining phase on over one million domain-specific papers, followed by an instruction-tuning process to refine the chatbot's capabilities. The chatbot is designed to assist researchers, educators, and students by providing instant, context-aware responses to queries in the field of materials science. We make the four trained checkpoints (7B, 13B, with or without chat ability) freely available to the research community at https://github.com/Xianjun-Yang/Quokka.
Authors: Ha Le, Bao Tran, Phuong Le, Tan Nguyen, Dac Nguyen, Ngoan Pham, Dang Huynh
Comparative opinion mining is a specialized field of sentiment analysis that aims to identify and extract sentiments expressed comparatively. To address this task, we propose an approach that consists of solving three sequential sub-tasks: (i) identifying comparative sentence, i.e., if a sentence has a comparative meaning, (ii) extracting comparative elements, i.e., what are comparison subjects, objects, aspects, predicates, and (iii) classifying comparison types which contribute to a deeper comprehension of user sentiments in Vietnamese product reviews. Our method is ranked fifth at the Vietnamese Language and Speech Processing (VLSP) 2023 challenge on Comparative Opinion Mining (ComOM) from Vietnamese Product Reviews.
Authors: Shujie Li, Liang Li, Ruiying Geng, Min Yang, Binhua Li, Guanghu Yuan, Wanwei He, Shao Yuan, Can Ma, Fei Huang, Yongbin Li
Data-to-text (D2T) generation aims to transform structured data into natural language text. Data-to-text pre-training has proved to be powerful in enhancing D2T generation and yields impressive performances. However, previous pre-training methods either oversimplified structured data into a sequence without considering input structures or designed training objectives tailored for a specific data structure (e.g., table or knowledge graph). In this paper, we unify different types of structured data (i.e., table, key-value data, knowledge graph) into the graph format and cast different data-to-text generation tasks as graph-to-text generation. To effectively exploit the structural information of the input graph, we propose a structure-enhanced pre-training method for D2T generation by designing a structure-enhanced Transformer. Concretely, we devise a position matrix for the Transformer, encoding relative positional information of connected nodes in the input graph. In addition, we propose a new attention matrix to incorporate graph structures into the original Transformer by taking the available explicit connectivity structure into account. Extensive experiments on six benchmark datasets show the effectiveness of our model. Our source codes are available at https://github.com/AlibabaResearch/DAMO-ConvAI/tree/main/unid2t.
Authors: Yury Orlovskiy, Camille Thibault, Anne Imouza, Jean-François Godbout, Reihaneh Rabbany, Kellin Pelrine
Misinformation poses a variety of risks, such as undermining public trust and distorting factual discourse. Large Language Models (LLMs) like GPT-4 have been shown effective in mitigating misinformation, particularly in handling statements where enough context is provided. However, they struggle to assess ambiguous or context-deficient statements accurately. This work introduces a new method to resolve uncertainty in such statements. We propose a framework to categorize missing information and publish category labels for the LIAR-New dataset, which is adaptable to cross-domain content with missing information. We then leverage this framework to generate effective user queries for missing context. Compared to baselines, our method improves the rate at which generated questions are answerable by the user by 38 percentage points and classification performance by over 10 percentage points macro F1. Thus, this approach may provide a valuable component for future misinformation mitigation pipelines.
Authors: Zhongkun Liu, Zheng Chen, Mengqi Zhang, Zhaochun Ren, Zhumin Chen, Pengjie Ren
Fine-tuning has been demonstrated to be an effective method to improve the domain performance of large language models (LLMs). However, LLMs might fit the dataset bias and shortcuts for prediction, leading to poor generation performance. Experimental result shows that LLMs are prone to exhibit position bias, i.e., leveraging information positioned at the beginning or end, or specific positional cues within the input. Existing works on mitigating position bias require external bias knowledge or annotated non-biased samples, which is unpractical in reality. In this work, we propose a zero-shot position debiasing (ZOE) framework to mitigate position bias for LLMs. ZOE leverages unsupervised responses from pre-trained LLMs for debiasing, thus without any external knowledge or datasets. To improve the quality of unsupervised responses, we propose a master-slave alignment (MSA) module to prune these responses. Experiments on eight datasets and five tasks show that ZOE consistently outperforms existing methods in mitigating four types of position biases. Besides, ZOE achieves this by sacrificing only a small performance on biased samples, which is simple and effective.
Authors: Fuchen Long, Zhaofan Qiu, Ting Yao, Tao Mei
The recent innovations and breakthroughs in diffusion models have significantly expanded the possibilities of generating high-quality videos for the given prompts. Most existing works tackle the single-scene scenario with only one video event occurring in a single background. Extending to generate multi-scene videos nevertheless is not trivial and necessitates to nicely manage the logic in between while preserving the consistent visual appearance of key content across video scenes. In this paper, we propose a novel framework, namely VideoDrafter, for content-consistent multi-scene video generation. Technically, VideoDrafter leverages Large Language Models (LLM) to convert the input prompt into comprehensive multi-scene script that benefits from the logical knowledge learnt by LLM. The script for each scene includes a prompt describing the event, the foreground/background entities, as well as camera movement. VideoDrafter identifies the common entities throughout the script and asks LLM to detail each entity. The resultant entity description is then fed into a text-to-image model to generate a reference image for each entity. Finally, VideoDrafter outputs a multi-scene video by generating each scene video via a diffusion process that takes the reference images, the descriptive prompt of the event and camera movement into account. The diffusion model incorporates the reference images as the condition and alignment to strengthen the content consistency of multi-scene videos. Extensive experiments demonstrate that VideoDrafter outperforms the SOTA video generation models in terms of visual quality, content consistency, and user preference.
Authors: Vincent Freiberger, Erik Buchmann
Natural Language Processing (NLP) plays an important role in our daily lives, particularly due to the enormous progress of Large Language Models (LLM). However, NLP has many fairness-critical use cases, e.g., as an expert system in recruitment or as an LLM-based tutor in education. Since NLP is based on human language, potentially harmful biases can diffuse into NLP systems and produce unfair results, discriminate against minorities or generate legal issues. Hence, it is important to develop a fairness certification for NLP approaches. We follow a qualitative research approach towards a fairness certification for NLP. In particular, we have reviewed a large body of literature on algorithmic fairness, and we have conducted semi-structured expert interviews with a wide range of experts from that area. We have systematically devised six fairness criteria for NLP, which can be further refined into 18 sub-categories. Our criteria offer a foundation for operationalizing and testing processes to certify fairness, both from the perspective of the auditor and the audited organization.
Authors: Quan Tu, Shilong Fan, Zihang Tian, Rui Yan
Recently, the advent of large language models (LLMs) has revolutionized generative agents. Among them, Role-Playing Conversational Agents (RPCAs) attract considerable attention due to their ability to emotionally engage users. However, the absence of a comprehensive benchmark impedes progress in this field. To bridge this gap, we introduce CharacterEval, a Chinese benchmark for comprehensive RPCA assessment, complemented by a tailored high-quality dataset. The dataset comprises 1,785 multi-turn role-playing dialogues, encompassing 23,020 examples and featuring 77 characters derived from Chinese novels and scripts. It was carefully constructed, beginning with initial dialogue extraction via GPT-4, followed by rigorous human-led quality control, and enhanced with in-depth character profiles sourced from Baidu Baike. CharacterEval employs a multifaceted evaluation approach, encompassing thirteen targeted metrics on four dimensions. Comprehensive experiments on CharacterEval demonstrate that Chinese LLMs exhibit more promising capabilities than GPT-4 in Chinese role-playing conversation. Source code, data source and reward model will be publicly accessible at https://github.com/morecry/CharacterEval.
Authors: Vilém Zouhar, Ondřej Bojar
Automatic machine translation metrics often use human translations to determine the quality system translations. Common wisdom in the field dictates that the human references should be of very high quality. However, there are no cost-benefit analyses that could be used to guide practitioners who plan to collect references for machine translation evaluation. We find that higher-quality references lead to better metric correlations with humans at the segment-level. Having up to 7 references per segment and taking their average helps all metrics. Interestingly, the references from vendors of different qualities can be mixed together and improve metric success. Higher quality references, however, cost more to create and we frame this as an optimization problem: given a specific budget, what references should be collected to maximize metric success. These findings can be used by evaluators of shared tasks when references need to be created under a certain budget.
Authors: Ningyu Zhang, Yunzhi Yao, Bozhong Tian, Peng Wang, Shumin Deng, Mengru Wang, Zekun Xi, Shengyu Mao, Jintian Zhang, Yuansheng Ni, Siyuan Cheng, Ziwen Xu, Xin Xu, Jia-Chen Gu, Yong Jiang, Pengjun Xie, Fei Huang, Lei Liang, Zhiqiang Zhang, Xiaowei Zhu, Jun Zhou, Huajun Chen
Large Language Models (LLMs) have shown extraordinary capabilities in understanding and generating text that closely mirrors human communication. However, a primary limitation lies in the significant computational demands during training, arising from their extensive parameterization. This challenge is further intensified by the dynamic nature of the world, necessitating frequent updates to LLMs to correct outdated information or integrate new knowledge, thereby ensuring their continued relevance. Note that many applications demand continual model adjustments post-training to address deficiencies or undesirable behaviors. There is an increasing interest in efficient, lightweight methods for on-the-fly model modifications. To this end, recent years have seen a burgeoning in the techniques of knowledge editing for LLMs, which aim to efficiently modify LLMs' behaviors within specific domains while preserving overall performance across various inputs. In this paper, we first define the knowledge editing problem and then provide a comprehensive review of cutting-edge approaches. Drawing inspiration from educational and cognitive research theories, we propose a unified categorization criterion that classifies knowledge editing methods into three groups: resorting to external knowledge, merging knowledge into the model, and editing intrinsic knowledge. Furthermore, we introduce a new benchmark, KnowEdit, for a comprehensive empirical evaluation of representative knowledge editing approaches. Additionally, we provide an in-depth analysis of knowledge location, which can provide a deeper understanding of the knowledge structures inherent within LLMs. Finally, we discuss several potential applications of knowledge editing, outlining its broad and impactful implications.
Authors: Matthew Dahl, Varun Magesh, Mirac Suzgun, Daniel E. Ho
Large language models (LLMs) have the potential to transform the practice of law, but this potential is threatened by the presence of legal hallucinations -- responses from these models that are not consistent with legal facts. We investigate the extent of these hallucinations using an original suite of legal queries, comparing LLMs' responses to structured legal metadata and examining their consistency. Our work makes four key contributions: (1) We develop a typology of legal hallucinations, providing a conceptual framework for future research in this area. (2) We find that legal hallucinations are alarmingly prevalent, occurring between 69% of the time with ChatGPT 3.5 and 88% with Llama 2, when these models are asked specific, verifiable questions about random federal court cases. (3) We illustrate that LLMs often fail to correct a user's incorrect legal assumptions in a contra-factual question setup. (4) We provide evidence that LLMs cannot always predict, or do not always know, when they are producing legal hallucinations. Taken together, these findings caution against the rapid and unsupervised integration of popular LLMs into legal tasks. Even experienced lawyers must remain wary of legal hallucinations, and the risks are highest for those who stand to benefit from LLMs the most -- pro se litigants or those without access to traditional legal resources.
Authors: S.M Towhidul Islam Tonmoy, S M Mehedi Zaman, Vinija Jain, Anku Rani, Vipula Rawte, Aman Chadha, Amitava Das
As Large Language Models (LLMs) continue to advance in their ability to write human-like text, a key challenge remains around their tendency to hallucinate generating content that appears factual but is ungrounded. This issue of hallucination is arguably the biggest hindrance to safely deploying these powerful LLMs into real-world production systems that impact people's lives. The journey toward widespread adoption of LLMs in practical settings heavily relies on addressing and mitigating hallucinations. Unlike traditional AI systems focused on limited tasks, LLMs have been exposed to vast amounts of online text data during training. While this allows them to display impressive language fluency, it also means they are capable of extrapolating information from the biases in training data, misinterpreting ambiguous prompts, or modifying the information to align superficially with the input. This becomes hugely alarming when we rely on language generation capabilities for sensitive applications, such as summarizing medical records, financial analysis reports, etc. This paper presents a comprehensive survey of over 32 techniques developed to mitigate hallucination in LLMs. Notable among these are Retrieval Augmented Generation (Lewis et al, 2021), Knowledge Retrieval (Varshney et al,2023), CoNLI (Lei et al, 2023), and CoVe (Dhuliawala et al, 2023). Furthermore, we introduce a detailed taxonomy categorizing these methods based on various parameters, such as dataset utilization, common tasks, feedback mechanisms, and retriever types. This classification helps distinguish the diverse approaches specifically designed to tackle hallucination issues in LLMs. Additionally, we analyze the challenges and limitations inherent in these techniques, providing a solid foundation for future research in addressing hallucinations and related phenomena within the realm of LLMs.
Authors: Hongye Jin, Xiaotian Han, Jingfeng Yang, Zhimeng Jiang, Zirui Liu, Chia-Yuan Chang, Huiyuan Chen, Xia Hu
This work elicits LLMs' inherent ability to handle long contexts without fine-tuning. The limited length of the training sequence during training may limit the application of Large Language Models (LLMs) on long input sequences for inference. In this work, we argue that existing LLMs themselves have inherent capabilities for handling long contexts. Based on this argument, we suggest extending LLMs' context window by themselves to fully utilize the inherent ability.We propose Self-Extend to stimulate LLMs' long context handling potential. The basic idea is to construct bi-level attention information: the group level and the neighbor level. The two levels are computed by the original model's self-attention, which means the proposed does not require any training. With only four lines of code modification, the proposed method can effortlessly extend existing LLMs' context window without any fine-tuning. We conduct comprehensive experiments and the results show that the proposed method can effectively extend existing LLMs' context window's length.
Authors: Zaratiana Urchade, Nadi Tomeh, Pierre Holat, Thierry Charnois
In this paper, we propose a novel method for joint entity and relation extraction from unstructured text by framing it as a conditional sequence generation problem. In contrast to conventional generative information extraction models that are left-to-right token-level generators, our approach is \textit{span-based}. It generates a linearized graph where nodes represent text spans and edges represent relation triplets. Our method employs a transformer encoder-decoder architecture with pointing mechanism on a dynamic vocabulary of spans and relation types. Our model can capture the structural characteristics and boundaries of entities and relations through span representations while simultaneously grounding the generated output in the original text thanks to the pointing mechanism. Evaluation on benchmark datasets validates the effectiveness of our approach, demonstrating competitive results. Code is available at https://github.com/urchade/ATG.
Authors: Mohammad Aliannejadi, Zahra Abbasiantaeb, Shubham Chatterjee, Jeffery Dalton, Leif Azzopardi
Conversational Information Seeking stands as a pivotal research area with significant contributions from previous works. The TREC Interactive Knowledge Assistance Track (iKAT) builds on the foundational work of the TREC Conversational Assistance Track (CAsT). However, iKAT distinctively emphasizes the creation and research of conversational search agents that adapt responses based on user's prior interactions and present context. The challenge lies in enabling Conversational Search Agents (CSA) to incorporate this personalized context to efficiency and effectively guide users through the relevant information to them. iKAT also emphasizes decisional search tasks, where users sift through data and information to weigh up options in order to reach a conclusion or perform an action. These tasks, prevalent in everyday information-seeking decisions -- be it related to travel, health, or shopping -- often revolve around a subset of high-level information operators where queries or questions about the information space include: finding options, comparing options, identifying the pros and cons of options, etc. Given the different personas and their information need (expressed through the sequence of questions), diverse conversation trajectories will arise -- because the answers to these similar queries will be very different. In this paper, we report on the first year of TREC iKAT, describing the task, topics, data collection, and evaluation framework. We further review the submissions and summarize the findings.
Authors: Zixiang Chen, Yihe Deng, Huizhuo Yuan, Kaixuan Ji, Quanquan Gu
Harnessing the power of human-annotated data through Supervised Fine-Tuning (SFT) is pivotal for advancing Large Language Models (LLMs). In this paper, we delve into the prospect of growing a strong LLM out of a weak one without the need for acquiring additional human-annotated data. We propose a new fine-tuning method called Self-Play fIne-tuNing (SPIN), which starts from a supervised fine-tuned model. At the heart of SPIN lies a self-play mechanism, where the LLM refines its capability by playing against instances of itself. More specifically, the LLM generates its own training data from its previous iterations, refining its policy by discerning these self-generated responses from those obtained from human-annotated data. Our method progressively elevates the LLM from a nascent model to a formidable one, unlocking the full potential of human-annotated demonstration data for SFT. Theoretically, we prove that the global optimum to the training objective function of our method is achieved only when the LLM policy aligns with the target data distribution. Empirically, we evaluate our method on several benchmark datasets including the HuggingFace Open LLM Leaderboard, MT-Bench, and datasets from Big-Bench. Our results show that SPIN can significantly improve the LLM's performance across a variety of benchmarks and even outperform models trained through direct preference optimization (DPO) supplemented with extra GPT-4 preference data. This sheds light on the promise of self-play, enabling the achievement of human-level performance in LLMs without the need for expert opponents.
Authors: Ping-Rui Tsai, Yen-Ting Chou, Nathan-Christopher Wang, Hui-Ling Chen, Hong-Yue Huang, Zih-Jia Luo, Tzay-Ming Hong
Music, enchanting and poetic, permeates every corner of human civilization. Although music is not unfamiliar to people, our understanding of its essence remains limited, and there is still no universally accepted scientific description. This is primarily due to music being regarded as a product of both reason and emotion, making it difficult to define. In this article, we focus on the fundamental elements of music and construct an evolutionary network from the perspective of music as a natural language, aligning with the statistical characteristics of texts. Through this approach, we aim to comprehend the structural differences in music across different periods, enabling a more scientific exploration of music. Relying on the advantages of structuralism, we can concentrate on the relationships and order between the physical elements of music, rather than getting entangled in the blurred boundaries of science and philosophy. The scientific framework we present not only conforms to past conclusions in music, but also serves as a bridge that connects music to natural language processing and knowledge graphs.
Authors: Zhengqing Yuan, Xinyi Wang, Kun Wang, Lichao Sun
In recent years, advancements in large language models have been remarkable, with models such as ChatGPT demonstrating exceptional proficiency in diverse linguistic tasks. The pre-training of large models with billions of parameters, poses a formidable challenge, primarily due to the scarcity of datasets of a commensurate scale for effective training. Nevertheless, innovative strategies have emerged, including methods to fine-tune these pre-trained models using fewer parameters set, as evidenced by models like MiniGPT-4 and LLaVA. Despite their potential in various domains, these models remain limited in their understanding of artistic imagery. They have yet to fully grasp the intricate nuances of art images or to provide an objective articulation of the emotions they evoke, in a manner akin to human perception. This work introduces ArtGPT-4, a pioneering large vision-language model tailored to address the deficiencies of contemporary models in artistic comprehension. ArtGPT-4 underwent training on image-text pairs utilizing a Tesla A100 device in a mere 2 hours, with a dataset comprising approximately 0.52M entries. Impressively, the model can render images with an artistic-understanding and convey the emotions they inspire, mirroring human interpretation. Additionally, this work presents a unique dataset designed to evaluate the efficacy of vision-language models. In subsequent evaluations, ArtGPT-4 not only achieved state-of-the-art performance on the ArtEmis and ArtEmis-v2.0 datasets but also exceeded the established benchmarks introduced in This study, lagging behind professional artists' descriptions by a negligible 0.15 points on a 6-point scale. The code and the pre-trained model are accessible in https://huggingface.co/Tyrannosaurus/ArtGPT-4.
Authors: Khanh Nguyen
How do language models "think"? This paper formulates a probabilistic cognitive model called the bounded pragmatic speaker, which can characterize the operation of different variations of language models. Specifically, we demonstrate that large language models fine-tuned with reinforcement learning from human feedback (Ouyang et al., 2022) embody a model of thought that conceptually resembles a fast-and-slow model (Kahneman, 2011), which psychologists have attributed to humans. We discuss the limitations of reinforcement learning from human feedback as a fast-and-slow model of thought and propose avenues for expanding this framework. In essence, our research highlights the value of adopting a cognitive probabilistic modeling approach to gain insights into the comprehension, evaluation, and advancement of language models.
Authors: Zilun Zhang, Tiancheng Zhao, Yulong Guo, Jianwei Yin
Pre-trained Vision-Language Models (VLMs) utilizing extensive image-text paired data have demonstrated unprecedented image-text association capabilities, achieving remarkable results across various downstream tasks. A critical challenge is how to make use of existing large-scale pre-trained VLMs, which are trained on common objects, to perform the domain-specific transfer for accomplishing domain-related downstream tasks. A critical challenge is how to make use of existing large-scale pre-trained VLMs, which are trained on common objects, to perform the domain-specific transfer for accomplishing domain-related downstream tasks. In this paper, we propose a new framework that includes the Domain pre-trained Vision-Language Model (DVLM), bridging the gap between the General Vision-Language Model (GVLM) and domain-specific downstream tasks. Moreover, we present an image-text paired dataset in the field of remote sensing (RS), RS5M, which has 5 million RS images with English descriptions. The dataset is obtained from filtering publicly available image-text paired datasets and captioning label-only RS datasets with pre-trained VLM. These constitute the first large-scale RS image-text paired dataset. Additionally, we fine-tuned the CLIP model and tried several Parameter-Efficient Fine-Tuning methods on RS5M to implement the DVLM. Experimental results show that our proposed dataset is highly effective for various tasks, and our model GeoRSCLIP improves upon the baseline or previous state-of-the-art model by $3\%\sim20\%$ in Zero-shot Classification (ZSC), $3\%\sim6\%$ in Remote Sensing Cross-Modal Text-Image Retrieval (RSCTIR) and $4\%\sim5\%$ in Semantic Localization (SeLo) tasks. Dataset and models have been released in: \url{https://github.com/om-ai-lab/RS5M}.
Authors: Viktor Kewenig, Andrew Lampinen, Samuel A. Nastase, Christopher Edwards, Quitterie Lacome DEstalenx, Akilles Rechardt, Jeremy I Skipper, Gabriella Vigliocco
The potential of multimodal generative artificial intelligence (mAI) to replicate human grounded language understanding, including the pragmatic, context-rich aspects of communication, remains to be clarified. Humans are known to use salient multimodal features, such as visual cues, to facilitate the processing of upcoming words. Correspondingly, multimodal computational models can integrate visual and linguistic data using a visual attention mechanism to assign next-word probabilities. To test whether these processes align, we tasked both human participants (N = 200) as well as several state-of-the-art computational models with evaluating the predictability of forthcoming words after viewing short audio-only or audio-visual clips with speech. During the task, the model's attention weights were recorded and human attention was indexed via eye tracking. Results show that predictability estimates from humans aligned more closely with scores generated from multimodal models vs. their unimodal counterparts. Furthermore, including an attention mechanism doubled alignment with human judgments when visual and linguistic context facilitated predictions. In these cases, the model's attention patches and human eye tracking significantly overlapped. Our results indicate that improved modeling of naturalistic language processing in mAI does not merely depend on training diet but can be driven by multimodality in combination with attention-based architectures. Humans and computational models alike can leverage the predictive constraints of multimodal information by attending to relevant features in the input.
Authors: Andreas Östling, Holli Sargeant, Huiyuan Xie, Ludwig Bull, Alexander Terenin, Leif Jonsson, Måns Magnusson, Felix Steffek
We introduce the Cambridge Law Corpus (CLC), a dataset for legal AI research. It consists of over 250 000 court cases from the UK. Most cases are from the 21st century, but the corpus includes cases as old as the 16th century. This paper presents the first release of the corpus, containing the raw text and meta-data. Together with the corpus, we provide annotations on case outcomes for 638 cases, done by legal experts. Using our annotated data, we have trained and evaluated case outcome extraction with GPT-3, GPT-4 and RoBERTa models to provide benchmarks. We include an extensive legal and ethical discussion to address the potentially sensitive nature of this material. As a consequence, the corpus will only be released for research purposes under certain restrictions.
Authors: Haikang Deng, Colin Raffel
While large language models have proven effective in a huge range of downstream applications, they often generate text that is problematic or lacks a desired attribute. In this paper, we introduce Reward-Augmented Decoding (RAD), a text generation procedure that uses a small unidirectional reward model to encourage a language model to generate text that has certain properties. Specifically, RAD uses the reward model to score generations as they are produced and rescales sampling probabilities to favor high-reward tokens. By using a unidirectional reward model, RAD can cache activations from prior generation steps to decrease computational overhead. Through experiments on generating non-toxic and sentiment-controlled text, we demonstrate that RAD performs best among methods that change only the generation procedure and matches the performance of state-of-the-art methods that involve re-training the language model. We further validate that RAD is effective on very large language models while incurring a minimal computational overhead.
Authors: Kai Chen, Chunwei Wang, Kuo Yang, Jianhua Han, Lanqing Hong, Fei Mi, Hang Xu, Zhengying Liu, Wenyong Huang, Zhenguo Li, Dit-Yan Yeung, Lifeng Shang, Xin Jiang, Qun Liu
The rapid development of large language models (LLMs) has not only provided numerous opportunities but also presented significant challenges. This becomes particularly evident when LLMs inadvertently generate harmful or toxic content, either unintentionally or because of intentional inducement. Existing alignment methods usually direct LLMs toward the favorable outcomes by utilizing human-annotated, flawless instruction-response pairs. Conversely, this study proposes a novel alignment technique based on mistake analysis, which deliberately exposes LLMs to erroneous content to learn the reasons for mistakes and how to avoid them. In this case, mistakes are repurposed into valuable data for alignment, effectively helping to avoid the production of erroneous responses. Without external models or human annotations, our method leverages a model's intrinsic ability to discern undesirable mistakes and improves the safety of its generated responses. Experimental results reveal that our method outperforms existing alignment approaches in enhancing model safety while maintaining the overall utility.
Authors: Liang Wang, Nan Yang, Xiaolong Huang, Linjun Yang, Rangan Majumder, Furu Wei
Modern search engines are built on a stack of different components, including query understanding, retrieval, multi-stage ranking, and question answering, among others. These components are often optimized and deployed independently. In this paper, we introduce a novel conceptual framework called large search model, which redefines the conventional search stack by unifying search tasks with one large language model (LLM). All tasks are formulated as autoregressive text generation problems, allowing for the customization of tasks through the use of natural language prompts. This proposed framework capitalizes on the strong language understanding and reasoning capabilities of LLMs, offering the potential to enhance search result quality while simultaneously simplifying the existing cumbersome search stack. To substantiate the feasibility of this framework, we present a series of proof-of-concept experiments and discuss the potential challenges associated with implementing this approach within real-world search systems.
Authors: Michael Günther, Jackmin Ong, Isabelle Mohr, Alaeddine Abdessalem, Tanguy Abel, Mohammad Kalim Akram, Susana Guzman, Georgios Mastrapas, Saba Sturua, Bo Wang, Maximilian Werk, Nan Wang, Han Xiao
Text embedding models have emerged as powerful tools for transforming sentences into fixed-sized feature vectors that encapsulate semantic information. While these models are essential for tasks like information retrieval, semantic clustering, and text re-ranking, most existing open-source models, especially those built on architectures like BERT, struggle to represent lengthy documents and often resort to truncation. One common approach to mitigate this challenge involves splitting documents into smaller paragraphs for embedding. However, this strategy results in a much larger set of vectors, consequently leading to increased memory consumption and computationally intensive vector searches with elevated latency.
To address these challenges, we introduce Jina Embeddings 2, an open-source text embedding model capable of accommodating up to 8192 tokens. This model is designed to transcend the conventional 512-token limit and adeptly process long documents. Jina Embeddings 2 not only achieves state-of-the-art performance on a range of embedding-related tasks in the MTEB benchmark but also matches the performance of OpenAI's proprietary ada-002 model. Additionally, our experiments indicate that an extended context can enhance performance in tasks such as NarrativeQA.
Authors: Sai Akash Bathini, Dagli Cihan
The application of Machine learning to finance has become a familiar approach, even more so in stock market forecasting. The stock market is highly volatile, and huge amounts of data are generated every minute globally. The extraction of effective intelligence from this data is of critical importance. However, a collaboration of numerical stock data with qualitative text data can be a challenging task. In this work, we accomplish this by providing an unprecedented, publicly available dataset with technical and fundamental data and sentiment that we gathered from news archives, TV news captions, radio transcripts, tweets, daily financial newspapers, etc. The text data entries used for sentiment extraction total more than 1.4 Million. The dataset consists of daily entries from January 2018 to December 2022 for eight companies representing diverse industrial sectors and the Dow Jones Industrial Average (DJIA) as a whole. Holistic Fundamental and Technical data is provided training ready for Model learning and deployment. Most importantly, the data generated could be used for incremental online learning with real-time data points retrieved daily since no stagnant data was utilized. All the data was retired from APIs or self-designed robust information retrieval technologies with extremely low latency and zero monetary cost. These adaptable technologies facilitate data extraction for any stock. Moreover, the utilization of Spearman's rank correlation over real-time data, linking stock returns with sentiment analysis has produced noteworthy results for the DJIA and the eight other stocks, achieving accuracy levels surpassing 60%. The dataset is made available at https://github.com/batking24/Huge-Stock-Dataset.
Authors: Hanlin Zhang, Yi-Fan Zhang, Yaodong Yu, Dhruv Madeka, Dean Foster, Eric Xing, Himabindu Lakkaraju, Sham Kakade
Accurate uncertainty quantification is crucial for the safe deployment of language models (LMs), and prior research has demonstrated improvements in the calibration of modern LMs. Our study focuses on in-context learning (ICL), a prevalent method for adapting static LMs through tailored prompts, and examines the balance between performance and calibration across a broad spectrum of natural language understanding and reasoning tasks. Through comprehensive experiments, we observe that, with an increasing number of ICL examples, models initially exhibit increased miscalibration before achieving better calibration and miscalibration tends to arise in low-shot settings. Moreover, we find that methods aimed at improving usability, such as fine-tuning and chain-of-thought (CoT) prompting, can lead to miscalibration and unreliable natural language explanations, suggesting that new methods may be required for scenarios where models are expected to be reliable.
Authors: Chenchen Gu, Xiang Lisa Li, Percy Liang, Tatsunori Hashimoto
Watermarking of language model outputs enables statistical detection of model-generated text, which has many applications in the responsible deployment of language models. Existing watermarking strategies operate by altering the decoder of an existing language model, and the ability for a language model to directly learn to generate the watermark would have significant implications for the real-world deployment of watermarks. First, learned watermarks could be used to build open models that naturally generate watermarked text, allowing for open models to benefit from watermarking. Second, if watermarking is used to determine the provenance of generated text, an adversary can hurt the reputation of a victim model by spoofing its watermark and generating damaging watermarked text. To investigate the learnability of watermarks, we propose watermark distillation, which trains a student model to behave like a teacher model that uses decoding-based watermarking. We test our approach on three distinct decoding-based watermarking strategies and various hyperparameter settings, finding that models can learn to generate watermarked text with high detectability. We also find limitations to learnability, including the loss of watermarking capabilities under fine-tuning on normal text and high sample complexity when learning low-distortion watermarks.
Authors: Yu Zhang, Rongjie Huang, Ruiqi Li, JinZheng He, Yan Xia, Feiyang Chen, Xinyu Duan, Baoxing Huai, Zhou Zhao
Style transfer for out-of-domain (OOD) singing voice synthesis (SVS) focuses on generating high-quality singing voices with unseen styles (such as timbre, emotion, pronunciation, and articulation skills) derived from reference singing voice samples. However, the endeavor to model the intricate nuances of singing voice styles is an arduous task, as singing voices possess a remarkable degree of expressiveness. Moreover, existing SVS methods encounter a decline in the quality of synthesized singing voices in OOD scenarios, as they rest upon the assumption that the target vocal attributes are discernible during the training phase. To overcome these challenges, we propose StyleSinger, the first singing voice synthesis model for zero-shot style transfer of out-of-domain reference singing voice samples. StyleSinger incorporates two critical approaches for enhanced effectiveness: 1) the Residual Style Adaptor (RSA) which employs a residual quantization module to capture diverse style characteristics in singing voices, and 2) the Uncertainty Modeling Layer Normalization (UMLN) to perturb the style attributes within the content representation during the training phase and thus improve the model generalization. Our extensive evaluations in zero-shot style transfer undeniably establish that StyleSinger outperforms baseline models in both audio quality and similarity to the reference singing voice samples. Access to singing voice samples can be found at https://stylesinger.github.io/.
Authors: Michael Dalvean
In English place name analysis, meanings are often derived from the resemblance of roots in place names to topographical features, proper names and/or habitation terms in one of the languages that have had an influence on English place names. The problem here is that it is sometimes difficult to determine the base language to use to interpret the roots. The purpose of this paper is to stochastically determine the resemblance between 18799 English place names and 84687 place names from Ireland, Scotland, Wales, Denmark, Norway, Sweden, France, Germany, the Netherlands and Ancient Rome. Each English place name is ranked according to the extent to which it resembles place names from the other countries, and this provides a basis for determining the likely language to use to interpret the place name. A number of observations can be made using the ranking provided. In particular, it is found that `Harlington' is the most archetypically English place name in the English sample, and `Anna' is the least. Furthermore, it is found that the place names in the non-English datasets are most similar to Norwegian place names and least similar to Welsh place names.
Authors: Konrad Staniszewski, Szymon Tworkowski, Sebastian Jaszczur, Henryk Michalewski, Łukasz Kuciński, Piotr Miłoś
Recent advances in long-context Large Language Models (LCLMs) have generated significant interest, especially in applications such as querying scientific research papers. However, their potential is often limited by inadequate context utilization. We identify the absence of long-range semantic dependencies in typical training data as a primary hindrance. To address this, we delve into the benefits of frequently incorporating related documents into training inputs. Using the inherent directory structure of code data as a source of training examples, we demonstrate improvements in perplexity, even for tasks unrelated to coding. Building on these findings, but with a broader focus, we introduce Structured Packing for Long Context (SPLiCe). SPLiCe is an innovative method for creating training examples by using a retrieval method to collate the most mutually relevant documents into a single training context. Our results indicate that \method{} enhances model performance and can be used to train large models to utilize long contexts better. We validate our results by training a large $3$B model, showing both perplexity improvements and better long-context performance on downstream tasks.