Authors: Zofia Malisz, Jan Foremski, Ma{\l}gorzata Kul
Abstract: We present a speech database and a phoneme-level language model of Polish. The database and model are designed for the analysis of prosodic and discourse factors and their impact on acoustic parameters in interaction with predictability effects. The database is also the first large, publicly available Polish speech corpus of excellent acoustic quality that can be used for phonetic analysis and training of multi-speaker speech technology systems. The speech in the database is processed in a pipeline that achieves a 90% degree of automation. It incorporates state-of-the-art, freely available tools enabling database expansion or adaptation to additional languages.
Authors: Neha Gupta, Harikrishna Narasimhan, Wittawat Jitkrittum, Ankit Singh Rawat, Aditya Krishna Menon, Sanjiv Kumar
Abstract: Recent advances in language models (LMs) have led to significant improvements in quality on complex NLP tasks, but at the expense of increased inference costs. Cascading offers a simple strategy to achieve more favorable cost-quality tradeoffs: here, a small model is invoked for most "easy" instances, while a few "hard" instances are deferred to the large model. While the principles underpinning cascading are well-studied for classification tasks - with deferral based on predicted class uncertainty favored theoretically and practically - a similar understanding is lacking for generative LM tasks. In this work, we initiate a systematic study of deferral rules for LM cascades. We begin by examining the natural extension of predicted class uncertainty to generative LM tasks, namely, the predicted sequence uncertainty. We show that this measure suffers from the length bias problem, either over- or under-emphasizing outputs based on their lengths. This is because LMs produce a sequence of uncertainty values, one for each output token; and moreover, the number of output tokens is variable across examples. To mitigate this issue, we propose to exploit the richer token-level uncertainty information implicit in generative LMs. We argue that naive predicted sequence uncertainty corresponds to a simple aggregation of these uncertainties. By contrast, we show that incorporating token-level uncertainty through learned post-hoc deferral rules can significantly outperform such simple aggregation strategies, via experiments on a range of natural language benchmarks with FLAN-T5 models. We further show that incorporating embeddings from the smaller model and intermediate layers of the larger model can give an additional boost in the overall cost-quality tradeoff.
Authors: Md Mahadi Hasan Nahid, Davood Rafiei
Abstract: Table reasoning is a challenging task that requires understanding both natural language questions and structured tabular data. Large language models (LLMs) have shown impressive capabilities in natural language understanding and generation, but they often struggle with large tables due to their limited input length. In this paper, we propose TabSQLify, a novel method that leverages text-to-SQL generation to decompose tables into smaller and relevant sub-tables, containing only essential information for answering questions or verifying statements, before performing the reasoning task. In our comprehensive evaluation on four challenging datasets, our approach demonstrates comparable or superior performance compared to prevailing methods reliant on full tables as input. Moreover, our method can reduce the input context length significantly, making it more scalable and efficient for large-scale table reasoning applications. Our method performs remarkably well on the WikiTQ benchmark, achieving an accuracy of 64.7%. Additionally, on the TabFact benchmark, it achieves a high accuracy of 79.5%. These results surpass other LLM-based baseline models on gpt-3.5-turbo (chatgpt). TabSQLify can reduce the table size significantly alleviating the computational load on LLMs when handling large tables without compromising performance.
Authors: Mauricio Gruppi, Soham Dan, Keerthiram Murugesan, Subhajit Chaudhury
Abstract: Text-based reinforcement learning involves an agent interacting with a fictional environment using observed text and admissible actions in natural language to complete a task. Previous works have shown that agents can succeed in text-based interactive environments even in the complete absence of semantic understanding or other linguistic capabilities. The success of these agents in playing such games suggests that semantic understanding may not be important for the task. This raises an important question about the benefits of LMs in guiding the agents through the game states. In this work, we show that rich semantic understanding leads to efficient training of text-based RL agents. Moreover, we describe the occurrence of semantic degeneration as a consequence of inappropriate fine-tuning of language models in text-based reinforcement learning (TBRL). Specifically, we describe the shift in the semantic representation of words in the LM, as well as how it affects the performance of the agent in tasks that are semantically similar to the training games. We believe these results may help develop better strategies to fine-tune agents in text-based RL scenarios.
Authors: Zelin Wu, Gan Song, Christopher Li, Pat Rondon, Zhong Meng, Xavier Velez, Weiran Wang, Diamantino Caseiro, Golan Pundak, Tsendsuren Munkhdalai, Angad Chandorkar, Rohit Prabhavalkar
Abstract: Contextual biasing enables speech recognizers to transcribe important phrases in the speaker's context, such as contact names, even if they are rare in, or absent from, the training data. Attention-based biasing is a leading approach which allows for full end-to-end cotraining of the recognizer and biasing system and requires no separate inference-time components. Such biasers typically consist of a context encoder; followed by a context filter which narrows down the context to apply, improving per-step inference time; and, finally, context application via cross attention. Though much work has gone into optimizing per-frame performance, the context encoder is at least as important: recognition cannot begin before context encoding ends. Here, we show the lightweight phrase selection pass can be moved before context encoding, resulting in a speedup of up to 16.1 times and enabling biasing to scale to 20K phrases with a maximum pre-decoding delay under 33ms. With the addition of phrase- and wordpiece-level cross-entropy losses, our technique also achieves up to a 37.5% relative WER reduction over the baseline without the losses and lightweight phrase selection pass.
Authors: Kevin Wu, Eric Wu, James Zou
Abstract: Retrieval augmented generation (RAG) is often used to fix hallucinations and provide up-to-date knowledge for large language models (LLMs). However, in cases when the LLM alone incorrectly answers a question, does providing the correct retrieved content always fix the error? Conversely, in cases where the retrieved content is incorrect, does the LLM know to ignore the wrong information, or does it recapitulate the error? To answer these questions, we systematically analyze the tug-of-war between a LLM's internal knowledge (i.e. its prior) and the retrieved information in settings when they disagree. We test GPT-4 and other LLMs on question-answering abilities across datasets with and without reference documents. As expected, providing the correct retrieved information fixes most model mistakes (94% accuracy). However, when the reference document is perturbed with increasing levels of wrong values, the LLM is more likely to recite the incorrect, modified information when its internal prior is weaker but is more resistant when its prior is stronger. Similarly, we also find that the more the modified information deviates from the model's prior, the less likely the model is to prefer it. These results highlight an underlying tension between a model's prior knowledge and the information presented in reference documents.
Authors: Huihan Li, Liwei Jiang, Nouha Dziri, Xiang Ren, Yejin Choi
Abstract: As the utilization of large language models (LLMs) has proliferated worldwide, it is crucial for them to have adequate knowledge and fair representation for diverse global cultures. In this work, we uncover culture perceptions of three SOTA models on 110 countries and regions on 8 culture-related topics through culture-conditioned generations, and extract symbols from these generations that are associated to each culture by the LLM. We discover that culture-conditioned generation consist of linguistic "markers" that distinguish marginalized cultures apart from default cultures. We also discover that LLMs have an uneven degree of diversity in the culture symbols, and that cultures from different geographic regions have different presence in LLMs' culture-agnostic generation. Our findings promote further research in studying the knowledge and fairness of global culture perception in LLMs. Code and Data can be found in: https://github.com/huihanlhh/Culture-Gen/
Authors: Jiaxuan Wu, Zhengxian Wu, Yiming Xue, Juan Wen, Wanli Peng
Abstract: Recent advances in large language models (LLMs) have blurred the boundary of high-quality text generation between humans and machines, which is favorable for generative text steganography. While, current advanced steganographic mapping is not suitable for LLMs since most users are restricted to accessing only the black-box API or user interface of the LLMs, thereby lacking access to the training vocabulary and its sampling probabilities. In this paper, we explore a black-box generative text steganographic method based on the user interfaces of large language models, which is called LLM-Stega. The main goal of LLM-Stega is that the secure covert communication between Alice (sender) and Bob (receiver) is conducted by using the user interfaces of LLMs. Specifically, We first construct a keyword set and design a new encrypted steganographic mapping to embed secret messages. Furthermore, to guarantee accurate extraction of secret messages and rich semantics of generated stego texts, an optimization mechanism based on reject sampling is proposed. Comprehensive experiments demonstrate that the proposed LLM-Stega outperforms current state-of-the-art methods.
Authors: Tunazzina Islam, Dan Goldwasser
Abstract: The widespread use of social media has led to a surge in popularity for automated methods of analyzing public opinion. Supervised methods are adept at text categorization, yet the dynamic nature of social media discussions poses a continual challenge for these techniques due to the constant shifting of the focus. On the other hand, traditional unsupervised methods for extracting themes from public discourse, such as topic modeling, often reveal overarching patterns that might not capture specific nuances. Consequently, a significant portion of research into social media discourse still depends on labor-intensive manual coding techniques and a human-in-the-loop approach, which are both time-consuming and costly. In this work, we study the problem of discovering arguments associated with a specific theme. We propose a generic LLMs-in-the-Loop strategy that leverages the advanced capabilities of Large Language Models (LLMs) to extract latent arguments from social media messaging. To demonstrate our approach, we apply our framework to contentious topics. We use two publicly available datasets: (1) the climate campaigns dataset of 14k Facebook ads with 25 themes and (2) the COVID-19 vaccine campaigns dataset of 9k Facebook ads with 14 themes. Furthermore, we analyze demographic targeting and the adaptation of messaging based on real-world events.
Authors: Yue Zhou, Barbara Di Eugenio, Brian Ziebart, Lisa Sharp, Bing Liu, Nikolaos Agadakos
Abstract: Health coaching helps patients achieve personalized and lifestyle-related goals, effectively managing chronic conditions and alleviating mental health issues. It is particularly beneficial, however cost-prohibitive, for low-socioeconomic status populations due to its highly personalized and labor-intensive nature. In this paper, we propose a neuro-symbolic goal summarizer to support health coaches in keeping track of the goals and a text-units-text dialogue generation model that converses with patients and helps them create and accomplish specific goals for physical activities. Our models outperform previous state-of-the-art while eliminating the need for predefined schema and corresponding annotation. We also propose a new health coaching dataset extending previous work and a metric to measure the unconventionality of the patient's response based on data difficulty, facilitating potential coach alerts during deployment.
Authors: Changmao Li, Jeffrey Flanigan
Abstract: Predicting the future is of great interest across many aspects of human activity. Businesses are interested in future trends, traders are interested in future stock prices, and companies are highly interested in future technological breakthroughs. While there are many automated systems for predicting future numerical data, such as weather, stock prices, and demand for products, there is relatively little work in automatically predicting textual data. Humans are interested in textual data predictions because it is a natural format for our consumption, and experts routinely make predictions in a textual format (Christensen et al., 2004; Tetlock & Gardner, 2015; Frick, 2015). However, there has been relatively little formalization of this general problem in the machine learning or natural language processing communities. To address this gap, we introduce the task of future language modeling: probabilistic modeling of texts in the future based on a temporal history of texts. To our knowledge, our work is the first work to formalize the task of predicting the future in this way. We show that it is indeed possible to build future language models that improve upon strong non-temporal language model baselines, opening the door to working on this important, and widely applicable problem.
Authors: Hengyuan Zhang, Yanru Wu, Dawei Li, Zacc Yang, Rui Zhao, Yong Jiang, Fei Tan
Abstract: Aligned Large Language Models (LLMs) showcase remarkable versatility, capable of handling diverse real-world tasks. Meanwhile, aligned LLMs are also expected to exhibit speciality, excelling in specific applications. However, fine-tuning with extra data, a common practice to gain speciality, often leads to catastrophic forgetting (CF) of previously acquired versatility, hindering the model's performance across diverse tasks. In response to this challenge, we propose CoFiTune, a coarse to fine framework in an attempt to strike the balance between speciality and versatility. At the coarse-grained level, an empirical tree-search algorithm is utilized to pinpoint and update specific modules that are crucial for speciality, while keeping other parameters frozen; at the fine-grained level, a soft-masking mechanism regulates the update to the LLMs, mitigating the CF issue without harming speciality. In an overall evaluation of both speciality and versatility, CoFiTune consistently outperforms baseline methods across diverse tasks and model scales. Compared to the full-parameter SFT, CoFiTune leads to about 14% versatility improvement and marginal speciality loss on a 13B model. Lastly, based on further analysis, we provide a speculative insight into the information forwarding process in LLMs, which helps explain the effectiveness of the proposed method. The code is available at https://github.com/rattlesnakey/CoFiTune.
Authors: Haixia Han, Tingyun Li, Shisong Chen, Jie Shi, Chengyu Du, Yanghua Xiao, Jiaqing Liang, Xin Lin
Abstract: Large Language Models (LLMs) have exhibited remarkable performance across various downstream tasks, but they may generate inaccurate or false information with a confident tone. One of the possible solutions is to empower the LLM confidence expression capability, in which the confidence expressed can be well-aligned with the true probability of the generated answer being correct. However, leveraging the intrinsic ability of LLMs or the signals from the output logits of answers proves challenging in accurately capturing the response uncertainty in LLMs. Therefore, drawing inspiration from cognitive diagnostics, we propose a method of Learning from Past experience (LePe) to enhance the capability for confidence expression. Specifically, we first identify three key problems: (1) How to capture the inherent confidence of the LLM? (2) How to teach the LLM to express confidence? (3) How to evaluate the confidence expression of the LLM? Then we devise three stages in LePe to deal with these problems. Besides, to accurately capture the confidence of an LLM when constructing the training data, we design a complete pipeline including question preparation and answer sampling. We also conduct experiments using the Llama family of LLMs to verify the effectiveness of our proposed method on four datasets.
Authors: Hyeonbin Hwang, Doyoung Kim, Seungone Kim, Seonghyeon Ye, Minjoon Seo
Abstract: Training on large amounts of rationales (i.e., CoT Fine-tuning) is effective at improving the reasoning capabilities of large language models (LLMs). However, acquiring human-authored rationales or augmenting rationales from proprietary models is costly and not scalable. In this paper, we study the problem of whether LLMs could self-improve their reasoning capabilities. To this end, we propose Self-Explore, where the LLM is tasked to explore the first wrong step (i.e., the first pit) within the rationale and use such signals as fine-grained rewards for further improvement. On the GSM8K and MATH test set, Self-Explore achieves 11.57% and 2.89% improvement on average across three LLMs compared to supervised fine-tuning (SFT). Our code is available at https://github.com/hbin0701/Self-Explore.
Authors: Yuqi Wang, Boran Jiang, Yi Luo, Dawei He, Peng Cheng, Liangcai Gao
Abstract: Large language models (LLMs), such as GPT3.5, GPT4 and LLAMA2 perform surprisingly well and outperform human experts on many tasks. However, in many domain-specific evaluations, these LLMs often suffer from hallucination problems due to insufficient training of relevant corpus. Furthermore, fine-tuning large models may face problems such as the LLMs are not open source or the construction of high-quality domain instruction is difficult. Therefore, structured knowledge databases such as knowledge graph can better provide domain back- ground knowledge for LLMs and make full use of the reasoning and analysis capabilities of LLMs. In some previous works, LLM was called multiple times to determine whether the current triplet was suitable for inclusion in the subgraph when retrieving subgraphs through a question. Especially for the question that require a multi-hop reasoning path, frequent calls to LLM will consume a lot of computing power. Moreover, when choosing the reasoning path, LLM will be called once for each step, and if one of the steps is selected incorrectly, it will lead to the accumulation of errors in the following steps. In this paper, we integrated and optimized a pipeline for selecting reasoning paths from KG based on LLM, which can reduce the dependency on LLM. In addition, we propose a simple and effective subgraph retrieval method based on chain of thought (CoT) and page rank which can returns the paths most likely to contain the answer. We conduct experiments on three datasets: GenMedGPT-5k [14], WebQuestions [2], and CMCQA [21]. Finally, RoK can demonstrate that using fewer LLM calls can achieve the same results as previous SOTAs models.
Authors: Zheng Yuan, \v{S}tefan Be\v{n}u\v{s}, Alessandro D'Ausilio
Abstract: This study explores F0 entrainment in second language (L2) English speech imitation during an Alternating Reading Task (ART). Participants with Italian, French, and Slovak native languages imitated English utterances, and their F0 entrainment was quantified using the Dynamic Time Warping (DTW) distance between the parameterized F0 contours of the imitated utterances and those of the model utterances. Results indicate a nuanced relationship between L2 English proficiency and entrainment: speakers with higher proficiency generally exhibit less entrainment in pitch variation and declination. However, within dyads, the more proficient speakers demonstrate a greater ability to mimic pitch range, leading to increased entrainment. This suggests that proficiency influences entrainment differently at individual and dyadic levels, highlighting the complex interplay between language skill and prosodic adaptation.
Authors: Yu Li, Zhihua Wei, Han Jiang, Chuanyang Gong
Abstract: Despite the remarkable achievements of language models (LMs) across a broad spectrum of tasks, their propensity for generating toxic outputs remains a prevalent concern. Current solutions involving fine-tuning or auxiliary models usually require extensive memory and computational resources, rendering them less practical for deployment in large language models (LLMs). In this paper, we propose DeStein, a novel method that detoxififies LMs by altering their internal representations in the activation space with lower resource and time cost. Specifically, we leverage self-induced steering pairs to identify detoxification vectors through arithmetic operations in the activation space. During inference, detoxification is achieved by blending the detoxification vectors with the original representations. Empirical results demonstrate that our method significantly outperforms previous state-of-the-art approaches on popular detoxification metrics, while also maintaining satisfactory generation quality and diversity. Furthermore, we extend our method to multiple LLMs, demonstrating its practicality and scalability. Warning: some example model outputs contain highly offensive or disturbing text.
Authors: Donya Rooein, Dirk Hovy
Abstract: Open conversations are one of the most engaging forms of teaching. However, creating those conversations in educational software is a complex endeavor, especially if we want to address the needs of different audiences. While language models hold great promise for educational applications, there are substantial challenges in training them to engage in meaningful and effective conversational teaching, especially when considering the diverse needs of various audiences. No official data sets exist for this task to facilitate the training of language models for conversational teaching, considering the diverse needs of various audiences. This paper presents a novel source for facilitating conversational teaching of scientific concepts at various difficulty levels (from preschooler to expert), namely dialogues taken from video transcripts. We analyse this data source in various ways to show that it offers a diverse array of examples that can be used to generate contextually appropriate and natural responses to scientific topics for specific target audiences. It is a freely available valuable resource for training and evaluating conversation models, encompassing organically occurring dialogues. While the raw data is available online, we provide additional metadata for conversational analysis of dialogues at each level in all available videos.
Authors: Chenggian Ma, Xiangyu Zhao, Chunhui Zhang, Yanzhao Qin, Wentao Zhang
Abstract: With the development of Large Language Models (LLM), numerous prompts have been proposed, each with a rich set of features and their own merits. This paper summarizes the prompt words for large language models (LLMs), categorizing them into stimulating and framework types, and proposes an Auto-Prompt Graphical Paradigm(APGP) that combines both stimulating and framework prompts to enhance the problem-solving capabilities of LLMs across multiple domains, then exemplifies it with a framework that adheres to this paradigm. The framework involves automated prompt generation and consideration of emotion-stimulus factors, guiding LLMs in problem abstraction, diversified solutions generation, comprehensive optimization, and self-verification after providing answers, ensuring solution accuracy. Compared to traditional stimuli and framework prompts, this framework integrates the advantages of both by adopting automated approaches inspired by APE work, overcoming the limitations of manually designed prompts. Test results on the ruozhiba and BBH datasets demonstrate that this framework can effectively improve the efficiency and accuracy of LLMs in problem-solving, paving the way for new applications of LLMs.
Authors: Yinan Chen
Abstract: The field of natural language processing (NLP) has made significant progress with the rapid development of deep learning technologies. One of the research directions in text sentiment analysis is sentiment analysis of medical texts, which holds great potential for application in clinical diagnosis. However, the medical field currently lacks sufficient text datasets, and the effectiveness of sentiment analysis is greatly impacted by different model design approaches, which presents challenges. Therefore, this paper focuses on the medical domain, using bidirectional encoder representations from transformers (BERT) as the basic pre-trained model and experimenting with modules such as convolutional neural network (CNN), fully connected network (FCN), and graph convolutional networks (GCN) at the output layer. Experiments and analyses were conducted on the METS-CoV dataset to explore the training performance after integrating different deep learning networks. The results indicate that CNN models outperform other networks when trained on smaller medical text datasets in combination with pre-trained models like BERT. This study highlights the significance of model selection in achieving effective sentiment analysis in the medical domain and provides a reference for future research to develop more efficient model architectures.
Authors: Yixin Wan, Kai-Wei Chang
Abstract: Social biases can manifest in language agency. For instance, White individuals and men are often described as "agentic" and achievement-oriented, whereas Black individuals and women are frequently described as "communal" and as assisting roles. This study establishes agency as an important aspect of studying social biases in both human-written and Large Language Model (LLM)-generated texts. To accurately measure "language agency" at sentence level, we propose a Language Agency Classification dataset to train reliable agency classifiers. We then use an agency classifier to reveal notable language agency biases in 6 datasets of human- or LLM-written texts, including biographies, professor reviews, and reference letters. While most prior NLP research on agency biases focused on single dimensions, we comprehensively explore language agency biases in gender, race, and intersectional identities. We observe that (1) language agency biases in human-written texts align with real-world social observations; (2) LLM-generated texts demonstrate remarkably higher levels of language agency bias than human-written texts; and (3) critical biases in language agency target people of minority groups--for instance, languages used to describe Black females exhibit the lowest level of agency across datasets. Our findings reveal intricate social biases in human- and LLM-written texts through the lens of language agency, warning against using LLM generations in social contexts without scrutiny.
Authors: Moshe Berchansky, Daniel Fleischer, Moshe Wasserblat, Peter Izsak
Abstract: State-of-the-art performance in QA tasks is currently achieved by systems employing Large Language Models (LLMs), however these models tend to hallucinate information in their responses. One approach focuses on enhancing the generation process by incorporating attribution from the given input to the output. However, the challenge of identifying appropriate attributions and verifying their accuracy against a source is a complex task that requires significant improvements in assessing such systems. We introduce an attribution-oriented Chain-of-Thought reasoning method to enhance the accuracy of attributions. This approach focuses the reasoning process on generating an attribution-centric output. Evaluations on two context-enhanced question-answering datasets using GPT-4 demonstrate improved accuracy and correctness of attributions. In addition, the combination of our method with finetuning enhances the response and attribution accuracy of two smaller LLMs, showing their potential to outperform GPT-4 in some cases.
Authors: Xiao Wang, Tianze Chen, Xianjun Yang, Qi Zhang, Xun Zhao, Dahua Lin
Abstract: The open-sourcing of large language models (LLMs) accelerates application development, innovation, and scientific progress. This includes both base models, which are pre-trained on extensive datasets without alignment, and aligned models, deliberately designed to align with ethical standards and human values. Contrary to the prevalent assumption that the inherent instruction-following limitations of base LLMs serve as a safeguard against misuse, our investigation exposes a critical oversight in this belief. By deploying carefully designed demonstrations, our research demonstrates that base LLMs could effectively interpret and execute malicious instructions. To systematically assess these risks, we introduce a novel set of risk evaluation metrics. Empirical results reveal that the outputs from base LLMs can exhibit risk levels on par with those of models fine-tuned for malicious purposes. This vulnerability, requiring neither specialized knowledge nor training, can be manipulated by almost anyone, highlighting the substantial risk and the critical need for immediate attention to the base LLMs' security protocols.
Authors: Masanori Hirano, Kentaro Imajo
Abstract: Large language models (LLMs) are now widely used in various fields, including finance. However, Japanese financial-specific LLMs have not been proposed yet. Hence, this study aims to construct a Japanese financial-specific LLM through continual pre-training. Before tuning, we constructed Japanese financial-focused datasets for continual pre-training. As a base model, we employed a Japanese LLM that achieved state-of-the-art performance on Japanese financial benchmarks among the 10-billion-class parameter models. After continual pre-training using the datasets and the base model, the tuned model performed better than the original model on the Japanese financial benchmarks. Moreover, the outputs comparison results reveal that the tuned model's outputs tend to be better than the original model's outputs in terms of the quality and length of the answers. These findings indicate that domain-specific continual pre-training is also effective for LLMs. The tuned model is publicly available on Hugging Face.
Authors: Haozheng Fan, Hao Zhou, Guangtai Huang, Parameswaran Raman, Xinwei Fu, Gaurav Gupta, Dhananjay Ram, Yida Wang, Jun Huan
Abstract: Getting large language models (LLMs) to perform well on the downstream tasks requires pre-training over trillions of tokens. This typically demands a large number of powerful computational devices in addition to a stable distributed training framework to accelerate the training. The growing number of applications leveraging AI/ML had led to a scarcity of the expensive conventional accelerators (such as GPUs), which begs the need for the alternative specialized-accelerators that are scalable and cost-efficient. AWS Trainium is the second-generation machine learning accelerator that has been purposely built for training large deep learning models. Its corresponding instance, Amazon EC2 trn1, is an alternative to GPU instances for LLM training. However, training LLMs with billions of parameters on trn1 is challenging due to its relatively nascent software ecosystem. In this paper, we showcase HLAT: a 7 billion parameter decoder-only LLM pre-trained using trn1 instances over 1.8 trillion tokens. The performance of HLAT is benchmarked against popular open source baseline models including LLaMA and OpenLLaMA, which have been trained on NVIDIA GPUs and Google TPUs, respectively. On various evaluation tasks, we show that HLAT achieves model quality on par with the baselines. We also share the best practice of using the Neuron Distributed Training Library (NDTL), a customized distributed training library for AWS Trainium to achieve efficient training. Our work demonstrates that AWS Trainium powered by the NDTL is able to successfully pre-train state-of-the-art LLM models with high performance and cost-effectiveness.
Authors: Pengyu Cheng, Tianhao Hu, Han Xu, Zhisong Zhang, Yong Dai, Lei Han, Nan Du
Abstract: We explore the self-play training procedure of large language models (LLMs) in a two-player adversarial language game called Adversarial Taboo. In this game, an attacker and a defender communicate with respect to a target word only visible to the attacker. The attacker aims to induce the defender to utter the target word unconsciously, while the defender tries to infer the target word from the attacker's utterances. To win the game, both players should have sufficient knowledge about the target word and high-level reasoning ability to infer and express in this information-reserved conversation. Hence, we are curious about whether LLMs' reasoning ability can be further enhanced by Self-Play in this Adversarial language Game (SPAG). With this goal, we let LLMs act as the attacker and play with a copy of itself as the defender on an extensive range of target words. Through reinforcement learning on the game outcomes, we observe that the LLMs' performance uniformly improves on a broad range of reasoning benchmarks. Furthermore, iteratively adopting this self-play process can continuously promote LLM's reasoning ability. The code is at https://github.com/Linear95/SPAG.
Authors: Quan Van Nguyen, Dan Quang Tran, Huy Quang Pham, Thang Kien-Bao Nguyen, Nghia Hieu Nguyen, Kiet Van Nguyen, Ngan Luu-Thuy Nguyen
Abstract: Visual Question Answering (VQA) is a complicated task that requires the capability of simultaneously processing natural language and images. Initially, this task was researched, focusing on methods to help machines understand objects and scene contexts in images. However, some text appearing in the image that carries explicit information about the full content of the image is not mentioned. Along with the continuous development of the AI era, there have been many studies on the reading comprehension ability of VQA models in the world. As a developing country, conditions are still limited, and this task is still open in Vietnam. Therefore, we introduce the first large-scale dataset in Vietnamese specializing in the ability to understand text appearing in images, we call it ViTextVQA (\textbf{Vi}etnamese \textbf{Text}-based \textbf{V}isual \textbf{Q}uestion \textbf{A}nswering dataset) which contains \textbf{over 16,000} images and \textbf{over 50,000} questions with answers. Through meticulous experiments with various state-of-the-art models, we uncover the significance of the order in which tokens in OCR text are processed and selected to formulate answers. This finding helped us significantly improve the performance of the baseline models on the ViTextVQA dataset. Our dataset is available at this \href{https://github.com/minhquan6203/ViTextVQA-Dataset}{link} for research purposes.
Authors: Pengcheng Lu, Massimo Poesio
Abstract: Resolving coreference and bridging relations in chemical patents is important for better understanding the precise chemical process, where chemical domain knowledge is very critical. We proposed an approach incorporating external knowledge into a multi-task learning model for both coreference and bridging resolution in the chemical domain. The results show that integrating external knowledge can benefit both chemical coreference and bridging resolution.
Authors: Vatsal Raina, Mark Gales
Abstract: Multiple-choice (MC) tests are an efficient method to assess English learners. It is useful for test creators to rank candidate MC questions by difficulty during exam curation. Typically, the difficulty is determined by having human test takers trial the questions in a pretesting stage. However, this is expensive and not scalable. Therefore, we explore automated approaches to rank MC questions by difficulty. However, there is limited data for explicit training of a system for difficulty scores. Hence, we compare task transfer and zero-shot approaches: task transfer adapts level classification and reading comprehension systems for difficulty ranking while zero-shot prompting of instruction finetuned language models contrasts absolute assessment against comparative. It is found that level classification transfers better than reading comprehension. Additionally, zero-shot comparative assessment is more effective at difficulty ranking than the absolute assessment and even the task transfer approaches at question difficulty ranking with a Spearman's correlation of 40.4%. Combining the systems is observed to further boost the correlation.
Authors: Yekun Chai, Qingyi Liu, Jingwu Xiao, Shuohuan Wang, Yu Sun, Hua Wu
Abstract: Harnessing visual texts represents a burgeoning frontier in the evolution of language modeling. In this paper, we introduce a novel pre-training framework for a suite of pixel-based autoregressive language models, pre-training on a corpus of over 400 million documents rendered as RGB images. Our approach is characterized by a dual-modality training regimen, engaging both visual data through next patch prediction with a regression head and textual data via next token prediction with a classification head. This study is particularly focused on investigating the synergistic interplay between visual and textual modalities of language. Our comprehensive evaluation across a diverse array of benchmarks reveals that the confluence of visual and textual data substantially augments the efficacy of pixel-based language models. Notably, our findings show that a unidirectional pixel-based model, devoid of textual data during training, can match the performance levels of advanced bidirectional pixel-based models on various language understanding benchmarks. This work highlights the considerable untapped potential of integrating visual and textual information for language modeling purposes. We will release our code, data, and checkpoints to inspire further research advancement.
Authors: Shusheng Xu, Wei Fu, Jiaxuan Gao, Wenjie Ye, Weilin Liu, Zhiyu Mei, Guangju Wang, Chao Yu, Yi Wu
Abstract: Reinforcement Learning from Human Feedback (RLHF) is currently the most widely used method to align large language models (LLMs) with human preferences. Existing RLHF methods can be roughly categorized as either reward-based or reward-free. Novel applications such as ChatGPT and Claude leverage reward-based methods that first learn a reward model and apply actor-critic algorithms, such as Proximal Policy Optimization (PPO). However, in academic benchmarks, state-of-the-art results are often achieved via reward-free methods, such as Direct Preference Optimization (DPO). Is DPO truly superior to PPO? Why does PPO perform poorly on these benchmarks? In this paper, we first conduct both theoretical and empirical studies on the algorithmic properties of DPO and show that DPO may have fundamental limitations. Moreover, we also comprehensively examine PPO and reveal the key factors for the best performances of PPO in fine-tuning LLMs. Finally, we benchmark DPO and PPO across various a collection of RLHF testbeds, ranging from dialogue to code generation. Experiment results demonstrate that PPO is able to surpass other alignment methods in all cases and achieve state-of-the-art results in challenging code competitions.
Authors: Liyan Tang, Philippe Laban, Greg Durrett
Abstract: Recognizing if LLM output can be grounded in evidence is central to many tasks in NLP: retrieval-augmented generation, summarization, document-grounded dialogue, and more. Current approaches to this kind of "fact-checking" are based on verifying each piece of a model generation against potential evidence using an LLM. However, this process can be very computationally expensive, requiring many calls to LLMs to check a single response. In this work, we show how to build small models that have GPT-4-level performance but for 400x lower cost. We do this by constructing synthetic training data with GPT-4, which involves creating realistic yet challenging instances of factual errors via a structured generation procedure. Training on this data teaches models to check each fact in the claim and recognize synthesis of information across sentences. For evaluation, we unify pre-existing datasets into a benchmark LLM-AggreFact, collected from recent work on fact-checking and grounding LLM generations. Our best system MiniCheck-FT5 (770M parameters) outperforms all systems of comparable size and reaches GPT-4 accuracy. We release LLM-AggreFact, code for data synthesis, and models.
Authors: Nuzhat Prova
Abstract: Recent advances in natural language processing (NLP) may enable artificial intelligence (AI) models to generate writing that is identical to human written form in the future. This might have profound ethical, legal, and social repercussions. This study aims to address this problem by offering an accurate AI detector model that can differentiate between electronically produced text and human-written text. Our approach includes machine learning methods such as XGB Classifier, SVM, BERT architecture deep learning models. Furthermore, our results show that the BERT performs better than previous models in identifying information generated by AI from information provided by humans. Provide a comprehensive analysis of the current state of AI-generated text identification in our assessment of pertinent studies. Our testing yielded positive findings, showing that our strategy is successful, with the BERT emerging as the most probable answer. We analyze the research's societal implications, highlighting the possible advantages for various industries while addressing sustainability issues pertaining to morality and the environment. The XGB classifier and SVM give 0.84 and 0.81 accuracy in this article, respectively. The greatest accuracy in this research is provided by the BERT model, which provides 0.93% accuracy.
Authors: Niyati Rawal, Roberto Bigazzi, Lorenzo Baraldi, Rita Cucchiara
Abstract: In the last few years, the research interest in Vision-and-Language Navigation (VLN) has grown significantly. VLN is a challenging task that involves an agent following human instructions and navigating in a previously unknown environment to reach a specified goal. Recent work in literature focuses on different ways to augment the available datasets of instructions for improving navigation performance by exploiting synthetic training data. In this work, we propose AIGeN, a novel architecture inspired by Generative Adversarial Networks (GANs) that produces meaningful and well-formed synthetic instructions to improve navigation agents' performance. The model is composed of a Transformer decoder (GPT-2) and a Transformer encoder (BERT). During the training phase, the decoder generates sentences for a sequence of images describing the agent's path to a particular point while the encoder discriminates between real and fake instructions. Experimentally, we evaluate the quality of the generated instructions and perform extensive ablation studies. Additionally, we generate synthetic instructions for 217K trajectories using AIGeN on Habitat-Matterport 3D Dataset (HM3D) and show an improvement in the performance of an off-the-shelf VLN method. The validation analysis of our proposal is conducted on REVERIE and R2R and highlights the promising aspects of our proposal, achieving state-of-the-art performance.
Authors: Tamay Besiroglu, Ege Erdil, Matthew Barnett, Josh You
Abstract: Hoffmann et al. (2022) propose three methods for estimating a compute-optimal scaling law. We attempt to replicate their third estimation procedure, which involves fitting a parametric loss function to a reconstruction of data from their plots. We find that the reported estimates are inconsistent with their first two estimation methods, fail at fitting the extracted data, and report implausibly narrow confidence intervals--intervals this narrow would require over 600,000 experiments, while they likely only ran fewer than 500. In contrast, our rederivation of the scaling law using the third approach yields results that are compatible with the findings from the first two estimation procedures described by Hoffmann et al.
Authors: Aashish Anantha Ramakrishnan, Sharon X. Huang, Dongwon Lee
Abstract: Text-to-Image (T2I) Synthesis has made tremendous strides in enhancing synthesized image quality, but current datasets evaluate model performance only on descriptive, instruction-based prompts. Real-world news image captions take a more pragmatic approach, providing high-level situational and Named-Entity (NE) information and limited physical object descriptions, making them abstractive. To evaluate the ability of T2I models to capture intended subjects from news captions, we introduce the Abstractive News Captions with High-level cOntext Representation (ANCHOR) dataset, containing 70K+ samples sourced from 5 different news media organizations. With Large Language Models (LLM) achieving success in language and commonsense reasoning tasks, we explore the ability of different LLMs to identify and understand key subjects from abstractive captions. Our proposed method Subject-Aware Finetuning (SAFE), selects and enhances the representation of key subjects in synthesized images by leveraging LLM-generated subject weights. It also adapts to the domain distribution of news images and captions through custom Domain Fine-tuning, outperforming current T2I baselines on ANCHOR. By launching the ANCHOR dataset, we hope to motivate research in furthering the Natural Language Understanding (NLU) capabilities of T2I models.
Authors: Elham J. Barezi, Parisa Kordjamshidi
Abstract: We analyze knowledge-based visual question answering, for which given a question, the models need to ground it into the visual modality and retrieve the relevant knowledge from a given large knowledge base (KB) to be able to answer. Our analysis has two folds, one based on designing neural architectures and training them from scratch, and another based on large pre-trained language models (LLMs). Our research questions are: 1) Can we effectively augment models by explicit supervised retrieval of the relevant KB information to solve the KB-VQA problem? 2) How do task-specific and LLM-based models perform in the integration of visual and external knowledge, and multi-hop reasoning over both sources of information? 3) Is the implicit knowledge of LLMs sufficient for KB-VQA and to what extent it can replace the explicit KB? Our results demonstrate the positive impact of empowering task-specific and LLM models with supervised external and visual knowledge retrieval models. Our findings show that though LLMs are stronger in 1-hop reasoning, they suffer in 2-hop reasoning in comparison with our fine-tuned NN model even if the relevant information from both modalities is available to the model. Moreover, we observed that LLM models outperform the NN model for KB-related questions which confirms the effectiveness of implicit knowledge in LLMs however, they do not alleviate the need for external KB.
Authors: Joshua Melton, Shannon Reid, Gabriel Terejanu, Siddharth Krishnan
Abstract: The high volume and rapid evolution of content on social media present major challenges for studying the stance of social media users. In this work, we develop a two stage stance labeling method that utilizes the user-hashtag bipartite graph and the user-user interaction graph. In the first stage, a simple and efficient heuristic for stance labeling uses the user-hashtag bipartite graph to iteratively update the stance association of user and hashtag nodes via a label propagation mechanism. This set of soft labels is then integrated with the user-user interaction graph to train a graph neural network (GNN) model using semi-supervised learning. We evaluate this method on two large-scale datasets containing tweets related to climate change from June 2021 to June 2022 and gun control from January 2022 to January 2023. Experiments demonstrate that our user-hashtag heuristic and the semi-supervised GNN method outperform zero-shot stance labeling using LLMs such as GPT4. Further analysis illustrates how the stance labeling information and interaction graph can be used for evaluating the polarization of social media interactions on divisive issues such as climate change and gun control.
Authors: Songtao Jiang, Tuo Zheng, Yan Zhang, Yeying Jin, Zuozhu Liu
Abstract: Mixture of Expert Tuning (MoE-Tuning) has effectively enhanced the performance of general MLLMs with fewer parameters, yet its application in resource-limited medical settings has not been fully explored. To address this gap, we developed MoE-TinyMed, a model tailored for medical applications that significantly lowers parameter demands. In evaluations on the VQA-RAD, SLAKE, and Path-VQA datasets, MoE-TinyMed outperformed LLaVA-Med in all Med-VQA closed settings with just 3.6B parameters. Additionally, a streamlined version with 2B parameters surpassed LLaVA-Med's performance in PathVQA, showcasing its effectiveness in resource-limited healthcare settings.
Authors: Vincent Conitzer, Rachel Freedman, Jobst Heitzig, Wesley H. Holliday, Bob M. Jacobs, Nathan Lambert, Milan Moss\'e, Eric Pacuit, Stuart Russell, Hailey Schoelkopf, Emanuel Tewolde, William S. Zwicker
Abstract: Foundation models such as GPT-4 are fine-tuned to avoid unsafe or otherwise problematic behavior, so that, for example, they refuse to comply with requests for help with committing crimes or with producing racist text. One approach to fine-tuning, called reinforcement learning from human feedback, learns from humans' expressed preferences over multiple outputs. Another approach is constitutional AI, in which the input from humans is a list of high-level principles. But how do we deal with potentially diverging input from humans? How can we aggregate the input into consistent data about ''collective'' preferences or otherwise use it to make collective choices about model behavior? In this paper, we argue that the field of social choice is well positioned to address these questions, and we discuss ways forward for this agenda, drawing on discussions in a recent workshop on Social Choice for AI Ethics and Safety held in Berkeley, CA, USA in December 2023.
Authors: Matthieu Futeral, Andrea Agostinelli, Marco Tagliasacchi, Neil Zeghidour, Eugene Kharitonov
Abstract: Generative spoken language models produce speech in a wide range of voices, prosody, and recording conditions, seemingly approaching the diversity of natural speech. However, the extent to which generated speech is acoustically diverse remains unclear due to a lack of appropriate metrics. We address this gap by developing lightweight metrics of acoustic diversity, which we collectively refer to as MAD Speech. We focus on measuring five facets of acoustic diversity: voice, gender, emotion, accent, and background noise. We construct the metrics as a composition of specialized, per-facet embedding models and an aggregation function that measures diversity within the embedding space. Next, we build a series of datasets with a priori known diversity preferences for each facet. Using these datasets, we demonstrate that our proposed metrics achieve a stronger agreement with the ground-truth diversity than baselines. Finally, we showcase the applicability of our proposed metrics across several real-life evaluation scenarios. MAD Speech will be made publicly accessible.
Authors: Ke Zhu, Liang Zhao, Zheng Ge, Xiangyu Zhang
Abstract: This paper makes the first attempt towards unsupervised preference alignment in Vision-Language Models (VLMs). We generate chosen and rejected responses with regard to the original and augmented image pairs, and conduct preference alignment with direct preference optimization. It is based on a core idea: properly designed augmentation to the image input will induce VLM to generate false but hard negative responses, which helps the model to learn from and produce more robust and powerful answers. The whole pipeline no longer hinges on supervision from GPT4 or human involvement during alignment, and is highly efficient with few lines of code. With only 8k randomly sampled unsupervised data, it achieves 90\% relative score to GPT-4 on complex reasoning in LLaVA-Bench, and improves LLaVA-7B/13B by 6.7\%/5.6\% score on complex multi-modal benchmark MM-Vet. Visualizations shows its improved ability to align with user-intentions. A series of ablations are firmly conducted to reveal the latent mechanism of the approach, which also indicates its potential towards further scaling. Code will be available.
Authors: Keqin Li, Peng Xirui, Jintong Song, Bo Hong, Jin Wang
Abstract: With the rapid advancement of technology, Augmented Reality (AR) technology, known for its ability to deeply integrate virtual information with the real world, is gradually transforming traditional work modes and teaching methods. Particularly in the realms of remote work and online education, AR technology demonstrates a broad spectrum of application prospects. This paper delves into the application potential and actual effects of AR technology in remote work and education. Through a systematic literature review, this study outlines the key features, advantages, and challenges of AR technology. Based on theoretical analysis, it discusses the scientific basis and technical support that AR technology provides for enhancing remote work efficiency and promoting innovation in educational teaching models. Additionally, by designing an empirical research plan and analyzing experimental data, this article reveals the specific performance and influencing factors of AR technology in practical applications. Finally, based on the results of the experiments, this research summarizes the application value of AR technology in remote work and education, looks forward to its future development trends, and proposes forward-looking research directions and strategic suggestions, offering empirical foundation and theoretical guidance for further promoting the in-depth application of AR technology in related fields.
Authors: Oliver Klingefjord, Ryan Lowe, Joe Edelman
Abstract: There is an emerging consensus that we need to align AI systems with human values (Gabriel, 2020; Ji et al., 2024), but there is very little work on what that means and how we actually do it. We split the problem of "aligning to human values" into three parts: first, eliciting values from people; second, reconciling those values into an alignment target for training ML models; and third, actually training the model. In this paper, we focus on the first two parts, and ask the question: what are "good" ways to synthesize diverse human inputs about values into a target for aligning language models? To answer this question, we first define a set of 6 criteria that we believe must be satisfied for an alignment target to shape model behavior in accordance with human values. We then propose a process for eliciting and reconciling values called Moral Graph Elicitation (MGE), which uses a large language model to interview participants about their values in particular contexts; our approach is inspired by the philosophy of values advanced by Taylor (1977), Chang (2004), and others. We trial MGE with a representative sample of 500 Americans, on 3 intentionally divisive prompts (e.g. advice about abortion). Our results demonstrate that MGE is promising for improving model alignment across all 6 criteria. For example, almost all participants (89.1%) felt well represented by the process, and (89%) thought the final moral graph was fair, even if their value wasn't voted as the wisest. Our process often results in "expert" values (e.g. values from women who have solicited abortion advice) rising to the top of the moral graph, without defining who is considered an expert in advance.
Authors: H. Laurie Jones, Brian C. Keegan
Abstract: The Arab Spring was a historic set of protests beginning in 2011 that toppled governments and led to major conflicts. Collective memories of events like these can vary significantly across social contexts in response to political, cultural, and linguistic factors. While Wikipedia plays an important role in documenting both historic and current events, little attention has been given to how Wikipedia articles, created in the aftermath of major events, continue to evolve over years or decades. Using the archived content of Arab Spring-related topics across the Arabic and English Wikipedias between 2011 and 2024, we define and evaluate multilingual measures of event salience, deliberation, contextualization, and consolidation of collective memory surrounding the Arab Spring. Our findings about the temporal evolution of the Wikipedia articles' content similarity across languages has implications for theorizing about online collective memory processes and evaluating linguistic models trained on these data.
Authors: Yu-Yang Li, Yu Bai, Cunshi Wang, Mengwei Qu, Ziteng Lu, Roberto Soria, Jifeng Liu
Abstract: Light curves serve as a valuable source of information on stellar formation and evolution. With the rapid advancement of machine learning techniques, it can be effectively processed to extract astronomical patterns and information. In this study, we present a comprehensive evaluation of deep-learning and large language model (LLM) based models for the automatic classification of variable star light curves, based on large datasets from the Kepler and K2 missions. Special emphasis is placed on Cepheids, RR Lyrae, and eclipsing binaries, examining the influence of observational cadence and phase distribution on classification precision. Employing AutoDL optimization, we achieve striking performance with the 1D-Convolution+BiLSTM architecture and the Swin Transformer, hitting accuracies of 94\% and 99\% correspondingly, with the latter demonstrating a notable 83\% accuracy in discerning the elusive Type II Cepheids-comprising merely 0.02\% of the total dataset.We unveil StarWhisper LightCurve (LC), an innovative Series comprising three LLM-based models: LLM, multimodal large language model (MLLM), and Large Audio Language Model (LALM). Each model is fine-tuned with strategic prompt engineering and customized training methods to explore the emergent abilities of these models for astronomical data. Remarkably, StarWhisper LC Series exhibit high accuracies around 90\%, significantly reducing the need for explicit feature engineering, thereby paving the way for streamlined parallel data processing and the progression of multifaceted multimodal models in astronomical applications. The study furnishes two detailed catalogs illustrating the impacts of phase and sampling intervals on deep learning classification accuracy, showing that a substantial decrease of up to 14\% in observation duration and 21\% in sampling points can be realized without compromising accuracy by more than 10\%.
Authors: Yuchi Wang, Shuhuai Ren, Rundong Gao, Linli Yao, Qingyan Guo, Kaikai An, Jianhong Bai, Xu Sun
Abstract: Diffusion models have exhibited remarkable capabilities in text-to-image generation. However, their performance in image-to-text generation, specifically image captioning, has lagged behind Auto-Regressive (AR) models, casting doubt on their applicability for such tasks. In this work, we revisit diffusion models, highlighting their capacity for holistic context modeling and parallel decoding. With these benefits, diffusion models can alleviate the inherent limitations of AR methods, including their slow inference speed, error propagation, and unidirectional constraints. Furthermore, we identify the prior underperformance of diffusion models stemming from the absence of an effective latent space for image-text alignment, and the discrepancy between continuous diffusion processes and discrete textual data. In response, we introduce a novel architecture, LaDiC, which utilizes a split BERT to create a dedicated latent space for captions and integrates a regularization module to manage varying text lengths. Our framework also includes a diffuser for semantic image-to-text conversion and a Back&Refine technique to enhance token interactivity during inference. LaDiC achieves state-of-the-art performance for diffusion-based methods on the MS COCO dataset with 38.2 BLEU@4 and 126.2 CIDEr, demonstrating exceptional performance without pre-training or ancillary modules. This indicates strong competitiveness with AR models, revealing the previously untapped potential of diffusion models in image-to-text generation.
Authors: Dominic Widdows, Aaranya Alexander, Daiwei Zhu, Chase Zimmerman, Arunava Majumder
Abstract: This paper describes experiments showing that some tasks in natural language processing (NLP) can already be performed using quantum computers, though so far only with small datasets. We demonstrate various approaches to topic classification. The first uses an explicit word-based approach, in which word-topic scoring weights are implemented as fractional rotations of individual qubit, and a new phrase is classified based on the accumulation of these weights in a scoring qubit using entangling controlled-NOT gates. This is compared with more scalable quantum encodings of word embedding vectors, which are used in the computation of kernel values in a quantum support vector machine: this approach achieved an average of 62% accuracy on classification tasks involving over 10000 words, which is the largest such quantum computing experiment to date. We describe a quantum probability approach to bigram modeling that can be applied to sequences of words and formal concepts, investigating a generative approximation to these distributions using a quantum circuit Born machine, and an approach to ambiguity resolution in verb-noun composition using single-qubit rotations for simple nouns and 2-qubit controlled-NOT gates for simple verbs. The smaller systems described have been run successfully on physical quantum computers, and the larger ones have been simulated. We show that statistically meaningful results can be obtained using real datasets, but this is much more difficult to predict than with easier artificial language examples used previously in developing quantum NLP systems. Other approaches to quantum NLP are compared, partly with respect to contemporary issues including informal language, fluency, and truthfulness.
Authors: Pai Liu, Wenyang Gao, Wenjie Dong, Lin Ai, Ziwei Gong, Songfang Huang, Zongsheng Li, Ehsan Hoque, Julia Hirschberg, Yue Zhang
Abstract: Open information extraction is an important NLP task that targets extracting structured information from unstructured text without limitations on the relation type or the domain of the text. This survey paper covers open information extraction technologies from 2007 to 2022 with a focus on new models not covered by previous surveys. We propose a new categorization method from the source of information perspective to accommodate the development of recent OIE technologies. In addition, we summarize three major approaches based on task settings as well as current popular datasets and model evaluation metrics. Given the comprehensive review, several future directions are shown from datasets, source of information, output form, method, and evaluation metric aspects.
Authors: Shreekanth M Prabhu, Abhisek Midye
Abstract: The study of spoken languages comprises phonology, morphology, and grammar. Analysis of a language can be based on its syntax, semantics, and pragmatics. The languages can be classified as root languages, inflectional languages, and stem languages. All these factors lead to the formation of vocabulary which has commonality/similarity as well as distinct and subtle differences across languages. In this paper, we make use of Paninian system of sounds to construct a phonetic map and then words are represented as state transitions on the phonetic map. Each group of related words that cut across languages is represented by a m-language (morphological language). Morphological Finite Automata (MFA) are defined that accept the words belonging to a given m-language. This exercise can enable us to better understand the inter-relationships between words in spoken languages in both language-agnostic and language-cognizant manner. Based on our study and analysis, we propose an Ecosystem Model for Linguistic Development with Sanskrit at the core, in place of the widely accepted family tree model.
Authors: Emil Carlsson, Devdatt Dubhashi, Terry Regier
Abstract: It has been argued that semantic systems reflect pressure for efficiency, and a current debate concerns the cultural evolutionary process that produces this pattern. We consider efficiency as instantiated in the Information Bottleneck (IB) principle, and a model of cultural evolution that combines iterated learning and communication. We show that this model, instantiated in neural networks, converges to color naming systems that are efficient in the IB sense and similar to human color naming systems. We also show that some other proposals such as iterated learning alone, communication alone, or the greater learnability of convex categories, do not yield the same outcome as clearly. We conclude that the combination of iterated learning and communication provides a plausible means by which human semantic systems become efficient.
Authors: Alexander Scarlatos, Andrew Lan
Abstract: Recent developments in large pre-trained language models have enabled unprecedented performance on a variety of downstream tasks. Achieving best performance with these models often leverages in-context learning, where a model performs a (possibly new) task given one or more examples. However, recent work has shown that the choice of examples can have a large impact on task performance and that finding an optimal set of examples is non-trivial. While there are many existing methods for selecting in-context examples, they generally score examples independently, ignoring the dependency between them and the order in which they are provided to the model. In this work, we propose Retrieval for In-Context Learning (RetICL), a learnable method for modeling and optimally selecting examples sequentially for in-context learning. We frame the problem of sequential example selection as a Markov decision process and train an example retriever using reinforcement learning. We evaluate RetICL on math word problem solving and scientific question answering tasks and show that it consistently outperforms or matches heuristic and learnable baselines. We also use case studies to show that RetICL implicitly learns representations of problem solving strategies.
Authors: Mourad Heddaya, Solomon Dworkin, Chenhao Tan, Rob Voigt, Alexander Zentefis
Abstract: Leveraging an established exercise in negotiation education, we build a novel dataset for studying how the use of language shapes bilateral bargaining. Our dataset extends existing work in two ways: 1) we recruit participants via behavioral labs instead of crowdsourcing platforms and allow participants to negotiate through audio, enabling more naturalistic interactions; 2) we add a control setting where participants negotiate only through alternating, written numeric offers. Despite the two contrasting forms of communication, we find that the average agreed prices of the two treatments are identical. But when subjects can talk, fewer offers are exchanged, negotiations finish faster, the likelihood of reaching agreement rises, and the variance of prices at which subjects agree drops substantially. We further propose a taxonomy of speech acts in negotiation and enrich the dataset with annotated speech acts. Our work also reveals linguistic signals that are predictive of negotiation outcomes.
Authors: Liunian Harold Li, Jack Hessel, Youngjae Yu, Xiang Ren, Kai-Wei Chang, Yejin Choi
Abstract: Chain-of-thought prompting (e.g., "Let's think step-by-step") primes large language models to verbalize rationalization for their predictions. While chain-of-thought can lead to dramatic performance gains, benefits appear to emerge only for sufficiently large models (beyond 50B parameters). We show that orders-of-magnitude smaller models (125M -- 1.3B parameters) can still benefit from chain-of-thought prompting. To achieve this, we introduce Symbolic Chain-of-Thought Distillation (SCoTD), a method to train a smaller student model on rationalizations sampled from a significantly larger teacher model. Experiments across several commonsense benchmarks show that: 1) SCoTD enhances the performance of the student model in both supervised and few-shot settings, and especially for challenge sets; 2) sampling many reasoning chains per instance from the teacher is paramount; and 3) after distillation, student chain-of-thoughts are judged by humans as comparable to the teacher, despite orders of magnitude fewer parameters. We test several hypotheses regarding what properties of chain-of-thought samples are important, e.g., diversity vs. teacher likelihood vs. open-endedness. We release our corpus of chain-of-thought samples and code.
Authors: Bo Liu, Liming Zhan, Zexin Lu, Yujie Feng, Lei Xue, Xiao-Ming Wu
Abstract: Out-of-distribution (OOD) detection plays a vital role in enhancing the reliability of machine learning (ML) models. The emergence of large language models (LLMs) has catalyzed a paradigm shift within the ML community, showcasing their exceptional capabilities across diverse natural language processing tasks. While existing research has probed OOD detection with relative small-scale Transformers like BERT, RoBERTa and GPT-2, the stark differences in scales, pre-training objectives, and inference paradigms call into question the applicability of these findings to LLMs. This paper embarks on a pioneering empirical investigation of OOD detection in the domain of LLMs, focusing on LLaMA series ranging from 7B to 65B in size. We thoroughly evaluate commonly-used OOD detectors, scrutinizing their performance in both zero-grad and fine-tuning scenarios. Notably, we alter previous discriminative in-distribution fine-tuning into generative fine-tuning, aligning the pre-training objective of LLMs with downstream tasks. Our findings unveil that a simple cosine distance OOD detector demonstrates superior efficacy, outperforming other OOD detectors. We provide an intriguing explanation for this phenomenon by highlighting the isotropic nature of the embedding spaces of LLMs, which distinctly contrasts with the anisotropic property observed in smaller BERT family models. The new insight enhances our understanding of how LLMs detect OOD data, thereby enhancing their adaptability and reliability in dynamic environments. We have released the source code at \url{https://github.com/Awenbocc/LLM-OOD} for other researchers to reproduce our results.
Authors: Juan Diego Rodriguez, Katrin Erk, Greg Durrett
Abstract: Understanding when two pieces of text convey the same information is a goal touching many subproblems in NLP, including textual entailment and fact-checking. This problem becomes more complex when those two pieces of text are in different languages. Here, we introduce X-PARADE (Cross-lingual Paragraph-level Analysis of Divergences and Entailments), the first cross-lingual dataset of paragraph-level information divergences. Annotators label a paragraph in a target language at the span level and evaluate it with respect to a corresponding paragraph in a source language, indicating whether a given piece of information is the same, new, or new but can be inferred. This last notion establishes a link with cross-language NLI. Aligned paragraphs are sourced from Wikipedia pages in different languages, reflecting real information divergences observed in the wild. Armed with our dataset, we investigate a diverse set of approaches for this problem, including token alignment from machine translation, textual entailment methods that localize their decisions, and prompting LLMs. Our results show that these methods vary in their capability to handle inferable information, but they all fall short of human performance.
Authors: S\'ebastien Ragot
Abstract: This work proposes to measure the scope of a patent claim as the reciprocal of self-information contained in this claim. Self-information is calculated based on a probability of occurrence of the claim, where this probability is obtained from a language model. Grounded in information theory, this approach is based on the assumption that an unlikely concept is more informative than a usual concept, insofar as it is more surprising. In turn, the more surprising the information required to define the claim, the narrower its scope. Seven language models are considered, ranging from simplest models (each word or character has an identical probability) to intermediate models (based on average word or character frequencies), to large language models (LLMs) such as GPT2 and davinci-002. Remarkably, when using the simplest language models to compute the probabilities, the scope becomes proportional to the reciprocal of the number of words or characters involved in the claim, a metric already used in previous works. Application is made to multiple series of patent claims directed to distinct inventions, where each series consists of claims devised to have a gradually decreasing scope. The performance of the language models is then assessed through several ad hoc tests. The LLMs outperform models based on word and character frequencies, which themselves outdo the simplest models based on word or character counts. Interestingly, however, the character count appears to be a more reliable indicator than the word count.
Authors: Letian Zhang, Xiaotong Zhai, Zhongkai Zhao, Yongshuo Zong, Xin Wen, Bingchen Zhao
Abstract: Counterfactual reasoning, a fundamental aspect of human cognition, involves contemplating alternatives to established facts or past events, significantly enhancing our abilities in planning and decision-making. In light of the advancements in current multi-modal large language models, we explore their effectiveness in counterfactual reasoning. To facilitate this investigation, we introduce a novel dataset, C-VQA, specifically designed to test the counterfactual reasoning capabilities of modern multi-modal large language models. This dataset is constructed by infusing original questions with counterfactual presuppositions, spanning various types such as numerical and boolean queries. It encompasses a mix of real and synthetic data, representing a wide range of difficulty levels. Our thorough evaluations of contemporary vision-language models using this dataset have revealed substantial performance drops, with some models showing up to a 40% decrease, highlighting a significant gap between current models and human-like vision reasoning capabilities. We hope our dataset will serve as a vital benchmark for evaluating the counterfactual reasoning capabilities of models. Code and dataset are publicly available at https://bzhao.me/C-VQA/.
URLs: https://bzhao.me/C-VQA/.
Authors: Zhiyuan Zeng, Jiatong Yu, Tianyu Gao, Yu Meng, Tanya Goyal, Danqi Chen
Abstract: As research in large language models (LLMs) continues to accelerate, LLM-based evaluation has emerged as a scalable and cost-effective alternative to human evaluations for comparing the ever increasing list of models. This paper investigates the efficacy of these ``LLM evaluators'', particularly in using them to assess instruction following, a metric that gauges how closely generated text adheres to the given instruction. We introduce a challenging meta-evaluation benchmark, LLMBar, designed to test the ability of an LLM evaluator in discerning instruction-following outputs. The authors manually curated 419 pairs of outputs, one adhering to instructions while the other diverging, yet may possess deceptive qualities that mislead an LLM evaluator, e.g., a more engaging tone. Contrary to existing meta-evaluation, we discover that different evaluators (i.e., combinations of LLMs and prompts) exhibit distinct performance on LLMBar and even the highest-scoring ones have substantial room for improvement. We also present a novel suite of prompting strategies that further close the gap between LLM and human evaluators. With LLMBar, we hope to offer more insight into LLM evaluators and foster future research in developing better instruction-following models.
Authors: Enric Boix-Adsera, Omid Saremi, Emmanuel Abbe, Samy Bengio, Etai Littwin, Joshua Susskind
Abstract: We investigate the capabilities of transformer models on relational reasoning tasks. In these tasks, models are trained on a set of strings encoding abstract relations, and are then tested out-of-distribution on data that contains symbols that did not appear in the training dataset. We prove that for any relational reasoning task in a large family of tasks, transformers learn the abstract relations and generalize to the test set when trained by gradient descent on sufficiently large quantities of training data. This is in contrast to classical fully-connected networks, which we prove fail to learn to reason. Our results inspire modifications of the transformer architecture that add only two trainable parameters per head, and that we empirically demonstrate improve data efficiency for learning to reason.
Authors: Xiaochong Lan, Chen Gao, Depeng Jin, Yong Li
Abstract: Stance detection automatically detects the stance in a text towards a target, vital for content analysis in web and social media research. Despite their promising capabilities, LLMs encounter challenges when directly applied to stance detection. First, stance detection demands multi-aspect knowledge, from deciphering event-related terminologies to understanding the expression styles in social media platforms. Second, stance detection requires advanced reasoning to infer authors' implicit viewpoints, as stance are often subtly embedded rather than overtly stated in the text. To address these challenges, we design a three-stage framework COLA (short for Collaborative rOle-infused LLM-based Agents) in which LLMs are designated distinct roles, creating a collaborative system where each role contributes uniquely. Initially, in the multidimensional text analysis stage, we configure the LLMs to act as a linguistic expert, a domain specialist, and a social media veteran to get a multifaceted analysis of texts, thus overcoming the first challenge. Next, in the reasoning-enhanced debating stage, for each potential stance, we designate a specific LLM-based agent to advocate for it, guiding the LLM to detect logical connections between text features and stance, tackling the second challenge. Finally, in the stance conclusion stage, a final decision maker agent consolidates prior insights to determine the stance. Our approach avoids extra annotated data and model training and is highly usable. We achieve state-of-the-art performance across multiple datasets. Ablation studies validate the effectiveness of each design role in handling stance detection. Further experiments have demonstrated the explainability and the versatility of our approach. Our approach excels in usability, accuracy, effectiveness, explainability and versatility, highlighting its value.
Authors: Mingchen Li, Huixue Zhou, Rui Zhang
Abstract: Biomedical triple extraction systems aim to automatically extract biomedical entities and relations between entities. The exploration of applying large language models (LLM) to triple extraction is still relatively unexplored. In this work, we mainly focus on sentence-level biomedical triple extraction. Furthermore, the absence of a high-quality biomedical triple extraction dataset impedes the progress in developing robust triple extraction systems. To address these challenges, initially, we compare the performance of various large language models. Additionally, we present GIT, an expert-annotated biomedical triple extraction dataset that covers a wider range of relation types.
Authors: Lang Cao
Abstract: Large language models (LLMs) have demonstrated impressive language understanding and generation capabilities, enabling them to answer a wide range of questions across various domains. However, these models are not flawless and often produce responses that contain errors or misinformation. These inaccuracies, commonly referred to as hallucinations, render LLMs unreliable and even unusable in many scenarios. In this paper, our focus is on mitigating the issue of hallucination in LLMs, particularly in the context of question-answering. Instead of attempting to answer all questions, we explore a refusal mechanism that instructs LLMs to refuse to answer challenging questions in order to avoid errors. We then propose a simple yet effective solution called Learn to Refuse (L2R), which incorporates the refusal mechanism to enable LLMs to recognize and refuse to answer questions that they find difficult to address. To achieve this, we utilize a structured knowledge base to represent all the LLM's understanding of the world, enabling it to provide traceable gold knowledge. This knowledge base is separate from the LLM and initially empty. It can be filled with validated knowledge and progressively expanded. When an LLM encounters questions outside its domain, the system recognizes its knowledge scope and determines whether it can answer the question independently. Additionally, we introduce a method for automatically and efficiently expanding the knowledge base of LLMs. Through qualitative and quantitative analysis, we demonstrate that our approach enhances the controllability and reliability of LLMs.
Authors: Ziyin Zhang, Chaoyu Chen, Bingchang Liu, Cong Liao, Zi Gong, Hang Yu, Jianguo Li, Rui Wang
Abstract: In this work we systematically review the recent advancements in code processing with language models, covering 50+ models, 30+ evaluation tasks, 170+ datasets, and 800 related works. We break down code processing models into general language models represented by the GPT family and specialized models that are specifically pretrained on code, often with tailored objectives. We discuss the relations and differences between these models, and highlight the historical transition of code modeling from statistical models and RNNs to pretrained Transformers and LLMs, which is exactly the same course that had been taken by NLP. We also discuss code-specific features such as AST, CFG, and unit tests, along with their application in training code language models, and identify key challenges and potential future directions in this domain. We keep the survey open and updated on GitHub at https://github.com/codefuse-ai/Awesome-Code-LLM.
Authors: Yuxia Wang, Revanth Gangi Reddy, Zain Muhammad Mujahid, Arnav Arora, Aleksandr Rubashevskii, Jiahui Geng, Osama Mohammed Afzal, Liangming Pan, Nadav Borenstein, Aditya Pillai, Isabelle Augenstein, Iryna Gurevych, Preslav Nakov
Abstract: The increased use of large language models (LLMs) across a variety of real-world applications calls for mechanisms to verify the factual accuracy of their outputs. In this work, we present a holistic end-to-end solution for annotating the factuality of LLM-generated responses, which encompasses a multi-stage annotation scheme designed to yield detailed labels concerning the verifiability and factual inconsistencies found in LLM outputs. We further construct an open-domain document-level factuality benchmark in three-level granularity: claim, sentence and document, aiming to facilitate the evaluation of automatic fact-checking systems. Preliminary experiments show that FacTool, FactScore and Perplexity.ai are struggling to identify false claims, with the best F1=0.63 by this annotation solution based on GPT-4. Annotation tool, benchmark and code are available at https://github.com/yuxiaw/Factcheck-GPT.
Authors: Yanzhu Guo, Guokan Shang, Michalis Vazirgiannis, Chlo\'e Clavel
Abstract: This study investigates the consequences of training language models on synthetic data generated by their predecessors, an increasingly prevalent practice given the prominence of powerful generative models. Diverging from the usual emphasis on performance metrics, we focus on the impact of this training methodology on linguistic diversity, especially when conducted recursively over time. To assess this, we adapt and develop a set of novel metrics targeting lexical, syntactic, and semantic diversity, applying them in recursive finetuning experiments across various natural language generation tasks in English. Our findings reveal a consistent decrease in the diversity of the model outputs through successive iterations, especially remarkable for tasks demanding high levels of creativity. This trend underscores the potential risks of training language models on synthetic text, particularly concerning the preservation of linguistic richness. Our study highlights the need for careful consideration of the long-term effects of such training approaches on the linguistic capabilities of language models.
Authors: Yan Cathy Hua, Paul Denny, Katerina Taskova, J\"org Wicker
Abstract: Aspect-based Sentiment Analysis (ABSA) is a fine-grained type of sentiment analysis that identifies aspects and their associated opinions from a given text. With the surge of digital opinionated text data, ABSA gained increasing popularity for its ability to mine more detailed and targeted insights. Many review papers on ABSA subtasks and solution methodologies exist, however, few focus on trends over time or systemic issues relating to research application domains, datasets, and solution approaches. To fill the gap, this paper presents a Systematic Literature Review (SLR) of ABSA studies with a focus on trends and high-level relationships among these fundamental components. This review is one of the largest SLRs on ABSA, and also, to our knowledge, the first that systematically examines the trends and inter-relations among ABSA research and data distribution across domains and solution paradigms and approaches. Our sample includes 519 primary studies screened from 4191 search results without time constraints via an innovative automatic filtering process. Our quantitative analysis not only identifies trends in nearly two decades of ABSA research development but also unveils a systemic lack of dataset and domain diversity as well as domain mismatch that may hinder the development of future ABSA research. We discuss these findings and their implications and propose suggestions for future research.
Authors: Zonghai Yao, Nandyala Siddharth Kantu, Guanghao Wei, Hieu Tran, Zhangqi Duan, Sunjae Kwon, Zhichao Yang, README annotation team, Hong Yu
Abstract: The advancement in healthcare has shifted focus toward patient-centric approaches, particularly in self-care and patient education, facilitated by access to Electronic Health Records (EHR). However, medical jargon in EHRs poses significant challenges in patient comprehension. To address this, we introduce a new task of automatically generating lay definitions, aiming to simplify complex medical terms into patient-friendly lay language. We first created the README dataset, an extensive collection of over 50,000 unique (medical term, lay definition) pairs and 300,000 mentions, each offering context-aware lay definitions manually annotated by domain experts. We have also engineered a data-centric Human-AI pipeline that synergizes data filtering, augmentation, and selection to improve data quality. We then used README as the training data for models and leveraged a Retrieval-Augmented Generation method to reduce hallucinations and improve the quality of model outputs. Our extensive automatic and human evaluations demonstrate that open-source mobile-friendly models, when fine-tuned with high-quality data, are capable of matching or even surpassing the performance of state-of-the-art closed-source large language models like ChatGPT. This research represents a significant stride in closing the knowledge gap in patient education and advancing patient-centric healthcare solutions.
Authors: Wei Liu, Weihao Zeng, Keqing He, Yong Jiang, Junxian He
Abstract: Instruction tuning is a standard technique employed to align large language models to end tasks and user preferences after the initial pretraining phase. Recent research indicates the critical role of data engineering in instruction tuning -- when appropriately selected, only limited data is necessary to achieve superior performance. However, we still lack a principled understanding of what makes good instruction tuning data for alignment, and how we should select data automatically and effectively. In this work, we delve deeply into automatic data selection strategies for alignment. We start with controlled studies to measure data across three dimensions: complexity, quality, and diversity, along which we examine existing methods and introduce novel techniques for enhanced data measurement. Subsequently, we propose a simple strategy to select data samples based on the measurement. We present deita (short for Data-Efficient Instruction Tuning for Alignment), a series of models fine-tuned from LLaMA and Mistral models using data samples automatically selected with our proposed approach. Empirically, deita performs better or on par with the state-of-the-art open-source alignment models with only 6K SFT training data samples -- over 10x less than the data used in the baselines. When further trained with direct preference optimization (DPO), deita-Mistral-7B + DPO trained with 6K SFT and 10K DPO samples achieve 7.55 MT-Bench and 90.06% AlpacaEval scores. We anticipate this work to provide tools on automatic data selection, facilitating data-efficient alignment. We release our models as well as the selected datasets for future researches to effectively align models more efficiently.
Authors: Gang Liu, Jinlong He, Pengfei Li, Genrong He, Zhaolin Chen, Shenjun Zhong
Abstract: Multimodal large language models (MLLMs) represent an evolutionary expansion in the capabilities of traditional large language models, enabling them to tackle challenges that surpass the scope of purely text-based applications. It leverages the knowledge previously encoded within these language models, thereby enhancing their applicability and functionality in the reign of multimodal contexts. Recent works investigate the adaptation of MLLMs as a universal solution to address medical multi-modal problems as a generative task. In this paper, we propose a parameter efficient framework for fine-tuning MLLMs, specifically validated on medical visual question answering (Med-VQA) and medical report generation (MRG) tasks, using public benchmark datasets. We also introduce an evaluation metric using the 5-point Likert scale and its weighted average value to measure the quality of the generated reports for MRG tasks, where the scale ratings are labelled by both humans manually and the GPT-4 model. We further assess the consistency of performance metrics across traditional measures, GPT-4, and human ratings for both VQA and MRG tasks. The results indicate that semantic similarity assessments using GPT-4 align closely with human annotators and provide greater stability, yet they reveal a discrepancy when compared to conventional lexical similarity measurements. This questions the reliability of lexical similarity metrics for evaluating the performance of generative models in Med-VQA and report generation tasks. Besides, our fine-tuned model significantly outperforms GPT-4v. This indicates that without additional fine-tuning, multi-modal models like GPT-4v do not perform effectively on medical imaging tasks. The code will be available here: https://github.com/jinlHe/PeFoMed.
Authors: Fanqi Wan, Xinting Huang, Leyang Cui, Xiaojun Quan, Wei Bi, Shuming Shi
Abstract: While large language models (LLMs) have demonstrated exceptional performance across various tasks following human alignment, they may still generate responses that sound plausible but contradict factual knowledge, a phenomenon known as \emph{hallucination}. In this paper, we demonstrate the feasibility of mitigating hallucinations by verifying and minimizing the inconsistency between external knowledge present in the alignment data and the intrinsic knowledge embedded within foundation LLMs. Specifically, we propose a novel approach called Knowledge Consistent Alignment (KCA), which employs a well-aligned LLM to automatically formulate assessments based on external knowledge to evaluate the knowledge boundaries of foundation LLMs. To address knowledge inconsistencies in the alignment data, KCA implements several specific strategies to deal with these data instances. We demonstrate the superior efficacy of KCA in reducing hallucinations across six benchmarks, utilizing foundation LLMs of varying backbones and scales. This confirms the effectiveness of mitigating hallucinations by reducing knowledge inconsistency. Our code, model weights, and data are openly accessible at \url{https://github.com/fanqiwan/KCA}.
Authors: Wolf Nuyts, Ruben Cartuyvels, Marie-Francine Moens
Abstract: Recognizing visual entities in a natural language sentence and arranging them in a 2D spatial layout require a compositional understanding of language and space. This task of layout prediction is valuable in text-to-image synthesis as it allows localized and controlled in-painting of the image. In this comparative study it is shown that we can predict layouts from language representations that implicitly or explicitly encode sentence syntax, if the sentences mention similar entity-relationships to the ones seen during training. To test compositional understanding, we collect a test set of grammatically correct sentences and layouts describing compositions of entities and relations that unlikely have been seen during training. Performance on this test set substantially drops, showing that current models rely on correlations in the training data and have difficulties in understanding the structure of the input sentences. We propose a novel structural loss function that better enforces the syntactic structure of the input sentence and show large performance gains in the task of 2D spatial layout prediction conditioned on text. The loss has the potential to be used in other generation tasks where a tree-like structure underlies the conditioning modality. Code, trained models and the USCOCO evaluation set are available via github.
Authors: David Peer, Philemon Sch\"opf, Volckmar Nebendahl, Alexander Rietzler, Sebastian Stabinger
Abstract: Traditionally, discriminative models have been the predominant choice for tasks like document classification and information extraction. These models make predictions that fall into a limited number of predefined classes, facilitating a binary true or false evaluation and enabling the direct calculation of metrics such as the F1 score. However, recent advancements in generative large language models (GLLMs) have prompted a shift in the field due to their enhanced zero-shot capabilities, which eliminate the need for a downstream dataset and computationally expensive fine-tuning. However, evaluating GLLMs presents a challenge as the binary true or false evaluation used for discriminative models is not applicable to the predictions made by GLLMs. This paper introduces a new metric for generative models called ANLS* for evaluating a wide variety of tasks, including information extraction and classification tasks. The ANLS* metric extends existing ANLS metrics as a drop-in-replacement and is still compatible with previously reported ANLS scores. An evaluation of 7 different datasets, 6 different GLLMs and 3 different prompting methods using the ANLS* metric is also provided, demonstrating the importance of the proposed metric. We also benchmark a novel approach to generate prompts for documents, called SFT, against other prompting techniques such as LATIN. In 27 out of 35 cases, SFT outperforms other techniques and improves the state-of-the-art, sometimes by as much as $18$ percentage points. Sources are available at https://github.com/deepopinion/anls_star_metric
Authors: Haiyan Zhao, Fan Yang, Bo Shen, Himabindu Lakkaraju, Mengnan Du
Abstract: Large language models (LLMs) have led to breakthroughs in language tasks, yet the internal mechanisms that enable their remarkable generalization and reasoning abilities remain opaque. This lack of transparency presents challenges such as hallucinations, toxicity, and misalignment with human values, hindering the safe and beneficial deployment of LLMs. This paper aims to uncover the mechanisms underlying LLM functionality through the lens of explainability. First, we review how knowledge is architecturally composed within LLMs and encoded in their internal parameters via mechanistic interpretability techniques. Then, we summarize how knowledge is embedded in LLM representations by leveraging probing techniques and representation engineering. Additionally, we investigate the training dynamics through a mechanistic perspective to explain phenomena such as grokking and memorization. Lastly, we explore how the insights gained from these explanations can enhance LLM performance through model editing, improve efficiency through pruning, and better align with human values.
Authors: Zhichao Xu, Jiepu Jiang
Abstract: Empathy is critical for effective and satisfactory conversational communication. Prior efforts to measure conversational empathy mostly focus on expressed communicative intents -- that is, the way empathy is expressed. Yet, these works ignore the fact that conversation is also a collaboration involving both speakers and listeners. In contrast, we propose a multi-dimensional empathy evaluation framework to measure both expressed intents from the speaker's perspective and perceived empathy from the listener's perspective. We apply our proposed framework to analyze our internal customer-service dialogue. We find the two dimensions (expressed intent types and perceived empathy) are inter-connected, and perceived empathy has a high correlation with dialogue satisfaction levels. To reduce the annotation cost, we explore different options to automatically measure conversational empathy: prompting LLMs and training language model-based classifiers. Our experiments show that prompting methods with even popular models like GPT-4 and Flan family models perform relatively poorly on both public and our internal datasets. In contrast, instruction-finetuned classifiers based on Flan-T5 family models outperform prior works and competitive baselines. We conduct a detailed ablation study to give more insights into instruction finetuning method's strong performance.
Authors: Aishik Rakshit, Smriti Singh, Shuvam Keshari, Arijit Ghosh Chowdhury, Vinija Jain, Aman Chadha
Abstract: Embeddings play a pivotal role in the efficacy of Large Language Models. They are the bedrock on which these models grasp contextual relationships and foster a more nuanced understanding of language and consequently perform remarkably on a plethora of complex tasks that require a fundamental understanding of human language. Given that these embeddings themselves often reflect or exhibit bias, it stands to reason that these models may also inadvertently learn this bias. In this work, we build on the seminal previous work and propose DeepSoftDebias, an algorithm that uses a neural network to perform 'soft debiasing'. We exhaustively evaluate this algorithm across a variety of SOTA datasets, accuracy metrics, and challenging NLP tasks. We find that DeepSoftDebias outperforms the current state-of-the-art methods at reducing bias across gender, race, and religion.
Authors: Renxi Wang, Haonan Li, Xudong Han, Yixuan Zhang, Timothy Baldwin
Abstract: Large language models (LLMs) have achieved success in acting as agents, which interact with environments through tools such as search engines. However, LLMs are optimized for language generation instead of tool use during training or alignment, limiting their effectiveness as agents. To resolve this problem, previous work has first collected interaction trajectories between LLMs and environments, using only trajectories that successfully finished the task to fine-tune smaller models, making fine-tuning data scarce and acquiring it both difficult and costly. Discarding failed trajectories also leads to significant wastage of data and resources and limits the possible optimization paths during fine-tuning. In this paper, we argue that unsuccessful trajectories offer valuable insights, and LLMs can learn from these trajectories through appropriate quality control and fine-tuning strategies. By simply adding a prefix or suffix that tells the model whether to generate a successful trajectory during training, we improve model performance by a large margin on mathematical reasoning, multi-hop question answering, and strategic question answering tasks. We further analyze the inference results and find that our method provides a better trade-off between valuable information and errors in unsuccessful trajectories. To our knowledge, we are the first to demonstrate the value of negative trajectories and their application in agent-tunning scenarios. Our findings offer guidance for developing better agent-tuning methods and low-resource data usage techniques.
Authors: Hanling Yi, Feng Lin, Hongbin Li, Peiyang Ning, Xiaotian Yu, Rong Xiao
Abstract: This research aims to accelerate the inference speed of large language models (LLMs) with billions of parameters. We propose \textbf{S}mart \textbf{P}arallel \textbf{A}uto-\textbf{C}orrect d\textbf{E}coding (SPACE), an innovative approach designed for achieving lossless acceleration of LLMs. By integrating semi-autoregressive inference and speculative decoding capabilities, SPACE uniquely enables autoregressive LLMs to parallelize token generation and verification. This is realized through a specialized semi-autoregressive supervised fine-tuning process that equips existing LLMs with the ability to simultaneously predict multiple tokens. Additionally, an auto-correct decoding algorithm facilitates the simultaneous generation and verification of token sequences within a single model invocation. Through extensive experiments on a range of LLMs, SPACE has demonstrated inference speedup ranging from 2.7x-4.0x on HumanEval-X while maintaining output quality.
Authors: Chang Zong, Yuchen Yan, Weiming Lu, Jian Shao, Eliot Huang, Heng Chang, Yueting Zhuang
Abstract: Recent progress with LLM-based agents has shown promising results across various tasks. However, their use in answering questions from knowledge bases remains largely unexplored. Implementing a KBQA system using traditional methods is challenging due to the shortage of task-specific training data and the complexity of creating task-focused model structures. In this paper, we present Triad, a unified framework that utilizes an LLM-based agent with three roles for KBQA tasks. The agent is assigned three roles to tackle different KBQA subtasks: agent as a generalist for mastering various subtasks, as a decision maker for the selection of candidates, and as an advisor for answering questions with knowledge. Our KBQA framework is executed in four phases, involving the collaboration of the agent's multiple roles. We evaluated the performance of our framework using three benchmark datasets, and the results show that our framework outperforms state-of-the-art systems on the LC-QuAD and YAGO-QA benchmarks, yielding F1 scores of 11.8% and 20.7%, respectively.
Authors: Bryan R Christ, Jonathan Kropko, Thomas Hartvigsen
Abstract: Math word problems are critical K-8 educational tools, but writing them is time-consuming and requires domain expertise. We suggest that language models can support K-8 math education by automatically generating problems. To be educational, generated problems must be 1) solvable, 2) accurate, and 3) appropriate. Existing datasets are unlabeled for these criteria, making them ill-suited for training problem generators. To address this gap, we use domain expert annotation to curate a high-quality synthetic training dataset for this task. We show the value of this data by using it to iteratively finetune Llama-2 (70B) to create MATHWELL, a K-8 word problem generator. Domain experts find MATHWELL has a 40% higher share of problems that have executable solutions and meet all criteria than existing open-source models, with 74% of its problems with executable solutions being solvable, accurate, and appropriate. MATHWELL achieves 94.9% of GPT-4 Turbo's performance on this task while outputting problems written at a more appropriate reading level for K-8 students. MATHWELL's performance despite being trained by finetuning only highlights the quality of our synthetic data for training age-appropriate word problem generators. We release our model, data, and annotations.
Authors: Rifki Afina Putri, Faiz Ghifari Haznitrama, Dea Adhista, Alice Oh
Abstract: Large Language Models (LLMs) are increasingly being used to generate synthetic data for training and evaluating models. However, it is unclear whether they can generate a good quality of question answering (QA) dataset that incorporates knowledge and cultural nuance embedded in a language, especially for low-resource languages. In this study, we investigate the effectiveness of using LLMs in generating culturally relevant commonsense QA datasets for Indonesian and Sundanese languages. To do so, we create datasets for these languages using various methods involving both LLMs and human annotators, resulting in ~4.5K questions per language (~9K in total), making our dataset the largest of its kind. Our experiments show that automatic data adaptation from an existing English dataset is less effective for Sundanese. Interestingly, using the direct generation method on the target language, GPT-4 Turbo can generate questions with adequate general knowledge in both languages, albeit not as culturally 'deep' as humans. We also observe a higher occurrence of fluency errors in the Sundanese dataset, highlighting the discrepancy between medium- and lower-resource languages.
Authors: Shaoyang Xu, Weilong Dong, Zishan Guo, Xinwei Wu, Deyi Xiong
Abstract: Prior research in representation engineering has revealed that LLMs encode concepts within their representation spaces, predominantly centered around English. In this study, we extend this philosophy to a multilingual scenario, delving into multilingual human value concepts in LLMs. Through our comprehensive exploration covering 7 types of human values, 16 languages and 3 LLM series with distinct multilinguality, we empirically substantiate the existence of multilingual human values in LLMs. Further cross-lingual analysis on these concepts discloses 3 traits arising from language resource disparities: cross-lingual inconsistency, distorted linguistic relationships, and unidirectional cross-lingual transfer between high- and low-resource languages, all in terms of human value concepts. Additionally, we validate the feasibility of cross-lingual control over value alignment capabilities of LLMs, leveraging the dominant language as a source language. Drawing from our findings on multilingual value alignment, we prudently provide suggestions on the composition of multilingual data for LLMs pre-training: including a limited number of dominant languages for cross-lingual alignment transfer while avoiding their excessive prevalence, and keeping a balanced distribution of non-dominant languages. We aspire that our findings would contribute to enhancing the safety and utility of multilingual AI.
Authors: Swagata Ashwani, Kshiteesh Hegde, Nishith Reddy Mannuru, Mayank Jindal, Dushyant Singh Sengar, Krishna Chaitanya Rao Kathala, Dishant Banga, Vinija Jain, Aman Chadha
Abstract: With the rise of Large Language Models(LLMs), it has become crucial to understand their capabilities and limitations in deciphering and explaining the complex web of causal relationships that language entails. Current methods use either explicit or implicit causal reasoning, yet there is a strong need for a unified approach combining both to tackle a wide array of causal relationships more effectively. This research proposes a novel architecture called Context Aware Reasoning Enhancement with Counterfactual Analysis(CARE CA) framework to enhance causal reasoning and explainability. The proposed framework incorporates an explicit causal detection module with ConceptNet and counterfactual statements, as well as implicit causal detection through LLMs. Our framework goes one step further with a layer of counterfactual explanations to accentuate LLMs understanding of causality. The knowledge from ConceptNet enhances the performance of multiple causal reasoning tasks such as causal discovery, causal identification and counterfactual reasoning. The counterfactual sentences add explicit knowledge of the not caused by scenarios. By combining these powerful modules, our model aims to provide a deeper understanding of causal relationships, enabling enhanced interpretability. Evaluation of benchmark datasets shows improved performance across all metrics, such as accuracy, precision, recall, and F1 scores. We also introduce CausalNet, a new dataset accompanied by our code, to facilitate further research in this domain.
Authors: Huixuan Zhang, Junzhe Zhang, Xiaojun Wan
Abstract: Large-scale vision-language models have demonstrated impressive skill in handling tasks that involve both areas. Nevertheless, these models frequently experience significant issues with generating inaccurate information, which is hallucination. In this study, we concentrate on a specific type of hallucination-number hallucination, referring to models incorrectly identifying the number of certain objects in pictures. We perform quantitative evaluations regarding number hallucination, showing it to be critical in major open-source large vision-language models. Furthermore, we utilizes two related tasks to conduct an in-depth analysis of number hallucination, revealing the severe inner and outer inconsistency among all tasks. Based on this examination, we devise a training approach aimed at improving consistency to reduce number hallucinations, which leads to an 8% enhancement in performance over direct finetuning methods. Our code and dataset will be released to the community.
Authors: Tanja Samardzic, Ximena Gutierrez, Christian Bentz, Steven Moran, Olga Pelloni
Abstract: Typologically diverse benchmarks are increasingly created to track the progress achieved in multilingual NLP. Linguistic diversity of these data sets is typically measured as the number of languages or language families included in the sample, but such measures do not consider structural properties of the included languages. In this paper, we propose assessing linguistic diversity of a data set against a reference language sample as a means of maximising linguistic diversity in the long run. We represent languages as sets of features and apply a version of the Jaccard index suitable for comparing sets of measures. In addition to the features extracted from typological data bases, we propose an automatic text-based measure, which can be used as a means of overcoming the well-known problem of data sparsity in manually collected features. Our diversity score is interpretable in terms of linguistic features and can identify the types of languages that are not represented in a data set. Using our method, we analyse a range of popular multilingual data sets (UD, Bible100, mBERT, XTREME, XGLUE, XNLI, XCOPA, TyDiQA, XQuAD). In addition to ranking these data sets, we find, for example, that (poly)synthetic languages are missing in almost all of them.
Authors: Jiatong Li, Wei Liu, Zhihao Ding, Wenqi Fan, Yuqiang Li, Qing Li
Abstract: Large Language Models (LLMs) have demonstrated exceptional performance in biochemical tasks, especially the molecule caption translation task, which aims to bridge the gap between molecules and natural language texts. However, previous methods in adapting LLMs to the molecule-caption translation task required extra domain-specific pre-training stages, suffered weak alignment between molecular and textual spaces, or imposed stringent demands on the scale of LLMs. To resolve the challenges, we propose In-Context Molecule Adaptation (ICMA), as a new paradigm allowing LLMs to learn the molecule-text alignment from context examples via In-Context Molecule Tuning. Specifically, ICMA incorporates the following three stages: Hybrid Context Retrieval, Post-retrieval Re-ranking, and In-context Molecule Tuning. Initially, Hybrid Context Retrieval utilizes BM25 Caption Retrieval and Molecule Graph Retrieval to retrieve informative context examples. Additionally, we also propose Post-retrieval Re-ranking with Sequence Reversal and Random Walk to further improve the quality of retrieval results. Finally, In-Context Molecule Tuning unlocks the in-context molecule learning capability of LLMs with retrieved examples and adapts the parameters of LLMs for the molecule-caption translation task. Experimental results demonstrate that ICMT can empower LLMs to achieve state-of-the-art or comparable performance without extra training corpora and intricate structures, showing that LLMs are inherently in-context molecule learners.
Authors: Watheq Mansour, Salam Albatarni, Sohaila Eltanbouly, Tamer Elsayed
Abstract: Although several methods were proposed to address the problem of automated essay scoring (AES) in the last 50 years, there is still much to desire in terms of effectiveness. Large Language Models (LLMs) are transformer-based models that demonstrate extraordinary capabilities on various tasks. In this paper, we test the ability of LLMs, given their powerful linguistic knowledge, to analyze and effectively score written essays. We experimented with two popular LLMs, namely ChatGPT and Llama. We aim to check if these models can do this task and, if so, how their performance is positioned among the state-of-the-art (SOTA) models across two levels, holistically and per individual writing trait. We utilized prompt-engineering tactics in designing four different prompts to bring their maximum potential to this task. Our experiments conducted on the ASAP dataset revealed several interesting observations. First, choosing the right prompt depends highly on the model and nature of the task. Second, the two LLMs exhibited comparable average performance in AES, with a slight advantage for ChatGPT. Finally, despite the performance gap between the two LLMs and SOTA models in terms of predictions, they provide feedback to enhance the quality of the essays, which can potentially help both teachers and students.
Authors: Akshat Gupta, Sidharth Baskaran, Gopala Anumanchipalli
Abstract: Recent work using Rank-One Model Editing (ROME), a popular model editing method, has shown that there are certain facts that the algorithm is unable to edit without breaking the model. Such edits have previously been called disabling edits. These disabling edits cause immediate model collapse and limits the use of ROME for sequential editing. In this paper, we show that disabling edits are an artifact of irregularities in the implementation of ROME. With this paper, we provide a more stable implementation ROME, which we call r-ROME and show that model collapse is no longer observed when making large scale sequential edits with r-ROME, while further improving generalization and locality of model editing compared to the original implementation of ROME. We also provide a detailed mathematical explanation of the reason behind disabling edits.
Authors: Gemma Team, Thomas Mesnard, Cassidy Hardin, Robert Dadashi, Surya Bhupatiraju, Shreya Pathak, Laurent Sifre, Morgane Rivi\`ere, Mihir Sanjay Kale, Juliette Love, Pouya Tafti, L\'eonard Hussenot, Pier Giuseppe Sessa, Aakanksha Chowdhery, Adam Roberts, Aditya Barua, Alex Botev, Alex Castro-Ros, Ambrose Slone, Am\'elie H\'eliou, Andrea Tacchetti, Anna Bulanova, Antonia Paterson, Beth Tsai, Bobak Shahriari, Charline Le Lan, Christopher A. Choquette-Choo, Cl\'ement Crepy, Daniel Cer, Daphne Ippolito, David Reid, Elena Buchatskaya, Eric Ni, Eric Noland, Geng Yan, George Tucker, George-Christian Muraru, Grigory Rozhdestvenskiy, Henryk Michalewski, Ian Tenney, Ivan Grishchenko, Jacob Austin, James Keeling, Jane Labanowski, Jean-Baptiste Lespiau, Jeff Stanway, Jenny Brennan, Jeremy Chen, Johan Ferret, Justin Chiu, Justin Mao-Jones, Katherine Lee, Kathy Yu, Katie Millican, Lars Lowe Sjoesund, Lisa Lee, Lucas Dixon, Machel Reid, Maciej Miku{\l}a, Mateo Wirth, Michael Sharman, Nikolai Chinaev, Nithum Thain, Olivier Bachem, Oscar Chang, Oscar Wahltinez, Paige Bailey, Paul Michel, Petko Yotov, Rahma Chaabouni, Ramona Comanescu, Reena Jana, Rohan Anil, Ross McIlroy, Ruibo Liu, Ryan Mullins, Samuel L Smith, Sebastian Borgeaud, Sertan Girgin, Sholto Douglas, Shree Pandya, Siamak Shakeri, Soham De, Ted Klimenko, Tom Hennigan, Vlad Feinberg, Wojciech Stokowiec, Yu-hui Chen, Zafarali Ahmed, Zhitao Gong, Tris Warkentin, Ludovic Peran, Minh Giang, Cl\'ement Farabet, Oriol Vinyals, Jeff Dean, Koray Kavukcuoglu, Demis Hassabis, Zoubin Ghahramani, Douglas Eck, Joelle Barral, Fernando Pereira, Eli Collins, Armand Joulin, Noah Fiedel, Evan Senter, Alek Andreev, Kathleen Kenealy
Abstract: This work introduces Gemma, a family of lightweight, state-of-the art open models built from the research and technology used to create Gemini models. Gemma models demonstrate strong performance across academic benchmarks for language understanding, reasoning, and safety. We release two sizes of models (2 billion and 7 billion parameters), and provide both pretrained and fine-tuned checkpoints. Gemma outperforms similarly sized open models on 11 out of 18 text-based tasks, and we present comprehensive evaluations of safety and responsibility aspects of the models, alongside a detailed description of model development. We believe the responsible release of LLMs is critical for improving the safety of frontier models, and for enabling the next wave of LLM innovations.
Authors: Chang Zong, Yuyan Chen, Weiming Lu, Jian Shao, Yueting Zhuang
Abstract: Large Language Models (LLMs) have demonstrated efficacy in various linguistic applications, including text summarization and controlled text generation. However, studies into their capacity of switching between styles via fine-tuning remain underexplored. This study concentrates on textual professionalism and introduces a novel methodology, named ProSwitch, which equips a language model with the ability to produce both professional and non-professional responses through knowledge-guided instruction tuning. ProSwitch unfolds across three phases: data preparation for gathering domain knowledge and training corpus; instruction tuning for optimizing language models with multiple levels of instruction formats; and comprehensive evaluation for assessing the professionalism discrimination and reference-based quality of generated text. Comparative analysis of ProSwitch against both general and specialized language models reveals that our approach outperforms baselines in switching between professional and non-professional text generation.
Authors: Zeyu Liu, Souvik Kundu, Anni Li, Junrui Wan, Lianghao Jiang, Peter Anthony Beerel
Abstract: We present a novel Parameter-Efficient Fine-Tuning (PEFT) method, dubbed as Adaptive Freezing of Low Rank Adaptation (AFLoRA). Specifically, for each pre-trained frozen weight tensor, we add a parallel path of trainable low-rank matrices, namely a down-projection and an up-projection matrix, each of which is followed by a feature transformation vector. Based on a novel freezing score, we the incrementally freeze these projection matrices during fine-tuning to reduce the computation and alleviate over-fitting. Our experimental results demonstrate that we can achieve state-of-the-art performance with an average improvement of up to $0.85\%$ as evaluated on GLUE benchmark while yeilding up to $9.5\times$ fewer average trainable parameters. While compared in terms of runtime, AFLoRA can yield up to $1.86\times$ improvement as opposed to similar PEFT alternatives. Besides the practical utility of our approach, we provide insights on the trainability requirements of LoRA paths at different modules and the freezing schedule for the different projection matrices. Code will be released.
Authors: Kun Sun, Rong Wang
Abstract: The majority of research in computational psycholinguistics has concentrated on the processing of words. This study introduces innovative methods for computing sentence-level metrics using multilingual large language models. The metrics developed sentence surprisal and sentence relevance and then are tested and compared to validate whether they can predict how humans comprehend sentences as a whole across languages. These metrics offer significant interpretability and achieve high accuracy in predicting human sentence reading speeds. Our results indicate that these computational sentence-level metrics are exceptionally effective at predicting and elucidating the processing difficulties encountered by readers in comprehending sentences as a whole across a variety of languages. Their impressive performance and generalization capabilities provide a promising avenue for future research in integrating LLMs and cognitive science.
Authors: Josef Valvoda, Ryan Cotterell
Abstract: Current legal outcome prediction models - a staple of legal NLP - do not explain their reasoning. However, to employ these models in the real world, human legal actors need to be able to understand the model's decisions. In the case of common law, legal practitioners reason towards the outcome of a case by referring to past case law, known as precedent. We contend that precedent is, therefore, a natural way of facilitating explainability for legal NLP models. In this paper, we contribute a novel method for identifying the precedent employed by legal outcome prediction models. Furthermore, by developing a taxonomy of legal precedent, we are able to compare human judges and neural models with respect to the different types of precedent they rely on. We find that while the models learn to predict outcomes reasonably well, their use of precedent is unlike that of human judges.
Authors: Jingyuan Ma, Damai Dai, Lei Sha, Zhifang Sui
Abstract: Large language models (LLMs) demonstrate substantial capabilities in solving math problems. However, they tend to produce hallucinations when given questions containing unreasonable errors. In this paper, we study the behavior of LLMs when faced with unreasonable math problems and further explore their potential to address these problems. We construct the Unreasonable Math Problem (UMP) benchmark to examine the error detection ability of LLMs. Experiments show that LLMs are able to detect unreasonable errors, but still fail in generating non-hallucinatory content. In order to improve their ability of error detection and correction, we further design a strategic prompt template called Critical Calculation and Conclusion(CCC). With CCC, LLMs can better self-evaluate and detect unreasonable errors in math questions, making them more reliable and safe in practical application scenarios.
Authors: Wei Chen, Zhiyuan Li
Abstract: Language models have shown effectiveness in a variety of software applications, particularly in tasks related to automatic workflow. These models possess the crucial ability to call functions, which is essential in creating AI agents. Despite the high performance of large-scale language models in cloud environments, they are often associated with concerns over privacy and cost. Current on-device models for function calling face issues with latency and accuracy. Our research presents a new method that empowers an on-device model with 2 billion parameters to surpass the performance of GPT-4 in both accuracy and latency, and decrease the context length by 95\%. When compared to Llama-7B with a RAG-based function calling mechanism, our method enhances latency by 35-fold. This method reduces the latency to levels deemed suitable for deployment across a variety of edge devices in production environments, aligning with the performance requisites for real-world applications.
Authors: Linan Yue, Qi Liu, Lili Zhao, Li Wang, Weibo Gao, Yanqing An
Abstract: With the development of legal intelligence, Criminal Court View Generation has attracted much attention as a crucial task of legal intelligence, which aims to generate concise and coherent texts that summarize case facts and provide explanations for verdicts. Existing researches explore the key information in case facts to yield the court views. Most of them employ a coarse-grained approach that partitions the facts into broad segments (e.g., verdict-related sentences) to make predictions. However, this approach fails to capture the complex details present in the case facts, such as various criminal elements and legal events. To this end, in this paper, we propose an Event Grounded Generation (EGG) method for criminal court view generation with cooperative (Large) Language Models, which introduces the fine-grained event information into the generation. Specifically, we first design a LLMs-based extraction method that can extract events in case facts without massive annotated events. Then, we incorporate the extracted events into court view generation by merging case facts and events. Besides, considering the computational burden posed by the use of LLMs in the extraction phase of EGG, we propose a LLMs-free EGG method that can eliminate the requirement for event extraction using LLMs in the inference phase. Extensive experimental results on a real-world dataset clearly validate the effectiveness of our proposed method.
Authors: Evan Shieh, Faye-Marie Vassel, Cassidy Sugimoto, Thema Monroe-White
Abstract: The rapid deployment of generative language models (LMs) has raised concerns about social biases affecting the well-being of diverse consumers. The extant literature on generative LMs has primarily examined bias via explicit identity prompting. However, prior research on bias in earlier language-based technology platforms, including search engines, has shown that discrimination can occur even when identity terms are not specified explicitly. Studies of bias in LM responses to open-ended prompts (where identity classifications are left unspecified) are lacking and have not yet been grounded in end-consumer harms. Here, we advance studies of generative LM bias by considering a broader set of natural use cases via open-ended prompting. In this "laissez-faire" setting, we find that synthetically generated texts from five of the most pervasive LMs (ChatGPT3.5, ChatGPT4, Claude2.0, Llama2, and PaLM2) perpetuate harms of omission, subordination, and stereotyping for minoritized individuals with intersectional race, gender, and/or sexual orientation identities (AI/AN, Asian, Black, Latine, MENA, NH/PI, Female, Non-binary, Queer). We find widespread evidence of bias to an extent that such individuals are hundreds to thousands of times more likely to encounter LM-generated outputs that portray their identities in a subordinated manner compared to representative or empowering portrayals. We also document a prevalence of stereotypes (e.g. perpetual foreigner) in LM-generated outputs that are known to trigger psychological harms that disproportionately affect minoritized individuals. These include stereotype threat, which leads to impaired cognitive performance and increased negative self-perception. Our findings highlight the urgent need to protect consumers from discriminatory harms caused by language models and invest in critical AI education programs tailored towards empowering diverse consumers.
Authors: Qingyi Liu, Yekun Chai, Shuohuan Wang, Yu Sun, Qiwei Peng, Keze Wang, Hua Wu
Abstract: Amidst the rapid advancements in generative language models, the investigation of how training data shapes the performance of GPT models is still emerging. This paper presents GPTfluence, a novel approach that leverages a featurized simulation to assess the impact of training examples on the training dynamics of GPT models. Our approach not only traces the influence of individual training instances on performance trajectories, such as loss and other key metrics, on targeted test points but also enables a comprehensive comparison with existing methods across various training scenarios in GPT models, ranging from 14 million to 2.8 billion parameters, across a range of downstream tasks. Contrary to earlier methods that struggle with generalization to new data, GPTfluence introduces a parameterized simulation of training dynamics, demonstrating robust generalization capabilities to unseen training data. This adaptability is evident across both fine-tuning and instruction-tuning scenarios, spanning tasks in natural language understanding and generation. We will make our code and data publicly available.
Authors: Chi Tran, Huong Le Thanh
Abstract: Large Language Models (LLMs) and Multimodal Large language models (MLLMs) have taken the world by storm with impressive abilities in complex reasoning and linguistic comprehension. Meanwhile there are plethora of works related to Vietnamese Large Language Models, the lack of high-quality resources in multimodality limits the progress of Vietnamese MLLMs. In this paper, we pioneer in address this by introducing LaVy, a state-of-the-art Vietnamese MLLM, and we also introduce LaVy-Bench benchmark designated for evaluating MLLMs's understanding on Vietnamese visual language tasks. Our project is public at https://github.com/baochi0212/LaVy
Authors: Kosuke Takahashi, Takahiro Omi, Kosuke Arima, Tatsuya Ishigaki
Abstract: Several previous studies have considered language- and domain-specific large language models (LLMs) as separate topics. This study explores the combination of a non-English language and a high-demand industry domain, focusing on a Japanese business-specific LLM. This type of a model requires expertise in the business domain, strong language skills, and regular updates of its knowledge. We trained a 13-billion-parameter LLM from scratch using a new dataset of business texts and patents, and continually pretrained it with the latest business documents. Further we propose a new benchmark for Japanese business domain question answering (QA) and evaluate our models on it. The results show that our pretrained model improves QA accuracy without losing general knowledge, and that continual pretraining enhances adaptation to new information. Our pretrained model and business domain benchmark are publicly available.
Authors: Ruixin Yang, Dheeraj Rajagopal, Shirley Anugrah Hayati, Bin Hu, Dongyeop Kang
Abstract: Uncertainty estimation is a significant issue for current large language models (LLMs) that are generally poorly calibrated and over-confident, especially with reinforcement learning from human feedback (RLHF). Unlike humans, whose decisions and confidences not only stem from intrinsic beliefs but can also be adjusted through daily observations, existing calibration methods for LLMs focus on estimating or eliciting individual confidence without taking full advantage of the "Collective Wisdom": the interaction among multiple LLMs that can collectively improve both accuracy and calibration. In this work, we propose Collaborative Calibration, a post-hoc training-free calibration strategy that leverages the collaborative and expressive capabilities of multiple tool-augmented LLM agents in a simulated group deliberation process. We demonstrate the effectiveness of Collaborative Calibration on generative QA tasks across various domains, showing its potential in harnessing the rationalization of collectively calibrated confidence assessments and improving the reliability of model predictions.
Authors: Xiaoshu Chen, Sihang Zhou, Ke Liang, Xinwang Liu
Abstract: Chain of thought finetuning aims to endow small student models with reasoning capacity to improve their performance towards a specific task by allowing them to imitate the reasoning procedure of large language models (LLMs) beyond simply predicting the answer to the question. However, the existing methods 1) generate rationale before the answer, making their answer correctness sensitive to the hallucination in the rationale;2) force the student model to repeat the exact LLMs rationale expression word-after-word, which could have the model biased towards learning the expression in rationale but count against the model from understanding the core logic behind it. Therefore, we propose a robust Post-Semantic-Thinking (PST) strategy to generate answers before rationale. Thanks to this answer-first setting, 1) the answering procedure can escape from the adverse effects caused by hallucinations in the rationale; 2) the complex reasoning procedure is tightly bound with the relatively concise answer, making the reasoning for questions easier with the prior information in the answer; 3) the efficiency of the method can also benefit from the setting since users can stop the generation right after answers are outputted when inference is conducted. Furthermore, the PST strategy loose the constraint against the generated rationale to be close to the LLMs gold standard in the hidden semantic space instead of the vocabulary space, thus making the small student model better comprehend the semantic reasoning logic in rationale. Extensive experiments conducted across 12 reasoning tasks demonstrate the effectiveness of PST.
Authors: Yuxi Li, Yi Liu, Gelei Deng, Ying Zhang, Wenjia Song, Ling Shi, Kailong Wang, Yuekang Li, Yang Liu, Haoyu Wang
Abstract: With the expanding application of Large Language Models (LLMs) in various domains, it becomes imperative to comprehensively investigate their unforeseen behaviors and consequent outcomes. In this study, we introduce and systematically explore the phenomenon of "glitch tokens", which are anomalous tokens produced by established tokenizers and could potentially compromise the models' quality of response. Specifically, we experiment on seven top popular LLMs utilizing three distinct tokenizers and involving a totally of 182,517 tokens. We present categorizations of the identified glitch tokens and symptoms exhibited by LLMs when interacting with glitch tokens. Based on our observation that glitch tokens tend to cluster in the embedding space, we propose GlitchHunter, a novel iterative clustering-based technique, for efficient glitch token detection. The evaluation shows that our approach notably outperforms three baseline methods on eight open-source LLMs. To the best of our knowledge, we present the first comprehensive study on glitch tokens. Our new detection further provides valuable insights into mitigating tokenization-related errors in LLMs.
Authors: Robik Shrestha, Kushal Kafle, Christopher Kanan
Abstract: Existing Visual Question Answering (VQA) methods tend to exploit dataset biases and spurious statistical correlations, instead of producing right answers for the right reasons. To address this issue, recent bias mitigation methods for VQA propose to incorporate visual cues (e.g., human attention maps) to better ground the VQA models, showcasing impressive gains. However, we show that the performance improvements are not a result of improved visual grounding, but a regularization effect which prevents over-fitting to linguistic priors. For instance, we find that it is not actually necessary to provide proper, human-based cues; random, insensible cues also result in similar improvements. Based on this observation, we propose a simpler regularization scheme that does not require any external annotations and yet achieves near state-of-the-art performance on VQA-CPv2.
Authors: Dan Andrei Iliescu, Devang Savita Ram Mohan, Tian Huey Teh, Zack Hodari
Abstract: We address the problem of human-in-the-loop control for generating prosody in the context of text-to-speech synthesis. Controlling prosody is challenging because existing generative models lack an efficient interface through which users can modify the output quickly and precisely. To solve this, we introduce a novel framework whereby the user provides partial inputs and the generative model generates the missing features. We propose a model that is specifically designed to encode partial prosodic features and output complete audio. We show empirically that our model displays two essential qualities of a human-in-the-loop control mechanism: efficiency and robustness. With even a very small number of input values (~4), our model enables users to improve the quality of the output significantly in terms of listener preference (4:1).
Authors: Shuyan Zhou, Frank F. Xu, Hao Zhu, Xuhui Zhou, Robert Lo, Abishek Sridhar, Xianyi Cheng, Tianyue Ou, Yonatan Bisk, Daniel Fried, Uri Alon, Graham Neubig
Abstract: With advances in generative AI, there is now potential for autonomous agents to manage daily tasks via natural language commands. However, current agents are primarily created and tested in simplified synthetic environments, leading to a disconnect with real-world scenarios. In this paper, we build an environment for language-guided agents that is highly realistic and reproducible. Specifically, we focus on agents that perform tasks on the web, and create an environment with fully functional websites from four common domains: e-commerce, social forum discussions, collaborative software development, and content management. Our environment is enriched with tools (e.g., a map) and external knowledge bases (e.g., user manuals) to encourage human-like task-solving. Building upon our environment, we release a set of benchmark tasks focusing on evaluating the functional correctness of task completions. The tasks in our benchmark are diverse, long-horizon, and designed to emulate tasks that humans routinely perform on the internet. We experiment with several baseline agents, integrating recent techniques such as reasoning before acting. The results demonstrate that solving complex tasks is challenging: our best GPT-4-based agent only achieves an end-to-end task success rate of 14.41%, significantly lower than the human performance of 78.24%. These results highlight the need for further development of robust agents, that current state-of-the-art large language models are far from perfect performance in these real-life tasks, and that WebArena can be used to measure such progress.
Authors: Albert Yu Sun, Eliott Zemour, Arushi Saxena, Udith Vaidyanathan, Eric Lin, Christian Lau, Vaikkunth Mugunthan
Abstract: Machine learning practitioners often fine-tune generative pre-trained models like GPT-3 to improve model performance at specific tasks. Previous works, however, suggest that fine-tuned machine learning models memorize and emit sensitive information from the original fine-tuning dataset. Companies such as OpenAI offer fine-tuning services for their models, but no prior work has conducted a memorization attack on any closed-source models. In this work, we simulate a privacy attack on GPT-3 using OpenAI's fine-tuning API. Our objective is to determine if personally identifiable information (PII) can be extracted from this model. We (1) explore the use of naive prompting methods on a GPT-3 fine-tuned classification model, and (2) we design a practical word generation task called Autocomplete to investigate the extent of PII memorization in fine-tuned GPT-3 within a real-world context. Our findings reveal that fine-tuning GPT3 for both tasks led to the model memorizing and disclosing critical personally identifiable information (PII) obtained from the underlying fine-tuning dataset. To encourage further research, we have made our codes and datasets publicly available on GitHub at: https://github.com/albertsun1/gpt3-pii-attacks
Authors: Forrest Huang, Gang Li, Tao Li, Yang Li
Abstract: Macros are building block tasks of our everyday smartphone activity (e.g., "login", or "booking a flight"). Effectively extracting macros is important for understanding mobile interaction and enabling task automation. These macros are however difficult to extract at scale as they can be comprised of multiple steps yet hidden within programmatic components of mobile apps. In this paper, we introduce a novel approach based on Large Language Models (LLMs) to automatically extract semantically meaningful macros from both random and user-curated mobile interaction traces. The macros produced by our approach are automatically tagged with natural language descriptions and are fully executable. We conduct multiple studies to validate the quality of extracted macros, including user evaluation, comparative analysis against human-curated tasks, and automatic execution of these macros. These experiments and analyses show the effectiveness of our approach and the usefulness of extracted macros in various downstream applications.
Authors: Eric J. Bigelow, Ekdeep Singh Lubana, Robert P. Dick, Hidenori Tanaka, Tomer D. Ullman
Abstract: Large language models (LLMs) trained on huge corpora of text datasets demonstrate intriguing capabilities, achieving state-of-the-art performance on tasks they were not explicitly trained for. The precise nature of LLM capabilities is often mysterious, and different prompts can elicit different capabilities through in-context learning. We propose a framework that enables us to analyze in-context learning dynamics to understand latent concepts underlying LLMs' behavioral patterns. This provides a more nuanced understanding than success-or-failure evaluation benchmarks, but does not require observing internal activations as a mechanistic interpretation of circuits would. Inspired by the cognitive science of human randomness perception, we use random binary sequences as context and study dynamics of in-context learning by manipulating properties of context data, such as sequence length. In the latest GPT-3.5+ models, we find emergent abilities to generate seemingly random numbers and learn basic formal languages, with striking in-context learning dynamics where model outputs transition sharply from seemingly random behaviors to deterministic repetition.
Authors: Andrew Szot, Max Schwarzer, Harsh Agrawal, Bogdan Mazoure, Walter Talbott, Katherine Metcalf, Natalie Mackraz, Devon Hjelm, Alexander Toshev
Abstract: We show that large language models (LLMs) can be adapted to be generalizable policies for embodied visual tasks. Our approach, called Large LAnguage model Reinforcement Learning Policy (LLaRP), adapts a pre-trained frozen LLM to take as input text instructions and visual egocentric observations and output actions directly in the environment. Using reinforcement learning, we train LLaRP to see and act solely through environmental interactions. We show that LLaRP is robust to complex paraphrasings of task instructions and can generalize to new tasks that require novel optimal behavior. In particular, on 1,000 unseen tasks it achieves 42% success rate, 1.7x the success rate of other common learned baselines or zero-shot applications of LLMs. Finally, to aid the community in studying language conditioned, massively multi-task, embodied AI problems we release a novel benchmark, Language Rearrangement, consisting of 150,000 training and 1,000 testing tasks for language-conditioned rearrangement. Video examples of LLaRP in unseen Language Rearrangement instructions are at https://llm-rl.github.io.
Authors: Nandan Thakur, Jianmo Ni, Gustavo Hern\'andez \'Abrego, John Wieting, Jimmy Lin, Daniel Cer
Abstract: There has been limited success for dense retrieval models in multilingual retrieval, due to uneven and scarce training data available across multiple languages. Synthetic training data generation is promising (e.g., InPars or Promptagator), but has been investigated only for English. Therefore, to study model capabilities across both cross-lingual and monolingual retrieval tasks, we develop SWIM-IR, a synthetic retrieval training dataset containing 33 (high to very-low resource) languages for fine-tuning multilingual dense retrievers without requiring any human supervision. To construct SWIM-IR, we propose SAP (summarize-then-ask prompting), where the large language model (LLM) generates a textual summary prior to the query generation step. SAP assists the LLM in generating informative queries in the target language. Using SWIM-IR, we explore synthetic fine-tuning of multilingual dense retrieval models and evaluate them robustly on three retrieval benchmarks: XOR-Retrieve (cross-lingual), MIRACL (monolingual) and XTREME-UP (cross-lingual). Our models, called SWIM-X, are competitive with human-supervised dense retrieval models, e.g., mContriever-X, finding that SWIM-IR can cheaply substitute for expensive human-labeled retrieval training data. SWIM-IR dataset and SWIM-X models are available at https://github.com/google-research-datasets/SWIM-IR.
Authors: Yuchen Zeng, Wonjun Kang, Yicong Chen, Hyung Il Koo, Kangwook Lee
Abstract: The evolution from Large Language Models (LLMs) to Multimodal Large Language Models (MLLMs) has spurred research into extending In-Context Learning (ICL) to its multimodal counterpart. Existing such studies have primarily concentrated on image-to-text ICL. However, the Text-to-Image ICL (T2I-ICL), with its unique characteristics and potential applications, remains underexplored. To address this gap, we formally define the task of T2I-ICL and present CoBSAT, the first T2I-ICL benchmark dataset, encompassing ten tasks. Utilizing our dataset to benchmark six state-of-the-art MLLMs, we uncover considerable difficulties MLLMs encounter in solving T2I-ICL. We identify the primary challenges as the inherent complexity of multimodality and image generation, and show that strategies such as fine-tuning and Chain-of-Thought prompting help to mitigate these difficulties, leading to notable improvements in performance. Our code and dataset are available at https://github.com/UW-Madison-Lee-Lab/CoBSAT.
Authors: Syed Mekael Wasti, Ken Q. Pu, Ali Neshati
Abstract: The evolution of Large Language Models (LLMs) has showcased remarkable capacities for logical reasoning and natural language comprehension. These capabilities can be leveraged in solutions that semantically and textually model complex problems. In this paper, we present our efforts toward constructing a framework that can serve as an intermediary between a user and their user interface (UI), enabling dynamic and real-time interactions. We employ a system that stands upon textual semantic mappings of UI components, in the form of annotations. These mappings are stored, parsed, and scaled in a custom data structure, supplementary to an agent-based prompting backend engine. Employing textual semantic mappings allows each component to not only explain its role to the engine but also provide expectations. By comprehending the needs of both the user and the components, our LLM engine can classify the most appropriate application, extract relevant parameters, and subsequently execute precise predictions of the user's expected actions. Such an integration evolves static user interfaces into highly dynamic and adaptable solutions, introducing a new frontier of intelligent and responsive user experiences.
Authors: Ziyi Guan, Hantao Huang, Yupeng Su, Hong Huang, Ngai Wong, Hao Yu
Abstract: Large Language Models (LLMs) have greatly advanced the natural language processing paradigm. However, the high computational load and huge model sizes pose a grand challenge for deployment on edge devices. To this end, we propose APTQ (Attention-aware Post-Training Mixed-Precision Quantization) for LLMs, which considers not only the second-order information of each layer's weights, but also, for the first time, the nonlinear effect of attention outputs on the entire model. We leverage the Hessian trace as a sensitivity metric for mixed-precision quantization, ensuring an informed precision reduction that retains model performance. Experiments show APTQ surpasses previous quantization methods, achieving an average of 4 bit width a 5.22 perplexity nearly equivalent to full precision in the C4 dataset. In addition, APTQ attains state-of-the-art zero-shot accuracy of 68.24\% and 70.48\% at an average bitwidth of 3.8 in LLaMa-7B and LLaMa-13B, respectively, demonstrating its effectiveness to produce high-quality quantized LLMs.
Authors: Yichi Zhang, Ziqiao Ma, Xiaofeng Gao, Suhaila Shakiah, Qiaozi Gao, Joyce Chai
Abstract: Most multimodal large language models (MLLMs) learn language-to-object grounding through causal language modeling where grounded objects are captured by bounding boxes as sequences of location tokens. This paradigm lacks pixel-level representations that are important for fine-grained visual understanding and diagnosis. In this work, we introduce GROUNDHOG, an MLLM developed by grounding Large Language Models to holistic segmentation. GROUNDHOG incorporates a masked feature extractor and converts extracted features into visual entity tokens for the MLLM backbone, which then connects groundable phrases to unified grounding masks by retrieving and merging the entity masks. To train GROUNDHOG, we carefully curated M3G2, a grounded visual instruction tuning dataset with Multi-Modal Multi-Grained Grounding, by harvesting a collection of segmentation-grounded datasets with rich annotations. Our experimental results show that GROUNDHOG achieves superior performance on various language grounding tasks without task-specific fine-tuning, and significantly reduces object hallucination. GROUNDHOG also demonstrates better grounding towards complex forms of visual input and provides easy-to-understand diagnosis in failure cases.
Authors: Gang Hu, Ke Qin, Chenhan Yuan, Min Peng, Alejandro Lopez-Lira, Benyou Wang, Sophia Ananiadou, Wanlong Yu, Jimin Huang, Qianqian Xie
Abstract: While the progression of Large Language Models (LLMs) has notably propelled financial analysis, their application has largely been confined to singular language realms, leaving untapped the potential of bilingual Chinese-English capacity. To bridge this chasm, we introduce ICE-PIXIU, seamlessly amalgamating the ICE-INTENT model and ICE-FLARE benchmark for bilingual financial analysis. ICE-PIXIU uniquely integrates a spectrum of Chinese tasks, alongside translated and original English datasets, enriching the breadth and depth of bilingual financial modeling. It provides unrestricted access to diverse model variants, a substantial compilation of diverse cross-lingual and multi-modal instruction data, and an evaluation benchmark with expert annotations, comprising 10 NLP tasks, 20 bilingual specific tasks, totaling 95k datasets. Our thorough evaluation emphasizes the advantages of incorporating these bilingual datasets, especially in translation tasks and utilizing original English data, enhancing both linguistic flexibility and analytical acuity in financial contexts. Notably, ICE-INTENT distinguishes itself by showcasing significant enhancements over conventional LLMs and existing financial LLMs in bilingual milieus, underscoring the profound impact of robust bilingual data on the accuracy and efficacy of financial NLP.
Authors: Smriti Singh, Shuvam Keshari, Vinija Jain, Aman Chadha
Abstract: Socioeconomic bias in society exacerbates disparities, influencing access to opportunities and resources based on individuals' economic and social backgrounds. This pervasive issue perpetuates systemic inequalities, hindering the pursuit of inclusive progress as a society. In this paper, we investigate the presence of socioeconomic bias, if any, in large language models. To this end, we introduce a novel dataset SilverSpoon, consisting of 3000 samples that illustrate hypothetical scenarios that involve underprivileged people performing ethically ambiguous actions due to their circumstances, and ask whether the action is ethically justified. Further, this dataset has a dual-labeling scheme and has been annotated by people belonging to both ends of the socioeconomic spectrum. Using SilverSpoon, we evaluate the degree of socioeconomic bias expressed in large language models and the variation of this degree as a function of model size. We also perform qualitative analysis to analyze the nature of this bias. Our analysis reveals that while humans disagree on which situations require empathy toward the underprivileged, most large language models are unable to empathize with the socioeconomically underprivileged regardless of the situation. To foster further research in this domain, we make SilverSpoon and our evaluation harness publicly available.
Authors: Santiago Cuervo, Ricard Marxer
Abstract: Speech Language Models (SLMs) aim to learn language from raw audio, without textual resources. Despite significant advances, our current models exhibit weak syntax and semantic abilities. However, if the scaling properties of neural language models hold for the speech modality, these abilities will improve as the amount of compute used for training increases. In this paper, we use models of this scaling behavior to estimate the scale at which our current methods will yield a SLM with the English proficiency of text-based Large Language Models (LLMs). We establish a strong correlation between pre-training loss and downstream syntactic and semantic performance in SLMs and LLMs, which results in predictable scaling of linguistic performance. We show that the linguistic performance of SLMs scales up to three orders of magnitude more slowly than that of text-based LLMs. Additionally, we study the benefits of synthetic data designed to boost semantic understanding and the effects of coarser speech tokenization.
Authors: Alexander Nemecek, Yuzhou Jiang, Erman Ayday
Abstract: Recent advancements of large language models (LLMs) have resulted in indistinguishable text outputs comparable to human-generated text. Watermarking algorithms are potential tools that offer a way to differentiate between LLM- and human-generated text by embedding detectable signatures within LLM-generated output. However, current watermarking schemes lack robustness against known attacks against watermarking algorithms. In addition, they are impractical considering an LLM generates tens of thousands of text outputs per day and the watermarking algorithm needs to memorize each output it generates for the detection to work. In this work, focusing on the limitations of current watermarking schemes, we propose the concept of a "topic-based watermarking algorithm" for LLMs. The proposed algorithm determines how to generate tokens for the watermarked LLM output based on extracted topics of an input prompt or the output of a non-watermarked LLM. Inspired from previous work, we propose using a pair of lists (that are generated based on the specified extracted topic(s)) that specify certain tokens to be included or excluded while generating the watermarked output of the LLM. Using the proposed watermarking algorithm, we show the practicality of a watermark detection algorithm. Furthermore, we discuss a wide range of attacks that can emerge against watermarking algorithms for LLMs and the benefit of the proposed watermarking scheme for the feasibility of modeling a potential attacker considering its benefit vs. loss.
Authors: Amir H. Abdi, Xinye Tang, Jeremias Eichelbaum, Mahan Das, Alex Klein, Nihal Irmak Pakis, William Blum, Daniel L Mace, Tanvi Raja, Namrata Padmanabhan, Ye Xing
Abstract: Data is growing rapidly in volume and complexity. Proficiency in database query languages is pivotal for crafting effective queries. As coding assistants become more prevalent, there is significant opportunity to enhance database query languages. The Kusto Query Language (KQL) is a widely used query language for large semi-structured data such as logs, telemetries, and time-series for big data analytics platforms. This paper introduces NL2KQL an innovative framework that uses large language models (LLMs) to convert natural language queries (NLQs) to KQL queries. The proposed NL2KQL framework includes several key components: Schema Refiner which narrows down the schema to its most pertinent elements; the Few-shot Selector which dynamically selects relevant examples from a few-shot dataset; and the Query Refiner which repairs syntactic and semantic errors in KQL queries. Additionally, this study outlines a method for generating large datasets of synthetic NLQ-KQL pairs which are valid within a specific database contexts. To validate NL2KQL's performance, we utilize an array of online (based on query execution) and offline (based on query parsing) metrics. Through ablation studies, the significance of each framework component is examined, and the datasets used for benchmarking are made publicly available. This work is the first of its kind and is compared with available baselines to demonstrate its effectiveness.
Authors: Jonathan D. Chang, Wenhao Zhan, Owen Oertell, Kiant\'e Brantley, Dipendra Misra, Jason D. Lee, Wen Sun
Abstract: Reinforcement Learning (RL) from Human Preference-based feedback is a popular paradigm for fine-tuning generative models, which has produced impressive models such as GPT-4 and Claude3 Opus. This framework often consists of two steps: learning a reward model from an offline preference dataset followed by running online RL to optimize the learned reward model. In this work, leveraging the idea of reset, we propose a new RLHF algorithm with provable guarantees. Motivated by the fact that offline preference dataset provides informative states (i.e., data that is preferred by the labelers), our new algorithm, Dataset Reset Policy Optimization (DR-PO), integrates the existing offline preference dataset into the online policy training procedure via dataset reset: it directly resets the policy optimizer to the states in the offline dataset, instead of always starting from the initial state distribution. In theory, we show that DR-PO learns to perform at least as good as any policy that is covered by the offline dataset under general function approximation with finite sample complexity. In experiments, we demonstrate that on both the TL;DR summarization and the Anthropic Helpful Harmful (HH) dataset, the generation from DR-PO is better than that from Proximal Policy Optimization (PPO) and Direction Preference Optimization (DPO), under the metric of GPT4 win-rate. Code for this work can be found at https://github.com/Cornell-RL/drpo.
Authors: Shreyas Chaudhari, Pranjal Aggarwal, Vishvak Murahari, Tanmay Rajpurohit, Ashwin Kalyan, Karthik Narasimhan, Ameet Deshpande, Bruno Castro da Silva
Abstract: State-of-the-art large language models (LLMs) have become indispensable tools for various tasks. However, training LLMs to serve as effective assistants for humans requires careful consideration. A promising approach is reinforcement learning from human feedback (RLHF), which leverages human feedback to update the model in accordance with human preferences and mitigate issues like toxicity and hallucinations. Yet, an understanding of RLHF for LLMs is largely entangled with initial design choices that popularized the method and current research focuses on augmenting those choices rather than fundamentally improving the framework. In this paper, we analyze RLHF through the lens of reinforcement learning principles to develop an understanding of its fundamentals, dedicating substantial focus to the core component of RLHF -- the reward model. Our study investigates modeling choices, caveats of function approximation, and their implications on RLHF training algorithms, highlighting the underlying assumptions made about the expressivity of reward. Our analysis improves the understanding of the role of reward models and methods for their training, concurrently revealing limitations of the current methodology. We characterize these limitations, including incorrect generalization, model misspecification, and the sparsity of feedback, along with their impact on the performance of a language model. The discussion and analysis are substantiated by a categorical review of current literature, serving as a reference for researchers and practitioners to understand the challenges of RLHF and build upon existing efforts.
Authors: Xuezhe Ma, Xiaomeng Yang, Wenhan Xiong, Beidi Chen, Lili Yu, Hao Zhang, Jonathan May, Luke Zettlemoyer, Omer Levy, Chunting Zhou
Abstract: The quadratic complexity and weak length extrapolation of Transformers limits their ability to scale to long sequences, and while sub-quadratic solutions like linear attention and state space models exist, they empirically underperform Transformers in pretraining efficiency and downstream task accuracy. We introduce Megalodon, a neural architecture for efficient sequence modeling with unlimited context length. Megalodon inherits the architecture of Mega (exponential moving average with gated attention), and further introduces multiple technical components to improve its capability and stability, including complex exponential moving average (CEMA), timestep normalization layer, normalized attention mechanism and pre-norm with two-hop residual configuration. In a controlled head-to-head comparison with Llama2, Megalodon achieves better efficiency than Transformer in the scale of 7 billion parameters and 2 trillion training tokens. Megalodon reaches a training loss of 1.70, landing mid-way between Llama2-7B (1.75) and 13B (1.67). Code: https://github.com/XuezheMax/megalodon
Authors: Francis McCann Ramirez, Luka Chkhetiani, Andrew Ehrenberg, Robert McHardy, Rami Botros, Yash Khare, Andrea Vanzo, Taufiquzzaman Peyash, Gabriel Oexle, Michael Liang, Ilya Sklyar, Enver Fakhan, Ahmed Etefy, Daniel McCrystal, Sam Flamini, Domenic Donato, Takuya Yoshioka
Abstract: This paper describes AssemblyAI's industrial-scale automatic speech recognition (ASR) system, designed to meet the requirements of large-scale, multilingual ASR serving various application needs. Our system leverages a diverse training dataset comprising unsupervised (12.5M hours), supervised (188k hours), and pseudo-labeled (1.6M hours) data across four languages. We provide a detailed description of our model architecture, consisting of a full-context 600M-parameter Conformer encoder pre-trained with BEST-RQ and an RNN-T decoder fine-tuned jointly with the encoder. Our extensive evaluation demonstrates competitive word error rates (WERs) against larger and more computationally expensive models, such as Whisper large and Canary-1B. Furthermore, our architectural choices yield several key advantages, including an improved code-switching capability, a 5x inference speedup compared to an optimized Whisper baseline, a 30% reduction in hallucination rate on speech data, and a 90% reduction in ambient noise compared to Whisper, along with significantly improved time-stamp accuracy. Throughout this work, we adopt a system-centric approach to analyzing various aspects of fully-fledged ASR models to gain practically relevant insights useful for real-world services operating at scale.
Authors: Navonil Majumder, Chia-Yu Hung, Deepanway Ghosal, Wei-Ning Hsu, Rada Mihalcea, Soujanya Poria
Abstract: Generative multimodal content is increasingly prevalent in much of the content creation arena, as it has the potential to allow artists and media personnel to create pre-production mockups by quickly bringing their ideas to life. The generation of audio from text prompts is an important aspect of such processes in the music and film industry. Many of the recent diffusion-based text-to-audio models focus on training increasingly sophisticated diffusion models on a large set of datasets of prompt-audio pairs. These models do not explicitly focus on the presence of concepts or events and their temporal ordering in the output audio with respect to the input prompt. Our hypothesis is focusing on how these aspects of audio generation could improve audio generation performance in the presence of limited data. As such, in this work, using an existing text-to-audio model Tango, we synthetically create a preference dataset where each prompt has a winner audio output and some loser audio outputs for the diffusion model to learn from. The loser outputs, in theory, have some concepts from the prompt missing or in an incorrect order. We fine-tune the publicly available Tango text-to-audio model using diffusion-DPO (direct preference optimization) loss on our preference dataset and show that it leads to improved audio output over Tango and AudioLDM2, in terms of both automatic- and manual-evaluation metrics.