Authors: Muhammad Aurangzeb Ahmad, Ilker Yaramis, Taposh Dutta Roy
Large language models have proliferated across multiple domains in as short period of time. There is however hesitation in the medical and healthcare domain towards their adoption because of issues like factuality, coherence, and hallucinations. Give the high stakes nature of healthcare, many researchers have even cautioned against its usage until these issues are resolved. The key to the implementation and deployment of LLMs in healthcare is to make these models trustworthy, transparent (as much possible) and explainable. In this paper we describe the key elements in creating reliable, trustworthy, and unbiased models as a necessary condition for their adoption in healthcare. Specifically we focus on the quantification, validation, and mitigation of hallucinations in the context in healthcare. Lastly, we discuss how the future of LLMs in healthcare may look like.
Authors: Manuel R. Ciosici, Alex Hedges, Yash Kankanampati, Justin Martin, Marjorie Freedman, Ralph Weischedel
We explore using a moderately sized large language model (GPT-J 6B parameters) to create a plan for a simulated robot to achieve 30 classes of goals in ScienceWorld, a text game simulator for elementary science experiments. Previously published empirical work claimed that large language models (LLMs) are a poor fit (Wang et al., 2022) compared to reinforcement learning. Using the Markov assumption (a single previous step), the LLM outperforms the reinforcement learning-based approach by a factor of 1.4. When we fill the LLM's input buffer with as many prior steps as possible, improvement rises to 3.5x. Even when training on only 6.5% of the training data, we observe a 2.2x improvement over the reinforcement-learning-based approach. Our experiments show that performance varies widely across the 30 classes of actions, indicating that averaging over tasks can hide significant performance issues. In work contemporaneous with ours, Lin et al. (2023) demonstrated a two-part approach (SwiftSage) that uses a small LLM (T5-large) complemented by OpenAI's massive LLMs to achieve outstanding results in ScienceWorld. Our 6-B parameter, single-stage GPT-J matches the performance of SwiftSage's two-stage architecture when it incorporates GPT-3.5 turbo which has 29-times more parameters than GPT-J.
Authors: Avalon Vinella, Margaret Capetz, Rebecca Pattichis, Christina Chance, Reshmi Ghosh
In recent years, climate change repercussions have increasingly captured public interest. Consequently, corporations are emphasizing their environmental efforts in sustainability reports to bolster their public image. Yet, the absence of stringent regulations in review of such reports allows potential greenwashing. In this study, we introduce a novel methodology to train a language model on generated labels for greenwashing risk. Our primary contributions encompass: developing a mathematical formulation to quantify greenwashing risk, a fine-tuned ClimateBERT model for this problem, and a comparative analysis of results. On a test set comprising of sustainability reports, our best model achieved an average accuracy score of 86.34% and F1 score of 0.67, demonstrating that our methods show a promising direction of exploration for this task.
Authors: Jaeff Hong, Duong Dung, Danielle Hutchinson, Zubair Akhtar, Rosalie Chen, Rebecca Dawson, Aditya Joshi, Samsung Lim, C Raina MacIntyre, Deepti Gurdasani
Relation Extraction from News Articles (RENA) is a browser-based tool designed to extract key entities and their semantic relationships in English language news articles related to infectious diseases. Constructed using the React framework, this system presents users with an elegant and user-friendly interface. It enables users to input a news article and select from a choice of two models to generate a comprehensive list of relations within the provided text. As a result, RENA allows real-time parsing of news articles to extract key information for epidemic surveillance, contributing to EPIWATCH, an open-source intelligence-based epidemic warning system.
Authors: Yifan Du, Hangyu Guo, Kun Zhou, Wayne Xin Zhao, Jinpeng Wang, Chuyuan Wang, Mingchen Cai, Ruihua Song, Ji-Rong Wen
Visual instruction tuning is an essential approach to improving the zero-shot generalization capability of Multi-modal Large Language Models (MLLMs). A surge of visual instruction datasets with various focuses and characteristics have been proposed recently, enabling MLLMs to achieve surprising results on evaluation benchmarks. To develop more capable MLLMs, in this paper, we aim to investigate a more fundamental question: ``what makes for good visual instructions?''. By conducting a comprehensive empirical study, we find that instructions focused on complex visual reasoning tasks are particularly effective in improving the performance of MLLMs on evaluation benchmarks. Building upon this finding, we design a systematic approach to automatically creating high-quality complex visual reasoning instructions. Our approach employs a synthesis-complication-reformulation paradigm, leveraging multiple stages to gradually increase the complexity of the instructions while guaranteeing quality. Based on this approach, we create the synthetic visual reasoning instruction dataset consisting of 32K examples, namely ComVint, and fine-tune four MLLMs on it. Experimental results demonstrate that our dataset consistently enhances the performance of all the compared MLLMs, e.g., improving the performance of MiniGPT-4 and BLIP-2 on MME-Cognition by 32.6% and 28.8%, respectively. Our code and data are publicly available at the link: https://github.com/RUCAIBox/ComVint.
Authors: Björn Deiseroth, Max Meuer, Nikolas Gritsch, Constantin Eichenberg, Patrick Schramowski, Matthias Aßenmacher, Kristian Kersting
Large Language Models (LLMs) have reshaped natural language processing with their impressive capabilities. Their ever-increasing size, however, raised concerns about their effective deployment and the need for LLM compressions. This study introduces the Divergent Token metrics (DTMs), a novel approach for assessing compressed LLMs, addressing the limitations of traditional measures like perplexity that fail to accurately reflect text generation quality. DTMs focus on token divergence, providing deeper insights into the subtleties of model compression. Our results indicate that significant levels of precision and sparsity can be achieved without compromising text generation quality. Moreover, DTMs offers a more precise evaluation of each component's impact individually. Utilizing the First Divergent Token metric (FDTM) in model sparsification reveals that nearly 20% of all components can be pruned over 90%. In terms of quantization, the FDTM suggests that over 80% of parameters can be straightforwardly transformed to int8 without special outlier management.
Authors: Weiwei Sun, Zheng Chen, Xinyu Ma, Lingyong Yan, Shuaiqiang Wang, Pengjie Ren, Zhumin Chen, Dawei Yin, Zhaochun Ren
Recent studies have demonstrated the great potential of Large Language Models (LLMs) serving as zero-shot relevance rankers. The typical approach involves making comparisons between pairs or lists of documents. Although effective, these listwise and pairwise methods are not efficient and also heavily rely on intricate prompt engineering. To tackle this problem, we introduce a novel instruction distillation method. The key idea is to distill the pairwise ranking ability of open-sourced LLMs to a simpler but more efficient pointwise ranking. Specifically, given the same LLM, we first rank documents using the effective pairwise approach with complex instructions, and then distill the teacher predictions to the pointwise approach with simpler instructions. Evaluation results on the BEIR, TREC, and ReDial datasets demonstrate that instruction distillation can improve efficiency by 10 to 100x and also enhance the ranking performance of LLMs. Furthermore, our approach surpasses the performance of existing supervised methods like monoT5 and is on par with the state-of-the-art zero-shot methods. The code to reproduce our results is available at www.github.com/sunnweiwei/RankGPT.
Authors: Mohammad Junayed Hasan, Suhra Noor, Mohammad Ashrafuzzaman Khan
Clinical texts, such as admission notes, discharge summaries, and progress notes, contain rich and valuable information that can be used for various clinical outcome prediction tasks. However, applying large language models, such as BERT-based models, to clinical texts poses two major challenges: the limitation of input length and the diversity of data sources. This paper proposes a novel method to preserve the knowledge of long clinical texts using aggregated ensembles of large language models. Unlike previous studies which use model ensembling or text aggregation methods separately, we combine ensemble learning with text aggregation and train multiple large language models on two clinical outcome tasks: mortality prediction and length of stay prediction. We show that our method can achieve better results than baselines, ensembling, and aggregation individually, and can improve the performance of large language models while handling long inputs and diverse datasets. We conduct extensive experiments on the admission notes from the MIMIC-III clinical database by combining multiple unstructured and high-dimensional datasets, demonstrating our method's effectiveness and superiority over existing approaches. We also provide a comprehensive analysis and discussion of our results, highlighting our method's applications and limitations for future research in the domain of clinical healthcare. The results and analysis of this study is supportive of our method assisting in clinical healthcare systems by enabling clinical decision-making with robust performance overcoming the challenges of long text inputs and varied datasets.
Authors: Guangyue Xu, Parisa Kordjamshidi, Joyce Chai
Humans have the ability to learn novel compositional concepts by recalling and generalizing primitive concepts acquired from past experiences. Inspired by this observation, in this paper, we propose MetaReVision, a retrieval-enhanced meta-learning model to address the visually grounded compositional concept learning problem. The proposed MetaReVision consists of a retrieval module and a meta-learning module which are designed to incorporate retrieved primitive concepts as a supporting set to meta-train vision-anguage models for grounded compositional concept recognition. Through meta-learning from episodes constructed by the retriever, MetaReVision learns a generic compositional representation that can be fast updated to recognize novel compositional concepts. We create CompCOCO and CompFlickr to benchmark the grounded compositional concept learning. Our experimental results show that MetaReVision outperforms other competitive baselines and the retrieval module plays an important role in this compositional learning process.
Authors: Gianluigi Lopardo, Frederic Precioso, Damien Garreau
Interpretability is essential for machine learning models to be trusted and deployed in critical domains. However, existing methods for interpreting text models are often complex, lack solid mathematical foundations, and their performance is not guaranteed. In this paper, we propose FRED (Faithful and Robust Explainer for textual Documents), a novel method for interpreting predictions over text. FRED identifies key words in a document that significantly impact the prediction when removed. We establish the reliability of FRED through formal definitions and theoretical analyses on interpretable classifiers. Additionally, our empirical evaluation against state-of-the-art methods demonstrates the effectiveness of FRED in providing insights into text models.
Authors: Gökberk Özsoy, Luis Salamanca, Matthew Connelly, Raymond Hicks, Fernando Pérez-Cruz
In the current paper, we present the KG-FRUS dataset, comprised of more than 300,000 US government diplomatic documents encoded in a Knowledge Graph (KG). We leverage the data of the Foreign Relations of the United States (FRUS) (available as XML files) to extract information about the documents and the individuals and countries mentioned within them. We use the extracted entities, and associated metadata, to create a graph-based dataset. Further, we supplement the created KG with additional entities and relations from Wikidata. The relations in the KG capture the synergies and dynamics required to study and understand the complex fields of diplomacy, foreign relations, and politics. This goes well beyond a simple collection of documents which neglects the relations between entities in the text. We showcase a range of possibilities of the current dataset by illustrating different approaches to probe the KG. In the paper, we exemplify how to use a query language to answer simple research questions and how to use graph algorithms such as Node2Vec and PageRank, that benefit from the complete graph structure. More importantly, the chosen structure provides total flexibility for continuously expanding and enriching the graph. Our solution is general, so the proposed pipeline for building the KG can encode other original corpora of time-dependent and complex phenomena. Overall, we present a mechanism to create KG databases providing a more versatile representation of time-dependent related text data and a particular application to the all-important FRUS database.
Authors: Ching-Feng Yeh, Po-Yao Huang, Vasu Sharma, Shang-Wen Li, Gargi Gosh
We propose Fast Language-Audio Pre-training (FLAP), a self-supervised approach that efficiently and effectively learns aligned audio and language representations through masking, contrastive learning and reconstruction. For efficiency, FLAP randomly drops audio spectrogram tokens, focusing solely on the remaining ones for self-supervision. Through inter-modal contrastive learning, FLAP learns to align paired audio and text representations in a shared latent space. Notably, FLAP leverages multiple augmented views via masking for inter-modal contrast and learns to reconstruct the masked portion of audio tokens. Moreover, FLAP leverages large language models (LLMs) to augment the text inputs, contributing to improved performance. These approaches lead to more robust and informative audio-text representations, enabling FLAP to achieve state-of-the-art (SoTA) performance on audio-text retrieval tasks on AudioCaps (achieving 53.0% R@1) and Clotho (achieving 25.5% R@1).
Authors: Te-Lin Wu, Zi-Yi Dou, Qingyuan Hu, Yu Hou, Nischal Reddy Chandra, Marjorie Freedman, Ralph M. Weischedel, Nanyun Peng
Multimodal counterfactual reasoning is a vital yet challenging ability for AI systems. It involves predicting the outcomes of hypothetical circumstances based on vision and language inputs, which enables AI models to learn from failures and explore hypothetical scenarios. Despite its importance, there are only a few datasets targeting the counterfactual reasoning abilities of multimodal models. Among them, they only cover reasoning over synthetic environments or specific types of events (e.g. traffic collisions), making them hard to reliably benchmark the model generalization ability in diverse real-world scenarios and reasoning dimensions. To overcome these limitations, we develop a video question answering dataset, ACQUIRED: it consists of 3.9K annotated videos, encompassing a wide range of event types and incorporating both first and third-person viewpoints, which ensures a focus on real-world diversity. In addition, each video is annotated with questions that span three distinct dimensions of reasoning, including physical, social, and temporal, which can comprehensively evaluate the model counterfactual abilities along multiple aspects. We benchmark our dataset against several state-of-the-art language-only and multimodal models and experimental results demonstrate a significant performance gap (>13%) between models and humans. The findings suggest that multimodal counterfactual reasoning remains an open challenge and ACQUIRED is a comprehensive and reliable benchmark for inspiring future research in this direction.
Authors: Shan Yu, Zhenting Zhu, Yu Chen, Hanchen Xu, Pengzhan Zhao, Yang Wang, Arthi Padmanabhan, Hugo Latapie, Harry Xu
Video analytics is widely used in contemporary systems and services. At the forefront of video analytics are video queries that users develop to find objects of particular interest. Building upon the insight that video objects (e.g., human, animals, cars, etc.), the center of video analytics, are similar in spirit to objects modeled by traditional object-oriented languages, we propose to develop an object-oriented approach to video analytics. This approach, named VQPy, consists of a frontend$\unicode{x2015}$a Python variant with constructs that make it easy for users to express video objects and their interactions$\unicode{x2015}$as well as an extensible backend that can automatically construct and optimize pipelines based on video objects. We have implemented and open-sourced VQPy, which has been productized in Cisco as part of its DeepVision framework.
Authors: Halim Cagri Ates, Shruti Bhargava, Site Li, Jiarui Lu, Siddhardha Maddula, Joel Ruben Antony Moniz, Anil Kumar Nalamalapu, Roman Hoang Nguyen, Melis Ozyildirim, Alkesh Patel, Dhivya Piraviperumal, Vincent Renkens, Ankit Samal, Thy Tran, Bo-Hsiang Tseng, Hong Yu, Yuan Zhang, Rong Zou
Successfully handling context is essential for any dialog understanding task. This context maybe be conversational (relying on previous user queries or system responses), visual (relying on what the user sees, for example, on their screen), or background (based on signals such as a ringing alarm or playing music). In this work, we present an overview of MARRS, or Multimodal Reference Resolution System, an on-device framework within a Natural Language Understanding system, responsible for handling conversational, visual and background context. In particular, we present different machine learning models to enable handing contextual queries; specifically, one to enable reference resolution, and one to handle context via query rewriting. We also describe how these models complement each other to form a unified, coherent, lightweight system that can understand context while preserving user privacy.
Authors: Shicheng Xu, Liang Pang, Jiangnan Li, Mo Yu, Fandong Meng, Huawei Shen, Xueqi Cheng, Jie Zhou
Retrieving relevant plots from the book for a query is a critical task, which can improve the reading experience and efficiency of readers. Readers usually only give an abstract and vague description as the query based on their own understanding, summaries, or speculations of the plot, which requires the retrieval model to have a strong ability to estimate the abstract semantic associations between the query and candidate plots. However, existing information retrieval (IR) datasets cannot reflect this ability well. In this paper, we propose Plot Retrieval, a labeled dataset to train and evaluate the performance of IR models on the novel task Plot Retrieval. Text pairs in Plot Retrieval have less word overlap and more abstract semantic association, which can reflect the ability of the IR models to estimate the abstract semantic association, rather than just traditional lexical or semantic matching. Extensive experiments across various lexical retrieval, sparse retrieval, dense retrieval, and cross-encoder methods compared with human studies on Plot Retrieval show current IR models still struggle in capturing abstract semantic association between texts. Plot Retrieval can be the benchmark for further research on the semantic association modeling ability of IR models.
Authors: Jiao Ou, Junda Lu, Che Liu, Yihong Tang, Fuzheng Zhang, Di Zhang, Zhongyuan Wang, Kun Gai
Large language models (LLMs) have achieved remarkable breakthroughs in new dialogue capabilities, refreshing human's impressions on dialogue systems. The long-standing goal of dialogue systems is to be human-like enough to establish long-term connections with users by satisfying the need for communication, affection and social belonging. Therefore, there has been an urgent need to evaluate LLMs as human-like dialogue systems. In this paper, we propose DialogBench, a dialogue evaluation benchmark that currently contains $12$ dialogue tasks to assess the capabilities of LLMs as human-like dialogue systems should have. Specifically, we prompt GPT-4 to generate evaluation instances for each task. We first design the basic prompt based on widely-used design principles and further mitigate the existing biases to generate higher-quality evaluation instances. Our extensive test over $28$ LLMs (including pre-trained and supervised instruction-tuning) shows that instruction fine-tuning benefits improve the human likeness of LLMs to a certain extent, but there is still much room to improve those capabilities for most LLMs as human-like dialogue systems. In addition, experimental results also indicate that LLMs perform differently in various abilities that human-like dialogue systems should have. We will publicly release DialogBench, along with the associated evaluation code for the broader research community.
Authors: Wenkai Chen, Sahithya Ravi, Vered Shwartz
LLMs have demonstrated impressive zero-shot performance on NLP tasks thanks to the knowledge they acquired in their training. In multiple-choice QA tasks, the LM probabilities are used as an imperfect measure of the plausibility of each answer choice. One of the major limitations of the basic score is that it treats all words as equally important. We propose CASE, a Commonsense-Augmented Score with an Expanded Answer Space. CASE addresses this limitation by assigning importance weights for individual words based on their semantic relations to other words in the input. The dynamic weighting approach outperforms basic LM scores, not only because it reduces noise from unimportant words, but also because it informs the model of implicit commonsense knowledge that may be useful for answering the question. We then also follow prior work in expanding the answer space by generating lexically-divergent answers that are conceptually-similar to the choices. When combined with answer space expansion, our method outperforms strong baselines on 5 commonsense benchmarks. We further show these two approaches are complementary and may be especially beneficial when using smaller LMs.
Authors: Zheyuan Bai, Xinduo Liu, Hailin Hu, Tianyu Guo, Qinghua Zhang, Yunhe Wang
Data-Free Knowledge Distillation (DFKD) plays a vital role in compressing the model when original training data is unavailable. Previous works for DFKD in NLP mainly focus on distilling encoder-only structures like BERT on classification tasks, which overlook the notable progress of generative language modeling. In this work, we propose a novel DFKD framework, namely DFKD-T$^{3}$, where the pretrained generative language model can also serve as a controllable data generator for model compression. This novel framework DFKD-T$^{3}$ leads to an end-to-end learnable text-to-text framework to transform the general domain corpus to compression-friendly task data, targeting to improve both the \textit{specificity} and \textit{diversity}. Extensive experiments show that our method can boost the distillation performance in various downstream tasks such as sentiment analysis, linguistic acceptability, and information extraction. Furthermore, we show that the generated texts can be directly used for distilling other language models and outperform the SOTA methods, making our method more appealing in a general DFKD setting. Our code is available at https://gitee.com/mindspore/models/tree/master/research/nlp/DFKD\_T3.
Authors: Beomjune Kim, Eunsun Lee, Dongbin Na
Many users reading online articles in various magazines may suffer considerable difficulty in distinguishing the implicit intents in texts. In this work, we focus on automatically recognizing the political intents of a given online newspaper by understanding the context of the text. To solve this task, we present a novel Korean text classification dataset that contains various articles. We also provide deep-learning-based text classification baseline models trained on the proposed dataset. Our dataset contains 12,000 news articles that may contain political intentions, from the politics section of six of the most representative newspaper organizations in South Korea. All the text samples are labeled simultaneously in two aspects (1) the level of political orientation and (2) the level of pro-government. To the best of our knowledge, our paper is the most large-scale Korean news dataset that contains long text and addresses multi-task classification problems. We also train recent state-of-the-art (SOTA) language models that are based on transformer architectures and demonstrate that the trained models show decent text classification performance. All the codes, datasets, and trained models are available at https://github.com/Kdavid2355/KoPolitic-Benchmark-Dataset.
Authors: Junxian Zhou, Haiqin Yang, Ye Junpeng, Yuxuan He, Hao Mou
Aspect sentiment quad prediction (ASQP) is a critical subtask of aspect-level sentiment analysis. Current ASQP datasets are characterized by their small size and low quadruple density, which hinders technical development. To expand capacity, we construct two large Chinese ASQP datasets crawled from multiple online platforms. The datasets hold several significant characteristics: larger size (each with 10,000+ samples) and rich aspect categories, more words per sentence, and higher density than existing ASQP datasets. Moreover, we are the first to evaluate the performance of Generative Pre-trained Transformer (GPT) series models on ASQP and exhibit potential issues. The experiments with state-of-the-art ASQP baselines underscore the need to explore additional techniques to address ASQP, as well as the importance of further investigation into methods to improve the performance of GPTs.
Authors: Sean Xie, Soroush Vosoughi, Saeed Hassanpour
Large Language Models (LLMs) have significantly advanced the field of Natural Language Processing (NLP), but their lack of interpretability has been a major concern. Current methods for interpreting LLMs are post hoc, applied after inference time, and have limitations such as their focus on low-level features and lack of explainability at higher level text units. In this work, we introduce proto-lm, a prototypical network-based white-box framework that allows LLMs to learn immediately interpretable embeddings during the fine-tuning stage while maintaining competitive performance. Our method's applicability and interpretability are demonstrated through experiments on a wide range of NLP tasks, and our results indicate a new possibility of creating interpretable models without sacrificing performance. This novel approach to interpretability in LLMs can pave the way for more interpretable models without the need to sacrifice performance.
Authors: Jiaxin Zhang, Zhuohang Li, Kamalika Das, Bradley A. Malin, Sricharan Kumar
Hallucination detection is a critical step toward understanding the trustworthiness of modern language models (LMs). To achieve this goal, we re-examine existing detection approaches based on the self-consistency of LMs and uncover two types of hallucinations resulting from 1) question-level and 2) model-level, which cannot be effectively identified through self-consistency check alone. Building upon this discovery, we propose a novel sampling-based method, i.e., semantic-aware cross-check consistency (SAC$^3$) that expands on the principle of self-consistency checking. Our SAC$^3$ approach incorporates additional mechanisms to detect both question-level and model-level hallucinations by leveraging advances including semantically equivalent question perturbation and cross-model response consistency checking. Through extensive and systematic empirical analysis, we demonstrate that SAC$^3$ outperforms the state of the art in detecting both non-factual and factual statements across multiple question-answering and open-domain generation benchmarks.
Authors: Letian Peng, Zilong Wang, Hang Liu, Zihan Wang, Jingbo Shang
With the rapid development of the internet, online social media welcomes people with different backgrounds through its diverse content. The increasing usage of emoji becomes a noticeable trend thanks to emoji's rich information beyond cultural or linguistic borders. However, the current study on emojis is limited to single emoji prediction and there are limited data resources available for further study of the interesting linguistic phenomenon. To this end, we synthesize a large text-emoji parallel corpus, Text2Emoji, from a large language model. Based on the parallel corpus, we distill a sequence-to-sequence model, EmojiLM, which is specialized in the text-emoji bidirectional translation. Extensive experiments on public benchmarks and human evaluation demonstrate that our proposed model outperforms strong baselines and the parallel corpus benefits emoji-related downstream tasks.
Authors: Randy Zakya Suchrady, Ayu Purwarianti
Aspect-based sentiment analysis is a method in natural language processing aimed at identifying and understanding sentiments related to specific aspects of an entity. Aspects are words or phrases that represent an aspect or attribute of a particular entity. Previous research has utilized generative pre-trained language models to perform aspect-based sentiment analysis. LEGO-ABSA is one framework that has successfully employed generative pre-trained language models in aspect-based sentiment analysis, particularly in English. LEGO-ABSA uses a multitask learning and prompting approach to enhance model performance. However, the application of this approach has not been done in the context of Bahasa Indonesia. Therefore, this research aims to implement the multitask learning and prompting approach in aspect-based sentiment analysis for Bahasa Indonesia using generative pre-trained language models. In this study, the Indo LEGO-ABSA model is developed, which is an aspect-based sentiment analysis model utilizing generative pre-trained language models and trained with multitask learning and prompting. Indo LEGO-ABSA is trained with a hotel domain dataset in the Indonesian language. The obtained results include an f1-score of 79.55% for the Aspect Sentiment Triplet Extraction task, 86.09% for Unified Aspect-based Sentiment Analysis, 79.85% for Aspect Opinion Pair Extraction, 87.45% for Aspect Term Extraction, and 88.09% for Opinion Term Extraction. Indo LEGO-ABSA adopts the LEGO-ABSA framework that employs the T5 model, specifically mT5, by applying multitask learning to train all tasks within aspect-based sentiment analysis.
Authors: Xin Yuan, Jie Guo, Weidong Qiu, Zheng Huang, Shujun Li
Mis- and disinformation online have become a major societal problem as major sources of online harms of different kinds. One common form of mis- and disinformation is out-of-context (OOC) information, where different pieces of information are falsely associated, e.g., a real image combined with a false textual caption or a misleading textual description. Although some past studies have attempted to defend against OOC mis- and disinformation through external evidence, they tend to disregard the role of different pieces of evidence with different stances. Motivated by the intuition that the stance of evidence represents a bias towards different detection results, we propose a stance extraction network (SEN) that can extract the stances of different pieces of multi-modal evidence in a unified framework. Moreover, we introduce a support-refutation score calculated based on the co-occurrence relations of named entities into the textual SEN. Extensive experiments on a public large-scale dataset demonstrated that our proposed method outperformed the state-of-the-art baselines, with the best model achieving a performance gain of 3.2% in accuracy.
Authors: Yiduo Guo, Zekai Zhang, Yaobo Liang, Dongyan Zhao, Duan Nan
Recent evaluations of Large Language Models (LLMs) have centered around testing their zero-shot/few-shot capabilities for basic natural language tasks and their ability to translate instructions into tool APIs. However, the evaluation of LLMs utilizing complex tools to finish multi-turn, multi-modal instructions in a complex multi-modal environment has not been investigated. To address this gap, we introduce the PowerPoint Task Completion (PPTC) benchmark to assess LLMs' ability to create and edit PPT files based on user instructions. It contains 279 multi-turn sessions covering diverse topics and hundreds of instructions involving multi-modal operations. We also propose the PPTX-Match Evaluation System that evaluates if LLMs finish the instruction based on the prediction file rather than the label API sequence, thus it supports various LLM-generated API sequences. We measure 3 closed LLMs and 6 open-source LLMs. The results show that GPT-4 outperforms other LLMs with 75.1\% accuracy in single-turn dialogue testing but faces challenges in completing entire sessions, achieving just 6\% session accuracy. We find three main error causes in our benchmark: error accumulation in the multi-turn session, long PPT template processing, and multi-modality perception. These pose great challenges for future LLM and agent systems. We release the data, code, and evaluation system of PPTC at \url{https://github.com/gydpku/PPTC}.
Authors: Yihao Wang, Ruiqi Song, Ru Zhang, Jianyi Liu
Linguistic steganalysis (LS) tasks aim to effectively detect stegos generated by linguistic steganography. Existing LS methods overlook the distinctive user characteristics, leading to weak performance in social networks. The limited occurrence of stegos further complicates detection. In this paper, we propose the UP4LS, a novel framework with the User Profile for enhancing LS performance. Specifically, by delving into post content, we explore user attributes like writing habits, psychological states, and focal areas, thereby building the user profile for LS. For each attribute, we design the identified feature extraction module. The extracted features are mapped to high-dimensional user features via deep-learning networks from existing methods. Then the language model is employed to extract content features. The user and content features are integrated to optimize feature representation. During the training phase, we prioritize the distribution of stegos. Experiments demonstrate that UP4LS can significantly enhance the performance of existing methods, and an overall accuracy improvement of nearly 25%. In particular, the improvement is especially pronounced with fewer stego samples. Additionally, UP4LS also sets the stage for studies on related tasks, encouraging extensive applications on LS tasks.
Authors: Guoxing Yang, Jianyu Shi, Zan Wang, Xiaohong Liu, Guangyu Wang
Pre-training and fine-tuning have emerged as a promising paradigm across various natural language processing (NLP) tasks. The effectiveness of pretrained large language models (LLM) has witnessed further enhancement, holding potential for applications in the field of medicine, particularly in the context of Traditional Chinese Medicine (TCM). However, the application of these general models to specific domains often yields suboptimal results, primarily due to challenges like lack of domain knowledge, unique objectives, and computational efficiency. Furthermore, their effectiveness in specialized domains, such as Traditional Chinese Medicine, requires comprehensive evaluation. To address the above issues, we propose a novel domain specific TCMDA (TCM Domain Adaptation) approach, efficient pre-training with domain-specific corpus. Specifically, we first construct a large TCM-specific corpus, TCM-Corpus-1B, by identifying domain keywords and retreving from general corpus. Then, our TCMDA leverages the LoRA which freezes the pretrained model's weights and uses rank decomposition matrices to efficiently train specific dense layers for pre-training and fine-tuning, efficiently aligning the model with TCM-related tasks, namely TCM-GPT-7B. We further conducted extensive experiments on two TCM tasks, including TCM examination and TCM diagnosis. TCM-GPT-7B archived the best performance across both datasets, outperforming other models by relative increments of 17% and 12% in accuracy, respectively. To the best of our knowledge, our study represents the pioneering validation of domain adaptation of a large language model with 7 billion parameters in TCM domain. We will release both TCMCorpus-1B and TCM-GPT-7B model once accepted to facilitate interdisciplinary development in TCM and NLP, serving as the foundation for further study.
Authors: Yijia Zhang, Sicheng Zhang, Shijie Cao, Dayou Du, Jianyu Wei, Ting Cao, Ningyi Xu
Large language models (LLMs) show great performance in various tasks, but face deployment challenges from limited memory capacity and bandwidth. Low-bit weight quantization can save memory and accelerate inference. Although floating-point (FP) formats show good performance in LLM quantization, they tend to perform poorly with small group sizes or sub-4 bits. We find the reason is that the absence of asymmetry in previous FP quantization makes it unsuitable for handling asymmetric value distribution of LLM weight tensors. In this work, we propose asymmetric FP quantization (AFPQ), which sets separate scales for positive and negative values. Our method leads to large accuracy improvements and can be easily plugged into other quantization methods, including GPTQ and AWQ, for better performance. Besides, no additional storage is needed compared with asymmetric integer (INT) quantization. The code is available at https://github.com/zhangsichengsjtu/AFPQ.
Authors: Yejin Bang, Nayeon Lee, Pascale Fung
Framing bias plays a significant role in exacerbating political polarization by distorting the perception of actual events. Media outlets with divergent political stances often use polarized language in their reporting of the same event. We propose a new loss function that encourages the model to minimize the polarity difference between the polarized input articles to reduce framing bias. Specifically, our loss is designed to jointly optimize the model to map polarity ends bidirectionally. Our experimental results demonstrate that incorporating the proposed polarity minimization loss leads to a substantial reduction in framing bias when compared to a BART-based multi-document summarization model. Notably, we find that the effectiveness of this approach is most pronounced when the model is trained to minimize the polarity loss associated with informational framing bias (i.e., skewed selection of information to report).
Authors: Isidor Konrad Maier, Johannes Kuhn, Jesse Beisegel, Markus Huber-Liebl, Matthias Wolff
In this paper we give instructions on how to write a minimalist grammar (MG). In order to present the instructions as an algorithm, we use a variant of context free grammars (CFG) as an input format. We can exclude overgeneration, if the CFG has no recursion, i.e. no non-terminal can (indirectly) derive to a right-hand side containing itself. The constructed MGs utilize licensors/-ees as a special way of exception handling. A CFG format for a derivation $A\_eats\_B\mapsto^* peter\_eats\_apples$, where $A$ and $B$ generate noun phrases, normally leads to overgeneration, e.\,g., $i\_eats\_apples$. In order to avoid overgeneration, a CFG would need many non-terminal symbols and rules, that mainly produce the same word, just to handle exceptions. In our MGs however, we can summarize CFG rules that produce the same word in one item and handle exceptions by a proper distribution of licensees/-ors. The difficulty with this technique is that in most generations the majority of licensees/-ors is not needed, but still has to be triggered somehow. We solve this problem with $\epsilon$-items called \emph{adapters}.
Authors: Mario Sänger, Ninon De Mecquenem, Katarzyna Ewa Lewińska, Vasilis Bountris, Fabian Lehmann, Ulf Leser, Thomas Kosch
Scientific workflow systems are increasingly popular for expressing and executing complex data analysis pipelines over large datasets, as they offer reproducibility, dependability, and scalability of analyses by automatic parallelization on large compute clusters. However, implementing workflows is difficult due to the involvement of many black-box tools and the deep infrastructure stack necessary for their execution. Simultaneously, user-supporting tools are rare, and the number of available examples is much lower than in classical programming languages. To address these challenges, we investigate the efficiency of Large Language Models (LLMs), specifically ChatGPT, to support users when dealing with scientific workflows. We performed three user studies in two scientific domains to evaluate ChatGPT for comprehending, adapting, and extending workflows. Our results indicate that LLMs efficiently interpret workflows but achieve lower performance for exchanging components or purposeful workflow extensions. We characterize their limitations in these challenging scenarios and suggest future research directions.
Authors: Shahar Jacob, Chen Shani, Dafna Shahaf
Analogy is one of the core capacities of human cognition; when faced with new situations, we often transfer prior experience from other domains. Most work on computational analogy relies heavily on complex, manually crafted input. In this work, we relax the input requirements, requiring only names of entities to be mapped. We automatically extract commonsense representations and use them to identify a mapping between the entities. Unlike previous works, our framework can handle partial analogies and suggest new entities to be added. Moreover, our method's output is easily interpretable, allowing for users to understand why a specific mapping was chosen.
Experiments show that our model correctly maps 81.2% of classical 2x2 analogy problems (guess level=50%). On larger problems, it achieves 77.8% accuracy (mean guess level=13.1%). In another experiment, we show our algorithm outperforms human performance, and the automatic suggestions of new entities resemble those suggested by humans. We hope this work will advance computational analogy by paving the way to more flexible, realistic input requirements, with broader applicability.
Authors: Yuhang Zhou, He Yu, Siyu Tian, Dan Chen, Liuzhi Zhou, Xinlin Yu, Chuanjun Ji, Sen Liu, Guangnan Ye, Hongfeng Chai
While current NL2SQL tasks constructed using Foundation Models have achieved commendable results, their direct application to Natural Language to Graph Query Language (NL2GQL) tasks poses challenges due to the significant differences between GQL and SQL expressions, as well as the numerous types of GQL. Our extensive experiments reveal that in NL2GQL tasks, larger Foundation Models demonstrate superior cross-schema generalization abilities, while smaller Foundation Models struggle to improve their GQL generation capabilities through fine-tuning. However, after fine-tuning, smaller models exhibit better intent comprehension and higher grammatical accuracy. Diverging from rule-based and slot-filling techniques, we introduce R3-NL2GQL, which employs both smaller and larger Foundation Models as reranker, rewriter and refiner. The approach harnesses the comprehension ability of smaller models for information reranker and rewriter, and the exceptional generalization and generation capabilities of larger models to transform input natural language queries and code structure schema into any form of GQLs. Recognizing the lack of established datasets in this nascent domain, we have created a bilingual dataset derived from graph database documentation and some open-source Knowledge Graphs (KGs). We tested our approach on this dataset and the experimental results showed that delivers promising performance and robustness.Our code and dataset is available at https://github.com/zhiqix/NL2GQL
Authors: Leonardo Rigutini, Tiziano Papini, Marco Maggini, Franco Scarselli
The problem of relevance ranking consists of sorting a set of objects with respect to a given criterion. Since users may prefer different relevance criteria, the ranking algorithms should be adaptable to the user needs. Two main approaches exist in literature for the task of learning to rank: 1) a score function, learned by examples, which evaluates the properties of each object yielding an absolute relevance value that can be used to order the objects or 2) a pairwise approach, where a "preference function" is learned using pairs of objects to define which one has to be ranked first. In this paper, we present SortNet, an adaptive ranking algorithm which orders objects using a neural network as a comparator. The neural network training set provides examples of the desired ordering between pairs of items and it is constructed by an iterative procedure which, at each iteration, adds the most informative training examples. Moreover, the comparator adopts a connectionist architecture that is particularly suited for implementing a preference function. We also prove that such an architecture has the universal approximation property and can implement a wide class of functions. Finally, the proposed algorithm is evaluated on the LETOR dataset showing promising performances in comparison with other state of the art algorithms.
Authors: Chen Shani, Jilles Vreeken, Dafna Shahaf
Concepts play a pivotal role in various human cognitive functions, including learning, reasoning and communication. However, there is very little work on endowing machines with the ability to form and reason with concepts. In particular, state-of-the-art large language models (LLMs) work at the level of tokens, not concepts.
In this work, we analyze how well contemporary LLMs capture human concepts and their structure. We then discuss ways to develop concept-aware LLMs, taking place at different stages of the pipeline. We sketch a method for pretraining LLMs using concepts, and also explore the simpler approach that uses the output of existing LLMs. Despite its simplicity, our proof-of-concept is shown to better match human intuition, as well as improve the robustness of predictions. These preliminary results underscore the promise of concept-aware LLMs.
Authors: Jinrui Yang, Timothy Baldwin, Trevor Cohn
We present Multi-EuP, a new multilingual benchmark dataset, comprising 22K multi-lingual documents collected from the European Parliament, spanning 24 languages. This dataset is designed to investigate fairness in a multilingual information retrieval (IR) context to analyze both language and demographic bias in a ranking context. It boasts an authentic multilingual corpus, featuring topics translated into all 24 languages, as well as cross-lingual relevance judgments. Furthermore, it offers rich demographic information associated with its documents, facilitating the study of demographic bias. We report the effectiveness of Multi-EuP for benchmarking both monolingual and multilingual IR. We also conduct a preliminary experiment on language bias caused by the choice of tokenization strategy.
Authors: Vitalii Fishchuk, Daniel Braun
Neural text detectors are models trained to detect whether a given text was generated by a language model or written by a human. In this paper, we investigate three simple and resource-efficient strategies (parameter tweaking, prompt engineering, and character-level mutations) to alter texts generated by GPT-3.5 that are unsuspicious or unnoticeable for humans but cause misclassification by neural text detectors. The results show that especially parameter tweaking and character-level mutations are effective strategies.
Authors: Xiaofei Sun, Xiaoya Li, Shengyu Zhang, Shuhe Wang, Fei Wu, Jiwei Li, Tianwei Zhang, Guoyin Wang
A standard paradigm for sentiment analysis is to rely on a singular LLM and makes the decision in a single round under the framework of in-context learning. This framework suffers the key disadvantage that the single-turn output generated by a single LLM might not deliver the perfect decision, just as humans sometimes need multiple attempts to get things right. This is especially true for the task of sentiment analysis where deep reasoning is required to address the complex linguistic phenomenon (e.g., clause composition, irony, etc) in the input.
To address this issue, this paper introduces a multi-LLM negotiation framework for sentiment analysis. The framework consists of a reasoning-infused generator to provide decision along with rationale, a explanation-deriving discriminator to evaluate the credibility of the generator. The generator and the discriminator iterate until a consensus is reached. The proposed framework naturally addressed the aforementioned challenge, as we are able to take the complementary abilities of two LLMs, have them use rationale to persuade each other for correction.
Experiments on a wide range of sentiment analysis benchmarks (SST-2, Movie Review, Twitter, yelp, amazon, IMDB) demonstrate the effectiveness of proposed approach: it consistently yields better performances than the ICL baseline across all benchmarks, and even superior performances to supervised baselines on the Twitter and movie review datasets.
Authors: Shahbaz Syed, Dominik Schwabe, Khalid Al-Khatib, Martin Potthast
Online forums encourage the exchange and discussion of different stances on many topics. Not only do they provide an opportunity to present one's own arguments, but may also gather a broad cross-section of others' arguments. However, the resulting long discussions are difficult to overview. This paper presents a novel unsupervised approach using large language models (LLMs) to generating indicative summaries for long discussions that basically serve as tables of contents. Our approach first clusters argument sentences, generates cluster labels as abstractive summaries, and classifies the generated cluster labels into argumentation frames resulting in a two-level summary. Based on an extensively optimized prompt engineering approach, we evaluate 19~LLMs for generative cluster labeling and frame classification. To evaluate the usefulness of our indicative summaries, we conduct a purpose-driven user study via a new visual interface called Discussion Explorer: It shows that our proposed indicative summaries serve as a convenient navigation tool to explore long discussions.
Authors: Valentin Knappich, Simon Razniewski, Annemarie Friedrich
Automatic simplification can help laypeople to comprehend complex scientific text. Language models are frequently applied to this task by translating from complex to simple language. In this paper, we describe our system based on Llama 2, which ranked first in the PLABA shared task addressing the simplification of biomedical text. We find that the large portion of shared tokens between input and output leads to weak training signals and conservatively editing models. To mitigate these issues, we propose sentence-level and token-level loss weights. They give higher weight to modified tokens, indicated by edit distance and edit operations, respectively. We conduct an empirical evaluation on the PLABA dataset and find that both approaches lead to simplifications closer to those created by human annotators (+1.8% / +3.5% SARI), simpler language (-1 / -1.1 FKGL) and more edits (1.6x / 1.8x edit distance) compared to the same model fine-tuned with standard cross entropy. We furthermore show that the hyperparameter $\lambda$ in token-level loss weights can be used to control the edit distance and the simplicity level (FKGL).
Authors: Mingze Yuan, Peng Bao, Jiajia Yuan, Yunhao Shen, Zifan Chen, Yi Xie, Jie Zhao, Yang Chen, Li Zhang, Lin Shen, Bin Dong
With the rapid development of artificial intelligence, large language models (LLMs) have shown promising capabilities in mimicking human-level language comprehension and reasoning. This has sparked significant interest in applying LLMs to enhance various aspects of healthcare, ranging from medical education to clinical decision support. However, medicine involves multifaceted data modalities and nuanced reasoning skills, presenting challenges for integrating LLMs. This paper provides a comprehensive review on the applications and implications of LLMs in medicine. It begins by examining the fundamental applications of general-purpose and specialized LLMs, demonstrating their utilities in knowledge retrieval, research support, clinical workflow automation, and diagnostic assistance. Recognizing the inherent multimodality of medicine, the review then focuses on multimodal LLMs, investigating their ability to process diverse data types like medical imaging and EHRs to augment diagnostic accuracy. To address LLMs' limitations regarding personalization and complex clinical reasoning, the paper explores the emerging development of LLM-powered autonomous agents for healthcare. Furthermore, it summarizes the evaluation methodologies for assessing LLMs' reliability and safety in medical contexts. Overall, this review offers an extensive analysis on the transformative potential of LLMs in modern medicine. It also highlights the pivotal need for continuous optimizations and ethical oversight before these models can be effectively integrated into clinical practice. Visit https://github.com/mingze-yuan/Awesome-LLM-Healthcare for an accompanying GitHub repository containing latest papers.
Authors: Tobias Katsch
Linear Recurrence has proven to be a powerful tool for modeling long sequences efficiently. In this work, we show that existing models fail to take full advantage of its potential. Motivated by this finding, we develop GateLoop, a foundational sequence model that generalizes linear recurrent models such as S4, S5, LRU and RetNet, by employing data-controlled state transitions. Utilizing this theoretical advance, GateLoop empirically outperforms existing models for auto-regressive language modeling. Our method comes with a low-cost $O(l)$ recurrent mode and an efficient $O(l \log_{2} l)$ parallel mode making use of highly optimized associative scan implementations. Furthermore, we derive an $O(l^2)$ surrogate attention mode, revealing remarkable implications for Transformer and recently proposed architectures. Specifically, we prove that our approach can be interpreted as providing data-controlled relative-positional information to Attention. While many existing models solely rely on data-controlled cumulative sums for context aggregation, our findings suggest that incorporating data-controlled complex cumulative products may be a crucial step towards more powerful sequence models.
Authors: Keunwoo Peter Yu
In natural language processing, interactive text-based games serve as a test bed for interactive AI systems. Prior work has proposed to play text-based games by acting based on discrete knowledge graphs constructed by the Discrete Graph Updater (DGU) to represent the game state from the natural language description. While DGU has shown promising results with high interpretability, it suffers from lower knowledge graph accuracy due to its lack of temporality and limited generalizability to complex environments with objects with the same label. In order to address DGU's weaknesses while preserving its high interpretability, we propose the Temporal Discrete Graph Updater (TDGU), a novel neural network model that represents dynamic knowledge graphs as a sequence of timestamped graph events and models them using a temporal point based graph neural network. Through experiments on the dataset collected from a text-based game TextWorld, we show that TDGU outperforms the baseline DGU. We further show the importance of temporal information for TDGU's performance through an ablation study and demonstrate that TDGU has the ability to generalize to more complex environments with objects with the same label. All the relevant code can be found at \url{https://github.com/yukw777/temporal-discrete-graph-updater}.
Authors: Yifan Wang, Qingyan Guo, Xinzhe Ni, Chufan Shi, Lemao Liu, Haiyun Jiang, Yujiu Yang
In-context learning (ICL) ability has emerged with the increasing scale of large language models (LLMs), enabling them to learn input-label mappings from demonstrations and perform well on downstream tasks. However, under the standard ICL setting, LLMs may sometimes neglect query-related information in demonstrations, leading to incorrect predictions. To address this limitation, we propose a new paradigm called Hint-enhanced In-Context Learning (HICL) to explore the power of ICL in open-domain question answering, an important form in knowledge-intensive tasks. HICL leverages LLMs' reasoning ability to extract query-related knowledge from demonstrations, then concatenates the knowledge to prompt LLMs in a more explicit way. Furthermore, we track the source of this knowledge to identify specific examples, and introduce a Hint-related Example Retriever (HER) to select informative examples for enhanced demonstrations. We evaluate HICL with HER on 3 open-domain QA benchmarks, and observe average performance gains of 2.89 EM score and 2.52 F1 score on gpt-3.5-turbo, 7.62 EM score and 7.27 F1 score on LLaMA-2-Chat-7B compared with standard setting.
Authors: Lukas Edman, Lisa Bylinina
This paper details the work of the University of Groningen for the BabyLM Challenge. We follow the idea that, like babies, language models should be introduced to simpler concepts first and build off of that knowledge to understand more complex concepts. We examine this strategy of simple-then-complex through a variety of lenses, namely context size, vocabulary, and overall linguistic complexity of the data. We find that only one, context size, is truly beneficial to training a language model. However this simple change to context size gives us improvements of 2 points on average on (Super)GLUE tasks, 1 point on MSGS tasks, and 12\% on average on BLiMP tasks. Our context-limited model outperforms the baseline that was trained on 10$\times$ the amount of data.
Authors: Kun Zhou, Yutao Zhu, Zhipeng Chen, Wentong Chen, Wayne Xin Zhao, Xu Chen, Yankai Lin, Ji-Rong Wen, Jiawei Han
Large language models~(LLMs) have greatly advanced the frontiers of artificial intelligence, attaining remarkable improvement in model capacity. To assess the model performance, a typical approach is to construct evaluation benchmarks for measuring the ability level of LLMs in different aspects. Despite that a number of high-quality benchmarks have been released, the concerns about the appropriate use of these benchmarks and the fair comparison of different models are increasingly growing. Considering these concerns, in this paper, we discuss the potential risk and impact of inappropriately using evaluation benchmarks and misleadingly interpreting the evaluation results. Specially, we focus on a special issue that would lead to inappropriate evaluation, \ie \emph{benchmark leakage}, referring that the data related to evaluation sets is occasionally used for model training. This phenomenon now becomes more common since pre-training data is often prepared ahead of model test. We conduct extensive experiments to study the effect of benchmark leverage, and find that it can dramatically boost the evaluation results, which would finally lead to an unreliable assessment of model performance. To improve the use of existing evaluation benchmarks, we finally present several guidelines for both LLM developers and benchmark maintainers. We hope this work can draw attention to appropriate training and evaluation of LLMs.
Authors: Alina Leidinger, Robert van Rooij, Ekaterina Shutova
The latest generation of LLMs can be prompted to achieve impressive zero-shot or few-shot performance in many NLP tasks. However, since performance is highly sensitive to the choice of prompts, considerable effort has been devoted to crowd-sourcing prompts or designing methods for prompt optimisation. Yet, we still lack a systematic understanding of how linguistic properties of prompts correlate with task performance. In this work, we investigate how LLMs of different sizes, pre-trained and instruction-tuned, perform on prompts that are semantically equivalent, but vary in linguistic structure. We investigate both grammatical properties such as mood, tense, aspect and modality, as well as lexico-semantic variation through the use of synonyms. Our findings contradict the common assumption that LLMs achieve optimal performance on lower perplexity prompts that reflect language use in pretraining or instruction-tuning data. Prompts transfer poorly between datasets or models, and performance cannot generally be explained by perplexity, word frequency, ambiguity or prompt length. Based on our results, we put forward a proposal for a more robust and comprehensive evaluation standard for prompting research.
Authors: Haotian Luo, Kunming Wu, Cheng Dai, Sixian Ding, Xinhao Chen
RNN-like language models are getting renewed attention from NLP researchers in recent years and several models have made significant progress, which demonstrates performance comparable to traditional transformers. However, due to the recurrent nature of RNNs, this kind of language model can only store information in a set of fixed-length state vectors. As a consequence, they still suffer from forgetfulness though after a lot of improvements and optimizations, when given complex instructions or prompts. As the prompted generation is the main and most concerned function of LMs, solving the problem of forgetting in the process of generation is no wonder of vital importance. In this paper, focusing on easing the prompt forgetting during generation, we proposed an architecture to teach the model memorizing prompt during generation by synthetic gradient. To force the model to memorize the prompt, we derive the states that encode the prompt, then transform it into model parameter modification using low-rank gradient approximation, which hard-codes the prompt into model parameters temporarily. We construct a dataset for experiments, and the results have demonstrated the effectiveness of our method in solving the problem of forgetfulness in the process of prompted generation. We will release all the code upon acceptance.
Authors: Gretel Liz De la Peña Sarracén, Paolo Rosso, Robert Litschko, Goran Glavaš, Simone Paolo Ponzetto
Cross-lingual transfer learning from high-resource to medium and low-resource languages has shown encouraging results. However, the scarcity of resources in target languages remains a challenge. In this work, we resort to data augmentation and continual pre-training for domain adaptation to improve cross-lingual abusive language detection. For data augmentation, we analyze two existing techniques based on vicinal risk minimization and propose MIXAG, a novel data augmentation method which interpolates pairs of instances based on the angle of their representations. Our experiments involve seven languages typologically distinct from English and three different domains. The results reveal that the data augmentation strategies can enhance few-shot cross-lingual abusive language detection. Specifically, we observe that consistently in all target languages, MIXAG improves significantly in multidomain and multilingual environments. Finally, we show through an error analysis how the domain adaptation can favour the class of abusive texts (reducing false negatives), but at the same time, declines the precision of the abusive language detection model.
Authors: Alexey Tikhonov, Ivan P. Yamshchikov
In the rapidly evolving landscape of Large Language Models (LLMs), introduction of well-defined and standardized evaluation methodologies remains a crucial challenge. This paper traces the historical trajectory of LLM evaluations, from the foundational questions posed by Alan Turing to the modern era of AI research. We categorize the evolution of LLMs into distinct periods, each characterized by its unique benchmarks and evaluation criteria. As LLMs increasingly mimic human-like behaviors, traditional evaluation proxies, such as the Turing test, have become less reliable. We emphasize the pressing need for a unified evaluation system, given the broader societal implications of these models. Through an analysis of common evaluation methodologies, we advocate for a qualitative shift in assessment approaches, underscoring the importance of standardization and objective criteria. This work serves as a call for the AI community to collaboratively address the challenges of LLM evaluation, ensuring their reliability, fairness, and societal benefit.
Authors: Alyssa Hwang, Andrew Head, Chris Callison-Burch
GPT-Vision has impressed us on a range of vision-language tasks, but it comes with the familiar new challenge: we have little idea of its capabilities and limitations. In our study, we formalize a process that many have instinctively been trying already to develop "grounded intuition" of this new model. Inspired by the recent movement away from benchmarking in favor of example-driven qualitative evaluation, we draw upon grounded theory and thematic analysis in social science and human-computer interaction to establish a rigorous framework for qualitative evaluation in natural language processing. We use our technique to examine alt text generation for scientific figures, finding that GPT-Vision is particularly sensitive to prompting, counterfactual text in images, and relative spatial relationships. Our method and analysis aim to help researchers ramp up their own grounded intuitions of new models while exposing how GPT-Vision can be applied to make information more accessible.
Authors: Uwe D. Reichel
The purposes of the CoPaSul toolkit are (1) automatic prosodic annotation and (2) prosodic feature extraction from syllable to utterance level. CoPaSul stands for contour-based, parametric, superpositional intonation stylization. In this framework intonation is represented as a superposition of global and local contours that are described parametrically in terms of polynomial coefficients. On the global level (usually associated but not necessarily restricted to intonation phrases) the stylization serves to represent register in terms of time-varying F0 level and range. On the local level (e.g. accent groups), local contour shapes are described. From this parameterization several features related to prosodic boundaries and prominence can be derived. Furthermore, by coefficient clustering prosodic contour classes can be obtained in a bottom-up way. Next to the stylization-based feature extraction also standard F0 and energy measures (e.g. mean and variance) as well as rhythmic aspects can be calculated. At the current state automatic annotation comprises: segmentation into interpausal chunks, syllable nucleus extraction, and unsupervised localization of prosodic phrase boundaries and prominent syllables. F0 and partly also energy feature sets can be derived for: standard measurements (as median and IQR), register in terms of F0 level and range, prosodic boundaries, local contour shapes, bottom-up derived contour classes, Gestalt of accent groups in terms of their deviation from higher level prosodic units, as well as for rhythmic aspects quantifying the relation between F0 and energy contours and prosodic event rates.
Authors: Weiwei Sun, Shuyu Guo, Shuo Zhang, Pengjie Ren, Zhumin Chen, Maarten de Rijke, Zhaochun Ren
Task-oriented dialogue systems (TDSs) are assessed mainly in an offline setting or through human evaluation. The evaluation is often limited to single-turn or is very time-intensive. As an alternative, user simulators that mimic user behavior allow us to consider a broad set of user goals to generate human-like conversations for simulated evaluation. Employing existing user simulators to evaluate TDSs is challenging as user simulators are primarily designed to optimize dialogue policies for TDSs and have limited evaluation capabilities. Moreover, the evaluation of user simulators is an open challenge.
In this work, we propose a metaphorical user simulator for end-to-end TDS evaluation, where we define a simulator to be metaphorical if it simulates user's analogical thinking in interactions with systems. We also propose a tester-based evaluation framework to generate variants, i.e., dialogue systems with different capabilities. Our user simulator constructs a metaphorical user model that assists the simulator in reasoning by referring to prior knowledge when encountering new items. We estimate the quality of simulators by checking the simulated interactions between simulators and variants. Our experiments are conducted using three TDS datasets. The proposed user simulator demonstrates better consistency with manual evaluation than an agenda-based simulator and a seq2seq model on three datasets; our tester framework demonstrates efficiency and has been tested on multiple tasks, such as conversational recommendation and e-commerce dialogues.
Authors: Yuxuan Ye, Edwin Simpson
This paper introduces a novel pipeline for summarising timelines of events reported by multiple news sources. Transformer-based models for abstractive summarisation generate coherent and concise summaries of long documents but can fail to outperform established extractive methods on specialised tasks such as timeline summarisation (TLS). While extractive summaries are more faithful to their sources, they may be less readable and contain redundant or unnecessary information. This paper proposes a preference-based reinforcement learning (PBRL) method for adapting pretrained abstractive summarisers to TLS, which can overcome the drawbacks of extractive timeline summaries. We define a compound reward function that learns from keywords of interest and pairwise preference labels, which we use to fine-tune a pretrained abstractive summariser via offline reinforcement learning. We carry out both automated and human evaluation on three datasets, finding that our method outperforms a comparable extractive TLS method on two of the three benchmark datasets, and participants prefer our method's summaries to those of both the extractive TLS method and the pretrained abstractive model. The method does not require expensive reference summaries and needs only a small number of preferences to align the generated summaries with human preferences.
Authors: Parishad BehnamGhader, Santiago Miret, Siva Reddy
Augmenting pretrained language models with retrievers has shown promise in effectively solving common NLP problems, such as language modeling and question answering. In this paper, we evaluate the strengths and weaknesses of popular retriever-augmented language models, namely kNN-LM, REALM, DPR + FiD, Contriever + ATLAS, and Contriever + Flan-T5, in reasoning over retrieved statements across different tasks. Our findings indicate that the simple similarity metric employed by retrievers is insufficient for retrieving all the necessary statements for reasoning. Additionally, the language models do not exhibit strong reasoning even when provided with only the required statements. Furthermore, when combined with imperfect retrievers, the performance of the language models becomes even worse, e.g., Flan-T5's performance drops by 28.6% when retrieving 5 statements using Contriever. While larger language models improve performance, there is still a substantial room for enhancement. Our further analysis indicates that multihop retrieve-and-read is promising for large language models like GPT-3.5, but does not generalize to other language models like Flan-T5-xxl.
Authors: Ben Prystawski, Michael Y. Li, Noah D. Goodman
Humans have a powerful and mysterious capacity to reason. Working through a set of mental steps enables us to make inferences we would not be capable of making directly even though we get no additional data from the world. Similarly, when large language models generate intermediate steps (a chain of thought) before answering a question, they often produce better answers than they would directly. We investigate why and how chain-of-thought reasoning is useful in language models, testing the hypothesis that reasoning is effective when training data consists of overlapping local clusters of variables that influence each other strongly. These training conditions enable the chaining of accurate local inferences to estimate relationships between variables that were not seen together in training. We prove that there will exist a "reasoning gap", where reasoning through intermediate variables reduces bias, for the simple case of an autoregressive density estimator trained on local samples from a chain-structured probabilistic model. We then test our hypothesis experimentally in more complex models, training an autoregressive language model on samples from Bayes nets but only including a subset of variables in each sample. We test language models' ability to match conditional probabilities with and without intermediate reasoning steps, finding that intermediate steps are only helpful when the training data is locally structured with respect to dependencies between variables. The combination of locally structured observations and reasoning is much more data-efficient than training on all variables. Our results illustrate how the effectiveness of reasoning step by step is rooted in the local statistical structure of the training data.
Authors: Yingqiang Ge, Wenyue Hua, Kai Mei, Jianchao Ji, Juntao Tan, Shuyuan Xu, Zelong Li, Yongfeng Zhang
Human Intelligence (HI) excels at combining basic skills to solve complex tasks. This capability is vital for Artificial Intelligence (AI) and should be embedded in comprehensive AI Agents, enabling them to harness expert models for complex task-solving towards Artificial General Intelligence (AGI). Large Language Models (LLMs) show promising learning and reasoning abilities, and can effectively use external models, tools, plugins, or APIs to tackle complex problems. In this work, we introduce OpenAGI, an open-source AGI research and development platform designed for solving multi-step, real-world tasks. Specifically, OpenAGI uses a dual strategy, integrating standard benchmark tasks for benchmarking and evaluation, and open-ended tasks including more expandable models, tools, plugins, or APIs for creative problem-solving. Tasks are presented as natural language queries to the LLM, which then selects and executes appropriate models. We also propose a Reinforcement Learning from Task Feedback (RLTF) mechanism that uses task results to improve the LLM's task-solving ability, which creates a self-improving AI feedback loop. While we acknowledge that AGI is a broad and multifaceted research challenge with no singularly defined solution path, the integration of LLMs with domain-specific expert models, inspired by mirroring the blend of general and specialized intelligence in humans, offers a promising approach towards AGI. We are open-sourcing the OpenAGI project's code, dataset, benchmarks, evaluation methods, and the UI demo to foster community involvement in AGI advancement: https://github.com/agiresearch/OpenAGI.
Authors: Mohammadreza Pourreza, Davood Rafiei
There is currently a significant gap between the performance of fine-tuned models and prompting approaches using Large Language Models (LLMs) on the challenging task of text-to-SQL, as evaluated on datasets such as Spider. To improve the performance of LLMs in the reasoning process, we study how decomposing the task into smaller sub-tasks can be effective. In particular, we show that breaking down the generation problem into sub-problems and feeding the solutions of those sub-problems into LLMs can be an effective approach for significantly improving their performance. Our experiments with three LLMs show that this approach consistently improves their simple few-shot performance by roughly 10%, pushing the accuracy of LLMs towards SOTA or surpassing it. On the holdout test set of Spider, the SOTA, in terms of execution accuracy, was 79.9 and the new SOTA at the time of this writing using our approach is 85.3. Our approach with in-context learning beats many heavily fine-tuned models by at least 5%. Additionally, when evaluated on the BIRD benchmark, our approach achieved an execution accuracy of 55.9%, setting a new SOTA on its holdout test set.
Authors: Sarik Ghazarian, Yijia Shao, Rujun Han, Aram Galstyan, Nanyun Peng
Commonsense reasoning is omnipresent in human communications and thus is an important feature for open-domain dialogue systems. However, evaluating commonsense in dialogue systems is still an open challenge. We take the first step by focusing on event commonsense that considers events and their relations, and is crucial in both dialogues and general commonsense reasoning. We propose ACCENT, an event commonsense evaluation metric empowered by commonsense knowledge bases (CSKBs). ACCENT first extracts event-relation tuples from a dialogue, and then evaluates the response by scoring the tuples in terms of their compatibility with the CSKB. To evaluate ACCENT, we construct the first public event commonsense evaluation dataset for open-domain dialogues. Our experiments show that ACCENT is an efficient metric for event commonsense evaluation, which achieves higher correlations with human judgments than existing baselines.
Authors: Xiaolei Wang, Xinyu Tang, Wayne Xin Zhao, Jingyuan Wang, Ji-Rong Wen
The recent success of large language models (LLMs) has shown great potential to develop more powerful conversational recommender systems (CRSs), which rely on natural language conversations to satisfy user needs. In this paper, we embark on an investigation into the utilization of ChatGPT for conversational recommendation, revealing the inadequacy of the existing evaluation protocol. It might over-emphasize the matching with the ground-truth items or utterances generated by human annotators, while neglecting the interactive nature of being a capable CRS. To overcome the limitation, we further propose an interactive Evaluation approach based on LLMs named iEvaLM that harnesses LLM-based user simulators. Our evaluation approach can simulate various interaction scenarios between users and systems. Through the experiments on two publicly available CRS datasets, we demonstrate notable improvements compared to the prevailing evaluation protocol. Furthermore, we emphasize the evaluation of explainability, and ChatGPT showcases persuasive explanation generation for its recommendations. Our study contributes to a deeper comprehension of the untapped potential of LLMs for CRSs and provides a more flexible and easy-to-use evaluation framework for future research endeavors. The codes and data are publicly available at https://github.com/RUCAIBox/iEvaLM-CRS.
Authors: Jinghan Yao, Nawras Alnaasan, Tian Chen, Aamir Shafi, Hari Subramoni, Dhabaleswar K. (DK) Panda
Autoregressive models, despite their commendable performance in a myriad of generative tasks, face challenges stemming from their inherently sequential structure. Inference on these models, by design, harnesses a temporal dependency, where the current token's probability distribution is conditioned on preceding tokens. This inherent characteristic severely impedes computational efficiency during inference as a typical inference request can require more than thousands of tokens, where generating each token requires a load of entire model weights, making the inference more memory-bound. The large overhead becomes profound in real deployment where requests arrive randomly, necessitating various generation lengths. Existing solutions, such as dynamic batching and concurrent instances, introduce significant response delays and bandwidth contention, falling short of achieving optimal latency and throughput. To address these shortcomings, we propose Flover -- a temporal fusion framework for efficiently inferring multiple requests in parallel. We deconstruct the general generation pipeline into pre-processing and token generation, and equip the framework with a dedicated work scheduler for fusing the generation process temporally across all requests. By orchestrating the token-level parallelism, Flover exhibits optimal hardware efficiency and significantly spares the system resources. By further employing a fast buffer reordering algorithm that allows memory eviction of finished tasks, it brings over 11x inference speedup on GPT and 16x on LLAMA compared to the cutting-edge solutions provided by NVIDIA FasterTransformer. Crucially, by leveraging the advanced tensor parallel technique, Flover proves efficacious across diverse computational landscapes, from single-GPU setups to distributed scenarios, thereby offering robust performance optimization that adapts to variable use cases.
Authors: Shuo Zhang, Liangming Pan, Junzhou Zhao, William Yang Wang
Large language models often necessitate grounding on external knowledge to generate faithful and reliable answers. Yet even with the correct groundings in the reference, they can ignore them and rely on wrong groundings or their inherent biases to hallucinate when users, being largely unaware of the specifics of the stored information, pose questions that might not directly correlate with the retrieved groundings. In this work, we formulate this knowledge alignment problem and introduce MixAlign, a framework that interacts with both the human user and the knowledge base to obtain and integrate clarifications on how the user question relates to the stored information. MixAlign employs a language model to achieve automatic knowledge alignment and, if necessary, further enhances this alignment through human user clarifications. Experimental results highlight the crucial role of knowledge alignment in boosting model performance and mitigating hallucination, with improvements noted up to 22.2% and 27.1% respectively. We also demonstrate the effectiveness of MixAlign in improving knowledge alignment by producing high-quality, user-centered clarifications.
Authors: Haoyi Qiu, Zi-Yi Dou, Tianlu Wang, Asli Celikyilmaz, Nanyun Peng
Model-based evaluation metrics (e.g., CLIPScore and GPTScore) have demonstrated decent correlations with human judgments in various language generation tasks. However, their impact on fairness remains largely unexplored. It is widely recognized that pretrained models can inadvertently encode societal biases, thus employing these models for evaluation purposes may inadvertently perpetuate and amplify biases. For example, an evaluation metric may favor the caption "a woman is calculating an account book" over "a man is calculating an account book," even if the image only shows male accountants. In this paper, we conduct a systematic study of gender biases in model-based automatic evaluation metrics for image captioning tasks. We start by curating a dataset comprising profession, activity, and object concepts associated with stereotypical gender associations. Then, we demonstrate the negative consequences of using these biased metrics, including the inability to differentiate between biased and unbiased generations, as well as the propagation of biases to generation models through reinforcement learning. Finally, we present a simple and effective way to mitigate the metric bias without hurting the correlations with human judgments. Our dataset and framework lay the foundation for understanding the potential harm of model-based evaluation metrics, and facilitate future works to develop more inclusive evaluation metrics.
Authors: Kangxi Wu, Liang Pang, Huawei Shen, Xueqi Cheng, Tat-Seng Chua
Generated texts from large language models (LLMs) are remarkably close to high-quality human-authored text, raising concerns about their potential misuse in spreading false information and academic misconduct. Consequently, there is an urgent need for a highly practical detection tool capable of accurately identifying the source of a given text. However, existing detection tools typically rely on access to LLMs and can only differentiate between machine-generated and human-authored text, failing to meet the requirements of fine-grained tracing, intermediary judgment, and rapid detection. Therefore, we propose LLMDet, a model-specific, secure, efficient, and extendable detection tool, that can source text from specific LLMs, such as GPT-2, OPT, LLaMA, and others. In LLMDet, we record the next-token probabilities of salient n-grams as features to calculate proxy perplexity for each LLM. By jointly analyzing the proxy perplexities of LLMs, we can determine the source of the generated text. Experimental results show that LLMDet yields impressive detection performance while ensuring speed and security, achieving 98.54% precision and x5.0 faster for recognizing human-authored text. Additionally, LLMDet can effortlessly extend its detection capabilities to a new open-source model. We will provide an open-source tool at https://github.com/TrustedLLM/LLMDet.
Authors: Benno Krojer, Elinor Poole-Dayan, Vikram Voleti, Christopher Pal, Siva Reddy
Text-conditioned image generation models have recently shown immense qualitative success using denoising diffusion processes. However, unlike discriminative vision-and-language models, it is a non-trivial task to subject these diffusion-based generative models to automatic fine-grained quantitative evaluation of high-level phenomena such as compositionality. Towards this goal, we perform two innovations. First, we transform diffusion-based models (in our case, Stable Diffusion) for any image-text matching (ITM) task using a novel method called DiffusionITM. Second, we introduce the Generative-Discriminative Evaluation Benchmark (GDBench) benchmark with 7 complex vision-and-language tasks, bias evaluation and detailed analysis. We find that Stable Diffusion + DiffusionITM is competitive on many tasks and outperforms CLIP on compositional tasks like like CLEVR and Winoground. We further boost its compositional performance with a transfer setup by fine-tuning on MS-COCO while retaining generative capabilities. We also measure the stereotypical bias in diffusion models, and find that Stable Diffusion 2.1 is, for the most part, less biased than Stable Diffusion 1.5. Overall, our results point in an exciting direction bringing discriminative and generative model evaluation closer. We will release code and benchmark setup soon.
Authors: Bailin Wang, Zi Wang, Xuezhi Wang, Yuan Cao, Rif A. Saurous, Yoon Kim
Large language models (LLMs) can learn to perform a wide range of natural language tasks from just a handful of in-context examples. However, for generating strings from highly structured languages (e.g., semantic parsing to complex domain-specific languages), it is challenging for the LLM to generalize from just a few exemplars. We propose \emph{grammar prompting}, a simple approach to enable LLMs to use external knowledge and domain-specific constraints, expressed through a grammar in Backus--Naur Form (BNF), during in-context learning. Grammar prompting augments each demonstration example with a specialized grammar that is minimally sufficient for generating the particular output example, where the specialized grammar is a subset of the full DSL grammar. For inference, the LLM first predicts a BNF grammar given a test input, and then generates the output according to the rules of the grammar. Experiments demonstrate that grammar prompting can enable LLMs to perform competitively on a diverse set of DSL generation tasks, including semantic parsing (SMCalFlow, Overnight, GeoQuery), PDDL planning, and SMILES-based molecule generation.
Authors: Banghua Zhu, Hiteshi Sharma, Felipe Vieira Frujeri, Shi Dong, Chenguang Zhu, Michael I. Jordan, Jiantao Jiao
Reinforcement learning from human feedback (RLHF) has emerged as a reliable approach to aligning large language models (LLMs) to human preferences. Among the plethora of RLHF techniques, proximal policy optimization (PPO) is of the most widely used methods. Despite its popularity, however, PPO may suffer from mode collapse, instability, and poor sample efficiency. We show that these issues can be alleviated by a novel algorithm that we refer to as Advantage-Induced Policy Alignment (APA), which leverages a squared error loss function based on the estimated advantages. We demonstrate empirically that APA consistently outperforms PPO in language tasks by a large margin, when a separate reward model is employed as the evaluator. In addition, compared with PPO, APA offers a more stable form of control over the deviation from the model's initial policy, ensuring that the model improves its performance without collapsing to deterministic output. In addition to empirical results, we also provide a theoretical justification supporting the design of our loss function.
Authors: Lakshya A Agrawal, Aditya Kanade, Navin Goyal, Shuvendu K. Lahiri, Sriram K. Rajamani
Language models of code (LMs) work well when the surrounding code provides sufficient context. This is not true when it becomes necessary to use types, functionality or APIs defined elsewhere in the repository or a linked library, especially those not seen during training. LMs suffer from limited awareness of such global context and end up hallucinating.
Integrated development environments (IDEs) assist developers in understanding repository context using static analysis. We extend this assistance, enjoyed by developers, to LMs. We propose monitor-guided decoding (MGD) where a monitor uses static analysis to guide the decoding. We construct a repository-level dataset PragmaticCode for method-completion in Java and evaluate MGD on it. On models of varying parameter scale, by monitoring for type-consistent object dereferences, MGD consistently improves compilation rates and agreement with ground truth. Further, LMs with fewer parameters, when augmented with MGD, can outperform larger LMs. With MGD, SantaCoder-1.1B achieves better compilation rate and next-identifier match than the much larger text-davinci-003 model.
We also conduct a generalizability study to evaluate the ability of MGD to generalize to multiple programming languages (Java, C# and Rust), coding scenarios (e.g., correct number of arguments to method calls), and to enforce richer semantic constraints (e.g., stateful API protocols). Our data and implementation are available at https://github.com/microsoft/monitors4codegen .
Authors: Saipraneeth Devunuri, Shirin Qiam, Lewis Lehe
The General Transit Feed Specification (GTFS) standard for publishing transit data is ubiquitous. GTFS being tabular data, with information spread across different files, necessitates specialized tools or packages to retrieve information. Concurrently, the use of Large Language Models(LLMs) for text and information retrieval is growing. The idea of this research is to see if the current widely adopted LLMs (ChatGPT) are able to understand GTFS and retrieve information from GTFS using natural language instructions without explicitly providing information. In this research, we benchmark OpenAI's GPT-3.5-Turbo and GPT-4 LLMs which are the backbone of ChatGPT. ChatGPT demonstrates a reasonable understanding of GTFS by answering 59.7% (GPT-3.5-Turbo) and 73.3% (GPT-4) of our multiple-choice questions (MCQ) correctly. Furthermore, we evaluated the LLMs on information extraction tasks using a filtered GTFS feed containing four routes. We found that program synthesis techniques outperformed zero-shot approaches, achieving up to 93% (90%) accuracy for simple queries and 61% (41%) for complex ones using GPT-4 (GPT-3.5-Turbo).
Authors: Javier Ferrando, Matthias Sperber, Hendra Setiawan, Dominic Telaar, Saša Hasan
Behavioral testing in NLP allows fine-grained evaluation of systems by examining their linguistic capabilities through the analysis of input-output behavior. Unfortunately, existing work on behavioral testing in Machine Translation (MT) is currently restricted to largely handcrafted tests covering a limited range of capabilities and languages. To address this limitation, we propose to use Large Language Models (LLMs) to generate a diverse set of source sentences tailored to test the behavior of MT models in a range of situations. We can then verify whether the MT model exhibits the expected behavior through matching candidate sets that are also generated using LLMs. Our approach aims to make behavioral testing of MT systems practical while requiring only minimal human effort. In our experiments, we apply our proposed evaluation framework to assess multiple available MT systems, revealing that while in general pass-rates follow the trends observable from traditional accuracy-based metrics, our method was able to uncover several important differences and potential bugs that go unnoticed when relying only on accuracy.
Authors: Yixin Wan, Jieyu Zhao, Aman Chadha, Nanyun Peng, Kai-Wei Chang
Recent advancements in Large Language Models empower them to follow freeform instructions, including imitating generic or specific demographic personas in conversations. We define generic personas to represent demographic groups, such as "an Asian person", whereas specific personas may take the form of specific popular Asian names like "Yumi". While the adoption of personas enriches user experiences by making dialogue systems more engaging and approachable, it also casts a shadow of potential risk by exacerbating social biases within model responses, thereby causing societal harm through interactions with users. In this paper, we systematically study "persona biases", which we define to be the sensitivity of dialogue models' harmful behaviors contingent upon the personas they adopt. We categorize persona biases into biases in harmful expression and harmful agreement, and establish a comprehensive evaluation framework to measure persona biases in five aspects: Offensiveness, Toxic Continuation, Regard, Stereotype Agreement, and Toxic Agreement. Additionally, we propose to investigate persona biases by experimenting with UNIVERSALPERSONA, a systematically constructed persona dataset encompassing various types of both generic and specific model personas. Through benchmarking on four different models -- including Blender, ChatGPT, Alpaca, and Vicuna -- our study uncovers significant persona biases in dialogue systems. Our findings also underscore the pressing need to revisit the use of personas in dialogue agents to ensure safe application.
Authors: Jonathan Kamp, Lisa Beinborn, Antske Fokkens
Feature attribution scores are used for explaining the prediction of a text classifier to users by highlighting a k number of tokens. In this work, we propose a way to determine the number of optimal k tokens that should be displayed from sequential properties of the attribution scores. Our approach is dynamic across sentences, method-agnostic, and deals with sentence length bias. We compare agreement between multiple methods and humans on an NLI task, using fixed k and dynamic k. We find that perturbation-based methods and Vanilla Gradient exhibit highest agreement on most method--method and method--human agreement metrics with a static k. Their advantage over other methods disappears with dynamic ks which mainly improve Integrated Gradient and GradientXInput. To our knowledge, this is the first evidence that sequential properties of attribution scores are informative for consolidating attribution signals for human interpretation.
Authors: Marco Bronzini, Carlo Nicolini, Bruno Lepri, Andrea Passerini, Jacopo Staiano
Over the last decade, several regulatory bodies have started requiring the disclosure of non-financial information from publicly listed companies, in light of the investors' increasing attention to Environmental, Social, and Governance (ESG) issues. Such information is publicly released in a variety of non-structured and multi-modal documentation. Hence, it is not straightforward to aggregate and consolidate such data in a cohesive framework to further derive insights about sustainability practices across companies and markets. Given these premises, it is natural to resort to Information Extraction (IE) techniques to provide concise, informative, and actionable data to the stakeholders. Moving beyond traditional text processing techniques, in this work we leverage Large Language Models (LLMs), along with the prominent in-context learning technique and the Retrieved Augmented Generation (RAG) paradigm, to extract semantically structured ESG-related information from companies' sustainability reports. We then adopt graph-based representations to conduct meaningful statistical, similarity and correlation analyses concerning the ESG-related actions disclosed by companies in their sustainability reports. These analyses unveiled that companies address ESG-related issues through several actions encompassing recognition, compliance, and partnerships; highlighting the complexity and joint efforts needed to address them. Moreover, disclosure similarities emerged among companies from the same region or sector. Lastly, we investigate which factual aspects impact the most on companies' ESG scores using our findings and other company information. This analysis unveiled that companies' disclosures affect ESG scores more than other financial or company characteristics.
Authors: Boyang Xue, Weichao Wang, Hongru Wang, Fei Mi, Rui Wang, Yasheng Wang, Lifeng Shang, Xin Jiang, Qun Liu, Kam-Fai Wong
Pretrained language models (PLMs) based knowledge-grounded dialogue systems are prone to generate responses that are factually inconsistent with the provided knowledge source. In such inconsistent responses, the dialogue models fail to accurately express the external knowledge they rely upon. Inspired by previous work which identified that feed-forward networks (FFNs) within Transformers are responsible for factual knowledge expressions, we investigate two methods to efficiently improve the factual expression capability {of FFNs} by knowledge enhancement and alignment respectively. We first propose \textsc{K-Dial}, which {explicitly} introduces {extended FFNs in Transformers to enhance factual knowledge expressions} given the specific patterns of knowledge-grounded dialogue inputs. Additionally, we apply the reinforcement learning for factual consistency (RLFC) method to implicitly adjust FFNs' expressions in responses by aligning with gold knowledge for the factual consistency preference. To comprehensively assess the factual consistency and dialogue quality of responses, we employ extensive automatic measures and human evaluations including sophisticated fine-grained NLI-based metrics. Experimental results on WoW and CMU\_DoG datasets demonstrate that our methods efficiently enhance the ability of the FFN module to convey factual knowledge, validating the efficacy of improving factual consistency for knowledge-grounded dialogue systems.
Authors: Yixin Wan, George Pu, Jiao Sun, Aparna Garimella, Kai-Wei Chang, Nanyun Peng
Large Language Models (LLMs) have recently emerged as an effective tool to assist individuals in writing various types of content, including professional documents such as recommendation letters. Though bringing convenience, this application also introduces unprecedented fairness concerns. Model-generated reference letters might be directly used by users in professional scenarios. If underlying biases exist in these model-constructed letters, using them without scrutinization could lead to direct societal harms, such as sabotaging application success rates for female applicants. In light of this pressing issue, it is imminent and necessary to comprehensively study fairness issues and associated harms in this real-world use case. In this paper, we critically examine gender biases in LLM-generated reference letters. Drawing inspiration from social science findings, we design evaluation methods to manifest biases through 2 dimensions: (1) biases in language style and (2) biases in lexical content. We further investigate the extent of bias propagation by analyzing the hallucination bias of models, a term that we define to be bias exacerbation in model-hallucinated contents. Through benchmarking evaluation on 2 popular LLMs- ChatGPT and Alpaca, we reveal significant gender biases in LLM-generated recommendation letters. Our findings not only warn against using LLMs for this application without scrutinization, but also illuminate the importance of thoroughly studying hidden biases and harms in LLM-generated professional documents.
Authors: Xiaoxi Kang, Lizhen Qu, Lay-Ki Soon, Adnan Trakic, Terry Yue Zhuo, Patrick Charles Emerton, Genevieve Grant
Large Language Models (LLMs), such as ChatGPT, have drawn a lot of attentions recently in the legal domain due to its emergent ability to tackle a variety of legal tasks. However, it is still unknown if LLMs are able to analyze a legal case and perform reasoning in the same manner as lawyers. Therefore, we constructed a novel corpus consisting of scenarios pertain to Contract Acts Malaysia and Australian Social Act for Dependent Child. ChatGPT is applied to perform analysis on the corpus using the IRAC method, which is a framework widely used by legal professionals for organizing legal analysis. Each scenario in the corpus is annotated with a complete IRAC analysis in a semi-structured format so that both machines and legal professionals are able to interpret and understand the annotations. In addition, we conducted the first empirical assessment of ChatGPT for IRAC analysis in order to understand how well it aligns with the analysis of legal professionals. Our experimental results shed lights on possible future research directions to improve alignments between LLMs and legal experts in terms of legal reasoning.
Authors: Weijia Shi, Anirudh Ajith, Mengzhou Xia, Yangsibo Huang, Daogao Liu, Terra Blevins, Danqi Chen, Luke Zettlemoyer
Although large language models (LLMs) are widely deployed, the data used to train them is rarely disclosed. Given the incredible scale of this data, up to trillions of tokens, it is all but certain that it includes potentially problematic text such as copyrighted materials, personally identifiable information, and test data for widely reported reference benchmarks. However, we currently have no way to know which data of these types is included or in what proportions. In this paper, we study the pretraining data detection problem: given a piece of text and black-box access to an LLM without knowing the pretraining data, can we determine if the model was trained on the provided text? To facilitate this study, we introduce a dynamic benchmark WIKIMIA that uses data created before and after model training to support gold truth detection. We also introduce a new detection method Min-K% Prob based on a simple hypothesis: an unseen example is likely to contain a few outlier words with low probabilities under the LLM, while a seen example is less likely to have words with such low probabilities. Min-K% Prob can be applied without any knowledge about the pretraining corpus or any additional training, departing from previous detection methods that require training a reference model on data that is similar to the pretraining data. Moreover, our experiments demonstrate that Min-K% Prob achieves a 7.4% improvement on WIKIMIA over these previous methods. We apply Min-K% Prob to three real-world scenarios, copyrighted book detection, contaminated downstream example detection and privacy auditing of machine unlearning, and find it a consistently effective solution.
Authors: Prakamya Mishra, Zonghai Yao, Shuwei Chen, Beining Wang, Rohan Mittal, Hong Yu
Large Language Models (LLMs) like the GPT and LLaMA families have demonstrated exceptional capabilities in capturing and condensing critical contextual information and achieving state-of-the-art performance in the summarization task. However, community concerns about these models' hallucination issues continue to rise. LLMs sometimes generate factually hallucinated summaries, which can be extremely harmful in the clinical domain NLP tasks (e.g., clinical note summarization), where factually incorrect statements can lead to critically erroneous diagnoses. Fine-tuning LLMs using human feedback has shown the promise of aligning LLMs to be factually consistent during generation, but such training procedure requires high-quality human-annotated data, which can be extremely expensive to get in the clinical domain. In this work, we propose a new pipeline using ChatGPT instead of human experts to generate high-quality feedback data for improving factual consistency in the clinical note summarization task. We focus specifically on edit feedback because recent work discusses the shortcomings of human alignment via preference feedback in complex situations (such as clinical NLP tasks that require extensive expert knowledge), as well as some advantages of collecting edit feedback from domain experts. In addition, although GPT has reached the expert level in many clinical NLP tasks (e.g., USMLE QA), there is not much previous work discussing whether GPT can generate expert-level edit feedback for LMs in the clinical note summarization task. We hope to fill this gap. Finally, our evaluations demonstrate the potential use of GPT edits in human alignment, especially from a factuality perspective.
Authors: Qiying Yu, Quan Sun, Xiaosong Zhang, Yufeng Cui, Fan Zhang, Yue Cao, Xinlong Wang, Jingjing Liu
Large multimodal models demonstrate remarkable generalist ability to perform diverse multimodal tasks in a zero-shot manner. Large-scale web-based image-text pairs contribute fundamentally to this success, but suffer from excessive noise. Recent studies use alternative captions synthesized by captioning models and have achieved notable benchmark performance. However, our experiments reveal significant Scalability Deficiency and World Knowledge Loss issues in models trained with synthetic captions, which have been largely obscured by their initial benchmark success. Upon closer examination, we identify the root cause as the overly-simplified language structure and lack of knowledge details in existing synthetic captions. To provide higher-quality and more scalable multimodal pretraining data, we propose CapsFusion, an advanced framework that leverages large language models to consolidate and refine information from both web-based image-text pairs and synthetic captions. Extensive experiments show that CapsFusion captions exhibit remarkable all-round superiority over existing captions in terms of model performance (e.g., 18.8 and 18.3 improvements in CIDEr score on COCO and NoCaps), sample efficiency (requiring 11-16 times less computation than baselines), world knowledge depth, and scalability. These effectiveness, efficiency and scalability advantages position CapsFusion as a promising candidate for future scaling of LMM training.
Authors: Mattia Opper, J. Morrison, N. Siddharth
This work explores the degree to which grammar acquisition is driven by language `simplicity' and the source modality (speech vs. text) of data. Using BabyBERTa as a probe, we find that grammar acquisition is largely driven by exposure to speech data, and in particular through exposure to two of the BabyLM training corpora: AO-Childes and Open Subtitles. We arrive at this finding by examining various ways of presenting input data to our model. First, we assess the impact of various sequence-level complexity based curricula. We then examine the impact of learning over `blocks' -- covering spans of text that are balanced for the number of tokens in each of the source corpora (rather than number of lines). Finally, we explore curricula that vary the degree to which the model is exposed to different corpora. In all cases, we find that over-exposure to AO-Childes and Open Subtitles significantly drives performance. We verify these findings through a comparable control dataset in which exposure to these corpora, and speech more generally, is limited by design. Our findings indicate that it is not the proportion of tokens occupied by high-utility data that aids acquisition, but rather the proportion of training steps assigned to such data. We hope this encourages future research into the use of more developmentally plausible linguistic data (which tends to be more scarce) to augment general purpose pre-training regimes.
Authors: Ryan Liu, Howard Yen, Raja Marjieh, Thomas L. Griffiths, Ranjay Krishna
How do we communicate with others to achieve our goals? We use our prior experience or advice from others, or construct a candidate utterance by predicting how it will be received. However, our experiences are limited and biased, and reasoning about potential outcomes can be difficult and cognitively challenging. In this paper, we explore how we can leverage Large Language Model (LLM) simulations to help us communicate better. We propose the Explore-Generate-Simulate (EGS) framework, which takes as input any scenario where an individual is communicating to an audience with a goal they want to achieve. EGS (1) explores the solution space by producing a diverse set of advice relevant to the scenario, (2) generates communication candidates conditioned on subsets of the advice, and (3) simulates the reactions from various audiences to determine both the best candidate and advice to use. We evaluate the framework on eight scenarios spanning the ten fundamental processes of interpersonal communication. For each scenario, we collect a dataset of human evaluations across candidates and baselines, and showcase that our framework's chosen candidate is preferred over popular generation mechanisms including Chain-of-Thought. We also find that audience simulations achieve reasonably high agreement with human raters across 5 of the 8 scenarios. Finally, we demonstrate the generality of our framework by applying it to real-world scenarios described by users on web forums. Through evaluations and demonstrations, we show that EGS enhances the effectiveness and outcomes of goal-oriented communication across a variety of situations, thus opening up new possibilities for the application of large language models in revolutionizing communication and decision-making processes.
Authors: Ke Hong, Guohao Dai, Jiaming Xu, Qiuli Mao, Xiuhong Li, Jun Liu, Kangdi Chen, Hanyu Dong, Yu Wang
As the Large Language Model (LLM) becomes increasingly important in various domains. However, the following challenges still remain unsolved in accelerating LLM inference: (1) Synchronized partial softmax update. The softmax operation requires a synchronized update operation among each partial softmax result, leading to ~20% overheads for the attention computation in LLMs. (2) Under-utilized computation of flat GEMM. The shape of matrices performing GEMM in LLM inference is flat, leading to under-utilized computation and >50% performance loss after padding zeros in previous designs. (3) Performance loss due to static dataflow. Kernel performance in LLM depends on varied input data features, hardware configurations, etc. A single and static dataflow may lead to a 50.25% performance loss for GEMMs of different shapes in LLM inference.
We present FlashDecoding++, a fast LLM inference engine supporting mainstream LLMs and hardware back-ends. To tackle the above challenges, FlashDecoding++ creatively proposes: (1) Asynchronized softmax with unified max value. FlashDecoding++ introduces a unified max value technique for different partial softmax computations to avoid synchronization. (2) Flat GEMM optimization with double buffering. FlashDecoding++ points out that flat GEMMs with different shapes face varied bottlenecks. Then, techniques like double buffering are introduced. (3) Heuristic dataflow with hardware resource adaptation. FlashDecoding++ heuristically optimizes dataflow using different hardware resource considering input dynamics. Due to the versatility of optimizations in FlashDecoding++, FlashDecoding++ can achieve up to 4.86x and 2.18x speedup on both NVIDIA and AMD GPUs compared to Hugging Face implementations. FlashDecoding++ also achieves an average speedup of 1.37x compared to state-of-the-art LLM inference engines on mainstream LLMs.
Authors: Yoshua Bengio, Geoffrey Hinton, Andrew Yao, Dawn Song, Pieter Abbeel, Yuval Noah Harari, Ya-Qin Zhang, Lan Xue, Shai Shalev-Shwartz, Gillian Hadfield, Jeff Clune, Tegan Maharaj, Frank Hutter, Atılım Güneş Baydin, Sheila McIlraith, Qiqi Gao, Ashwin Acharya, David Krueger, Anca Dragan, Philip Torr, Stuart Russell, Daniel Kahneman, Jan Brauner, Sören Mindermann
In this short consensus paper, we outline risks from upcoming, advanced AI systems. We examine large-scale social harms and malicious uses, as well as an irreversible loss of human control over autonomous AI systems. In light of rapid and continuing AI progress, we propose priorities for AI R&D and governance.