In embodied instruction-following (EIF), the integration of pretrained language models (LMs) as task planners emerges as a significant branch, where tasks are planned at the skill level by prompting LMs with pretrained skills and user instructions. However, grounding these pretrained skills in different domains remains challenging due to their intricate entanglement with the domain-specific knowledge. To address this challenge, we present a semantic skill grounding (SemGro) framework that leverages the hierarchical nature of semantic skills. SemGro recognizes the broad spectrum of these skills, ranging from short-horizon low-semantic skills that are universally applicable across domains to long-horizon rich-semantic skills that are highly specialized and tailored for particular domains. The framework employs an iterative skill decomposition approach, starting from the higher levels of semantic skill hierarchy and then moving downwards, so as to ground each planned skill to an executable level within the target domain. To do so, we use the reasoning capabilities of LMs for composing and decomposing semantic skills, as well as their multi-modal extension for assessing the skill feasibility in the target domain. Our experiments in the VirtualHome benchmark show the efficacy of SemGro in 300 cross-domain EIF scenarios.
论文检索
输入标题、作者或关键词，从 9,460 篇学术成果中精准定位
Enhancing Semantic Consistency of Large Language Models through Model Editing: An Interpretability-Oriented Approach
PDF ↗A Large Language Model (LLM) tends to generate inconsistent and sometimes contradictory outputs when presented with a prompt that has equivalent semantics but is expressed differently from the original prompt. To achieve semantic consistency of an LLM, one of the key approaches is to finetune the model with prompt-output pairs with semantically equivalent meanings. Despite its effectiveness, a data-driven finetuning method incurs substantial computation costs in data preparation and model optimization. In this regime, an LLM is treated as a “black box”, restricting our ability to gain deeper insights into its internal mechanism. In this paper, we are motivated to enhance the semantic consistency of LLMs through a more interpretable method (i.e., model editing) to this end. We first identify the model components (i.e., attention heads) that have a key impact on the semantic consistency of an LLM. We subsequently inject biases into the output of these model components along the semantic-consistency activation direction. It is noteworthy that these modifications are cost-effective, without reliance on mass manipulations of the original model parameters. Through comprehensive experiments on the constructed NLU and open-source NLG datasets, our method demonstrates significant improvements in the semantic consistency and task performance of LLMs. Additionally, our method exhibits promising generalization capabilities by performing well on tasks beyond the primary tasks.
VLMs (Vision-Language Models) extend the capabilities of LLMs (Large Language Models) to accept multimodal inputs. Since it has been verified that LLMs can be induced to generate harmful or inaccurate content through specific test cases (termed as Red Teaming), how VLMs perform in similar scenarios, especially with their combination of textual and visual inputs, remains a question. To explore this problem, we present a novel red teaming dataset RTVLM, which encompasses 12 subtasks (e.g., image misleading, multi-modal jailbreaking, face fairness, etc) under 4 primary aspects (faithfulness, privacy, safety, fairness). Our RTVLM is the first red teaming dataset to benchmark current VLMs in terms of these 4 different aspects. Detailed analysis shows that 10 prominent open-sourced VLMs struggle with the red teaming in different degrees and have up to 31% performance gap with GPT-4V. Additionally, we simply apply red teaming alignment to LLaVA-v1.5 with Supervised Fine-tuning (SFT) using RTVLM, and this bolsters the models’ performance with 10% in RTVLM test set, 13% in MM-hallu, and without noticeable decline in MM-Bench, overpassing other LLaVA-based models in similar size with regular alignment data. This reveals that current open-sourced VLMs still lack red teaming alignment. Our code and datasets will be open-sourced.
TimeChara: Evaluating Point-in-Time Character Hallucination of Role-Playing Large Language Models
PDF ↗While Large Language Models (LLMs) can serve as agents to simulate human behaviors (i.e., role-playing agents), we emphasize the importance of point-in-time role-playing. This situates characters at specific moments in the narrative progression for three main reasons: (i) enhancing users’ narrative immersion, (ii) avoiding spoilers, and (iii) fostering engagement in fandom role-playing. To accurately represent characters at specific time points, agents must avoid character hallucination, where they display knowledge that contradicts their characters’ identities and historical timelines. We introduce TimeChara, a new benchmark designed to evaluate point-in-time character hallucination in role-playing LLMs. Comprising 10,895 instances generated through an automated pipeline, this benchmark reveals significant hallucination issues in current state-of-the-art LLMs (e.g., GPT-4o). To counter this challenge, we propose Narrative-Experts, a method that decomposes the reasoning steps and utilizes narrative experts to reduce point-in-time character hallucinations effectively. Still, our findings with TimeChara highlight the ongoing challenges of point-in-time character hallucination, calling for further study.
Drama is a form of storytelling inspired by human creativity, proceeding with a predefined storyline, carrying emotions and thoughts.This paper introduces LLM-based interactive drama, which endows traditional drama with an unprecedented immersion, where a person is allowed to walk into it and interact with the characters and scenes.We define this new artistic genre by 6 essential elements—plot, character, thought, diction, spectacle and interaction—and study the entire pipeline to forge a backbone drama LLM to drive the playing process, which is challenged by limited drama resources, uncontrollable narrative development, and complicated instruction following.We propose Narrative Chain to offer finer control over the narrative progression during interaction with players;Auto-Drama to synthesize drama scripts given arbitrary stories;Sparse Instruction Tuning to allow the model to follow sophisticated instructions.We manually craft 3 scripts, Detective Conan, Harry Potter, Romeo and Juliet, and design a 5-dimension principle to evaluate the drama LLM comprehensively.
Large Language Models (LLMs) have shown remarkable comprehension abilities but face challenges in GPU memory usage during inference, hindering their scalability for real-time applications like chatbots. To accelerate inference, we store computed keys and values (KV cache) in the GPU memory. Existing methods study the KV cache compression to reduce memory by pruning the pre-computed KV cache. However, they neglect the inter-layer dependency between layers and huge memory consumption in pre-computation. To explore these deficiencies, we find that the number of crucial keys and values that influence future generations decreases layer by layer and we can extract them by the consistency in attention weights. Based on the findings, we propose PyramidInfer, a method that compresses the KV cache by layer-wise retaining crucial context. PyramidInfer saves significant memory by computing fewer keys and values without sacrificing performance. Experimental results show PyramidInfer improves 2.2x throughput compared to Accelerate with over 54% GPU memory reduction in KV cache.
The Music Maestro or The Musically Challenged, A Massive Music Evaluation Benchmark for Large Language Models
PDF ↗Benchmark plays a pivotal role in assessing the advancements of large language models (LLMs). While numerous benchmarks have been proposed to evaluate LLMs’ capabilities, there is a notable absence of a dedicated benchmark for assessing their musical abilities. To address this gap, we present ZIQI-Eval, a comprehensive and large-scale music benchmark specifically designed to evaluate the music-related capabilities of LLMs.ZIQI-Eval encompasses a wide range of questions, covering 10 major categories and 56 subcategories, resulting in over 14,000 meticulously curated data entries. By leveraging ZIQI-Eval, we conduct a comprehensive evaluation over 16 LLMs to evaluate and analyze LLMs’ performance in the domain of music.Results indicate that all LLMs perform poorly on the ZIQI-Eval benchmark, suggesting significant room for improvement in their musical capabilities.With ZIQI-Eval, we aim to provide a standardized and robust evaluation framework that facilitates a comprehensive assessment of LLMs’ music-related abilities. The dataset is available at GitHub and HuggingFace.
Large Language Models (LLMs) are widely used to evaluate natural language generation tasks as automated metrics.However, the likelihood, a measure of LLM’s plausibility for a sentence, can vary due to superficial differences in sentences, such as word order and sentence structure.It is therefore possible that there might be a likelihood bias if LLMs are used for evaluation: they might overrate sentences with higher likelihoods while underrating those with lower likelihoods.In this paper, we investigate the presence and impact of likelihood bias in LLM-based evaluators.We also propose a method to mitigate the likelihood bias.Our method utilizes highly biased instances as few-shot examples for in-context learning.Our experiments in evaluating the data-to-text and grammatical error correction tasks reveal that several LLMs we test display a likelihood bias.Furthermore, our proposed method successfully mitigates this bias, also improving evaluation performance (in terms of correlation of models with human scores) significantly.
Gender bias has been widely observed in NLP models, which has the potential to perpetuate harmful stereotypes and discrimination. In this paper, we construct a dataset GenderStance of 36k samples to measure gender bias in stance detection, determining whether models consistently predict the same stance for a particular gender group. We find that all models are gender-biased and prone to classify sentences that contain male nouns as Against and those with female nouns as Favor. Moreover, extensive experiments indicate that sources of gender bias stem from the fine-tuning data and the foundation model itself. We will publicly release our code and dataset.
Refine, Align, and Aggregate: Multi-view Linguistic Features Enhancement for Aspect Sentiment Triplet Extraction
PDF ↗Aspect Sentiment Triplet Extraction (ASTE) aims to extract the triplets of aspect terms, their associated sentiment and opinion terms. Previous works based on different modeling paradigms have achieved promising results. However, these methods struggle to comprehensively explore the various specific relations between sentiment elements in multi-view linguistic features, which is the prior indication effect for facilitating sentiment triplets extraction, requiring to align and aggregate them to capture the complementary higher-order interactions. In this paper, we propose Multi-view Linguistic Features Enhancement (MvLFE) to explore the aforementioned prior indication effect in the “Refine, Align, and Aggregate” learning process. Specifically, we first introduce the relational graph attention network to encode the word-pair relations represented by each linguistic feature and refine them to pay more attention to the aspect-opinion pairs. Next, we employ the multi-view contrastive learning to align them at a fine-grained level in the contextual semantic space to maintain semantic consistency. Finally, we utilize the multi-semantic cross attention to capture and aggregate the complementary higher-order interactions between diverse linguistic features to enhance the aspect-opinion relations. Experimental results on several benchmark datasets show the effectiveness and robustness of our model, which achieves state-of-the-art performance.
Document revision is a crucial aspect of the writing process, particularly in collaborative environments where multiple authors contribute simultaneously. However, current tools lack an efficient way to provide a comprehensive overview of changes between versions, leading to difficulties in understanding revisions. To address this, we propose a novel task of providing thematic summary of changes between document versions, organizing individual edits based on shared themes. We assess capabilities of LLMs on this task and further introduce three strategies to tackle this task: (i) representing the input of two documents along with edits in the ‘diff’ format (ii) a two-stage task decomposition with individual edit description generation as an intermediate task and (iii) clustering based chunking and subsequent merging techniques for handling longer documents. Our experiments demonstrate the effectiveness of our approach in improving the model’s capacity to handle this complex task. Additionally, we introduce ChangeSumm, a curated dataset comprising human-written thematic summaries for pairs of document versions, to facilitate evaluation and further research in this direction.
Artificial agents that learn to communicate in order to accomplish a given task acquire communication protocols that are typically opaque to a human. A large body of work has attempted to evaluate the emergent communication via various evaluation measures, with **compositionality** featuring as a prominent desired trait. However, current evaluation procedures do not directly expose the compositionality of the emergent communication. We propose a procedure to assess the compositionality of emergent communication by finding the best-match between emerged words and natural language concepts.The best-match algorithm provides both a global score and a translation-map from emergent words to natural language concepts. To the best of our knowledge, it is the first time that such direct and interpretable mapping between emergent words and human concepts is provided.
GENDEX: Generative Data Augmentation Strategy Leveraging External Data for Abstractive Dialogue Summarization
PDF ↗With the proliferation of digital communication, dialogue summarization has become increasingly important. However, it still faces a shortage of data. To address this issue, we developed **Gen**erative **D**ata Augmentation Strategy Leveraging **Ex**ternal Data for Abstractive Dialogue Summarization (**GENDEX**), which is based on the hypothetical foundation that texts containing people and their interpersonal interactions can potentially serve as summaries of corresponding dialogues. We filter short texts containing people and resolve coreferences for better contextual analysis. We then identify the semantic roles of words within the texts and filter them based on the patterns observed in the dialogue summarization datasets. Using these texts, we generate synthetic dialogues through a controlled generation method. To better leverage the augmented data, we utilize noise-tolerant training to fine-tune the summarization model. The experimental results demonstrate the effectiveness of our proposed method, showing its robust performance, generalizability, and scalability. Moreover, performance improvements by *GENDEX* were observed regardless of complexity of dialogues. The code is available at https://github.com/DMCB-GIST/GENDEX.
Autonomous graphical user interface (GUI) agents aim to facilitate task automation by interacting with the user interface without manual intervention. Recent studies have investigated eliciting the capabilities of large language models (LLMs) for effective engagement in diverse environments. To align with the input-output requirement of LLMs, most existing approaches are developed under a sandbox setting where they rely on external tools and application-specific APIs to parse the environment into textual elements and interpret the predicted actions. Consequently, those approaches often grapple with inference inefficiency and error propagation risks. To mitigate the challenges, we introduce Auto-GUI, a multimodal solution that directly interacts with the interface, bypassing the need for environment parsing or reliance on application-dependent APIs. Moreover, we propose a chain-of-action technique—leveraging a series of intermediate previous action histories and future action plans—to help the agent decide what action to execute. We evaluate our approach on a new device-control benchmark AITW with 30K unique instructions, spanning multi-step tasks such as application operation, web searching, and web shopping. Experimental results show that Auto-GUI achieves state-of-the-art performance with an action type prediction accuracy of 90% and an overall action success rate of 74%. Code is publicly available at https://github.com/cooelf/Auto-GUI.
LLMFactor: Extracting Profitable Factors through Prompts for Explainable Stock Movement Prediction
PDF ↗Recently, Large Language Models (LLMs) have attracted significant attention for their exceptional performance across a broad range of tasks, particularly in text analysis. However, the finance sector presents a distinct challenge due to its dependence on time-series data for complex forecasting tasks. In this study, we introduce a novel framework called LLMFactor, which employs Sequential Knowledge-Guided Prompting (SKGP) to identify factors that influence stock movements using LLMs. Unlike previous methods that relied on keyphrases or sentiment analysis, this approach focuses on extracting factors more directly related to stock market dynamics, providing clear explanations for complex temporal changes. Our framework directs the LLMs to create background knowledge through a fill-in-the-blank strategy and then discerns potential factors affecting stock prices from related news. Guided by background knowledge and identified factors, we leverage historical stock prices in textual format to predict stock movement. An extensive evaluation of the LLMFactor framework across four benchmark datasets from both the U.S. and Chinese stock markets demonstrates its superiority over existing state-of-the-art methods and its effectiveness in financial time-series forecasting.
Many studies have demonstrated that large language models (LLMs) can produce harmful responses, exposing users to unexpected risks. Previous studies have proposed comprehensive taxonomies of LLM risks, as well as corresponding prompts that can be used to examine LLM safety. However, the focus has been almost exclusively on English. We aim to broaden LLM safety research by introducing a dataset for the safety evaluation of Chinese LLMs, and extending it to better identify false negative and false positive examples in terms of risky prompt rejections. We further present a set of fine-grained safety assessment criteria for each risk type, facilitating both manual annotation and automatic evaluation in terms of LLM response harmfulness. Our experiments over five LLMs show that region-specific risks are the prevalent risk type. Warning: this paper contains example data that may be offensive, harmful, or biased. Our data is available at https://github.com/Libr-AI/do-not-answer.
Modeling Dynamic Topics in Chain-Free Fashion by Evolution-Tracking Contrastive Learning and Unassociated Word Exclusion
PDF ↗Dynamic topic models track the evolution of topics in sequential documents, which have derived various applications like trend analysis. However, existing models suffer from repetitive topic and unassociated topic issues, failing to reveal the evolution and hindering further applications. To address these issues, we break the tradition of simply chaining topics in existing work and propose a novel neural Chain-Free Dynamic Topic Model. We introduce a new evolution-tracking contrastive learning method that builds the similarity relations among dynamic topics. This not only tracks topic evolution but also maintains topic diversity, mitigating the repetitive topic issue. To avoid unassociated topics, we further present an unassociated word exclusion method that consistently excludes unassociated words from discovered topics. Extensive experiments demonstrate our model significantly outperforms state-of-the-art baselines, tracking topic evolution with high-quality topics, showing better performance on downstream tasks, and remaining robust to the hyperparameter for evolution intensities.
Recently, the use of pre-trained generation models for extracting sentiment elements has resulted in significant advancements in aspect-based sentiment analysis benchmarks. However, these approaches often overlook the importance of explicitly modeling structure among sentiment elements. To address this limitation, we present a study that aims to integrate general pre-trained sequence-to-sequence language models with a structure-aware transition-based approach. Therefore, we propose a transition system for opinion tree generation, designed to better exploit pre-trained language models for structured fine-tuning. Our proposed transition system ensures the structural integrity of the generated opinion tree. By leveraging pre-trained generation models and simplifying the transition set, we are able to maximize the accuracy of opinion tree generation. Extensive experiments show that our model significantly advances the state-of-the-art performance on several benchmark datasets. In addition, the empirical studies also indicate that the proposed opinion tree generation with transition system is more effective in capturing the sentiment structure than other generation models.
Open large language models (LLMs) with great performance in various tasks have significantly advanced the development of LLMs. However, they are far inferior to commercial models such as ChatGPT and GPT-4 when acting as agents to tackle complex tasks in the real world. These agent tasks employ LLMs as the central controller responsible for planning, memorization, and tool utilization, necessitating both fine-grained prompting methods and robust LLMs to achieve satisfactory performance. Though many prompting methods have been proposed to complete particular agent tasks, there is lack of research focusing on improving the agent capabilities of LLMs themselves without compromising their general abilities. In this work, we present AgentTuning, a simple and general method to enhance the agent abilities of LLMs while maintaining their general LLM capabilities. We construct AgentInstruct, a lightweight instruction-tuning dataset containing high-quality interaction trajectories. We employ a hybrid instruction-tuning strategy by combining AgentInstruct with open-source instructions from general domains. AgentTuning is used to instruction-tune the Llama 2 series, resulting in AgentLM. Our evaluations show that AgentTuning enables LLMs’ agent capabilities without compromising general abilities. The AgentLM-70B is comparable to GPT-3.5-turbo on unseen agent tasks, demonstrating generalized agent capabilities. We open source the AgentInstruct and AgentLM-7B, 13B, and 70B models at https://anonymous.4open.science/r/AgentTuning, serving open and powerful alternatives to commercial LLMs for agent tasks.