Textual style transfer is the task of transforming stylistic properties of text while preserving meaning. Target "styles" can be defined in numerous ways, ranging from single attributes (e.g. formality) to authorship (e.g. Shakespeare). Previous unsupervised style-transfer approaches generally rely on significant amounts of labeled data for only a fixed set of styles or require large language models. In contrast, we introduce a novel diffusion-based framework for general-purpose style transfer that can be flexibly adapted to arbitrary target styles at inference time. Our parameter-efficient approach, ParaGuide, leverages paraphrase-conditioned diffusion models alongside gradient-based guidance from both off-the-shelf classifiers and strong existing style embedders to transform the style of text while preserving semantic information. We validate the method on the Enron Email Corpus, with both human and automatic evaluations, and find that it outperforms strong baselines on formality, sentiment, and even authorship style transfer.
论文检索
输入标题、作者或关键词，从 11,272 篇学术成果中精准定位
Text2Analysis: A Benchmark of Table Question Answering with Advanced Data Analysis and Unclear Queries
PDF ↗Tabular data analysis is crucial in various fields, and large language models show promise in this area. However, current research mostly focuses on rudimentary tasks like Text2SQL and TableQA, neglecting advanced analysis like forecasting and chart generation. To address this gap, we developed the Text2Analysis benchmark, incorporating advanced analysis tasks that go beyond the SQL-compatible operations and require more in-depth analysis. We also develop five innovative and effective annotation methods, harnessing the capabilities of large language models to enhance data quality and quantity. Additionally, we include unclear queries that resemble real-world user questions to test how well models can understand and tackle such challenges. Finally, we collect 2249 query-result pairs with 347 tables. We evaluate five state-of-the-art models using three different metrics and the results show that our benchmark presents introduces considerable challenge in the field of tabular data analysis, paving the way for more advanced research opportunities.
Factual error correction (FEC) aims to revise factual errors in false claims with minimal editing, making them faithful to the provided evidence. This task is crucial for alleviating the hallucination problem encountered by large language models. Given the lack of paired data (i.e., false claims and their corresponding correct claims), existing methods typically adopt the ‘mask-then-correct’ paradigm. This paradigm relies solely on unpaired false claims and correct claims, thus being referred to as distantly supervised methods. These methods require a masker to explicitly identify factual errors within false claims before revising with a corrector. However, the absence of paired data to train the masker makes accurately pinpointing factual errors within claims challenging. To mitigate this, we propose to improve FEC by Learning to Inject Factual Errors (LIFE), a three-step distantly supervised method: ‘mask-corrupt-correct’. Specifically, we first train a corruptor using the ‘mask-then-corrupt’ procedure, allowing it to deliberately introduce factual errors into correct text. The corruptor is then applied to correct claims, generating a substantial amount of paired data. After that, we filter out low-quality data, and use the remaining data to train a corrector. Notably, our corrector does not require a masker, thus circumventing the bottleneck associated with explicit factual error identification. Our experiments on a public dataset verify the effectiveness of LIFE in two key aspects: Firstly, it outperforms the previous best-performing distantly supervised method by a notable margin of 10.59 points in SARI Final (19.3% improvement). Secondly, even compared to ChatGPT prompted with in-context examples, LIFE achieves a superiority of 7.16 points in SARI Final.
Large language models (LLMs) can understand human instructions, showing their potential for pragmatic applications beyond traditional NLP tasks. However, they still struggle with complex instructions, which can be either complex task descriptions that require multiple tasks and constraints, or complex input that contains long context, noise, heterogeneous information and multi-turn format. Due to these features, LLMs often ignore semantic constraints from task descriptions, generate incorrect formats, violate length or sample count constraints, and be unfaithful to the input text. Existing benchmarks are insufficient to assess LLMs’ ability to understand complex instructions, as they are close-ended and simple. To bridge this gap, we propose CELLO, a benchmark for evaluating LLMs' ability to follow complex instructions systematically. We design eight features for complex instructions and construct a comprehensive evaluation dataset from real-world scenarios. We also establish four criteria and develop corresponding metrics, as current ones are inadequate, biased or too strict and coarse-grained. We compare the performance of representative Chinese-oriented and English-oriented models in following complex instructions through extensive experiments. Resources of CELLO are publicly available at https://github.com/Abbey4799/CELLO.
Chain-of-thought (CoT) reasoning has exhibited impressive performance in language models for solving complex tasks and answering questions. However, many real-world questions require multi-modal information, such as text and images. Previous research on multi-modal CoT has primarily focused on extracting fixed image features from off-the-shelf vision models and then fusing them with text using attention mechanisms. This approach has limitations because these vision models were not designed for complex reasoning tasks and do not align well with language thoughts. To overcome this limitation, we introduce a novel approach for multi-modal CoT reasoning that utilizes latent space learning via diffusion processes to generate effective image features that align with language thoughts. Our method fuses image features and text representations at a deep level and improves the complex reasoning ability of multi-modal CoT. We demonstrate the efficacy of our proposed method on multi-modal ScienceQA and machine translation benchmarks, achieving state-of-the-art performance on ScienceQA. Overall, our approach offers a more robust and effective solution for multi-modal reasoning in language models, enhancing their ability to tackle complex real-world problems.
Few-shot intent classification and slot filling are important but challenging tasks due to the scarcity of finely labeled data. Therefore, current works first train a model on source domains with sufficiently labeled data, and then transfer the model to target domains where only rarely labeled data is available. However, experience transferring as a whole usually suffers from gaps that exist among source domains and target domains. For instance, transferring domain-specific-knowledge-related experience is difficult. To tackle this problem, we propose a new method that explicitly decouples the transferring of general-semantic-representation-related experience and the domain-specific-knowledge-related experience. Specifically, for domain-specific-knowledge-related experience, we design two modules to capture intent-slot relation and slot-slot relation respectively. Extensive experiments on Snips and FewJoint datasets show that our method achieves state-of-the-art performance. The method improves the joint accuracy metric from 27.72% to 42.20% in the 1-shot setting, and from 46.54% to 60.79% in the 5-shot setting.
Generative Language Models (LMs) such as ChatGPT have exhibited remarkable performance across various downstream tasks. Nevertheless, one of their most prominent drawbacks is generating inaccurate or false information with a confident tone. Previous studies have devised sophisticated pipelines and prompts to induce large LMs to exhibit the capability for self-correction. However, large LMs are explicitly prompted to verify and modify their answers separately rather than completing all steps spontaneously like humans. Moreover, these complex prompts are extremely challenging for small LMs to follow. In this paper, we introduce the Intrinsic Self-Correction (ISC) in generative language models, aiming to correct the initial output of LMs in a self-triggered manner, even for those small LMs with 6 billion parameters. Specifically, we devise a pipeline for constructing self-correction data and propose Partial Answer Masking (PAM), aiming to endow the model with the capability for intrinsic self-correction through fine-tuning. We conduct experiments using LMs with parameters sizes ranging from 6 billion to 13 billion in two tasks, including commonsense reasoning and factual knowledge reasoning. Our experiments demonstrate that the outputs generated using ISC outperform those generated without self-correction. We believe that the output quality of even small LMs can be further improved by empowering them with the ability to intrinsic self-correct.
Text-based audio generation models have limitations as they cannot encompass all the information in audio, leading to restricted controllability when relying solely on text. To address this issue, we propose a novel model that enhances the controllability of existing pre-trained text-to-audio models by incorporating additional conditions including content (timestamp) and style (pitch contour and energy contour) as supplements to the text. This approach achieves fine-grained control over the temporal order, pitch, and energy of generated audio. To preserve the diversity of generation, we employ a trainable control condition encoder that is enhanced by a large language model and a trainable Fusion-Net to encode and fuse the additional conditions while keeping the weights of the pre-trained text-to-audio model frozen. Due to the lack of suitable datasets and evaluation metrics, we consolidate existing datasets into a new dataset comprising the audio and corresponding conditions and use a series of evaluation metrics to evaluate the controllability performance. Experimental results demonstrate that our model successfully achieves fine-grained control to accomplish controllable audio generation.
Deep learning is now widely used in drug discovery, providing significant acceleration and cost reduction. As the most fundamental building block, molecular representation is essential for predicting molecular properties to enable various downstream applications. Most existing methods attempt to incorporate more information to learn better representations. However, not all features are equally important for a specific task. Ignoring this would potentially compromise the training efficiency and predictive accuracy. To address this issue, we propose a novel approach, which treats language models as an agent and molecular pretraining models as a knowledge base. The agent accentuates task-relevant features in the molecular representation by understanding the natural language description of the task, just as a tailor customizes clothes for clients. Thus, we call this approach MolTailor. Evaluations demonstrate MolTailor's superior performance over baselines, validating the efficacy of enhancing relevance for molecular representation learning. This illustrates the potential of language model guided optimization to better exploit and unleash the capabilities of existing powerful molecular representation methods. Our code and appendix are available at https://github.com/SCIR-HI/MolTailor.
Instruction tuned Large Vision Language Models (LVLMs) have significantly advanced in generalizing across a diverse set of multi-modal tasks, especially for Visual Question Answering (VQA). However, generating detailed responses that are visually grounded is still a challenging task for these models. We find that even the current state-of-the-art LVLMs (InstructBLIP) still contain a staggering 30 percent of the hallucinatory text in the form of non-existent objects, unfaithful descriptions, and inaccurate relationships. To address this, we introduce M-HalDetect, a Multimodal Hallucination Detection Dataset that can be used to train and benchmark models for hallucination detection and prevention. M-HalDetect consists of 16k fine-grained annotations on VQA examples, making it the first comprehensive multi-modal hallucination detection dataset for detailed image descriptions. Unlike previous work that only consider object hallucination, we additionally annotate both entity descriptions and relationships that are unfaithful. To demonstrate the potential of this dataset for hallucination prevention, we optimize InstructBLIP through our novel Fine-grained Direct Preference Optimization (FDPO). We also train fine-grained multi-modal reward models from InstructBLIP and evaluate their effectiveness with best-of-n rejection sampling (RS). We perform human evaluation on both FDPO and rejection sampling, and find that they reduce hallucination rates in InstructBLIP by 41% and 55% respectively. We also find that our reward model generalizes to other multi-modal models, reducing hallucinations in LLaVA and mPLUG-OWL by 15% and 57% respectively, and has strong correlation with human evaluated accuracy scores. The dataset is available at https://github.com/hendryx-scale/mhal-detect.
Mitigating Large Language Model Hallucinations via Autonomous Knowledge Graph-Based Retrofitting
PDF ↗Incorporating factual knowledge in knowledge graph is regarded as a promising approach for mitigating the hallucination of large language models (LLMs). Existing methods usually only use the user's input to query the knowledge graph, thus failing to address the factual hallucination generated by LLMs during its reasoning process. To address this problem, this paper proposes Knowledge Graph-based Retrofitting (KGR), a new framework that incorporates LLMs with KGs to mitigate factual hallucination during the reasoning process by retrofitting the initial draft responses of LLMs based on the factual knowledge stored in KGs. Specifically, KGR leverages LLMs to extract, select, validate, and retrofit factual statements within the model-generated responses, which enables an autonomous knowledge verifying and refining procedure without any additional manual efforts. Experiments show that KGR can significantly improve the performance of LLMs on factual QA benchmarks especially when involving complex reasoning processes, which demonstrates the necessity and effectiveness of KGR in mitigating hallucination and enhancing the reliability of LLMs.
The style transfer task in Text-to-Speech (TTS) refers to the process of transferring style information into text content to generate corresponding speech with a specific style. However, most existing style transfer approaches are either based on fixed emotional labels or reference speech clips, which cannot achieve flexible style transfer. Recently, some methods have adopted text descriptions to guide style transfer. In this paper, we propose a more flexible multi-modal and style controllable TTS framework named MM-TTS. It can utilize any modality as the prompt in unified multi-modal prompt space, including reference speech, emotional facial images, and text descriptions, to control the style of the generated speech in a system. The challenges of modeling such a multi-modal style controllable TTS mainly lie in two aspects: 1) aligning the multi-modal information into a unified style space to enable the input of arbitrary modality as the style prompt in a single system, and 2) efficiently transferring the unified style representation into the given text content, thereby empowering the ability to generate prompt style-related voice. To address these problems, we propose an aligned multi-modal prompt encoder that embeds different modalities into a unified style space, supporting style transfer for different modalities. Additionally, we present a new adaptive style transfer method named Style Adaptive Convolutions (SAConv) to achieve a better style representation. Furthermore, we design a Rectified Flow based Refiner to solve the problem of over-smoothing Mel-spectrogram and generate audio of higher fidelity. Since there is no public dataset for multi-modal TTS, we construct a dataset named MEAD-TTS, which is related to the field of expressive talking head. Our experiments on the MEAD-TTS dataset and out-of-domain datasets demonstrate that MM-TTS can achieve satisfactory results based on multi-modal prompts. The audio samples and constructed dataset are available at https://multimodal-tts.github.io.
Instruction-following is particularly crucial for large language models (LLMs) to support diverse user requests. While existing work has made progress in aligning LLMs with human preferences, evaluating their capabilities on instruction-following remains a challenge due to complexity and diversity of real-world user instructions. While existing evaluation methods focus on general skills, they suffer from two main shortcomings, i.e., lack of fine-grained task-level evaluation and reliance on singular instruction expression. To address these problems, this paper introduces DINGO, a fine-grained and diverse instruction-following evaluation dataset that has two main advantages: (1) DINGO is based on a manual annotated, fine-grained and multi-level category tree with 130 nodes derived from real-world user requests; (2) DINGO includes diverse instructions, generated by both GPT-4 and human experts. Through extensive experiments, we demonstrate that DINGO can not only provide more challenging and comprehensive evaluation for LLMs, but also provide task-level fine-grained directions to further improve LLMs.
New Natural Langauge Process~(NLP) benchmarks are urgently needed to align with the rapid development of large language models (LLMs). We present Xiezhi, the most comprehensive evaluation suite designed to assess holistic domain knowledge.Xiezhi comprises multiple-choice questions across 516 diverse disciplines ranging from 13 different subjects with 249,587 questions and accompanied by Xiezhi-Specialty with 14,041 questions and Xiezhi-Interdiscipline with 10,746 questions. We conduct evaluation of the 47 cutting-edge LLMs on Xiezhi. Results indicate that LLMs exceed average performance of humans in science, engineering, agronomy, medicine, and art, but fall short in economics, jurisprudence, pedagogy, literature, history, and management. All the evaluation code and data are open sourced in https://github.com/MikeGu721/XiezhiBenchmark
Recent advances in multimodal learning has resulted in powerful vision-language models, whose representations are generalizable across a variety of downstream tasks. Recently, their generalization ability has been further extended by incorporating trainable prompts, borrowed from the natural language processing literature. While such prompt learning techniques have shown impressive results, we identify that these prompts are trained based on global image features which limits itself in two aspects: First, by using global features, these prompts could be focusing less on the discriminative foreground image, resulting in poor generalization to various out-of-distribution test cases. Second, existing work weights all prompts equally whereas intuitively, prompts should be reweighed according to the semantics of the image. We address these as part of our proposed Contextual Prompt Learning (CoPL) framework, capable of aligning the prompts to the localized features of the image. Our key innovations over earlier works include using local image features as part of the prompt learning process, and more crucially, learning to weight these prompts based on local features that are appropriate for the task at hand. This gives us dynamic prompts that are both aligned to local image features as well as aware of local contextual relationships. Our extensive set of experiments on a variety of standard and few-shot datasets show that our method produces substantially improved performance when compared to the current state of the art methods. We also demonstrate both few-shot and out-of-distribution performance to establish the utility of learning dynamic prompts that are aligned to local image features.
What Makes Quantization for Large Language Model Hard? An Empirical Study from the Lens of Perturbation
PDF ↗Quantization has emerged as a promising technique for improving the memory and computational efficiency of large language models (LLMs). Though the trade-off between performance and efficiency is well-known, there is still much to be learned about the relationship between quantization and LLM performance. To shed light on this relationship, we propose a new perspective on quantization, viewing it as perturbations added to the weights and activations of LLMs. We call this approach ``the lens of perturbation". Using this lens, we conduct experiments with various artificial perturbations to explore their impact on LLM performance. Our findings reveal several connections between the properties of perturbations and LLM performance, providing insights into the failure cases of uniform quantization and suggesting potential solutions to improve the robustness of LLM quantization. To demonstrate the significance of our findings, we implement a simple non-uniform quantization approach based on our insights. Our experiments show that this approach achieves minimal performance degradation on both 4-bit weight quantization and 8-bit quantization for weights and activations. These results validate the correctness of our approach and highlight its potential to improve the efficiency of LLMs without sacrificing performance.
One of the prominent methods for explaining the decision of a machine-learning classifier is by a counterfactual example. Most current algorithms for generating such examples in the textual domain are based on generative language models. Generative models, however, are trained to minimize a specific loss function in order to fulfill certain requirements for the generated texts. Any change in the requirements may necessitate costly retraining, thus potentially limiting their applicability. In this paper, we present a general search-based framework for generating counterfactual explanations in the textual domain. Our framework is model-agnostic, domain-agnostic, anytime, and does not require retraining in order to adapt to changes in the user requirements. We model the task as a search problem in a space where the initial state is the classified text, and the goal state is a text in a given target class. Our framework includes domain-independent modification operators, but can also exploit domain-specific knowledge through specialized operators. The search algorithm attempts to find a text from the target class with minimal user-specified distance from the original classified object.
Who Knows the Answer? Finding the Best Model and Prompt for Each Query Using Confidence-Based Search
PDF ↗There are increasingly many large language models (LLMs) available to the public. While these LLMs have exhibited impressive abilities on a variety of task, any individual LLM in particular may do well on some tasks and worse on others. Additionally, the performance of these models is heavily dependent on the choice of prompt template used. For instance, they exhibit sensitivity to the few shot examples chosen or brittleness to the wording of instructions. Moreover, a prompt template that makes a model perform well for one input may not be the optimal template for another input. This necessitates an approach for adaptively selecting LLM and prompt template pairs for each input. Recent work has shown that the accuracy of LLM's responses is correlated with the LLM's confidence in the response. Thus, a natural choice for selecting which model and prompt template to use is to select the pair that is most confident in its response. However, existing confidence metrics are expensive to calculate - necessitating multiple calls to each LLm and prompt pair. We thus propose an approach to predict the confidence of each pair using an auxiliary regression model that is inexpensive to run. Using this auxiliary model, we select the LLM and prompt template with the highest predicted confidence for a given input. Results on a range of benchmark datasets show that our confidence-based instance-level prompt search method consistently improves the performance of LLMs.
Discrepancy and Uncertainty Aware Denoising Knowledge Distillation for Zero-Shot Cross-Lingual Named Entity Recognition
PDF ↗The knowledge distillation-based approaches have recently yielded state-of-the-art (SOTA) results for cross-lingual NER tasks in zero-shot scenarios. These approaches typically employ a teacher network trained with the labelled source (rich-resource) language to infer pseudo-soft labels for the unlabelled target (zero-shot) language, and force a student network to approximate these pseudo labels to achieve knowledge transfer. However, previous works have rarely discussed the issue of pseudo-label noise caused by the source-target language gap, which can mislead the training of the student network and result in negative knowledge transfer. This paper proposes an discrepancy and uncertainty aware Denoising Knowledge Distillation model (DenKD) to tackle this issue. Specifically, DenKD uses a discrepancy-aware denoising representation learning method to optimize the class representations of the target language produced by the teacher network, thus enhancing the quality of pseudo labels and reducing noisy predictions. Further, DenKD employs an uncertainty-aware denoising method to quantify the pseudo-label noise and adjust the focus of the student network on different samples during knowledge distillation, thereby mitigating the noise's adverse effects. We conduct extensive experiments on 28 languages including 4 languages not covered by the pre-trained models, and the results demonstrate the effectiveness of our DenKD.
Multi-Source cross-lingual transfer learning deals with the transfer of task knowledge from multiple labelled source languages to an unlabeled target language under the language shift. Existing methods typically focus on weighting the predictions produced by language-specific classifiers of different sources that follow a shared encoder. However, all source languages share the same encoder, which is updated by all these languages. The extracted representations inevitably contain different source languages' information, which may disturb the learning of the language-specific classifiers. Additionally, due to the language gap, language-specific classifiers trained with source labels are unable to make accurate predictions for the target language. Both facts impair the model's performance. To address these challenges, we propose a Disentangled and Adaptive Network ~(DA-Net). Firstly, we devise a feedback-guided collaborative disentanglement method that seeks to purify input representations of classifiers, thereby mitigating mutual interference from multiple sources. Secondly, we propose a class-aware parallel adaptation method that aligns class-level distributions for each source-target language pair, thereby alleviating the language pairs' language gap. Experimental results on three different tasks involving 38 languages validate the effectiveness of our approach.