Recent advances in large language models (LLMs) have blurred the boundary of high-quality text generation between humans and machines, which is favorable for generative text steganography. While, current advanced steganographic mapping is not suitable for LLMs since most users are restricted to accessing only the black-box API or user interface of the LLMs, thereby lacking access to the training vocabulary and its sampling probabilities. In this paper, we explore a black-box generative text steganographic method based on the user interfaces of large language models, which is called LLM-Stega. The main goal of LLM-Stega is that the secure covert communication between Alice (sender) and Bob (receiver) is conducted by using the user interfaces of LLMs. Specifically, We first construct a keyword set and design a new encrypted steganographic mapping to embed secret messages. Furthermore, to guarantee accurate extraction of secret messages and rich semantics of generated stego texts, an optimization mechanism based on reject sampling is proposed. Comprehensive experiments demonstrate that the proposed LLM-Stega outperforms current state-of-the-art methods.
Diffusion models have exhibited remarkable capabilities in text-to-image generation. However, their performance in image-to-text generation, specifically image captioning, has lagged behind Auto-Regressive (AR) models, casting doubt on their applicability for such tasks. In this work, we revisit diffusion models, highlighting their capacity for holistic context modeling and parallel decoding. With these benefits, diffusion models can alleviate the inherent limitations of AR methods, including their slow inference speed, error propagation, and unidirectional constraints. Furthermore, we identify the prior underperformance of diffusion models stemming from the absence of an effective latent space for image-text alignment, and the discrepancy between continuous diffusion processes and discrete textual data. In response, we introduce a novel architecture, LaDiC, which utilizes a split BERT to create a dedicated latent space for captions and integrates a regularization module to manage varying text lengths. Our framework also includes a diffuser for semantic image-to-text conversion and a Back&Refine technique to enhance token interactivity during inference. LaDiC achieves state-of-the-art performance for diffusion-based methods on the MS COCO dataset with 38.2 BLEU@4 and 126.2 CIDEr, demonstrating exceptional performance without pre-training or ancillary modules. This indicates strong competitiveness with AR models, revealing the previously untapped potential of diffusion models in image-to-text generation.
Social biases can manifest in language agency. However, very limited research has investigated such biases in Large Language Model (LLM)-generated content. In addition, previous works often rely on string-matching techniques to identify agentic and communal words within texts, falling short of accurately classifying language agency. We introduce the Language Agency Bias Evaluation (LABE) benchmark, which comprehensively evaluates biases in LLMs by analyzing agency levels attributed to different demographic groups in model generations. LABE tests for gender, racial, and intersectional language agency biases in LLMs on 3 text generation tasks: biographies, professor reviews, and reference letters. Using LABE, we unveil language agency social biases in 3 recent LLMs: ChatGPT, Llama3, and Mistral. We observe that: (1) LLM generations tend to demonstrate greater gender bias than human-written texts; (2) Models demonstrate remarkably higher levels of intersectional bias than the other bias aspects. (3) Prompt-based mitigation is unstable and frequently leads to bias exacerbation. Based on our observations, we propose Mitigation via Selective Rewrite (MSR), a novel bias mitigation strategy that leverages an agency classifier to identify and selectively revise parts of generated texts that demonstrate communal traits. Empirical results prove MSR to be more effective and reliable than prompt-based mitigation method, showing a promising research direction.
Citation Text Generation (CTG) is a task in natural language processing (NLP) that aims to produce text that accurately cites or references a cited document within a source document. In CTG, the generated text draws upon contextual cues from both the source document and the cited paper, ensuring accurate and relevant citation information is provided. Previous work in the field of citation generation is mainly based on the text summarization of documents. Following this, this paper presents a framework, and a comparative study to demonstrate the use of Large Language Models (LLMs) for the task of citation generation. Also, we have shown the improvement in the results of citation generation by incorporating the knowledge graph relations of the papers in the prompt for the LLM to better learn the relationship between the papers. To assess how well our model is performing, we have used a subset of standard S2ORC dataset, which only consists of computer science academic research papers in the English Language. Vicuna performs best for this task with 14.15 Meteor, 12.88 Rouge-1, 1.52 Rouge-2, and 10.94 Rouge-L. Also, Alpaca performs best, and improves the performance by 36.98% in Rouge-1, and 33.14% in Meteor by including knowledge graphs.
Ensembling different large language models (LLMs) to unleash their complementary potential and harness their individual strengths is highly valuable. Nevertheless, vocabulary discrepancies among various LLMs have constrained previous studies to either selecting or blending completely generated outputs. This limitation hinders the dynamic correction and enhancement of outputs during the generation process, resulting in a limited capacity for effective ensemble. To address this issue, we propose a novel method to Ensemble LLMs via Vocabulary Alignment (EVA). EVA bridges the lexical gap among various LLMs, enabling meticulous ensemble at each generation step. Specifically, we first learn mappings between the vocabularies of different LLMs with the assistance of overlapping tokens. Subsequently, these mappings are employed to project output distributions of LLMs into a unified space, facilitating a fine-grained ensemble. Finally, we design a filtering strategy to exclude models that generate unfaithful tokens. Experimental results on commonsense reasoning, arithmetic reasoning, machine translation, and data-to-text generation tasks demonstrate the superiority of our approach compared with individual LLMs and previous ensemble methods conducted on complete outputs. Further analyses confirm that our approach can leverage knowledge from different language models and yield consistent improvement.
Natural Language Processing (NLP) is witnessing a remarkable breakthrough driven by the success of Large Language Models (LLMs). LLMs have gained significant attention across academia and industry for their versatile applications in text generation, question answering, and text summarization. As the landscape of NLP evolves with an increasing number of domain-specific LLMs employing diverse techniques and trained on various corpus, evaluating performance of these models becomes paramount. To quantify the performance, it's crucial to have a comprehensive grasp of existing metrics. Among the evaluation, metrics which quantifying the performance of LLMs play a pivotal role. This paper offers a comprehensive exploration of LLM evaluation from a metrics perspective, providing insights into the selection and interpretation of metrics currently in use. Our main goal is to elucidate their mathematical formulations and statistical interpretations. We shed light on the application of these metrics using recent Biomedical LLMs. Additionally, we offer a succinct comparison of these metrics, aiding researchers in selecting appropriate metrics for diverse tasks. The overarching goal is to furnish researchers with a pragmatic guide for effective LLM evaluation and metric selection, thereby advancing the understanding and application of these large language models.
The task of Split and Rephrase, which splits a complex sentence into multiple simple sentences with the same meaning, improves readability and enhances the performance of downstream tasks in natural language processing (NLP). However, while Split and Rephrase can be improved using a text-to-text generation approach that applies encoder-decoder models fine-tuned with a large-scale dataset, it still suffers from hallucinations and under-splitting. To address these issues, this paper presents a simple and strong data refinement approach. Here, we create WikiSplit++ by removing instances in WikiSplit where complex sentences do not entail at least one of the simpler sentences and reversing the order of reference simple sentences. Experimental results show that training with WikiSplit++ leads to better performance than training with WikiSplit, even with fewer training instances. In particular, our approach yields significant gains in the number of splits and the entailment ratio, a proxy for measuring hallucinations.
The Parallel Meaning Bank (PMB) serves as a corpus for semantic processing with a focus on semantic parsing and text generation. Currently, we witness an excellent performance of neural parsers and generators on the PMB. This might suggest that such semantic processing tasks have by and large been solved. We argue that this is not the case and that performance scores from the past on the PMB are inflated by non-optimal data splits and test sets that are too easy. In response, we introduce several changes. First, instead of the prior random split, we propose a more systematic splitting approach to improve the reliability of the standard test data. Second, except for the standard test set, we also propose two challenge sets: one with longer texts including discourse structure, and one that addresses compositional generalization. We evaluate five neural models for semantic parsing and meaning-to-text generation. Our results show that model performance declines (in some cases dramatically) on the challenge sets, revealing the limitations of neural models when confronting such challenges.
As large language models (LLMs) have gained popularity for a variety of use cases, making them adaptable and controllable has become increasingly important, especially for user-facing applications. While the existing literature on LLM adaptation primarily focuses on finding a model (or models) that optimizes a single predefined objective, here we focus on the challenging case where the model must dynamically adapt to diverse -- and often changing -- user preferences. For this, we leverage adaptation methods based on linear weight interpolation, casting them as continuous multi-domain interpolators that produce models with specific prescribed generation characteristics on-the-fly. Specifically, we use low-rank updates to fine-tune a base model to various different domains, yielding a set of anchor models with distinct generation profiles. Then, we use the weight updates of these anchor models to parametrize the entire (infinite) class of models contained within their convex hull. We empirically show that varying the interpolation weights yields predictable and consistent change in the model outputs with respect to all of the controlled attributes. We find that there is little entanglement between most attributes and identify and discuss the pairs of attributes for which this is not the case. Our results suggest that linearly interpolating between the weights of fine-tuned models facilitates predictable, fine-grained control of model outputs with respect to multiple stylistic characteristics simultaneously.
Pretrained Language Models (PLMs) benefit from external knowledge stored in graph structures for various downstream tasks. However, bridging the modality gap between graph structures and text remains a significant challenge. Traditional methods like linearizing graphs for PLMs lose vital graph connectivity, whereas Graph Neural Networks (GNNs) require cumbersome processes for integration into PLMs. In this work, we propose a novel graph-guided self-attention mechanism, GraSAME. GraSAME seamlessly incorporates token-level structural information into PLMs without necessitating additional alignment or concatenation efforts. As an end-to-end, lightweight multimodal module, GraSAME follows a multi-task learning strategy and effectively bridges the gap between graph and textual modalities, facilitating dynamic interactions between GNNs and PLMs. Our experiments on the graph-to-text generation task demonstrate that GraSAME outperforms baseline models and achieves results comparable to state-of-the-art (SOTA) models on WebNLG datasets. Furthermore, compared to SOTA models, GraSAME eliminates the need for extra pre-training tasks to adjust graph inputs and reduces the number of trainable parameters by over 100 million.
The Directed Acyclic Transformer is a fast non-autoregressive (NAR) model that performs well in Neural Machine Translation. Two issues prevent its application to general Natural Language Generation (NLG) tasks: frequent Out-Of-Vocabulary (OOV) errors and the inability to faithfully generate entity names. We introduce Control-DAG, a constrained decoding algorithm for our Directed Acyclic T5 (DA-T5) model which offers lexical, vocabulary and length control. We show that Control-DAG significantly enhances DA-T5 on the Schema Guided Dialogue and the DART datasets, establishing strong NAR results for Task-Oriented Dialogue and Data-to-Text NLG.
Although current large language models are complex, the most basic specifications of the underlying language generation problem itself are simple to state: given a finite set of training samples from an unknown language, produce valid new strings from the language that don't already appear in the training data. Here we ask what we can conclude about language generation using only this specification, without further assumptions. In particular, suppose that an adversary enumerates the strings of an unknown target language L that is known only to come from one of a possibly infinite list of candidates. A computational agent is trying to learn to generate from this language; we say that the agent generates from L in the limit if after some finite point in the enumeration of L, the agent is able to produce new elements that come exclusively from L and that have not yet been presented by the adversary. Our main result is that there is an agent that is able to generate in the limit for every countable list of candidate languages. This contrasts dramatically with negative results due to Gold and Angluin in a well-studied model of language learning where the goal is to identify an unknown language from samples; the difference between these results suggests that identifying a language is a fundamentally different problem than generating from it.
Assessing the factual consistency of automatically generated texts in relation to source context is crucial for developing reliable natural language generation applications. Recent literature proposes AlignScore which uses a unified alignment model to evaluate factual consistency and substantially outperforms previous methods across many benchmark tasks. In this paper, we take a closer look of datasets used in AlignScore and uncover an unexpected finding: utilizing a smaller number of data points can actually improve performance. We process the original AlignScore training dataset to remove noise, augment with robustness-enhanced samples, and utilize a subset comprising 10\% of the data to train an improved factual consistency evaluation model, we call LIM-RA (Less Is More for Robust AlignScore). LIM-RA demonstrates superior performance, consistently outperforming AlignScore and other strong baselines like ChatGPT across four benchmarks (two utilizing traditional natural language generation datasets and two focused on large language model outputs). Our experiments show that LIM-RA achieves the highest score on 24 of the 33 test datasets, while staying competitive on the rest, establishing the new state-of-the-art benchmarks.
Large language models (LLMs) have become increasingly integrated with various applications. To ensure that LLMs do not generate unsafe responses, they are aligned with safeguards that specify what content is restricted. However, such alignment can be bypassed to produce prohibited content using a technique commonly referred to as jailbreak. Different systems have been proposed to perform the jailbreak automatically. These systems rely on evaluation methods to determine whether a jailbreak attempt is successful. However, our analysis reveals that current jailbreak evaluation methods have two limitations. (1) Their objectives lack clarity and do not align with the goal of identifying unsafe responses. (2) They oversimplify the jailbreak result as a binary outcome, successful or not. In this paper, we propose three metrics, safeguard violation, informativeness, and relative truthfulness, to evaluate language model jailbreak. Additionally, we demonstrate how these metrics correlate with the goal of different malicious actors. To compute these metrics, we introduce a multifaceted approach that extends the natural language generation evaluation method after preprocessing the response. We evaluate our metrics on a benchmark dataset produced from three malicious intent datasets and three jailbreak systems. The benchmark dataset is labeled by three annotators. We compare our multifaceted approach with three existing jailbreak evaluation methods. Experiments demonstrate that our multifaceted evaluation outperforms existing methods, with F1 scores improving on average by 17% compared to existing baselines. Our findings motivate the need to move away from the binary view of the jailbreak problem and incorporate a more comprehensive evaluation to ensure the safety of the language model.
In this work, we explicitly show that modern LLMs tend to generate correct facts first, then "drift away" and generate incorrect facts later: this was occasionally observed but never properly measured. We develop a semantic drift score that measures the degree of separation between correct and incorrect facts in generated texts and confirm our hypothesis when generating Wikipedia-style biographies. This correct-then-incorrect generation pattern suggests that factual accuracy can be improved by knowing when to stop generation. Therefore, we explore the trade-off between information quantity and factual accuracy for several early stopping methods and manage to improve factuality by a large margin. We further show that reranking with semantic similarity can further improve these results, both compared to the baseline and when combined with early stopping. Finally, we try calling external API to bring the model back to the right generation path, but do not get positive results. Overall, our methods generalize and can be applied to any long-form text generation to produce more reliable information, by balancing trade-offs between factual accuracy, information quantity and computational cost.
Transformer-based Large Language Models (LLMs) have shown exceptional language generation capabilities in response to text-based prompts. However, controlling the direction of generation via textual prompts has been challenging, especially with smaller models. In this work, we explore the use of Prompt Tuning to achieve controlled language generation. Generated text is steered using prompt embeddings, which are trained using a small language model, used as a discriminator. Moreover, we demonstrate that these prompt embeddings can be trained with a very small dataset, with as low as a few hundred training examples. Our method thus offers a data and parameter efficient solution towards controlling language model outputs. We carry out extensive evaluation on four datasets: SST-5 and Yelp (sentiment analysis), GYAFC (formality) and JIGSAW (toxic language). Finally, we demonstrate the efficacy of our method towards mitigating harmful, toxic, and biased text generated by language models.
The rapid advancement of Large Language Models (LLMs) has inaugurated a transformative epoch in natural language processing, fostering unprecedented proficiency in text generation, comprehension, and contextual scrutiny. Nevertheless, effectively handling extensive contexts, crucial for myriad applications, poses a formidable obstacle owing to the intrinsic constraints of the models' context window sizes and the computational burdens entailed by their operations. This investigation presents an innovative framework that strategically tailors LLMs for streamlined context processing by harnessing the synergies among natural language summarization, soft prompt compression, and augmented utility preservation mechanisms. Our methodology, dubbed SoftPromptComp, amalgamates natural language prompts extracted from summarization methodologies with dynamically generated soft prompts to forge a concise yet semantically robust depiction of protracted contexts. This depiction undergoes further refinement via a weighting mechanism optimizing information retention and utility for subsequent tasks. We substantiate that our framework markedly diminishes computational overhead and enhances LLMs' efficacy across various benchmarks, while upholding or even augmenting the caliber of the produced content. By amalgamating soft prompt compression with sophisticated summarization, SoftPromptComp confronts the dual challenges of managing lengthy contexts and ensuring model scalability. Our findings point towards a propitious trajectory for augmenting LLMs' applicability and efficiency, rendering them more versatile and pragmatic for real-world applications. This research enriches the ongoing discourse on optimizing language models, providing insights into the potency of soft prompts and summarization techniques as pivotal instruments for the forthcoming generation of NLP solutions.
Honeyfiles are security assets designed to attract and detect intruders on compromised systems. Honeyfiles are a type of honeypot that mimic real, sensitive documents, creating the illusion of the presence of valuable data. Interaction with a honeyfile reveals the presence of an intruder, and can provide insights into their goals and intentions. Their practical use, however, is limited by the time, cost and effort associated with manually creating realistic content. The introduction of large language models has made high-quality text generation accessible, but honeyfiles contain a variety of content including charts, tables and images. This content needs to be plausible and realistic, as well as semantically consistent both within honeyfiles and with the real documents they mimic, to successfully deceive an intruder. In this paper, we focus on an important component of the honeyfile content generation problem: document charts. Charts are ubiquitous in corporate documents and are commonly used to communicate quantitative and scientific data. Existing image generation models, such as DALL-E, are rather prone to generating charts with incomprehensible text and unconvincing data. We take a multi-modal approach to this problem by combining two purpose-built generative models: a multitask Transformer and a specialized multi-head autoencoder. The Transformer generates realistic captions and plot text, while the autoencoder generates the underlying tabular data for the plot. To advance the field of automated honeyplot generation, we also release a new document-chart dataset and propose a novel metric Keyword Semantic Matching (KSM). This metric measures the semantic consistency between keywords of a corpus and a smaller bag of words. Extensive experiments demonstrate excellent performance against multiple large language models, including ChatGPT and GPT4.
Finetuning pretrained models on downstream generation tasks often leads to catastrophic forgetting in zero-shot conditions. In this work, we focus on summarization and tackle the problem through the lens of language-independent representations. After training on monolingual summarization, we perform zero-shot transfer to new languages or language pairs. We first show naively finetuned models are highly language-specific in both output behavior and internal representations, resulting in poor zero-shot performance. Next, we propose query-key (QK) finetuning to decouple task-specific knowledge from the pretrained language generation abilities. Then, after showing downsides of the standard adversarial language classifier, we propose a balanced variant that more directly enforces language-agnostic representations. Moreover, our qualitative analyses show removing source language identity correlates to zero-shot summarization performance. Our code is openly available.
Compositional generalization, representing the model's ability to generate text with new attribute combinations obtained by recombining single attributes from the training data, is a crucial property for multi-aspect controllable text generation (MCTG) methods. Nonetheless, a comprehensive compositional generalization evaluation benchmark of MCTG is still lacking. We propose CompMCTG, a benchmark encompassing diverse multi-aspect labeled datasets and a crafted three-dimensional evaluation protocol, to holistically evaluate the compositional generalization of MCTG approaches. We observe that existing MCTG works generally confront a noticeable performance drop in compositional testing. To mitigate this issue, we introduce Meta-MCTG, a training framework incorporating meta-learning, where we enable models to learn how to generalize by simulating compositional generalization scenarios in the training phase. We demonstrate the effectiveness of Meta-MCTG through achieving obvious improvement (by at most 3.64%) for compositional testing performance in 94.4% cases.