Decision-making, a complicated task requiring various types of abilities, presents an excellent framework for assessing Large Language Models (LLMs). Our research investigates LLMs' decision-making capabilities through the lens of a well-established field, Game Theory. We focus specifically on games that support the participation of more than two agents simultaneously. Subsequently, we introduce our framework, GAMA-Bench, including eight classical multi-agent games. We design a scoring scheme to assess a model's performance in these games quantitatively. Through GAMA-Bench, we investigate LLMs' robustness, generalizability, and enhancement strategies. Results reveal that while GPT-3.5 shows satisfying robustness, its generalizability is relatively limited. However, its performance can be improved through approaches such as Chain-of-Thought. Additionally, we conduct evaluations across various LLMs and find that GPT-4 outperforms other models on GAMA-Bench, achieving a score of 72.5. Moreover, the increasingly higher scores across the three iterations of GPT-3.5 (0613, 1106, 0125) demonstrate marked advancements in the model's intelligence with each update. The code and experimental results are made publicly available via https://github.com/CUHK-ARISE/GAMABench.
Text watermarking technology aims to tag and identify content produced by large language models (LLMs) to prevent misuse. In this study, we introduce the concept of ''cross-lingual consistency'' in text watermarking, which assesses the ability of text watermarks to maintain their effectiveness after being translated into other languages. Preliminary empirical results from two LLMs and three watermarking methods reveal that current text watermarking technologies lack consistency when texts are translated into various languages. Based on this observation, we propose a Cross-lingual Watermark Removal Attack (CWRA) to bypass watermarking by first obtaining a response from an LLM in a pivot language, which is then translated into the target language. CWRA can effectively remove watermarks by reducing the Area Under the Curve (AUC) from 0.95 to 0.67 without performance loss. Furthermore, we analyze two key factors that contribute to the cross-lingual consistency in text watermarking and propose a defense method that increases the AUC from 0.67 to 0.88 under CWRA.
Embedding entities and relations into continuous vector spaces has attracted a surge of interest in recent years. Most embedding methods assume that all test entities are available during training, which makes it time-consuming to retrain embeddings for newly emerging entities. To address this issue, recent works apply the graph neural network on the existing neighbors of the unseen entities. In this paper, we propose a novel framework, namely Virtual Neighbor (VN) network, to address three key challenges. Firstly, to reduce the neighbor sparsity problem, we introduce the concept of the virtual neighbors inferred by rules. And we assign soft labels to these neighbors by solving a rule-constrained problem, rather than simply regarding them as unquestionably true. Secondly, many existing methods only use one-hop or two-hop neighbors for aggregation and ignore the distant information that may be helpful. Instead, we identify both logic and symmetric path rules to capture complex patterns. Finally, instead of one-time injection of rules, we employ an iterative learning scheme between the embedding method and virtual neighbor prediction to capture the interactions within. Experimental results on two knowledge graph completion tasks demonstrate that our VN network significantly outperforms state-of-the-art baselines. Furthermore, results on Subject/Object-R show that our proposed VN network is highly robust to the neighbor sparsity problem.
Motivated by the success of unsupervised neural machine translation (UNMT), we introduce an unsupervised sign language translation and generation network (USLNet), which learns from abundant single-modality (text and video) data without parallel sign language data. USLNet comprises two main components: single-modality reconstruction modules (text and video) that rebuild the input from its noisy version in the same modality and cross-modality back-translation modules (text-video-text and video-text-video) that reconstruct the input from its noisy version in the different modality using back-translation procedure.Unlike the single-modality back-translation procedure in text-based UNMT, USLNet faces the cross-modality discrepancy in feature representation, in which the length and the feature dimension mismatch between text and video sequences. We propose a sliding window method to address the issues of aligning variable-length text with video sequences. To our knowledge, USLNet is the first unsupervised sign language translation and generation model capable of generating both natural language text and sign language video in a unified manner. Experimental results on the BBC-Oxford Sign Language dataset (BOBSL) and Open-Domain American Sign Language dataset (OpenASL) reveal that USLNet achieves competitive results compared to supervised baseline models, indicating its effectiveness in sign language translation and generation.
In this paper, we re-examine the Markov property in the context of neural machine translation. We design a Markov Autoregressive Transformer~(MAT) and undertake a comprehensive assessment of its performance across four WMT benchmarks. Our findings indicate that MAT with an order larger than 4 can generate translations with quality on par with that of conventional autoregressive transformers. In addition, counter-intuitively, we also find that the advantages of utilizing a higher-order MAT do not specifically contribute to the translation of longer sentences.
Speculative decoding is a relatively new decoding framework that leverages small and efficient draft models to reduce the latency of LLMs. In this study, we introduce GliDe and CaPE, two low-hassle modifications to vanilla speculative decoding to further improve the decoding speed of a frozen LLM. Specifically, GliDe is a modified draft model architecture that reuses the cached keys and values from the target LLM, while CaPE is a proposal expansion method that uses the draft model's confidence scores to help select additional candidate tokens for verification. Extensive experiments on different benchmarks demonstrate that our proposed GliDe draft model significantly reduces the expected decoding latency. Additional evaluation using walltime reveals that GliDe can accelerate Vicuna models up to 2.17x and further extend the improvement to 2.61x with CaPE. We will release our code, data, and the trained draft models.
Insufficient modeling of human preferences within the reward model is a major obstacle for leveraging human feedback to improve translation quality. Fortunately, quality estimation (QE), which predicts the quality of a given translation without reference, has achieved impressive alignment with human evaluations in the last two years. In this work, we investigate the potential of employing the QE model as the reward model (the QE-based reward model) to predict human preferences for feedback training. We first identify the overoptimization problem during QE-based feedback training, manifested as an increase in reward while translation quality declines. We examine the problem and argue that the vulnerability of the QE model might lead to high rewards for incorrect translations, resulting in overoptimization and error propagation. To address the problem, we adopt a simple yet effective method that uses heuristic rules to detect the incorrect translations and assigns a penalty term to the QE-based rewards for the detected incorrect translations. Experimental results show that the proposed QE-based feedback training achieves consistent and significant improvements across various settings, further verified through human preference studies. Our subsequent analysis demonstrates the high data efficiency of the proposed QE-based feedback training: the proposed approach using a small amount of monolingual data can outperform systems using larger parallel corpora.
The proliferation of open-source Large Language Models (LLMs) from various institutions has highlighted the urgent need for comprehensive evaluation methods. However, current evaluation platforms, such as the widely recognized HuggingFace open LLM leaderboard, neglect a crucial aspect -- uncertainty, which is vital for thoroughly assessing LLMs. To bridge this gap, we introduce a new benchmarking approach for LLMs that integrates uncertainty quantification. Our examination involves eight LLMs (LLM series) spanning five representative natural language processing tasks. Additionally, we introduce an uncertainty-aware evaluation metric, UAcc, which takes into account both prediction accuracy and prediction uncertainty. Our findings reveal that: I) LLMs with higher accuracy may exhibit lower certainty; II) Larger-scale LLMs may display greater uncertainty compared to their smaller counterparts; and III) Instruction-finetuning tends to increase the uncertainty of LLMs. By taking uncertainty into account, our new UAcc metric can either amplify or diminish the relative improvement of one LLM over another and may even change the relative ranking of two LLMs. These results underscore the significance of incorporating uncertainty in the evaluation of LLMs.
The evolution of Neural Machine Translation (NMT) has been significantly influenced by six core challenges (Koehn and Knowles, 2017), which have acted as benchmarks for progress in this field. This study revisits these challenges, offering insights into their ongoing relevance in the context of advanced Large Language Models (LLMs): domain mismatch, amount of parallel data, rare word prediction, translation of long sentences, attention model as word alignment, and sub-optimal beam search. Our empirical findings indicate that LLMs effectively lessen the reliance on parallel data for major languages in the pretraining phase. Additionally, the LLM-based translation system significantly enhances the translation of long sentences that contain approximately 80 words and shows the capability to translate documents of up to 512 words. However, despite these significant improvements, the challenges of domain mismatch and prediction of rare words persist. While the challenges of word alignment and beam search, specifically associated with NMT, may not apply to LLMs, we identify three new challenges for LLMs in translation tasks: inference efficiency, translation of low-resource languages in the pretraining phase, and human-aligned evaluation. The datasets and models are released at https://github.com/pangjh3/LLM4MT.
Large Language Models (LLMs) like ChatGPT are foundational in various applications due to their extensive knowledge from pre-training and fine-tuning. Despite this, they are prone to generating factual and commonsense errors, raising concerns in critical areas like healthcare, journalism, and education to mislead users. Current methods for evaluating LLMs' veracity are limited by test data leakage or the need for extensive human labor, hindering efficient and accurate error detection. To tackle this problem, we introduce a novel, automatic testing framework, FactChecker, aimed at uncovering factual inaccuracies in LLMs. This framework involves three main steps: First, it constructs a factual knowledge graph by retrieving fact triplets from a large-scale knowledge database. Then, leveraging the knowledge graph, FactChecker employs a rule-based approach to generates three types of questions (Yes-No, Multiple-Choice, and WH questions) that involve single-hop and multi-hop relations, along with correct answers. Lastly, it assesses the LLMs' responses for accuracy using tailored matching strategies for each question type. Our extensive tests on six prominent LLMs, including text-davinci-002, text-davinci-003, ChatGPT~(gpt-3.5-turbo, gpt-4), Vicuna, and LLaMA-2, reveal that FactChecker can trigger factual errors in up to 45\% of questions in these models. Moreover, we demonstrate that FactChecker's test cases can improve LLMs' factual accuracy through in-context learning and fine-tuning (e.g., llama-2-13b-chat's accuracy increase from 35.3\% to 68.5\%). We are making all code, data, and results available for future research endeavors.