Thailand

To mitigate the potential misuse of large language models (LLMs), recent research has developed watermarking algorithms, which restrict the generation process to leave an invisible trace for watermark detection. Due to the two-stage nature of the task, most studies evaluate the generation and detection separately, thereby presenting a challenge in unbiased, thorough, and applicable evaluations. In this paper, we introduce WaterBench, the first comprehensive benchmark for LLM watermarks, in which we design three crucial factors: (1) For benchmarking procedure, to ensure an apples-to-apples comparison, we first adjust each watermarking method&#39;s hyper-parameter to reach the same watermarking strength, then jointly evaluate their generation and detection performance. (2) For task selection, we diversify the input and output length to form a five-category taxonomy, covering $9$ tasks. (3) For evaluation metric, we adopt the GPT4-Judge for automatically evaluating the decline of instruction-following abilities after watermarking. We evaluate $4$ open-source watermarks on $2$ LLMs under $2$ watermarking strengths and observe the common struggles for current methods on maintaining the generation quality. The code and data are available at https://github.com/THU-KEG/WaterBench.

ACL 2024

WaterBench: Towards Holistic Evaluation of Watermarks for Large Language Models

evaluation methodologies

benchmarking

evaluation

To mitigate the potential misuse of large language models (LLMs), recent research has developed watermarking algorithms, which restrict the generation process to leave an invisible trace for watermark detection. Due to the two-stage nature of the task, most studies evaluate the generation and detection separately, thereby presenting a challenge in unbiased, thorough, and applicable evaluations. In this paper, we introduce WaterBench, the first comprehensive benchmark for LLM watermarks, in which we design three crucial factors: (1) For benchmarking procedure, to ensure an apples-to-apples comparison, we first adjust each watermarking method's hyper-parameter to reach the same watermarking strength, then jointly evaluate their generation and detection performance. (2) For task selection, we diversify the input and output length to form a five-category taxonomy, covering $9$ tasks. (3) For evaluation metric, we adopt the GPT4-Judge for automatically evaluating the decline of instruction-following abilities after watermarking. We evaluate $4$ open-source watermarks on $2$ LLMs under $2$ watermarking strengths and observe the common struggles for current methods on maintaining the generation quality. The code and data are available at https://github.com/THU-KEG/WaterBench.

poster

### Welcome!
The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. Our Virtual Poster Sessions will take place online Thursday, August 22, 2024.

You are required to register for this event. **Please register [here](https://2024.aclweb.org/registration). **

If you have already registered, please check your inbox for an email from Underline granting you access to ACL 2024 content.

Please register!

The 62nd Annual Meeting of the Association for Computational Linguistics (ACL 2024) will take place in Bangkok, Thailand from August 11th to 16th, 2024. More information will be announced soon.

Role-playing agents (RPAs), powered by large language models, have emerged as a flourishing field of applications. However, a key challenge lies in assessing whether RPAs accurately reproduce the personas of target characters, namely their character fidelity. Existing methods mainly focus on the knowledge and linguistic patterns of characters. This paper, instead, introduces a novel perspective to evaluate the personality fidelity of RPAs with psychological scales. Overcoming drawbacks of previous self-report assessments on RPAs, we propose InCharacter, namely **In**terviewing **Character** agents for personality tests. Experiments include various types of RPAs and LLMs, covering 32 distinct characters on 14 widely used psychological scales. The results validate the effectiveness of InCharacter in measuring RPA personalities. Then, with InCharacter, we show that state-of-the-art RPAs exhibit personalities highly aligned with the human-perceived personalities of the characters, achieving an accuracy up to 80.7%.

InCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviews

Advances towards more faithful and traceable answers of Large Language Models (LLMs) are crucial for various research and practical endeavors. One avenue in reaching this goal is basing the answers on reliable sources. However, this Evidence-Based QA has proven to work insufficiently with LLMs in terms of citing the correct sources (source quality) and truthfully representing the information within sources (answer attributability). In this work, we systematically investigate how to robustly fine-tune LLMs for better source quality and answer attributability. Specifically, we introduce a data generation pipeline with automated data quality filters, which can synthesize diversified high-quality training and testing data at scale. We further introduce four test sets to benchmark the robustness of fine-tuned specialist models. Extensive evaluation shows that fine-tuning on synthetic data improves performance on both in- and out-of-distribution. Furthermore, we show that data quality, which can be drastically improved by proposed quality filters, matters more than quantity in improving Evidence-Based QA.

Towards Faithful and Robust LLM Specialists for Evidence-Based Question-Answering

Ten years ago a single metric, BLEU, governed progress in machine translation research. For better or worse, there is no such consensus today, and consequently it is difficult for researchers to develop and retain intuitions about metric deltas that drove earlier research and deployment decisions. This paper investigates the “dynamic range” of a number of modern metrics in an effort to provide a collective understanding of the meaning of differences in scores both within and among metrics; in other words, we ask "what point difference x in metric y is required between two systems for humans to notice?". We conduct our evaluation on a new large dataset, ToShip23, using it to discover deltas at which metrics achieve system-level differences that are meaningful to humans, which we measure by pairwise system accuracy. We additionally show that this method of establishing delta-accuracy is more stable than the standard use of statistical p-values in regards to testset size. Where data size permits, we also explore the effect of metric deltas and accuracy across finer-grained features such as translation direction, domain, and system closeness.

Navigating the Metrics Maze: Reconciling Score Magnitudes and Accuracies

Large Language Models (LLMs) are transforming diverse fields and gaining increasing influence as human proxies. This development underscores the urgent need for evaluating value orientations and understanding of LLMs to ensure their responsible integration into public-facing applications. This work introduces ValueBench, the first comprehensive psychometric benchmark for evaluating value orientations and understanding in LLMs. ValueBench collects data from 44 established psychometric inventories, encompassing 453 multifaceted value dimensions. We propose an evaluation pipeline grounded in realistic human-AI interactions to probe value orientations, along with novel tasks for evaluating value understanding in an open-ended value space. With extensive experiments conducted on six representative LLMs, we unveil their shared and distinctive value orientations and exhibit their ability to approximate expert conclusions in value-related extraction and generation tasks.

ValueBench: Towards Comprehensively Evaluating Value Orientations and Understanding of Large Language Models

The paper introduces SceMQA, a novel benchmark for scientific multimodal question answering at the college entrance level. It addresses a critical educational phase often overlooked in existing benchmarks, spanning high school to pre-college levels. SceMQA focuses on core science subjects including Mathematics, Physics, Chemistry, and Biology. It features a blend of multiple-choice and free-response formats, ensuring a comprehensive evaluation of AI models' abilities. Additionally, our benchmark provides specific knowledge points for each problem and detailed explanations for each answer. SceMQA also uniquely presents problems with identical contexts but varied questions to facilitate a more thorough and accurate assessment of reasoning capabilities. In the experiment, we evaluate both open-source and close-source state-of-the-art Multimodal Large Language Models (MLLMs), across various experimental settings. The results show that further research and development are needed in developing more capable MLLM, as highlighted by only 50% to 60% accuracy achieved by the strongest models.

SceMQA: A Scientific College Entrance Level Multimodal Question Answering Benchmark

Extending large language models (LLMs) to process longer inputs is crucial for numerous applications. However, the considerable computational cost of transformers, coupled with limited generalization of positional encoding, restricts the size of their context window. We introduce Cross-Attention to Parallel Encodings (CAPE), a framework that can be applied to any existing decoder-only LLMs for context expansion. CAPE leverages a small encoder to process a long input chunk by chunk and enables the frozen decoder to cross-attend to the additional contexts. CAPE is efficient, generalizable, and versatile: trained with 8K-token documents, CAPE extends the context window of LLaMA-2 to 128K tokens, offering $10\times$ of the throughput with only 1/6 of the memory. CAPE yields strong performance on language modeling and in-context learning. CAPE also excels in retrieval-augmented applications, while existing long-context models degenerate with retrieved contexts. We further introduce a CAPE variant that can extend the context window of instruction-tuned models with only unlabeled data, and showcase its effectiveness on LLaMA-2-Chat, leading to a strong instruction-following model that can leverage very long context on downstream tasks.

Long-Context Language Modeling with Parallel Context Encoding

Machine learning models have made incredible progress, but they still struggle when applied to examples from unseen domains. This study focuses on a specific problem of domain generalization, where a model is trained on one source domain and tested on multiple target domains that are unseen during training. We propose IMO: Invariant features Masks for Out-of-Distribution text classification, to achieve OOD generalization by learning invariant features. During training, IMO would learn sparse mask layers to remove irrelevant features for prediction, where the remaining features keep invariant. Additionally, IMO has an attention module at the token level to focus on tokens that are useful for prediction. Our comprehensive experiments show that IMO substantially outperforms strong baselines in terms of various evaluation metrics and settings.

IMO: Greedy Layer-Wise Sparse Representation Learning for Out-of-Distribution Text Classification with Pre-trained Models

In this work, we present the largest benchmark to date on linguistic acceptability: Multilingual Evaluation of Linguistic Acceptability---MELA, with 46K samples covering 10 languages from a diverse set of language families. We establish LLM baselines on this benchmark, and investigate cross-lingual transfer in acceptability judgements with XLM-R. In pursuit of multilingual interpretability, we conduct probing experiments with fine-tuned XLM-R to explore the process of syntax capability acquisition. Our results show that GPT-4o exhibits a strong multilingual ability, outperforming fine-tuned XLM-R, while open-source multilingual models lag behind by a noticeable gap. Cross-lingual transfer experiments show that transfer in acceptability judgment is non-trivial: 500 Icelandic fine-tuning examples lead to 23 MCC performance in a completely unrelated language---Chinese. Results of our probing experiments indicate that training on MELA improves the performance of XLM-R on syntax-related tasks.

MELA: Multilingual Evaluation of Linguistic Acceptability

The current charge prediction datasets mostly focus on single-defendant criminal cases.
However, real-world criminal cases usually involve multiple defendants whose criminal facts are intertwined. 
In an early attempt to fill this gap, we introduce a new benchmark that encompasses legal cases involving multiple defendants, where each defendant is labeled with a charge and four types of crime elements, \emph{i.e.,} \textit{Object Element}, \textit{Objective Element}, \textit{Subject Element}, and \textit{Subjective Element}. 
Based on the dataset, we further develop an interpretable model called EJudge that incorporates crime elements and legal rules to infer charges. We observe that predicting crime charges while providing corresponding rationales benefits the interpretable AI system. Extensive experiments show that EJudge significantly surpasses state-of-the-art methods, which verify the importance of crime elements and legal rules in multi-defendant charge prediction. The source code and dataset are available at \url{https://anonymous.4open.science/r/MCP_1-6010}.

Through the MUD: A Multi-Defendant Charge Prediction Benchmark with Linked Crime Elements

The widespread use of large language models (LLMs) is increasing the demand for methods that detect machine-generated text to prevent misuse. The goal of our study is to stress test the detectors’ robustness to malicious attacks under realistic scenarios. We comprehensively study the robustness of popular machine-generated text detectors under attacks from diverse categories: editing, paraphrasing, co-generating, and prompting. Our attacks assume limited access to the generator LLMs, and we compare the performance of detectors on different attacks under different budget levels. Our experiments reveal that almost none of the existing detectors remain robust under all the attacks, and all detectors exhibit different loopholes. Averaging all detectors, the performance drops by 35% across all attacks. Further, we investigate the reasons behind these defects and propose initial out-of-the-box patches.

Premium content

WaterBench: Towards Holistic Evaluation of Watermarks for Large Language Models

Downloads

Next from ACL 2024

InCharacter: Evaluating Personality Fidelity in Role-Playing Agents through Psychological Interviews

Stay up to date with the latest Underline news!

PRESENTATIONS

CONFERENCES

COMPANY

RESOURCES