Image captioning has been a longstanding challenge in vision-language research. With the rise of LLMs, modern Vision-Language Models (VLMs) generate detailed and comprehensive image descriptions. However, benchmarking the quality of such captions remains unresolved. This paper addresses two key questions: (1) How well do VLMs actually perform on image captioning, particularly compared to humans? We built CapArena, a platform with over 6000 pairwise caption battles and high-quality human preference votes. Our Arena-style evaluation marks a milestone, showing that leading models like GPT-4o achieve or even surpass human performance, while most open-source models lag behind. (2) Can automated metrics reliably assess caption quality? Using human annotations from CapArena, we evaluate traditional and recent captioning metrics, as well as VLM-as-a-Judge. Our analysis reveals that while some metrics (e.g., METEOR) show high caption-level agreement with humans, their systematic biases lead to inconsistencies in model ranking. In contrast, VLM-as-a-Judge demonstrates robust discernment at both the caption and model levels. Building on these insights, we release CapArena-Auto, an accurate and efficient automated benchmark for detailed captioning, achieving 93.4% correlation with human rankings at just $4 per test. All data and evaluation resources have been open-sourced.

Existing retrieval-based methods have made significant strides in maintaining long-term conversations. However, these approaches face challenges in memory database management and accurate memory retrieval, hindering their efficacy in dynamic, real-world interactions. This study introduces a novel framework, COmpressive Memory-Enhanced Dialogue sYstems (COMEDY), which eschews traditional retrieval modules and memory databases. Instead, COMEDY adopts a “One-for-All” approach, utilizing a single language model to manage memory generation, compression, and response generation. Central to this framework is the concept of compressive memory, which integrates session-specific summaries, user-bot dynamics, and past events into a concise memory format. To support COMEDY, we collect the biggest Chinese long-term conversation dataset, Dolphin, derived from real user-chatbot interactions. Comparative evaluations demonstrate COMEDY’s superiority over traditional retrieval-based methods in producing more nuanced and human-like conversational experiences.

Large language models (LLMs) have shown increasing power on various natural language processing (NLP) tasks. However, tuning these models for downstream tasks usually needs exorbitant costs or is unavailable due to commercial considerations. Recently, black-box tuning has been proposed to address this problem by optimizing task-specific prompts without accessing the gradients and hidden representations. However, most existing works have yet fully exploited the potential of gradient-free optimization under the scenario of few-shot learning. In this paper, we describe BBT-RGB, a suite of straightforward and complementary techniques for enhancing the efficiency and performance of black-box optimization. Specifically, our method includes three plug-and-play components: (1) Two-stage derivative-free optimization strategy that facilitates fast convergence and mitigates overfitting; (2) Automatic verbalizer construction with its novel usage under few-shot settings; (3) Better prompt initialization policy based on instruction search and auto-selected demonstration. Extensive experiments across various tasks on natural language understanding and inference demonstrate the effectiveness of our method. Our codes are available at https://github.com/QiushiSun/BBT-RGB.

Code pre-trained models (CodePTMs) have recently demonstrated a solid capacity to process various code intelligence tasks, e.g., code clone detection, code translation, and code summarization. The current mainstream method that deploys these models to downstream tasks is to fine-tune them on individual tasks, which is generally costly and needs sufficient data for large models. To tackle the issue, in this paper, we present TransCoder, a unified Transferable fine-tuning strategy for Code representation learning. Inspired by human inherent skills of knowledge generalization, TransCoder drives the model to learn better code-related knowledge like human programmers. Specifically, we employ a tunable prefix encoder to first capture cross-task and cross-language transferable knowledge, subsequently applying the acquired knowledge for optimized downstream adaptation. Besides, our approach confers benefits for tasks with minor training sample sizes and languages with smaller corpora, underscoring versatility and efficacy. Extensive experiments conducted on representative datasets clearly demonstrate that our method can lead to superior performance on various code-related tasks and encourage mutual reinforcement, especially in low-resource scenarios. Our codes are available at https://github.com/QiushiSun/TransCoder.

Over the past few years, we have witnessed remarkable advancements in Code Pre-trained Models (CodePTMs). These models achieved excellent representation capabilities by designing structure-based pre-training tasks for code. However, how to enhance the absorption of structural knowledge when fine-tuning CodePTMs still remains a significant challenge. To fill this gap, in this paper, we present SAT, a novel structure-enhanced and plug-and-play fine-tuning method for CodePTMs. We first propose a structure loss to quantify the difference between the information learned by CodePTMs and the knowledge extracted from code structure. Specifically, we use the attention scores from Transformer layer as the learned information, and the shortest path length between leaves in abstract syntax trees as the structural knowledge. Subsequently, multi-task learning is introduced to improve the performance of fine-tuning. Experiments conducted on four pre-trained models and two generation tasks demonstrate the effectiveness of our proposed method as a plug-and-play solution. Furthermore, we observed that SAT can benefit CodePTMs more with limited training data.

Recent Retrieval Augmented Generation (RAG) aims to enhance Large Language Models (LLMs) by incorporating extensive knowledge retrieved from external sources. However, such approach encounters some challenges: Firstly, the original queries may not be suitable for precise retrieval, resulting in erroneous contextual knowledge; Secondly, the language model can easily generate inconsistent answer with external references due to their knowledge boundary limitation. To address these issues, we propose the chain-of-verification (CoV-RAG) to enhance the external retrieval correctness and internal generation consistency. Specifically, we integrate the verification module into the RAG, engaging in scoring, judgment, and rewriting. To correct external retrieval errors, CoV-RAG retrieves new knowledge using a revised query. To correct internal generation errors, we unify QA and verification tasks with a Chain-of-Thought (CoT) reasoning during training. Our comprehensive experiments across various LLMs demonstrate the effectiveness and adaptability compared with other strong baselines. Especially, our CoV-RAG can significantly surpass the state-of-the-art baselines using different LLM backbones.

Software development is a complex task that necessitates cooperation among multiple members with diverse skills. Numerous studies used deep learning to improve specific phases in a waterfall model, such as design, coding, and testing. However, the deep learning model in each phase requires unique designs, leading to technical inconsistencies across various phases, which results in a fragmented and ineffective development process. In this paper, we introduce ChatDev, a chat-powered software development framework in which specialized agents driven by large language models (LLMs) are guided in what to communicate (via chat chain) and how to communicate (via communicative dehallucination). These agents actively contribute to the design, coding, and testing phases through unified language-based communication, with solutions derived from their multi-turn dialogues. We found their utilization of natural language is advantageous for system design, and communicating in programming language proves helpful in debugging. This paradigm demonstrates how linguistic communication facilitates multi-agent collaboration, establishing language as a unifying bridge for autonomous task-solving among LLM agents. The code and data are available at https://github.com/OpenBMB/ChatDev.

Code pre-trained models (CodePTMs) have significantly advanced the field of neural code intelligence. Despite their capabilities, these models are susceptible to adversarial attacks that subtly modify the model inputs, resulting in incorrect outputs or predictions. Previous methods of robustness evaluation for CodePTMs primarily stem from a textual perspective, without explicitly taking into account the structure of the code. Furthermore, prior studies fail to encompass a broad enough spectrum of tasks and models. In this paper, we propose a set of novel robustness evaluation methods based on the intrinsic structure of the code. Specifically, we first launch adversarial attacks on crucial identifier tokens and sub-tree structures to explore the impact of imperceptible perturbation. Then, we perform global restructuring of the code using different traversal methods for abstract syntax trees, aiming to explore the model’s sensitivity to input samples with equivalent information. Moreover, for each scenario, we employ adversarial training methods to explore the possibility of restoring the performance of perturbed models. For both code understanding and generation, our proposed method has demonstrated its effectiveness across a wide range of models and tasks, thereby allowing us to make one step forward in our understanding of the inner mechanisms of CodePTMs.

Machine reading comprehension (MRC) is an important area of conversation agents and draws a lot of attention. However, there is a notable limitation to current MRC benchmarks: The labeled answers are mostly either spans extracted from the target corpus or the choices of the given candidates, ignoring the natural aspect of high-quality responses. As a result, MRC models trained on these datasets can not generate human-like responses in real QA scenarios. To this end, we construct a new dataset called Penguin to promote the research of MRC, providing a training and test bed for natural response generation to real scenarios. Concretely, Penguin consists of 200k training data with high-quality fluent, and well-informed responses. Penguin is the first benchmark towards natural response generation in Chinese MRC on a relatively large scale. To address the challenges in Penguin, we develop two strong baselines: end-to-end and two-stage frameworks. Following that, we further design Prompt-BART: fine-tuning the pre-trained generative language models with a mixture of prefix prompts in Penguin. Extensive experiments validated the effectiveness of this design.

In recent years, Dialogue-style Large Language Models (LLMs) such as ChatGPT and GPT4 have demonstrated immense potential in constructing open-domain dialogue agents. However, aligning these agents with specific characters or individuals remains a considerable challenge due to the complexities of character representation and the lack of comprehensive annotations. In this paper, we introduce the Harry Potter Dialogue (HPD) dataset, designed to advance the study of dialogue agents and character alignment. The dataset encompasses all dialogue sessions (in both English and Chinese) from the Harry Potter series and is annotated with vital background information, including dialogue scenes, speakers, character relationships, and attributes. These extensive annotations may empower LLMs to unlock character-driven dialogue capabilities. Furthermore, it can serve as a universal benchmark for evaluating how well can a LLM aligning with a specific character. We benchmark LLMs on HPD using both fine-tuning and in-context learning settings. Evaluation results reveal that although there is substantial room for improvement in generating high-quality, character-aligned responses, the proposed dataset is valuable in guiding models toward responses that better align with the character of Harry Potter.

The recent success of large pre-trained language models (PLMs) heavily hinges on massive labeled data, which typically produces inferior performance in low-resource scenarios. To remedy this dilemma, we study self-training as one of the predominant semi-supervised learning (SSL) approaches, which utilizes large-scale unlabeled data to generate synthetic examples. However, too many noisy labels will hurt the model performance, and the self-training procedure requires multiple training iterations making it more expensive if all the model parameters of the PLM are updated. This paper presents UPET, a novel Uncertainty-aware Parameter-Efficient self-Training framework to effectively and efficiently address the labeled data scarcity issue. Specifically, we incorporate Monte Carlo (MC) dropout in Bayesian neural network (BNN) to perform uncertainty estimation for the teacher model and then judiciously select reliable pseudo-labeled examples based on confidence and certainty. During the student training, we introduce multiple parameter-efficient learning (PEL) paradigms that allow optimizes only a small percentage of parameters. We also propose a novel Easy-Hard Contrastive Tuning to enhance the robustness and generalization. Extensive experiments over multiple downstream tasks demonstrate that UPET achieves a substantial improvement in terms of performance and efficiency. Our codes and data are released at https://github.com/wjn1996/UPET.

Code pre-trained models (CodePTMs) have recently become the de-facto paradigm for various tasks in the domain of code intelligence. To achieve excellent performance, the widely used strategy is to fine-tune all the parameters of CodePTMs. However, as the model size increases along with the number of downstream tasks, this strategy becomes excessively expensive. There are also some prior works that utilize Parameter-Efficient Learning (PEL) methods for model tuning in natural language processing to mitigate similar problems, but applying them directly to CodePTMs fails to capture the inherent structural characteristics of codes. To address the problem, in this paper, we propose Pass-Tuning for structure-aware Parameter-Efficient code representation learning. Specifically, a plug-and-play graph neural network module that can learn from Abstract Syntax Tree (AST) is employed as a tunable prefix. On the one hand, Pass-Tuning can further exploit the structural information of source code. On the other hand, it could serve as a replacement for full fine-tuning. We evaluate our method on multiple tasks across eight programming languages, including code understanding and generation. These results demonstrate the effectiveness, robustness, and universality of our method.

The conversational machine reading comprehension (CMRC) task aims to answer questions in conversations, which has been a hot research topic in recent years because of its wide applications. However, existing CMRC benchmarks in which each conversation is assigned a static passage are inconsistent with real scenarios. Thus, model’s comprehension ability towards real scenarios are hard to evaluate reasonably. To this end, we propose the first Chinese CMRC benchmark Orca and further provide zero-shot/few-shot settings to evaluate model’s generalization ability towards diverse domains. We collect 831 hot-topic driven conversations with 4,742 turns in total. Each turn of a conversation is assigned with a response-related passage, aiming to evaluate model’s comprehension ability more reasonably. The topics of conversations are collected from social media platform and cover 33 domains, trying to be consistent with real scenarios. Importantly, answers in Orca are all well-annotated natural responses rather than the specific spans or short phrase in previous datasets. Besides, we implement three strong baselines to tackle the challenge in Orca. The results indicate the great challenge of our CMRC benchmark.

To present, multilingual language models trained using various pre-training tasks like mask language modeling (MLM) have yielded encouraging results on a wide range of downstream tasks. Despite the promising performances, structural knowledge in cross-lingual corpus is less explored in current works, leading to the semantic misalignment. In this paper, we propose a new pre-training task named Structural Contrast Pretraining (SCP) to align the structural words in a parallel sentence, enhancing the models’ ability to comprehend cross-lingual representations. Concretely, each structural word in source and target languages is regarded as a positive pair in SCP. Since contrastive learning compares positive and negative pairs, an increase in the frequency of negative pairings could enhance the performance of the resulting model. Therefore, we further propose Cross-lingual Momentum Contrast (CL-MoCo) to increase the number of negative pairs by maintaining a large size of the queue. CL-MoCo extends the original Moco approach into cross-lingual training and jointly optimizes the source-to-target language and target-to-source language representations, resulting in a more suitable encoder for cross-lingual transfer. We conduct extensive experiments to validate the proposed approach on three cross-lingual tasks across five datasets such as MLQA, WikiAnn, etc, and results prove the effectiveness of our method.

Currently, learning better unsupervised sentence representations is the pursuit of many natural language processing communities. Lots of approaches based on pre-trained language models (PLMs) and contrastive learning have achieved promising results on this task. Experimentally, we observe that the over-smoothing problem reduces the capacity of these powerful PLMs, leading to sub-optimal sentence representations. In this paper, we present a Simple method named Self-Contrastive Learning (SSCL) to alleviate this issue, which samples negatives from PLMs intermediate layers, improving the quality of the sentence representation. Our proposed method is quite simple and can be easily extended to various state-of-the-art models for performance boosting, which can be seen as a plug-and-play contrastive framework for learning unsupervised sentence representation. Extensive results prove that SSCL brings the superior performance improvements of different strong baselines (e.g., BERT and SimCSE) on Semantic Textual Similarity and Transfer datasets

Code pre-trained models (CodePTMs) have recently demonstrated significant success in code intelligence. To interpret these models, some probing methods have been applied. However, these methods fail to consider the inherent characteristics of codes. In this paper, to address the problem, we propose a novel probing method CAT-probing to quantitatively interpret how CodePTMs attend code structure. We first denoise the input code sequences based on the token types pre-defined by the compilers to filter those tokens whose attention scores are too small. After that, we define a new metric CAT-score to measure the commonality between the token-level attention scores generated in CodePTMs and the pair-wise distances between corresponding AST nodes. The higher the CAT-score, the stronger the ability of CodePTMs to capture code structure. We conduct extensive experiments to integrate CAT-probing with representative CodePTMs for different programming languages. Experimental results show the effectiveness of CAT-probing in CodePTM interpretation. Our codes and data are publicly available at https://github.com/nchen909/CodeAttention.

Multi-intent natural language understanding (NLU) has recently gained attention. It detects multiple intents in an utterance, which is better suited to real-world scenarios. However, the state-of-the-art joint NLU models mainly detect multiple intents on threshold-based strategy, resulting in one main issue: the model is extremely sensitive to the threshold settings. In this paper, we propose a transformer-based Threshold-Free Multi-intent NLU model (TFMN) with multi-task learning (MTL). Specifically, we first leverage multiple layers of a transformer-based encoder to generate multi-grain representations. Then we exploit the information of the number of multiple intents in each utterance without additional manual annotations and propose an auxiliary detection task: Intent Number detection (IND). Furthermore, we propose a threshold-free intent multi-intent classifier that utilizes the output of IND task and detects the multiple intents without depending on the threshold. Extensive experiments demonstrate that our proposed model achieves superior results on two public multi-intent datasets.

Spoken question answering (SQA) requires fine-grained understanding of both spoken documents and questions for the optimal answer prediction. In this paper, we propose novel training schemes for spoken question answering with a self-supervised training stage and a contrastive representation learning stage. In the self-supervised stage, we propose three auxiliary self-supervised tasks, including utterance restoration, utterance insertion, and question discrimination, and jointly train the model to capture consistency and coherence among speech documents without any additional data or annotations. We then propose to learn noise-invariant utterance representations in a contrastive objective by adopting multiple augmentation strategies, including span deletion and span substitution. Besides, we design a Temporal-Alignment attention to semantically align the speech-text clues in the learned common space and benefit the SQA tasks. By this means, the training schemes can more effectively guide the generation model to predict more proper answers. Experimental results show that our model achieves state-of-the-art results on three SQA benchmarks. Our code will be publicly available after publication.