Abstract:Through reading the documentation in the context, tool-using language models can dynamically extend their capability using external tools. The cost is that we have to input lengthy documentation every time the model needs to use the tool, occupying the input window as well as slowing down the decoding process. Given the progress in general-purpose compression, soft context compression is a suitable approach to alleviate the problem. However, when compressing tool documentation, existing methods suffer from the weaknesses of key information loss (specifically, tool/parameter name errors) and difficulty in adjusting the length of compressed sequences based on documentation lengths. To address these problems, we propose two strategies for compressing tool documentation into concise and precise summary sequences for tool-using language models. 1) Selective compression strategy mitigates key information loss by deliberately retaining key information as raw text tokens. 2) Block compression strategy involves dividing tool documentation into short chunks and then employing a fixed-length compression model to achieve variable-length compression. This strategy facilitates the flexible adjustment of the compression ratio. Results on API-Bank and APIBench show that our approach reaches a performance comparable to the upper-bound baseline under up to 16x compression ratio.
Abstract:Human and model-generated texts can be distinguished by examining the magnitude of likelihood in language. However, it is becoming increasingly difficult as language model's capabilities of generating human-like texts keep evolving. This study provides a new perspective by using the relative likelihood values instead of absolute ones, and extracting useful features from the spectrum-view of likelihood for the human-model text detection task. We propose a detection procedure with two classification methods, supervised and heuristic-based, respectively, which results in competitive performances with previous zero-shot detection methods and a new state-of-the-art on short-text detection. Our method can also reveal subtle differences between human and model languages, which find theoretical roots in psycholinguistics studies. Our code is available at https://github.com/CLCS-SUSTech/FourierGPT
Abstract:Much of the success of modern language models depends on finding a suitable prompt to instruct the model. Until now, it has been largely unknown how variations in the linguistic expression of prompts affect these models. This study systematically and empirically evaluates which linguistic features influence models through paraphrase types, i.e., different linguistic changes at particular positions. We measure behavioral changes for five models across 120 tasks and six families of paraphrases (i.e., morphology, syntax, lexicon, lexico-syntax, discourse, and others). We also control for other prompt engineering factors (e.g., prompt length, lexical diversity, and proximity to training data). Our results show a potential for language models to improve tasks when their prompts are adapted in specific paraphrase types (e.g., 6.7% median gain in Mixtral 8x7B; 5.5% in LLaMA 3 8B). In particular, changes in morphology and lexicon, i.e., the vocabulary used, showed promise in improving prompts. These findings contribute to developing more robust language models capable of handling variability in linguistic expression.
Abstract:Decompilation transforms compiled code back into a high-level programming language for analysis when source code is unavailable. Previous work has primarily focused on enhancing decompilation performance by increasing the scale of model parameters or training data for pre-training. Based on the characteristics of the decompilation task, we propose two methods: (1) Without fine-tuning, the Self-Constructed Context Decompilation (sc$^2$dec) method recompiles the LLM's decompilation results to construct pairs for in-context learning, helping the model improve decompilation performance. (2) Fine-grained Alignment Enhancement (FAE), which meticulously aligns assembly code with source code at the statement level by leveraging debugging information, is employed during the fine-tuning phase to achieve further improvements in decompilation. By integrating these two methods, we achieved a Re-Executability performance improvement of approximately 7.35\% on the Decompile-Eval benchmark, establishing a new state-of-the-art performance of 55.03\%.
Abstract:Large Language Models (LLMs) have shown impressive capabilities but also a concerning tendency to hallucinate. This paper presents RefChecker, a framework that introduces claim-triplets to represent claims in LLM responses, aiming to detect fine-grained hallucinations. In RefChecker, an extractor generates claim-triplets from a response, which are then evaluated by a checker against a reference. We delineate three task settings: Zero, Noisy and Accurate Context, to reflect various real-world use cases. We curated a benchmark spanning various NLP tasks and annotated 11k claim-triplets from 2.1k responses by seven LLMs. RefChecker supports both proprietary and open-source models as the extractor and checker. Experiments demonstrate that claim-triplets enable superior hallucination detection, compared to other granularities such as response, sentence and sub-sentence level claims. RefChecker outperforms prior methods by 6.8 to 26.1 points on our benchmark and the checking results of RefChecker are strongly aligned with human judgments. This work is open sourced at https://github.com/amazon-science/RefChecker
Abstract:State estimation is a critical foundational module in robotics applications, where robustness and performance are paramount. Although in recent years, many works have been focusing on improving one of the most widely adopted state estimation methods, visual inertial odometry (VIO), by incorporating multiple cameras, these efforts predominantly address synchronous camera systems. Asynchronous cameras, which offer simpler hardware configurations and enhanced resilience, have been largely overlooked. To fill this gap, this paper presents VINS-Multi, a novel multi-camera-IMU state estimator for asynchronous cameras. The estimator comprises parallel front ends, a front end coordinator, and a back end optimization module capable of handling asynchronous input frames. It utilizes the frames effectively through a dynamic feature number allocation and a frame priority coordination strategy. The proposed estimator is integrated into a customized quadrotor platform and tested in multiple realistic and challenging scenarios to validate its practicality. Additionally, comprehensive benchmark results are provided to showcase the robustness and superior performance of the proposed estimator.
Abstract:In view of the fact that most of the existing machine translation evaluation algorithms only consider the lexical and syntactic information, but ignore the deep semantic information contained in the sentence, this paper proposes a computational method for evaluating the semantic correctness of machine translations based on reference translations and incorporating semantic dependencies and sentence keyword information. Use the language technology platform developed by the Social Computing and Information Retrieval Research Center of Harbin Institute of Technology to conduct semantic dependency analysis and keyword analysis on sentences, and obtain semantic dependency graphs, keywords, and weight information corresponding to keywords. It includes all word information with semantic dependencies in the sentence and keyword information that affects semantic information. Construct semantic association pairs including word and dependency multi-features. The key semantics of the sentence cannot be highlighted in the semantic information extracted through semantic dependence, resulting in vague semantics analysis. Therefore, the sentence keyword information is also included in the scope of machine translation semantic evaluation. To achieve a comprehensive and in-depth evaluation of the semantic correctness of sentences, the experimental results show that the accuracy of the evaluation algorithm has been improved compared with similar methods, and it can more accurately measure the semantic correctness of machine translation.
Abstract:Human action recognition and performance assessment have been hot research topics in recent years. Recognition problems have mature solutions in the field of sign language, but past research in performance analysis has focused on competitive sports and medical training, overlooking the scoring assessment ,which is an important part of sign language teaching digitalization. In this paper, we analyze the existing technologies for performance assessment and adopt methods that perform well in human pose reconstruction tasks combined with motion rotation embedded expressions, proposing a two-stage sign language performance evaluation pipeline. Our analysis shows that choosing reconstruction tasks in the first stage can provide more expressive features, and using smoothing methods can provide an effective reference for assessment. Experiments show that our method provides good score feedback mechanisms and high consistency with professional assessments compared to end-to-end evaluations.
Abstract:Traditional sign language teaching methods face challenges such as limited feedback and diverse learning scenarios. Although 2D resources lack real-time feedback, classroom teaching is constrained by a scarcity of teacher. Methods based on VR and AR have relatively primitive interaction feedback mechanisms. This study proposes an innovative teaching model that uses real-time monocular vision and mixed reality technology. First, we introduce an improved hand-posture reconstruction method to achieve sign language semantic retention and real-time feedback. Second, a ternary system evaluation algorithm is proposed for a comprehensive assessment, maintaining good consistency with experts in sign language. Furthermore, we use mixed reality technology to construct a scenario-based 3D sign language classroom and explore the user experience of scenario teaching. Overall, this paper presents a novel teaching method that provides an immersive learning experience, advanced posture reconstruction, and precise feedback, achieving positive feedback on user experience and learning effectiveness.
Abstract:Recent advancement in large language models (LLMs) has offered a strong potential for natural language systems to process informal language. A representative form of informal language is slang, used commonly in daily conversations and online social media. To date, slang has not been comprehensively evaluated in LLMs due partly to the absence of a carefully designed and publicly accessible benchmark. Using movie subtitles, we construct a dataset that supports evaluation on a diverse set of tasks pertaining to automatic processing of slang. For both evaluation and finetuning, we show the effectiveness of our dataset on two core applications: 1) slang detection, and 2) identification of regional and historical sources of slang from natural sentences. We also show how our dataset can be used to probe the output distributions of LLMs for interpretive insights. We find that while LLMs such as GPT-4 achieve good performance in a zero-shot setting, smaller BERT-like models finetuned on our dataset achieve comparable performance. Furthermore, we show that our dataset enables finetuning of LLMs such as GPT-3.5 that achieve substantially better performance than strong zero-shot baselines. Our work offers a comprehensive evaluation and a high-quality benchmark on English slang based on the OpenSubtitles corpus, serving both as a publicly accessible resource and a platform for applying tools for informal language processing.