Abstract:Present Large Language Models (LLM) self-training methods always under-sample on challenging queries, leading to inadequate learning on difficult problems which limits LLMs' ability. Therefore, this work proposes a difficulty-aware self-training (DAST) framework that focuses on improving both the quantity and quality of self-generated responses on challenging queries during self-training. DAST is specified in three components: 1) sampling-based difficulty level estimation, 2) difficulty-aware data augmentation, and 3) the self-training algorithm using SFT and DPO respectively. Experiments on mathematical tasks demonstrate the effectiveness and generalization of DAST, highlighting the critical role of difficulty-aware strategies in advancing LLM self-training.
Abstract:Current evaluations of commonsense reasoning in LLMs are hindered by the scarcity of natural language corpora with structured annotations for reasoning tasks. To address this, we introduce KnowLogic, a benchmark generated through a knowledge-driven synthetic data strategy. KnowLogic integrates diverse commonsense knowledge, plausible scenarios, and various types of logical reasoning. One of the key advantages of KnowLogic is its adjustable difficulty levels, allowing for flexible control over question complexity. It also includes fine-grained labels for in-depth evaluation of LLMs' reasoning abilities across multiple dimensions. Our benchmark consists of 3,000 bilingual (Chinese and English) questions across various domains, and presents significant challenges for current LLMs, with the highest-performing model achieving only 69.57\%. Our analysis highlights common errors, such as misunderstandings of low-frequency commonsense, logical inconsistencies, and overthinking. This approach, along with our benchmark, provides a valuable tool for assessing and enhancing LLMs' commonsense reasoning capabilities and can be applied to a wide range of knowledge domains.
Abstract:In order to deeply understand the capability of pretrained language models in text generation and conduct a diagnostic evaluation, we propose TGEA, an error-annotated dataset with multiple benchmark tasks for text generation from pretrained language models (PLMs). We use carefully selected prompt words to guide GPT-2 to generate candidate sentences, from which we select 47K for error annotation. Crowdsourced workers manually check each of these sentences and detect 12k erroneous sentences. We create an error taxonomy to cover 24 types of errors occurring in these erroneous sentences according to the nature of errors with respect to linguistics and knowledge (eg, common sense). For each erroneous span in PLM-generated sentences, we also detect another span that is closely associated with it. Each error is hence manually labeled with comprehensive annotations, including the span of the error, the associated span, minimal correction to the error, the type of the error, and rationale behind the error. Apart from the fully annotated dataset, we also present a detailed description of the data collection procedure, statistics and analysis of the dataset. This is the first dataset with comprehensive annotations for PLM-generated texts, which facilitates the diagnostic evaluation of PLM-based text generation. Furthermore, we use TGEA as a benchmark dataset and propose a series of automatic diagnosis tasks, including error detection, error type classification, associated span detection, error rationale generation, to further promote future study on the automatic error detection and correction on texts generated by pretrained language models.
Abstract:Does neural machine translation yield translations that are congenial with common sense? In this paper, we present a test suite to evaluate the commonsense reasoning capability of neural machine translation. The test suite consists of three test sets, covering lexical and contextless/contextual syntactic ambiguity that requires commonsense knowledge to resolve. We manually create 1,200 triples, each of which contain a source sentence and two contrastive translations, involving 7 different common sense types. Language models pretrained on large-scale corpora, such as BERT, GPT-2, achieve a commonsense reasoning accuracy of lower than 72% on target translations of this test suite. We conduct extensive experiments on the test suite to evaluate commonsense reasoning in neural machine translation and investigate factors that have impact on this capability. Our experiments and analyses demonstrate that neural machine translation performs poorly on commonsense reasoning of the three ambiguity types in terms of both reasoning accuracy (60.1%) and reasoning consistency (31%). The built commonsense test suite is available at https://github.com/tjunlp-lab/CommonMT.
Abstract:Large Language Models (LLMs) are being used more and more extensively for automated evaluation in various scenarios. Previous studies have attempted to fine-tune open-source LLMs to replicate the evaluation explanations and judgments of powerful proprietary models, such as GPT-4. However, these methods are largely limited to text-based analyses under predefined general criteria, resulting in reduced adaptability for unseen instructions and demonstrating instability in evaluating adherence to quantitative and structural constraints. To address these limitations, we propose a novel evaluation framework, ARJudge, that adaptively formulates evaluation criteria and synthesizes both text-based and code-driven analyses to evaluate LLM responses. ARJudge consists of two components: a fine-tuned Analyzer that generates multi-faceted evaluation analyses and a tuning-free Refiner that combines and refines all analyses to make the final judgment. We construct a Composite Analysis Corpus that integrates tasks for evaluation criteria generation alongside text-based and code-driven analysis generation to train the Analyzer. Our results demonstrate that ARJudge outperforms existing fine-tuned evaluators in effectiveness and robustness. Furthermore, it demonstrates the importance of multi-faceted evaluation and code-driven analyses in enhancing evaluation capabilities.
Abstract:Long context is an important topic in Natural Language Processing (NLP), running through the development of NLP architectures, and offers immense opportunities for Large Language Models (LLMs) giving LLMs the lifelong learning potential akin to humans. Unfortunately, the pursuit of a long context is accompanied by numerous obstacles. Nevertheless, long context remains a core competitive advantage for LLMs. In the past two years, the context length of LLMs has achieved a breakthrough extension to millions of tokens. Moreover, the research on long-context LLMs has expanded from length extrapolation to a comprehensive focus on architecture, infrastructure, training, and evaluation technologies. Inspired by the symphonic poem, Thus Spake Zarathustra, we draw an analogy between the journey of extending the context of LLM and the attempts of humans to transcend its mortality. In this survey, We will illustrate how LLM struggles between the tremendous need for a longer context and its equal need to accept the fact that it is ultimately finite. To achieve this, we give a global picture of the lifecycle of long-context LLMs from four perspectives: architecture, infrastructure, training, and evaluation, showcasing the full spectrum of long-context technologies. At the end of this survey, we will present 10 unanswered questions currently faced by long-context LLMs. We hope this survey can serve as a systematic introduction to the research on long-context LLMs.
Abstract:Visual Instruction Tuning (VIT) enhances Multimodal Large Language Models (MLLMs) but it is hindered by corrupted datasets containing hallucinated content, incorrect responses, and poor OCR quality. While prior works focus on dataset refinement through high-quality data collection or rule-based filtering, they are costly or limited to specific types of corruption. To deeply understand how corrupted data affects MLLMs, in this paper, we systematically investigate this issue and find that while corrupted data degrades the performance of MLLMs, its effects are largely superficial in that the performance of MLLMs can be largely restored by either disabling a small subset of parameters or post-training with a small amount of clean data. Additionally, corrupted MLLMs exhibit improved ability to distinguish clean samples from corrupted ones, enabling the dataset cleaning without external help. Based on those insights, we propose a corruption-robust training paradigm combining self-validation and post-training, which significantly outperforms existing corruption mitigation strategies.
Abstract:Existing approaches to mathematical reasoning with large language models (LLMs) rely on Chain-of-Thought (CoT) for generalizability or Tool-Integrated Reasoning (TIR) for precise computation. While efforts have been made to combine these methods, they primarily rely on post-selection or predefined strategies, leaving an open question: whether LLMs can autonomously adapt their reasoning strategy based on their inherent capabilities. In this work, we propose TATA (Teaching LLMs According to Their Aptitude), an adaptive framework that enables LLMs to personalize their reasoning strategy spontaneously, aligning it with their intrinsic aptitude. TATA incorporates base-LLM-aware data selection during supervised fine-tuning (SFT) to tailor training data to the model's unique abilities. This approach equips LLMs to autonomously determine and apply the appropriate reasoning strategy at test time. We evaluate TATA through extensive experiments on six mathematical reasoning benchmarks, using both general-purpose and math-specialized LLMs. Empirical results demonstrate that TATA effectively combines the complementary strengths of CoT and TIR, achieving superior or comparable performance with improved inference efficiency compared to TIR alone. Further analysis underscores the critical role of aptitude-aware data selection in enabling LLMs to make effective and adaptive reasoning decisions and align reasoning strategies with model capabilities.
Abstract:Multi-modal multi-party conversation (MMC) is a less studied yet important topic of research due to that it well fits real-world scenarios and thus potentially has more widely-used applications. Compared with the traditional multi-modal conversations, MMC requires stronger character-centered understanding abilities as there are many interlocutors appearing in both the visual and textual context. To facilitate the study of this problem, we present Friends-MMC in this paper, an MMC dataset that contains 24,000+ unique utterances paired with video context. To explore the character-centered understanding of the dialogue, we also annotate the speaker of each utterance, the names and bounding bboxes of faces that appear in the video. Based on this Friends-MMC dataset, we further study two fundamental MMC tasks: conversation speaker identification and conversation response prediction, both of which have the multi-party nature with the video or image as visual context. For conversation speaker identification, we demonstrate the inefficiencies of existing methods such as pre-trained models, and propose a simple yet effective baseline method that leverages an optimization solver to utilize the context of two modalities to achieve better performance. For conversation response prediction, we fine-tune generative dialogue models on Friend-MMC, and analyze the benefits of speaker information. The code and dataset is publicly available at https://github.com/yellow-binary-tree/Friends-MMC and thus we call for more attention on modeling speaker information when understanding conversations.
Abstract:Large language models (LLMs) possess vast semantic knowledge but often struggle with complex reasoning tasks, particularly in relational reasoning problems such as kinship or spatial reasoning. In this paper, we present Path-of-Thoughts (PoT), a novel framework designed to tackle relation reasoning by decomposing the task into three key stages: graph extraction, path identification, and reasoning. Unlike previous approaches, PoT efficiently extracts a task-agnostic graph that identifies crucial entities, relations, and attributes within the problem context. Subsequently, PoT identifies relevant reasoning chains within the graph corresponding to the posed question, facilitating inference of potential answers. Experimental evaluations on four benchmark datasets, demanding long reasoning chains, demonstrate that PoT surpasses state-of-the-art baselines by a significant margin (maximum 21.3%) without necessitating fine-tuning or extensive LLM calls. Furthermore, as opposed to prior neuro-symbolic methods, PoT exhibits improved resilience against LLM errors by leveraging the compositional nature of graphs.