Abstract:Multi-hop reasoning, which requires multi-step reasoning based on the supporting documents within a given context, remains challenging for large language models (LLMs). LLMs often struggle to filter out irrelevant documents within the context, and their performance is sensitive to the position of supporting documents within that context. In this paper, we identify an additional challenge: LLMs' performance is also sensitive to the order in which the supporting documents are presented. We refer to this as the misordered context problem. To address this issue, we propose a simple yet effective method called context repetition (CoRe), which involves prompting the model by repeatedly presenting the context to ensure the supporting documents are presented in the optimal order for the model. Using CoRe, we improve the F1 score by up to 30%p on multi-hop QA tasks and increase accuracy by up to 70%p on a synthetic task. Additionally, CoRe helps mitigate the well-known "lost-in-the-middle" problem in LLMs and can be effectively combined with retrieval-based approaches utilizing Chain-of-Thought (CoT) reasoning.
Abstract:A binary decision task, like yes-no questions or answer verification, reflects a significant real-world scenario such as where users look for confirmation about the correctness of their decisions on specific issues. In this work, we observe that language models exhibit a negative bias in the binary decisions of complex reasoning tasks. Based on our observations and the rationale about attention-based model dynamics, we propose a negative attention score (NAS) to systematically and quantitatively formulate negative bias. Based on NAS, we identify attention heads that attend to negative tokens provided in the instructions as answer candidate of binary decisions, regardless of the question in the prompt, and validate their association with the negative bias. Additionally, we propose the negative attention score alignment (NASA) method, which is a parameter-efficient fine-tuning technique to address the extracted negatively biased attention heads. Experimental results from various domains of reasoning tasks and large model search space demonstrate that NASA significantly reduces the gap between precision and recall caused by negative bias while preserving their generalization abilities. Our codes are available at \url{https://github.com/ysw1021/NASA}.
Abstract:In this paper, we identify a new category of bias that induces input-conflicting hallucinations, where large language models (LLMs) generate responses inconsistent with the content of the input context. This issue we have termed the false negative problem refers to the phenomenon where LLMs are predisposed to return negative judgments when assessing the correctness of a statement given the context. In experiments involving pairs of statements that contain the same information but have contradictory factual directions, we observe that LLMs exhibit a bias toward false negatives. Specifically, the model presents greater overconfidence when responding with False. Furthermore, we analyze the relationship between the false negative problem and context and query rewriting and observe that both effectively tackle false negatives in LLMs.
Abstract:In this paper, we primarily address the issue of dialogue-form context query within the interactive text-to-image retrieval task. Our methodology, PlugIR, actively utilizes the general instruction-following capability of LLMs in two ways. First, by reformulating the dialogue-form context, we eliminate the necessity of fine-tuning a retrieval model on existing visual dialogue data, thereby enabling the use of any arbitrary black-box model. Second, we construct the LLM questioner to generate non-redundant questions about the attributes of the target image, based on the information of retrieval candidate images in the current context. This approach mitigates the issues of noisiness and redundancy in the generated questions. Beyond our methodology, we propose a novel evaluation metric, Best log Rank Integral (BRI), for a comprehensive assessment of the interactive retrieval system. PlugIR demonstrates superior performance compared to both zero-shot and fine-tuned baselines in various benchmarks. Additionally, the two methodologies comprising PlugIR can be flexibly applied together or separately in various situations. Our codes are available at https://github.com/Saehyung-Lee/PlugIR.
Abstract:Despite recent progress in language models, generating constrained text for specific domains remains a challenge, particularly when utilizing black-box models that lack domain-specific knowledge. In this paper, we introduce ScoPE (Score-based Progressive Editor) generation, a novel approach for controlled text generation for black-box language models. We employ ScoPE to facilitate text generation in the target domain by integrating it with language models through a cascading approach. Trained to enhance the target domain score of the edited text, ScoPE progressively edits intermediate output discrete tokens to align with the target attributes throughout the auto-regressive generation process of the language model. This iterative process guides subsequent steps to produce desired output texts for the target domain. Our experimental results on diverse controlled generations demonstrate that ScoPE effectively facilitates controlled text generation for black-box language models in both in-domain and out-of-domain conditions, which is challenging for existing methods.
Abstract:Despite advances in neural network language model, the representation degeneration problem of embeddings is still challenging. Recent studies have found that the learned output embeddings are degenerated into a narrow-cone distribution which makes the similarity between each embeddings positive. They analyzed the cause of the degeneration problem has been demonstrated as common to most embeddings. However, we found that the degeneration problem is especially originated from the training of embeddings of rare words. In this study, we analyze the intrinsic mechanism of the degeneration of rare word embeddings with respect of their gradient about the negative log-likelihood loss function. Furthermore, we theoretically and empirically demonstrate that the degeneration of rare word embeddings causes the degeneration of non-rare word embeddings, and that the overall degeneration problem can be alleviated by preventing the degeneration of rare word embeddings. Based on our analyses, we propose a novel method, Adaptive Gradient Partial Scaling(AGPS), to address the degeneration problem. Experimental results demonstrate the effectiveness of the proposed method qualitatively and quantitatively.