ACL 2024 论文列表

Whether embedding spaces use all their dimensions equally, i.e., whether they are isotropic, has been a recent subject of discussion. Evidence has been accrued both for and against enforcing isotropy in embedding spaces. In the present paper, we stress that isotropy imposes requirements on the embedding space that are not compatible with the presence of clusters—which also negatively impacts linear classification objectives. We demonstrate this fact both empirically and mathematically and use it to shed light on previous results from the literature.

Humans often express their communicative intents indirectly or non-literally, which requires their interlocutors—human or AI—to understand beyond the literal meaning of words. While most existing work has focused on discriminative evaluations, we present a new approach to generatively evaluate large language models’ (LLMs’) intention understanding by examining their responses to non-literal utterances. Ideally, an LLM should respond in line with the true intention of a non-literal utterance, not its literal interpretation. Our findings show that LLMs struggle to generate contextually relevant responses to non-literal language. We also find that providing oracle intentions substantially improves response appropriateness, but using chain-of-thought to make models spell out intentions before responding improves much less. These findings suggest that LLMs are not yet pragmatic interlocutors, and that explicitly modeling intention could improve LLM responses to non-literal language.

Retrieving relevant tables containing the necessary information to accurately answer a given question over tables is critical to open-domain question-answering (QA) systems. Previous methods assume the answer to such a question can be found either in a single table or multiple tables identified through question decomposition or rewriting. However, neither of these approaches is sufficient, as many questions require retrieving multiple tables and joining them through a join plan that cannot be discerned from the user query itself. If the join plan is not considered in the retrieval stage, the subsequent steps of reasoning and answering based on those retrieved tables are likely to be incorrect. To address this problem, we introduce a method that uncovers useful join relations for any query and database during table retrieval. We use a novel re-ranking method formulated as a mixed-integer program that considers not only table-query relevance but also table-table relevance that requires inferring join relationships. Our method outperforms the state-of-the-art approaches for table retrieval by up to 9.3% in F1 score and for end-to-end QA by up to 5.4% in accuracy.

Explainable AI methods facilitate the understanding of model behaviour, yet, small, imperceptible perturbations to inputs can vastly distort explanations. As these explanations are typically evaluated holistically, before model deployment, it is difficult to assess when a particular explanation is trustworthy. Some studies have tried to create confidence estimators for explanations, but none have investigated an existing link between uncertainty and explanation quality. We artificially simulate epistemic uncertainty in text input by introducing noise at inference time. In this large-scale empirical study, we insert different levels of noise perturbations and measure the effect on the output of pre-trained language models and different uncertainty metrics. Realistic perturbations have minimal effect on performance and explanations, yet masking has a drastic effect. We find that high uncertainty doesn’t necessarily imply low explanation plausibility; the correlation between the two metrics can be moderately positive when noise is exposed during the training process. This suggests that noise-augmented models may be better at identifying salient tokens when uncertain. Furthermore, when predictive and epistemic uncertainty measures are over-confident, the robustness of a saliency map to perturbation can indicate model stability issues. Integrated Gradients shows the overall greatest robustness to perturbation, while still showing model-specific patterns in performance; however, this phenomenon is limited to smaller Transformer-based language models.

Understanding textual description to generate code seems to be an achieved capability of instruction-following Large Language Models (LLMs) in zero-shot scenario. However, there is a severe possibility that this translation ability may be influenced by having seen target textual descriptions and the related code. This effect is known as Data Contamination.In this study, we investigate the impact of Data Contamination on the performance of GPT-3.5 in the Text-to-SQL code-generating tasks. Hence, we introduce a novel method to detect Data Contamination in GPTs and examine GPT-3.5’s Text-to-SQL performances using the known Spider Dataset and our new unfamiliar dataset Termite. Furthermore, we analyze GPT-3.5’s efficacy on databases with modified information via an adversarial table disconnection (ATD) approach, complicating Text-to-SQL tasks by removing structural pieces of information from the database. Our results indicate a significant performance drop in GPT-3.5 on the unfamiliar Termite dataset, even with ATD modifications, highlighting the effect of Data Contamination on LLMs in Text-to-SQL translation tasks.

Though advanced in understanding visual information with human languages, Large Vision-Language Models (LVLMs) still suffer from multimodal hallucinations. A natural concern is that during multimodal interaction, the generated hallucinations could influence the LVLMs’ subsequent generation. Thus, we raise a question: When presented with a query relevant to the previously generated hallucination, will LVLMs be misled and respond incorrectly, even though the ground visual information exists? To answer this, we propose a framework called \\textitMMHalSnowball to evaluate LVLMs’ behaviors when encountering generated hallucinations, where LVLMs are required to answer specific visual questions within a curated hallucinatory conversation. Crucially, our experiment shows that the performance of open-source LVLMs drops by at least 31\\%, indicating that LVLMs are prone to accept the generated hallucinations and make false claims that they would not have supported without distractions. We term this Multimodal Hallucination Snowballing. To mitigate this issue, we further propose a training-free method called Residual Visual Decoding, where we revise the output distribution of LVLMs with the one derived from the residual visual input, providing models with direct access to the visual information. Experiments show that our method can mitigate more than 24\\% of the snowballed multimodal hallucination while maintaining capabilities.

LLMs are increasingly powerful and widely used to assist users in a variety of tasks. This use risks introducing LLM biases into consequential decisions such as job hiring, human performance evaluation, and criminal sentencing. Bias in NLP systems along the lines of gender and ethnicity has been widely studied, especially for specific stereotypes (e.g., Asians are good at math). In this paper, we investigate bias along less-studied but still consequential, dimensions, such as age and beauty, measuring subtler correlated decisions that LLMs make between social groups and unrelated positive and negative attributes. Although these subtler biases are understudied they follow people as much as gender and ethnicity do. So, we want to see whether they also follow one with LLMs.We introduce a template-generated dataset of sentence completion tasks that asks the model to select the most appropriate attribute to complete an evaluative statement about a person described as a member of a specific social group. We also reverse the completion task to select the social group based on an attribute. We report the correlations that we find for 4 cutting-edge LLMs. This dataset can be used as a benchmark to evaluate progress in more generalized biases and the templating technique can be used to expand the benchmark with minimal additional human annotation.

Recent work has showcased the powerful capability of large language models (LLMs) in recalling knowledge and reasoning. However, the reliability of LLMs in combining these two capabilities into reasoning through multi-hop facts has not been widely explored. This paper systematically investigates the possibilities for LLMs to utilize shortcuts based on direct connections between the initial and terminal entities of multi-hop knowledge. We first explore the existence of factual shortcuts through Knowledge Neurons, revealing that: (i) the strength of factual shortcuts is highly correlated with the frequency of co-occurrence of initial and terminal entities in the pre-training corpora; (ii) few-shot prompting leverage more shortcuts in answering multi-hop questions compared to chain-of-thought prompting. Then, we analyze the risks posed by factual shortcuts from the perspective of multi-hop knowledge editing. Analysis shows that approximately 20% of the failures are attributed to shortcuts, and the initial and terminal entities in these failure instances usually have higher co-occurrences in the pre-training corpus. Finally, we propose erasing shortcut neurons to mitigate the associated risks and find that this approach significantly reduces failures in multiple-hop knowledge editing caused by shortcuts. Code is publicly available at https://github.com/Jometeorie/MultiHopShortcuts.

The intricate relationship between language and culture has long been a subject of exploration within the realm of linguistic anthropology. Large Language Models (LLMs), promoted as repositories of collective human knowledge, raise a pivotal question: do these models genuinely encapsulate the diverse knowledge adopted by different cultures? Our study reveals that these models demonstrate greater cultural alignment along two dimensions—firstly, when prompted with the dominant language of a specific culture, and secondly, when pretrained with a refined mixture of languages employed by that culture. We quantify cultural alignment by simulating sociological surveys, comparing model responses to those of actual survey participants as references. Specifically, we replicate a survey conducted in various regions of Egypt and the United States through prompting LLMs with different pretraining data mixtures in both Arabic and English with the personas of the real respondents and the survey questions. Further analysis reveals that misalignment becomes more pronounced for underrepresented personas and for culturally sensitive topics, such as those probing social values. Finally, we introduce Anthropological Prompting, a novel method leveraging anthropological reasoning to enhance cultural alignment. Our study emphasizes the necessity for a more balanced multilingual pretraining dataset to better represent the diversity of human experience and the plurality of different cultures with many implications on the topic of cross-lingual transfer.

This study investigates the behaviors of Large Language Models (LLMs) when faced with conflicting prompts versus their internal memory. This will not only help to understand LLMs’ decision mechanism but also benefit real-world applications, such as retrieval-augmented generation (RAG).Drawing on cognitive theory, we target the first scenario of decision-making styles where there is no superiority in the conflict and categorize LLMs’ preference into dependent, intuitive, and rational/irrational styles.Another scenario of factual robustness considers the correctness of prompt and memory in knowledge-intensive tasks, which can also distinguish if LLMs behave rationally or irrationally in the first scenario.To quantify them, we establish a complete benchmarking framework including a dataset, a robustness evaluation pipeline, and corresponding metrics. Extensive experiments with seven LLMs reveal their varying behaviors. And, with role play intervention, we can change the styles, but different models present distinct adaptivity and upper-bound. One of our key takeaways is to optimize models or the prompts according to the identified style. For instance, RAG models with high role play adaptability may dynamically adjust the interventions according to the quality of retrieval results — being dependent to better leverage informative context; and, being intuitive when external prompt is noisy.

Recently, a human evaluation study of Referring Expression Generation (REG) models had an unexpected conclusion: on WEBNLG, Referring Expressions (REs) generated by the state-of-the-art neural models were not only indistinguishable from the REs in WEBNLG but also from the REs generated by a simple rule-based system. Here, we argue that this limitation could stem from the use of a purely ratings-based human evaluation (which is a common practice in Natural Language Generation). To investigate these issues, we propose an intrinsic task-based evaluation for REG models, in which, in addition to rating the quality of REs, participants were asked to accomplish two meta-level tasks. One of these tasks concerns the referential success of each RE; the other task asks participants to suggest a better alternative for each RE. The outcomes suggest that, in comparison to previous evaluations, the new evaluation protocol assesses the performance of each REG model more comprehensively and makes the participants’ ratings more reliable and discriminable.

Similar to vision-and-language navigation (VLN) tasks that focus on bridging the gap between vision and language for embodied navigation, the new Rendezvous (RVS) task requires reasoning over allocentric spatial relationships using non-sequential navigation instructions and maps. However, performance substantially drops in new environments with no training data.Using opensource descriptions paired with coordinates (e.g., Wikipedia) provides training data but suffers from limited spatially-oriented text resulting in low geolocation resolution. We propose a large-scale augmentation method for generating high-quality synthetic data for new environments using readily available geospatial data. Our method constructs a grounded knowledge-graph, capturing entity relationships. Sampled entities and relations (“shop north of school”) generate navigation instructions via (i) generating numerous templates using context-free grammar (CFG) to embed specific entities and relations; (ii) feeding the entities and relation into a large language model (LLM) for instruction generation. A comprehensive evaluation on RVS, showed that our approach improves the 100-meter accuracy by 45.83% on unseen environments. Furthermore, we demonstrate that models trained with CFG-based augmentation achieve superior performance compared with those trained with LLM-based augmentation, both in unseen and seen environments. These findings suggest that the potential advantages of explicitly structuring spatial information for text-based geospatial reasoning in previously unknown, can unlock data-scarce scenarios.

Despite the many advances of Large Language Models (LLMs) and their unprecedented rapid evolution, their impact and integration into every facet of our daily lives is limited due to various reasons. One critical factor hindering their widespread adoption is the occurrence of hallucinations, where LLMs invent answers that sound realistic, yet drift away from factual truth. In this paper, we present a novel method for detecting hallucinations in large language models, which tackles a critical issue in the adoption of these models in various real-world scenarios. Through extensive evaluations across multiple datasets and LLMs, including Llama-2, we study the hallucination levels of various recent LLMs and demonstrate the effectiveness of our method to automatically detect them. Notably, we observe up to 87% hallucinations for Llama-2 in a specific experiment, where our method achieves a Balanced Accuracy of 81%, all without relying on external knowledge.

Conversational dense retrieval has shown to be effective in conversational search. However, a major limitation of conversational dense retrieval is their lack of interpretability, hindering intuitive understanding of model behaviors for targeted improvements. This paper presents CONVINV, a simple yet effective approach to shed light on interpretable conversational dense retrieval models. CONVINV transforms opaque conversational session embeddings into explicitly interpretable text while faithfully maintaining their original retrieval performance as much as possible. Such transformation is achieved by training a recently proposed Vec2Text model based on the ad-hoc query encoder, leveraging the fact that the session and query embeddings share the same space in existing conversational dense retrieval.To further enhance interpretability, we propose to incorporate external interpretable query rewrites into the transformation process. Extensive evaluations on three conversational search benchmarks demonstrate that CONVINV can yield more interpretable text and faithfully preserve original retrieval performance than baselines. Our work connects opaque session embeddings with transparent query rewriting, paving the way toward trustworthy conversational search.

Accurate and interpretable user satisfaction estimation (USE) is critical for understanding, evaluating, and continuously improving conversational systems. Users express their satisfaction or dissatisfaction with diverse conversational patterns in both general-purpose (ChatGPT and Bing Copilot) and task-oriented (customer service chatbot) conversational systems. Existing approaches based on featurized ML models or text embeddings fall short in extracting generalizable patterns and are hard to interpret. In this work, we show that LLMs can extract interpretable signals of user satisfaction from their natural language utterances more effectively than embedding-based approaches. Moreover, an LLM can be tailored for USE via an iterative prompting framework using supervision from labeled examples. Our proposed method, Supervised Prompting for User satisfaction Rubrics (SPUR), not only has higher accuracy but is more interpretable as it scores user satisfaction via learned rubrics with a detailed breakdown.

The usual way to interpret language models (LMs) is to test their performance on different benchmarks and subsequently infer their internal processes.In this paper, we present an alternative approach, concentrating on the _quality_ of LM processing, with a focus on their language abilities.To this end, we construct ‘linguistic task spaces’ – representations of an LM’s language conceptualisation – that shed light on the connections LMs draw between language phenomena.Task spaces are based on the interactions of the learning signals from different linguistic phenomena, which we assess via a method we call ‘similarity probing’.To disentangle the learning signals of linguistic phenomena, we further introduce a method called ‘fine-tuning via gradient differentials’ (FTGD).We apply our methods to language models of three different scales and find that larger models generalise better to overarching general concepts for linguistic tasks, making better use of their shared structure. Further, the distributedness of linguistic processing increases with pre-training through increased parameter sharing between related linguistic tasks. The overall generalisation patterns are mostly stable throughout training and not marked by incisive stages, potentially explaining the lack of successful curriculum strategies for LMs.

This study explores the realm of knowledge base question answering (KBQA). KBQA is considered a challenging task, particularly in parsing intricate questions into executable logical forms. Traditional semantic parsing (SP)-based methods require extensive data annotations, which result in significant costs. Recently, the advent of few-shot in-context learning, powered by large language models (LLMs), has showcased promising capabilities. Yet, fully leveraging LLMs to parse questions into logical forms in low-resource scenarios poses a substantial challenge. To tackle these hurdles, we introduce Interactive-KBQA, a framework designed to generate logical forms through direct interaction with knowledge bases (KBs). Within this framework, we have developed three generic APIs for KB interaction. For each category of complex question, we devised exemplars to guide LLMs through the reasoning processes. Our method achieves competitive results on the WebQuestionsSP, ComplexWebQuestions, KQA Pro, and MetaQA datasets with a minimal number of examples (shots). Importantly, our approach supports manual intervention, allowing for the iterative refinement of LLM outputs. By annotating a dataset with step-wise reasoning processes, we showcase our model’s adaptability and highlight its potential for contributing significant enhancements to the field.

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.

Advances in machine learning have made it possible to perform various text and speech processing tasks, such as automatic speech recognition (ASR), in an end-to-end (E2E) manner. E2E approaches utilizing pre-trained models are gaining attention for conserving training data and resources. However, most of their applications in ASR involve only one of either a pre-trained speech or a language model. This paper proposes integrating a pre-trained speech representation model and a large language model (LLM) for E2E ASR. The proposed model enables the optimization of the entire ASR process, including acoustic feature extraction and acoustic and language modeling, by combining pre-trained models with a bridge network and also enables the application of remarkable developments in LLM utilization, such as parameter-efficient domain adaptation and inference optimization. Experimental results demonstrate that the proposed model achieves a performance comparable to that of modern E2E ASR models by utilizing powerful pre-training models with the proposed integrated approach.