EMNLP 2024 论文列表

Large language model (LLM) leads to a surge of autonomous GUI agents for smartphone, which completes a task triggered by natural language through predicting a sequence of actions of API. Even though the task highly relies on past actions and visual observations, existing studies typically consider little semantic information carried out by intermediate screenshots and screen operations. To address this, this work presents Chain-of-Action-Thought (dubbed CoAT), which takes the description of the previous actions, the current screen, and more importantly the action thinking of what actions should be performed and the outcomes led by the chosen action. We demonstrate that, in a zero-shot setting upon three off-the-shelf LMMs, CoAT significantly improves the action prediction compared to previous proposed context modeling. To further facilitate the research in this line, we construct a dataset Android-In-The-Zoo (AitZ), which contains 18,643 screen-action pairs together with chain-of-action-thought annotations. Experiments show that fine-tuning a 1B model (i.e. AUTO-UI-base) on our AitZ dataset achieves on-par performance with CogAgent-Chat-18B.

Emotional intelligence (EI) in artificial intelligence (AI), which refers to the ability of an AI to understand and respond appropriately to human emotions, has emerged as a crucial research topic. Recent studies have shown that large language models (LLMs) and vision large language models (VLLMs) possess EI and the ability to understand emotional stimuli in the form of text and images, respectively. However, factors influencing the emotion prediction performance of VLLMs in real-world conversational contexts have not been sufficiently explored. This study aims to analyze the key elements affecting the emotion prediction performance of VLLMs in conversational contexts systematically. To achieve this, we reconstructed the MELD dataset, which is based on the popular TV series Friends, and conducted experiments through three sub-tasks: overall emotion tone prediction, character emotion prediction, and contextually appropriate emotion expression selection. We evaluated the performance differences based on various model architectures (e.g., image encoders, modality alignment, and LLMs) and image scopes (e.g., entire scene, person, and facial expression). In addition, we investigated the impact of providing persona information on the emotion prediction performance of the models and analyzed how personality traits and speaking styles influenced the emotion prediction process. We conducted an in-depth analysis of the impact of various other factors, such as gender and regional biases, on the emotion prediction performance of VLLMs. The results revealed that these factors significantly influenced the model performance.

Large language models (LLMs) have achieved state-of-the-art performance in machine translation (MT) and demonstrated the ability to leverage in-context learning through few-shot examples. However, the mechanisms by which LLMs use different parts of the input context remain largely unexplored. In this work, we provide a comprehensive analysis of context utilization in MT, studying how LLMs use various context parts, such as few-shot examples and the source text, when generating translations. We highlight several key findings: (1) the source part of few-shot examples appears to contribute more than its corresponding targets, irrespective of translation direction; (2) finetuning LLMs with parallel data alters the contribution patterns of different context parts; and (3) there is a positional bias where earlier few-shot examples have higher contributions to the translated sequence. Finally, we demonstrate that inspecting anomalous context contributions can potentially uncover pathological translations, such as hallucinations. Our findings shed light on the internal workings of LLM-based MT which go beyond those known for standard encoder-decoder MT models.

In text generation, hallucinations refer to the generation of seemingly coherent text that contradicts established knowledge. One compelling hypothesis is that hallucinations occur when a language model is given a generation task outside its parametric knowledge (due to rarity, recency, domain, etc.). A common strategy to address this limitation is to infuse the language models with retrieval mechanisms, providing the model with relevant knowledge for the task. In this paper, we leverage the planning capabilities of instruction-tuned LLMs and analyze how planning can be used to guide retrieval to further reduce the frequency of hallucinations. We empirically evaluate several variations of our proposed approach on long-form text generation tasks. By improving the coverage of relevant facts, plan-guided retrieval and generation can produce more informative responses while providing a higher rate of attribution to source documents.

Humans regularly engage in analogical thinking, relating personal experiences to current situations (X is analogous to Y because of Z). Analogical thinking allows humans to solve problems in creative ways, grasp difficult concepts, and articulate ideas more effectively. Can language models (LMs) do the same? To answer this question, we propose AnaloBench, a benchmark to determine analogical reasoning ability in LMs. Our benchmarking approach focuses on aspects of this ability that are common among humans: (i) recalling related experiences from a large amount of information, and (ii) applying analogical reasoning to complex and lengthy scenarios. We collect a set of 340 high quality, human written analogies for use in our benchmark, which constitutes the largest such collection to date. We then test a broad collection of models consisting of 12 open source and 3 proprietary in various sizes and architectures. As in prior results, scaling up LMs results in some performance boosts. Surprisingly, scale offers minimal gains when, (i) analogies involve lengthy scenarios, or (ii) recalling relevant scenarios from a large pool of information, a process analogous to finding a needle in a haystack. We hope these observations encourage further research in this field.

Given the rise of multimedia content, human translators increasingly focus on culturally adapting not only words but also other modalities such as images to convey the same meaning. While several applications stand to benefit from this, machine translation systems remain confined to dealing with language in speech and text. In this work, we introduce a new task of translating images to make them culturally relevant. First, we build three pipelines comprising state-of-the-art generative models to do the task. Next, we build a two-part evaluation dataset – (i) concept: comprising 600 images that are cross-culturally coherent, focusing on a single concept per image; and (ii) application: comprising 100 images curated from real-world applications. We conduct a multi-faceted human evaluation of translated images to assess for cultural relevance and meaning preservation. We find that as of today, image-editing models fail at this task, but can be improved by leveraging LLMs and retrievers in the loop. Best pipelines can only translate 5% of images for some countries in the easier concept dataset and no translation is successful for some countries in the application dataset, highlighting the challenging nature of the task. Our project webpage is here: https://machine-transcreation.github.io/image-transcreation and our code, data and model outputs can be found here: https://github.com/simran-khanuja/image-transcreation.

Electronic healthcare records are vital for patient safety as they document conditions, plans, and procedures in both free text and medical codes. Language models have significantly enhanced the processing of such records, streamlining workflows and reducing manual data entry, thereby saving healthcare providers significant resources. However, the black-box nature of these models often leaves healthcare professionals hesitant to trust them. State-of-the-art explainability methods increase model transparency but rely on human-annotated evidence spans, which are costly. In this study, we propose an approach to produce plausible and faithful explanations without needing such annotations. We demonstrate on the automated medical coding task that adversarial robustness training improves explanation plausibility and introduce AttInGrad, a new explanation method superior to previous ones. By combining both contributions in a fully unsupervised setup, we produce explanations of comparable quality, or better, to that of a supervised approach. We release our code and model weights.

Data contamination in model evaluation has become increasingly prevalent with the growing popularity of large language models. It allows models to “cheat” via memorisation instead of displaying true capabilities. Therefore, contamination analysis has become an crucial part of reliable model evaluation to validate results. However, existing contamination analysis is usually conducted internally by large language model developers and often lacks transparency and completeness. This paper presents an extensive data contamination report for over 15 popular large language models across six popular multiple-choice QA benchmarks. We also introduce an open-source pipeline that enables the community to perform contamination analysis on customised data and models. Our experiments reveal varying contamination levels ranging from 1% to 45% across benchmarks, with the contamination degree increasing rapidly over time. Performance analysis of large language models indicates that data contamination does not necessarily lead to increased model metrics: while significant accuracy boosts of up to 14% and 7% are observed on contaminated C-Eval and Hellaswag benchmarks, only a minimal increase is noted on contaminated MMLU. We also find larger models seem able to gain more advantages than smaller models on contaminated test sets.

Stance detection aims to identify the attitudes toward specific targets from text, which is an important research area in text mining and social media analytics. Existing research is mainly conducted in monolingual setting on English datasets. To tackle the data scarcity problem in low-resource languages, cross-lingual stance detection (CLSD) transfers the knowledge from high-resource (source) language to low-resource (target) language. The CLSD task is the most challenging in zero-shot setting when no training data is available in target language, and transferring stance-relevant knowledge learned from high-resource language to bridge the language gap is the key for improving the performance of zero-shot CLSD. In this paper, we leverage the capability of large language model (LLM) for stance knowledge acquisition, and propose KEAR, a knowledge elicitation and retrieval framework. The knowledge elicitation module in KEAR first derives different types of stance knowledge from LLM’s reasoning process. Then, the knowledge retrieval module in KEAR matches the target language input to the most relevant stance knowledge for enhancing text representations. Experiments on multilingual datasets show the effectiveness of KEAR compared with competitive baselines as well as the CLSD approaches trained with labeled data in target language.

Research on dialogue constructiveness assessment focuses on (i) analysing conversational factors that influence individuals to take specific actions, win debates, change their perspectives or broaden their open-mindedness and (ii) predicting constructiveness outcomes following dialogues for such use cases. These objectives can be achieved by training either interpretable feature-based models (which often involve costly human annotations) or neural models such as pre-trained language models (which have empirically shown higher task accuracy but lack interpretability). In this paper we propose an LLM feature-based framework for dialogue constructiveness assessment that combines the strengths of feature-based and neural approaches, while mitigating their downsides. The framework first defines a set of dataset-independent and interpretable linguistic features, which can be extracted by both prompting an LLM and simple heuristics. Such features are then used to train LLM feature-based models. We apply this framework to three datasets of dialogue constructiveness and find that our LLM feature-based models outperform or performs at least as well as standard feature-based models and neural models. We also find that the LLM feature-based model learns more robust prediction rules instead of relying on superficial shortcuts, which often trouble neural models.

Extracting finite state automata (FSAs) fromblack-box models offers a powerful approachto gaining interpretable insights into complexmodel behaviors. To support this pursuit, wepresent a weighted variant of Angluin’s (1987)L* algorithm for learning FSAs. We stay faithful to the original formulation, devising a wayto exactly learn deterministic weighted FSAswhose weights support division. Furthermore,we formulate the learning process in a mannerthat highlights the connection with FSA minimization, showing how L* directly learns aminimal automaton for the target language.

With the rapidly-growing deployment of large language model (LLM) inference services, privacy concerns have arisen regarding to the user input data. Recent studies are exploring transforming user inputs to obfuscated embedded vectors, so that the data will not be eavesdropped by service provides. However, in this paper we show that again, without a solid and deliberate security design and analysis, such embedded vector obfuscation failed to protect users’ privacy. We demonstrate the conclusion via conducting a novel inversion attack called Element-wise Differential Nearest Neighbor (EDNN) on the glide-reflection proposed in (CITATION), and the result showed that the original user input text can be 100% recovered from the obfuscated embedded vectors. We further analyze security requirements on embedding obfuscation and present several remedies to our proposed attack.

Aspect sentiment quad prediction (ASQP) is crucial in aspect-based sentiment analysis (ABSA). It involves identifying a text’s aspect,sentiment, opinion, and category. Existing methods have insufficiently explored how to effectively leverage the knowledge of pre-trainedlanguage models (PLMs) to handle implicit aspects and opinions, particularly in combinations such as implicit aspect & explicit opinion, explicit aspect & implicit opinion, and implicit aspect & implicit opinion. We introduce ITSCL, a framework leveraging Instruction Tuning and Supervised Contrastive Learning to improve aspect sentiment quad predictions, especially for implicit aspects and opinions. Implementing this approach presents several challenges. First, designing effective instructions and prompts to optimize the model’s training is difficult. Second, creating sentiment combination vectors with contrastive learning to enhance the model’s discrimination requires further investigation. To address these challenges, ITSCL combines instruction tuning with aligned PLM templates, enabling better knowledge acquisition and identification of implicit sentiments. Additionally, the contrastive learning framework enhances performance by using four fully connected layers to combine sentiments, aspects, opinions, and combinations, maximizing similarity for same-label representationsand minimizing it for different labels. Experimental results show our method significantly outperforms previous methods on benchmark datasets.

The patent citation count is a good indicator of patent quality. This often generates monetary value for the inventors and organizations. However, the factors that influence a patent receiving high citations over the year are still not well understood. With the patents over the past two decades, we study the problem of patent citation prediction and formulate this as a binary classification problem. We create a semantic graph of patents based on their semantic similarities, enabling the use of Graph Neural Network (GNN)-based approaches for predicting citations. Our experimental results demonstrate the effectiveness of our GNN-based methods when applied to the semantic graph, showing that they can accurately predict patent citations using only patent text. More specifically, these methods produce up to 94% recall for patents with high citations and outperform existing baselines. Furthermore, we leverage this constructed graph to gain insights and explanations for the predictions made by the GNNs.

LLM-based agents can greatly extend the abilities of LLMs and thus attract sharply increased studies. An ambitious vision – serving users by manipulating massive API-based tools – has been proposed and explored. However, we find a widely accepted evaluation mechanism for generic agents is still missing. This work aims to fill this gap. We decompose tool use capability into seven aspects and form a thorough evaluation schema. In addition, we design and release an instruction dataset and a toolset – the two sides that the agents bridge between – following the principle of reflecting real-world challenges. Furthermore, we evaluate multiple generic agents. Our findings can inspire future research in improving LLM-based agents and rethink the philosophy of API design.

The development of state-of-the-art generative large language models (LLMs) disproportionately relies on English-centric tokenizers, vocabulary and pre-training data. Despite the fact that some LLMs have multilingual capabilities, recent studies have shown that their inference efficiency deteriorates when generating text in languages other than English. This results in increased inference time and costs. Cross-lingual vocabulary adaptation (CVA) methods have been proposed for adapting models to a target language aiming to improve downstream performance. However, the effectiveness of these methods on increasing inference efficiency of generative LLMs has yet to be explored. In this paper, we perform an empirical study of five CVA methods on four generative LLMs (including monolingual and multilingual models) across four typologically-diverse languages and four natural language understanding tasks. We find that CVA substantially contributes to LLM inference speedups of up to 271.5%. We also show that adapting LLMs that have been pre-trained on more balanced multilingual data results in downstream performance comparable to the original models.

Reasoning is one crucial capability in Large Language Models (LLMs), allowing them to perform complex tasks such as solving math problems and multi-step planning. While reasoning capability can emerge in larger models, smaller ones usually have to rely on distillation to transfer this capability from a larger model. However, recent efforts to distill reasoning capabilities have focused mainly on English, leaving multilingual distillation underexplored. To address this gap, this paper examines existing English reasoning distillation methods that utilize a variety of positive rationales in multilingual settings and proposes d-CoT-nR, a novel approach that incorporates incorrect rationales as additional guidance. Empirical results from multilingual high-school examinations show that d-CoT-nR significantly surpasses the baseline, improving accuracy in unseen languages and correctness in step-by-step reasoning.

Large Multimodal Models (LMMs) have achieved strong performance across a range of vision and language tasks. However, their spatial reasoning capabilities are under-investigated. In this paper, we construct a novel VQA dataset, Spatial-MM, to comprehensively study LMMs’ spatial understanding and reasoning capabilities. Our analyses on object-relationship and multi-hop reasoning reveal several important findings. Firstly, bounding boxes and scene graphs, even synthetic ones, can significantly enhance LMMs’ spatial reasoning. Secondly, LMMs struggle more with questions posed from the human perspective than the camera perspective about the image. Thirdly, chain of thought (CoT) prompting does not improve model performance on complex multi-hop questions involving spatial relations. Moreover, spatial reasoning steps are much less accurate than non-spatial ones across MLLMs. Lastly, our perturbation analysis on GQA-spatial reveals that LMMs are much stronger at basic object detection than complex spatial reasoning. We believe our new benchmark dataset and in-depth analyses can spark further research on LMMs spatial reasoning.

Modern large language models (LLMs) have exhibited cooperative synergy on complex task-solving, and collective decision-making (CDM) is a pivotal component in LLM-based multi-agent collaboration frameworks. Our survey on 52 recent such systems uncovers a severe lack of diversity, with a heavy reliance on dictatorial and plurality voting for CDM. Through the lens of social choice theory, we scrutinize widely-adopted CDM methods and identify their limitations. To enrich current landscape of LLM-based CDM, we present GEDI, an electoral CDM module that incorporates various ordinal preferential voting mechanisms. Our empirical case study across three benchmarks shows that the integration of certain CDM methods can markedly improve the reasoning capabilities and robustness of some leading LLMs, all without requiring intricate system designs. Additionally, we find that some CDM mechanisms generate positive synergies even with as few as three agents. The voting-based methods also demonstrate robustness against single points of failure, as well as diversity in terms of hit-rate@k and subject-wise impacts.