EMNLP 2024 Papers

Rotary Position Embedding (RoPE) is an efficient position encoding approach and is widely utilized in numerous large language models (LLMs). Recently, a lot of methods have been put forward to further expand the context window based on RoPE. The core concept of those methods is to predefine or search for a set of factors to rescale the base frequencies of RoPE. Nevertheless, it is quite a challenge for existing methods to predefine an optimal factor due to the exponential search space. In view of this, we introduce PSC (Phase Shift Calibration), a small module for calibrating the frequencies predefined by existing methods. With the employment of PSC, we demonstrate that many existing methods can be further enhanced, like PI, YaRN, and LongRoPE. We conducted extensive experiments across multiple models and tasks. The results demonstrate that (1) when PSC is enabled, the comparative reductions in perplexity increase as the context window size is varied from 16k, to 32k, and up to 64k. (2) Our approach is broadly applicable and exhibits robustness across a variety of models and tasks.

Prompt optimization aims to find the best prompt to a large language model (LLM) for a given task. LLMs have been successfully used to help find and improve prompt candidates for single-step tasks. However, realistic tasks for agents are multi-step and introduce new challenges: (1) Prompt content is likely to be more extensive and complex, making it more difficult for LLMs to analyze errors, (2) the impact of an individual step is difficult to evaluate, and (3) different people may have varied preferences about task execution. While humans struggle to optimize prompts, they are good at providing feedback about LLM outputs; we therefore introduce a new LLM-driven discrete prompt optimization framework PROMST that incorporates human-designed feedback rules to automatically offer direct suggestions for improvement. We also use an extra learned heuristic model that predicts prompt performance to efficiently sample from prompt candidates. This approach significantly outperforms both human-engineered prompts and several other prompt optimization methods across 11 representative multi-step tasks (an average 10.6%-29.3% improvement to current best methods on five LLMs respectively). We believe our work can serve as a benchmark for automatic prompt optimization for LLM-driven multi-step tasks.

Spoken Language Understanding (SLU) is a critical component of voice assistants; it consists of converting speech to semantic parses for task execution. Previous works have explored end-to-end models to improve the quality and robustness of SLU models with Deliberation, however these models have remained autoregressive, resulting in higher latencies. In this work we introduce PRoDeliberation, a novel method leveraging a Connectionist Temporal Classification-based decoding strategy as well as a denoising objective to train robust non-autoregressive deliberation models. We show that PRoDeliberation achieves the latency reduction of parallel decoding (2-10x improvement over autoregressive models) while retaining the ability to correct Automatic Speech Recognition (ASR) mistranscriptions of autoregressive deliberation systems. We further show that the design of the denoising training allows PRoDeliberation to overcome the limitations of small ASR devices, and we provide analysis on the necessity of each component of the system.

While image-text pre-trained models, such as CLIP, have demonstrated impressive capabilities in learning robust text and image representations, a critical area for substantial improvement remains—precise color understanding. In this paper, we address this limitation by introducing PRISM, a simple yet highly effective method that extends CLIP’s capability to grasp the nuances of precise colors. PRISM seamlessly adapts to both recognized HTML colors and out-of-vocabulary RGB inputs through the utilization of our curated dataset of 100 image-text pairs, which can be effortlessly repurposed for fine-tuning with any desired color. Importantly, PRISM achieves these enhancements without compromising CLIP’s performance on established benchmarks. Furthermore, we introduce a novel evaluation framework, ColorLens, featuring both seen and unseen test sets that can be readily repurposed to assess a model’s precision in understanding precise colors. Our comprehensive evaluation and results demonstrate significant improvements over baseline models.

Multimodal Large Language Models (MLLMs) have seen growing adoption across various scientific disciplines. These advancements encourage the investigation of molecule-text modeling within synthetic chemistry, a field dedicated to designing and conducting chemical reactions to synthesize new compounds with desired properties and applications. Current approaches, however, often neglect the critical role of multi-molecule graph interaction in understanding chemical reactions, leading to suboptimal performance in synthetic chemistry tasks. This study introduces PRESTO (Progressive Pretraining Enhances Synthetic Chemistry Outcomes), a new framework that bridges the molecule-text modality gap by integrating a comprehensive benchmark of pretraining strategies and dataset configurations. It progressively improves multimodal LLMs through cross-modal alignment and multi-graph understanding. Our extensive experiments demonstrate that PRESTO offers competitive results in downstream synthetic chemistry tasks. The code can be found at https://github.com/IDEA-XL/PRESTO.

While a few public benchmarks have been proposed for training hate speech detection models, the differences in labeling criteria between these benchmarks pose challenges for generalized learning, limiting the applicability of the models. Previous research has presented methods to generalize models through data integration or augmentation, but overcoming the differences in labeling criteria between datasets remains a limitation. To address these challenges, we propose PREDICT, a novel framework that uses the notion of multi-agent for hate speech detection. PREDICT consists of two phases: (1) PRE (Perspective-based REasoning): Multiple agents are created based on the induced labeling criteria of given datasets, and each agent generates stances and reasons; (2) DICT (Debate using InCongruenT references): Agents representing hate and non-hate stances conduct the debate, and a judge agent classifies hate or non-hate and provides a balanced reason. Experiments on five representative public benchmarks show that PREDICT achieves superior cross-evaluation performance compared to methods that focus on specific labeling criteria or majority voting methods. Furthermore, we validate that PREDICT effectively mediates differences between agents’ opinions and appropriately incorporates minority opinions to reach a consensus. Our code is available at https://github.com/Hanyang-HCC-Lab/PREDICT

The growing dependence on Large Language Models (LLMs) for finishing user instructions necessitates a comprehensive understanding of their robustness to complex task completion in real-world situations. To address this critical need, we propose the PowerPoint Task Completion-Robustness (PPTC-R) benchmark to measure LLMs’ robustness to the user PPT task instruction and software version (Powerpoint). Specifically, we construct adversarial user instructions by attacking user instructions at sentence, semantic, and multi-language levels. To assess the robustness of Language Models to software versions, we vary the number of provided APIs to simulate both the newest version and earlier version settings. Subsequently, we test 3 closed-source and 4 open-source LLMs using a benchmark that incorporates these robustness settings, aiming to evaluate how deviations impact LLMs’ API calls for task completion. We find that GPT-4 exhibits the highest performance and strong robustness in our benchmark, particularly in the version update and the multilingual settings. However, we find that all LLMs lose their robustness when confronted with multiple challenges (e.g., multi-turn) simultaneously, leading to significant performance drops. We further analyze the robustness behavior and error reasons of LLMs in our benchmark, which provide valuable insights for researchers to understand the LLM’s robustness in task completion and develop more robust LLMs and agents.

Despite their remarkable capabilities, Large Language Models (LLMs) are found to be surprisingly sensitive to minor variations in prompts, often generating significantly divergent outputs in response to minor variations in the prompts, such as spelling errors, alteration of wording or the prompt template. However, while assessing the quality of an LLM, the focus often tends to be solely on its performance on downstream tasks, while very little to no attention is paid to prompt sensitivity. To fill this gap, we propose POSIX – a novel PrOmpt Sensitivity IndeX as a reliable measure of prompt sensitivity, thereby offering a more comprehensive evaluation of LLM performance. The key idea behind POSIX is to capture the relative change in loglikelihood of a given response upon replacing the corresponding prompt with a different intent-preserving prompt. We provide thorough empirical evidence demonstrating the efficacy of POSIX in capturing prompt sensitivity and subsequently use it to measure and thereby compare prompt sensitivity of various open source LLMs. We find that merely increasing the parameter count or instruction tuning does not necessarily reduce prompt sensitivity whereas adding some few-shot exemplars, even just one, almost always leads to significant decrease in prompt sensitivity. We also find that alterations to prompt template lead to the highest sensitivity in the case of MCQ type tasks, whereas paraphrasing results in the highest sensitivity in open-ended generation tasks. The code for reproducing our results is open-sourced at https://github.com/kowndinya-renduchintala/POSIX.

In textual backdoor attacks, attackers insert poisoned samples with triggered inputs and target labels into training datasets to manipulate model behavior, threatening the model’s security and reliability. Current defense methods can generally be categorized into inference-time and training-time ones. The former often requires a part of clean samples to set detection thresholds, which may be hard to obtain in practical application scenarios, while the latter usually requires an additional retraining or unlearning process to get a clean model, significantly increasing training costs. To avoid these drawbacks, we focus on developing a practical defense method before model training without using any clean samples. Our analysis reveals that with the help of a pre-trained language model (PLM), poisoned samples, different from clean ones, exhibit mismatched relationship and shared characteristics. Based on these observations, we further propose a two-stage poison detection strategy solely leveraging insights from PLM before model training. Extensive experiments confirm our approach’s effectiveness, achieving better performance than current leading methods more swiftly. Our code is available at https://github.com/Ascian/PKAD.

AI systems’ ability to interpret human emotions and adapt to variations is becoming more crucial as AI gets embedded into everyone’s daily lives. Emotion Recognition in Conversations (ERC) is based on this fundamental challenge. Current state-of-the-art technologies in ERC are limited due to the need for future information. We introduce High-Dimensional Temporal Fusion Transformer (HiTFT), a time-series forecasting transformer that predicts pseudo-future information to overcome this constraint. This retains the models’ dynamic nature and provides future information more efficiently than other methods. Our proposed method combines pseudo future embeddings with an encoder that models the speaker’s emotional state using past and pseudo-future information as well as inter and intra speaker interactions; these speaker states are then passed through a decoder block that predicts the inferred emotion of that utterance. We further evaluate our method and show that it achieves state of the art performance on three ERC datasets - MELD, EmoryNLP, and IEMOCap.

Question answering (QA) can only make progress if we know if an answer is correct, but current answer correctness (AC) metrics struggle with verbose, free-form answers from large language models (LLMs). There are two challenges with current short-form QA evaluations: a lack of diverse styles of evaluation data and an over-reliance on expensive and slow LLMs. LLM-based scorers correlate better with humans, but this expensive task has only been tested on limited QA datasets. We rectify these issues by providing rubrics and datasets for evaluating machine QA adopted from the Trivia community. We also propose an efficient, and interpretable QA evaluation that is more stable than an exact match and neural methods (BERTScore).

Identifying preferences of customers in their shopping journey is a pivotal aspect in providing product recommendations. The task becomes increasingly challenging when there is a multi-turn conversation between the user and a shopping assistant chatbot. In this paper, we tackle a novel and complex problem of identifying customer preferences in the form of key-value filters on an e-commerce website in a multi-turn conversational setting. Existing systems specialize in extracting customer preferences from standalone customer queries which makes them unsuitable to multi-turn setup. We propose PEARL (Preference Extraction with ICL Augmentation and Retrieval with LLM Agents) that leverages collaborative LLM agents, generates in-context learning exemplars and dynamically retrieves relevant exemplars during inference time to extract customer preferences as a combination of key-value filters. Our experiments on proprietary and public datasets show that PEARL not only improves performance on exact match by ~10% compared to competitive LLM-based baselines but additionally improves inference latency by ~110%.

The black-box nature of deep learning models in NLP hinders their widespread application. The research focus has shifted to Hierarchical Attribution (HA) for its ability to model feature interactions. Recent works model non-contiguous combinations with a time-costly greedy search in Eculidean spaces, neglecting underlying linguistic information in feature representations. In this work, we introduce a novel method, namely Poincare Explanation (PE), for modeling feature interactions with hyperbolic spaces in a time efficient manner.Specifically, we take building text hierarchies as finding spanning trees in hyperbolic spaces. First we project the embeddings into hyperbolic spaces to elicit inherit semantic and syntax hierarchical structures. Then we propose a simple yet effective strategy to calculate Shapley score. Finally we build the the hierarchy with proving the constructing process in the projected space could be viewed as building a minimum spanning tree and introduce a time efficient building algorithm. Experimental results demonstrate the effectiveness of our approach. Our code is available at https://anonymous.4open.science/r/PE-747B.

Large Language Models (LLMs) have issues with document question answering (QA) in situations where the document is unable to fit in the small context length of an LLM. To overcome this issue, most existing works focus on retrieving the relevant context from the document, representing them as plain text. However, documents such as PDFs, web pages, and presentations are naturally structured with different pages, tables, sections, and so on. Representing such structured documents as plain text is incongruous with the user’s mental model of these documents with rich structure. When a system has to query the document for context, this incongruity is brought to the fore, and seemingly trivial questions can trip up the QA system. To bridge this fundamental gap in handling structured documents, we propose an approach called PDFTriage that enables models to retrieve the context based on either structure or content. Our experiments demonstrate the effectiveness of the proposed PDFTriage-augmented models across several classes of questions where existing retrieval-augmented LLMs fail. To facilitate further research on this fundamental problem, we release our benchmark dataset consisting of 900+ human-generated questions over 80 structured documents from 10 different categories of question types for document QA. Our code and datasets will be released soon on Github.

In many PDF documents, the reading order of text blocks is missing, which can hinder machine understanding of the document’s content.Existing works try to extract one universal reading order for a PDF file.However, applications, like Retrieval Augmented Generation (RAG), require breaking long articles into sections and subsections for better indexing.For this reason, this paper introduces a new task and dataset, PDF-to-Tree, which organizes the text blocks of a PDF into a tree structure.Since a PDF may contain thousands of text blocks, far exceeding the number of words in a sentence, this paper proposes a transition-based parser that uses a greedy strategy to build the tree structure.Compared to parser for plain text, we also use multi-modal features to encode the parser state.Experiments show that our approach achieves an accuracy of 93.93%, surpassing the performance of baseline methods by an improvement of 6.72%.

Conversational Question Generation (CQG) enhances the interactivity of conversational question-answering systems in fields such as education, customer service, and entertainment. However, traditional CQG, focusing primarily on the immediate context, lacks the conversational foresight necessary to guide conversations toward specified conclusions. This limitation significantly restricts their ability to achieve conclusion-oriented conversational outcomes. In this work, we redefine the CQG task as Conclusion-driven Conversational Question Generation (CCQG) by focusing on proactivity, not merely reacting to the unfolding conversation but actively steering it towards a conclusion-oriented question-answer pair. To address this, we propose a novel approach, called Proactive Conversational Question Planning with self-Refining (PCQPR). Concretely, by integrating a planning algorithm inspired by Monte Carlo Tree Search (MCTS) with the analytical capabilities of large language models (LLMs), PCQPR predicts future conversation turns and continuously refines its questioning strategies. This iterative self-refining mechanism ensures the generation of contextually relevant questions strategically devised to reach a specified outcome. Our extensive evaluations demonstrate that PCQPR significantly surpasses existing CQG methods, marking a paradigm shift towards conclusion-oriented conversational question-answering systems.

Mental illness remains one of the most critical public health issues. Despite its importance, many mental health professionals highlight a disconnect between their training and actual real-world patient practice. To help bridge this gap, we propose PATIENT-𝜓, a novel patient simulation framework for cognitive behavior therapy (CBT) training. To build PATIENT-𝜓, we construct diverse patient cognitive models based on CBT principles and use large language models (LLMs) programmed with these cognitive models to act as a simulated therapy patient. We propose an interactive training scheme, PATIENT-𝜓-TRAINER, for mental health trainees to practice a key skill in CBT – formulating the cognitive model of the patient – through role-playing a therapy session with PATIENT-𝜓. To evaluate PATIENT-𝜓, we conducted a comprehensive user study of 13 mental health trainees and 20 experts. The results demonstrate that practice using PATIENT-𝜓-TRAINER enhances the perceived skill acquisition and confidence of the trainees beyond existing forms of training such as textbooks, videos, and role-play with non-patients. Based on the experts’ perceptions, PATIENT-𝜓 is perceived to be closer to real patient interactions than GPT-4, and PATIENT-𝜓-TRAINER holds strong promise to improve trainee competencies. Our code and data are released at https://github.com/ruiyiw/patient-psi.

Evaluation of multilingual Large Language Models (LLMs) is challenging due to a variety of factors – the lack of benchmarks with sufficient linguistic diversity, contamination of popular benchmarks into LLM pre-training data and the lack of local, cultural nuances in translated benchmarks. In this work, we study human and LLM-based evaluation in a multilingual, multi-cultural setting. We evaluate 30 models across 10 Indic languages by conducting 90K human evaluations and 30K LLM-based evaluations and find that models such as GPT-4o and Llama-3 70B consistently perform best for most Indic languages. We build leaderboards for two evaluation settings - pairwise comparison and direct assessment and analyse the agreement between humans and LLMs. We find that humans and LLMs agree fairly well in the pairwise setting but the agreement drops for direct assessment evaluation especially for languages such as Bengali and Odia. We also check for various biases in human and LLM-based evaluation and find evidence of self-bias in the GPT-based evaluator. Our work presents a significant step towards scaling up multilingual evaluation of LLMs.

In the realm of parameter-efficient fine-tuning (PEFT) methods, while options like LoRA are available, there is a persistent demand in the industry for a PEFT approach that excels in both efficiency and performance within the context of single-backbone multi-tenant applications. This paper introduces a new and straightforward PEFT technique, termed Prompt Aware Representation Adjustment (PARA). The core of our proposal is to integrate a lightweight vector generator within each Transformer layer. This generator produces vectors that are responsive to input prompts, thereby adjusting the hidden representations accordingly. Our extensive experimentation across diverse tasks has yielded promising results. Firstly, the PARA method has been shown to surpass current PEFT benchmarks in terms of performance, despite having a similar number of adjustable parameters. Secondly, it has proven to be more efficient than LoRA in the single-backbone multi-tenant scenario, highlighting its significant potential for industrial adoption.