EMNLP 2024 Papers

The primary aim of Knowledge Graph Embeddings (KGE) is to learn low-dimensional representations of entities and relations for predicting missing facts. While rotation-based methods like RotatE and QuatE perform well in KGE, they face two challenges: limited model flexibility requiring proportional increases in relation size with entity dimension, and difficulties in generalizing the model for higher-dimensional rotations. To address these issues, we introduce OrthogonalE, a novel KGE model employing matrices for entities and block-diagonal orthogonal matrices with Riemannian optimization for relations. This approach not only enhances the generality and flexibility of KGE models but also captures several relation patterns that rotation-based methods can identify. Experimental results indicate that our new KGE model, OrthogonalE, offers generality and flexibility, captures several relation patterns, and significantly outperforms state-of-the-art KGE models while substantially reducing the number of relation parameters.

Retrieval-augmented Large Language Models (LLMs) offer substantial benefits in enhancing performance across knowledge-intensive scenarios. However, these methods often struggle with complex inputs and encounter difficulties due to noisy knowledge retrieval, notably hindering model effectiveness. To address this issue, we introduce BlendFilter, a novel approach that elevates retrieval-augmented LLMs by integrating query generation blending with knowledge filtering. BlendFilter proposes the blending process through its query generation method, which integrates both external and internal knowledge augmentation with the original query, ensuring comprehensive information gathering. Additionally, our distinctive knowledge filtering module capitalizes on the intrinsic capabilities of the LLM, effectively eliminating extraneous data. We conduct extensive experiments on three open-domain question answering benchmarks, and the findings clearly indicate that our innovative BlendFilter surpasses state-of-the-art baselines significantly.

Efficient state space models (SSMs), including linear recurrent neural networks and linear attention variants, have emerged as potential alternative language models to Transformers. While efficient, SSMs struggle with tasks requiring in-context retrieval, such as text copying and associative recall, limiting their usefulness in practical settings. Prior work on how to meet this challenge has focused on the internal model architecture and not investigated the role of the training procedure. This paper proposes a new training procedure that improve the performance of SSMs on retrieval-intensive tasks. This novel pre-training procedure combines a bidirectional processing of the input with dynamic mixtures of pre-training objectives to improve the utilization of the SSM’s fixed-size state. Our experimental evaluations show that this procedure significantly improves performance on retrieval-intensive tasks that challenge current SSMs, such as phone book lookup, long paragraph question-answering, and infilling tasks. Our findings offer insights into a new direction to advance the training of SSMs to close the performance gap with Transformers.

The ever-growing biomedical publications magnify the challenge of extracting structured data from unstructured texts. This task involves two components: biomedical entity identification (Named Entity Recognition, NER) and their interrelation determination (Relation Extraction, RE). However, existing methods often neglect unique features of the biomedical literature, such as ambiguous entities, nested proper nouns, and overlapping relation triplets, and underutilize prior knowledge, leading to an intolerable performance decline in the biomedical domain, especially with limited annotated training data. In this paper, we propose the Biomedical Relation-First eXtraction (Bio-RFX) model by leveraging sentence-level relation classification before entity extraction to tackle entity ambiguity. Moreover, we exploit structural constraints between entities and relations to guide the model’s hypothesis space, enhancing extraction performance across different training scenarios. Comprehensive experimental results on biomedical datasets show that Bio-RFX achieves significant improvements on both NER and RE tasks. Even under the low-resource training scenarios, it outperforms all baselines in NER and has highly competitive performance compared to the state-of-the-art fine-tuned baselines in RE.

Toxic content detection plays a vital role in addressing the misuse of social media platforms to harm people or groups due to their race, gender or ethnicity. However, due to the nature of the datasets, systems develop an unintended bias due to the over-generalization of the model to the training data. This compromises the fairness of the systems, which can impact certain groups due to their race, gender, etc.Existing methods mitigate bias using data augmentation, adversarial learning, etc., which require re-training and adding extra parameters to the model.In this work, we present a robust and generalizable technique BiasWipe to mitigate unintended bias in language models. BiasWipe utilizes model interpretability using Shapley values, which achieve fairness by pruning the neuron weights responsible for unintended bias. It first identifies the neuron weights responsible for unintended bias and then achieves fairness by pruning them without loss of original performance. It does not require re-training or adding extra parameters to the model. To show the effectiveness of our proposed technique for bias unlearning, we perform extensive experiments for Toxic content detection for BERT, RoBERTa, and GPT models. .

Existing works examining Vision-Language Models (VLMs) for social biases predominantly focus on a limited set of documented bias associations, such as gender-profession or race-crime. This narrow scope often overlooks a vast range of unexamined implicit associations, restricting the identification and, hence, mitigation of such biases. We address this gap by probing VLMs to (1) uncover hidden, implicit associations across 9 bias dimensions. We systematically explore diverse input and output modalities and (2) demonstrate how biased associations vary in their negativity, toxicity, and extremity. Our work (3) identifies subtle and extreme biases that are typically not recognized by existing methodologies. We make the **D**ataset **o**f **r**etrieved **a**ssociations (**Dora**) publicly available.

Evaluating the bias of LLMs becomes more crucial with their rapid development. However, existing evaluation approaches rely on fixed-form outputs and cannot adapt to the flexible open-text generation scenarios of LLMs (e.g., sentence completion and question answering). To address this, we introduce BiasAlert, a plug-and-play tool designed to detect social bias in open-text generations of LLMs. BiasAlert integrates external human knowledge with its inherent reasoning capabilities to detect bias reliably. Extensive experiments demonstrate that BiasAlert significantly outperforms existing state-of-the-art methods like GPT-4-as-Judge in detecting bias. Furthermore, through application studies, we showcase the utility of BiasAlert in reliable LLM fairness evaluation and bias mitigation across various scenarios. Model and code will be publicly released.

In this paper, we introduce BiMediX, the first bilingual medical mixture of experts LLM designed for seamless interaction in both English and Arabic. Our model facilitates a wide range of medical interactions in English and Arabic, including multi-turn chats to inquire about additional details such as patient symptoms and medical history, multiple-choice question answering, and open-ended question answering. We propose a semi-automated English-to-Arabic translation pipeline with human refinement to ensure high-quality translations. We also introduce a comprehensive evaluation benchmark for Arabic medical LLMs. Furthermore, we introduce BiMed1.3M, an extensive Arabic-English bilingual instruction set that covers 1.3 Million diverse medical interactions, including 200k synthesized multi-turn doctor-patient chats, in a 1:2 Arabic-to-English ratio. Our model outperforms state-of-the-art Med42 and Meditron by average absolute gains of 2.5% and 4.1%, respectively, computed across multiple medical evaluation benchmarks in English, while operating at 8-times faster inference. Moreover, our BiMediX outperforms the generic Arabic-English bilingual LLM, Jais-30B, by average absolute gains of 10% on our Arabic and 15% on our bilingual evaluations across multiple datasets. Additionally, BiMediX exceeds the accuracy of GPT4 by 4.4% in open-ended question UPHILL evaluation and largely outperforms state-of-the-art open source medical LLMs in human evaluations of multi-turn conversations. Our trained models, instruction set, and source code are available at https://github.com/mbzuai-oryx/BiMediX.

Adapting pretrained models to downstream tasks is important in practical applications. Existing frameworks adapt from an initial pretrained model to each downstream task directly, but ignore the sequential nature of the downstream tasks and their feedback effect on the pretrained model. In this work, we propose a new framework, called BiKT, to enable bidirectional knowledge transfer between pretrained models and downstream tasks in rounds. We model each downstream task in the current round as a target task for adaptation and treat all the tasks in the previous rounds as source tasks for feedback. We design a feedback algorithm by multi-task learning over the labeled data of the source tasks, where task-specific prompts are plugged into the backbone network for decoupling task-exclusive knowledge from task-shared knowledge. We further utilize the good initiation of the new backbone network updated in the feedback phase and the trained prompts of the source tasks for adaptation. Evaluation over 9 GLUE datasets, 6 SuperGLUE datasets, and 8 other datasets using models with different pretraining levels and different parameter scales shows remarkable improvement in full-shot and few-shot adaptation settings.

Chinese Spelling Check (CSC) aims to detect and correct potentially misspelled characters in Chinese sentences. Naturally, it involves the detection and correction subtasks, which interact with each other dynamically. Such interactions are bi-directional, i.e., the detection result would help reduce the risk of over-correction and under-correction while the knowledge learnt from correction would help prevent false detection. Current CSC approaches are of two types: correction-only or single-directional detection-to-correction interactive frameworks. Nonetheless, they overlook the bi-directional interactions between detection and correction. This paper aims to fill the gap by proposing a Bi-directional Detector-Corrector framework for CSC (Bi-DCSpell). Notably, Bi-DCSpell contains separate detection and correction encoders, followed by a novel interactive learning module facilitating bi-directional feature interactions between detection and correction to improve each other’s representation learning. Extensive experimental results demonstrate a robust correction performance of Bi-DCSpell on widely used benchmarking datasets while possessing a satisfactory detection ability.

As large language models (LLMs) increasingly permeate daily lives, there is a growing demand for real-time interactions that mirror human conversations. Traditional turn-based chat systems driven by LLMs prevent users from verbally interacting with the system while generating responses.To overcome these limitations, we adapt existing LLMs to duplex models so that they can listen to users while generating output and dynamically adjust themselves to provide instant feedback.Specifically, we divide the queries and responses of conversations into several time slices and then adopt a time-division-multiplexing (TDM) encoding-decoding strategy to process these slices pseudo-simultaneously.Furthermore, to make LLMs proficient enough to handle real-time conversations, we build a fine-tuning dataset consisting of alternating time slices of queries and responses and covering typical feedback types in instantaneous interactions.Our experiments show that although the queries and responses of conversations are segmented into incomplete slices for processing, LLMs can preserve their original performance on standard benchmarks with a few fine-tuning steps on our dataset. Automatic and human evaluation indicate that duplex models make user-AI interactions more natural and human-like, and greatly improve user satisfaction compared to vanilla LLMs. Our duplex model and dataset will be released soon.

Despite broad interest in modeling spoken dialogue agents, most approaches are inherently “half-duplex” – restricted to turn-based interaction with responses requiring explicit prompting by the user or implicit tracking of interruption or silence events. Human dialogue, by contrast, is “full-duplex” allowing for rich synchronicity in the form of quick and dynamic turn-taking, overlapping speech, and backchanneling. Technically, the challenge of achieving full-duplex dialogue with LLMs lies in modeling synchrony as pre-trained LLMs do not have a sense of “time”. To bridge this gap, we propose Synchronous LLMs for full-duplex spoken dialogue modeling. We design a novel mechanism to integrate time information into Llama3-8b so that they run synchronously with the real-world clock. We also introduce a training recipe that uses 212k hours of synthetic spoken dialogue data generated from text dialogue data to create a model that generates meaningful and natural spoken dialogue, with just 2k hours of real-world spoken dialogue data. Synchronous LLMs outperform state-of-the-art in dialogue meaningfulness while maintaining naturalness. Finally, we demonstrate the model’s ability to participate in full-duplex dialogue by simulating interaction between two agents trained on different datasets, while considering Internet-scale latencies of up to 240 ms.

Various audio-LLMs (ALLMs) have been explored recently for tackling different audio tasks simultaneously using a single, unified model. While existing evaluations of ALLMs primarily focus on single-audio tasks, real-world applications often involve processing multiple audio streams simultaneously. To bridge this gap, we propose the first multi-audio evaluation (MAE) benchmark that consists of 20 datasets from 11 multi-audio tasks encompassing both speech and sound scenarios. Comprehensive experiments on MAE demonstrate that the existing ALLMs, while being powerful in comprehending primary audio elements in individual audio inputs, struggling to handle multi-audio scenarios. To this end, we propose a novel multi-audio-LLM (MALLM) to capture audio context among multiple similar audios using discriminative learning on our proposed synthetic data. The results demonstrate that the proposed MALLM outperforms all baselines and achieves high data efficiency using synthetic data without requiring human annotations. The proposed MALLM opens the door for ALLMs towards multi-audio processing era and brings us closer to replicating human auditory capabilities in machines.

In Machine Translation (MT) evaluations, the conventional approach is to compare a translated sentence against its human-created reference sentence. MT metrics provide an absolute score (e.g., from 0 to 1) to a candidate sentence based on the similarity with the reference sentence. Thus, existing MT metrics give the maximum score to the reference sentence. However, this approach overlooks the potential for a candidate sentence to exceed the reference sentence in terms of quality. In particular, recent advancements in Large Language Models (LLMs) have highlighted this issue, as LLM-generated sentences often exceed the quality of human-written sentences. To address the problem, we introduce the Residual score Metric (ResuMe), which evaluates the relative quality between reference and candidate sentences. ResuMe assigns a positive score to candidate sentences that outperform their reference sentences, and a negative score when they fall short. By adding the residual scores from ResuMe to the absolute scores from MT metrics, it can be possible to allocate higher scores to candidate sentences than what reference sentences are received from MT metrics. Experimental results demonstrate that ResuMe enhances the alignments between MT metrics and human judgments both at the segment-level and the system-level.

With the aid of large language models, current conversational recommender system (CRS) has gaining strong abilities to persuade users to accept recommended items. While these CRSs are highly persuasive, they can mislead users by incorporating incredible information in their explanations, ultimately damaging the long-term trust between users and the CRS. To address this, we propose a simple yet effective method, called PC-CRS, to enhance the credibility of CRS’s explanations during persuasion. It guides the explanation generation through our proposed credibility-aware persuasive strategies and then gradually refines explanations via post-hoc self-reflection. Experimental results demonstrate the efficacy of PC-CRS in promoting persuasive and credible explanations. Further analysis reveals the reason behind current methods producing incredible explanations and the potential of credible explanations to improve recommendation accuracy.

Increasingly, model compression techniques enable large language models (LLMs) to be deployed in real-world applications. As a result of this momentum towards local deployment, compressed LLMs will interact with a large population. Prior work on compression typically prioritize preserving perplexity, which is directly analogous to training loss. The impact of compression method on other critical aspects of model behavior—particularly safety—requires systematic assessment. To this end, we investigate the impact of model compression along four dimensions: (1) degeneration harm, i.e., bias and toxicity in generation; (2) representational harm, i.e., biases in discriminative tasks; (3) dialect bias; and (4) language modeling and downstream task performance. We examine a wide spectrum of LLM compression techniques, including unstructured pruning, semi-structured pruning, and quantization. Our analysis reveals that compression can lead to unexpected consequences. Although compression may unintentionally alleviate LLMs’ degeneration harm, it can still exacerbate representational harm. Furthermore, increasing compression produces a divergent impact on different protected groups. Finally, different compression methods have drastically different safety impacts: for example, quantization mostly preserves bias while pruning degrades quickly. Our findings underscore the importance of integrating safety assessments into the development of compressed LLMs to ensure their reliability across real-world applications.

Natural language (NL) has long been the predominant format for human cognition and communication, and by extension, has been similarly pivotal in the development and application of Large Language Models (LLMs). Yet, besides NL, LLMs have seen various non-NL formats during pre-training, such as code and logical expression. NL’s status as the optimal format for LLMs, particularly in single-LLM reasoning and multi-agent communication, has not been thoroughly examined. In this work, we challenge the default use of NL by exploring the utility of non-NL formats in these contexts. We show that allowing LLMs to autonomously select the most suitable format before reasoning or communicating leads to a 3.3 to 5.7% improvement in reasoning efficiency for different LLMs, and up to a 72.7% reduction in token usage in multi-agent communication, all while maintaining communicative effectiveness. Our comprehensive analysis further reveals that LLMs can devise a format from limited task instructions and that the devised format is effectively transferable across different LLMs. Intriguingly, the structured communication format decided by LLMs exhibits notable parallels with established agent communication languages, suggesting a natural evolution towards efficient, structured communication in agent communication. Our code will be released to facilitate further exploration.

Large Language Models (LLMs) demonstrate ever-increasing abilities in mathematical and algorithmic tasks, yet their geometric reasoning skills are underexplored. We investigate LLMs’ abilities in constructive geometric problem-solving, – one of the most fundamental steps in developing human mathematical reasoning, revealing notable challenges in this domain. LLMs exhibit biases in variable names, struggle with 2D spatial relationships and planning, and hallucinate object placements. To this end, we introduce a framework that enhances LLMs’ reasoning potential through a multi-agent system conducting internal dialogue. This work underscores LLMs’ limitations in geometric reasoning and improves their capabilities through self-correction, collaboration, and diverse role specializations.

Medical coding, the translation of unstructured clinical text into standardized medical codes, is a crucial but time-consuming healthcare practice. Though large language models (LLM) could automate the coding process and improve the efficiency of such tasks, interpretability remains paramount for maintaining patient trust. Current efforts in interpretability of medical coding applications rely heavily on label attention mechanisms, which often leads to the highlighting of extraneous tokens irrelevant to the ICD code. To facilitate accurate interpretability in medical language models, this paper leverages dictionary learning that can efficiently extract sparsely activated representations from dense language model embeddings in superposition. Compared with common label attention mechanisms, our model goes beyond token-level representations by building an interpretable dictionary which enhances the mechanistic-based explanations for each ICD code prediction, even when the highlighted tokens are medically irrelevant. We show that dictionary features are human interpretable, can elucidate the hidden meanings of upwards of 90% of medically irrelevant tokens, and steer model behavior.