论文检索

Cheng-Han Chiang, Hung-yi Lee

Long-form generations from large language models (LLMs) contain a mix of factual and non-factual claims, making evaluating factuality difficult.Prior works evaluate the factuality of a long paragraph by decomposing it into multiple facts, verifying those facts independently, and aggregating the results.Such methods assume that combining factual claims forms a factual paragraph.The above assumption can be violated: we show that strong open-source models like Llama-chat can generate paragraphs that contain verifiable facts, but the facts are combined into a non-factual paragraph due to entity ambiguity.We further reveal that existing factuality metrics, including FActScore and citation recall, cannot properly evaluate these non-factual paragraphs and overestimate their factuality.To address this, we introduce an enhanced metric, **D-FActScore**, specifically designed for content with ambiguous entities.We evaluate the D-FActScores of people biographies generated by retrieval-augmented LLMs.We show that D-FActScore can better assess the factuality of paragraphs with entity ambiguity than FActScore.We also find that four widely used open-source LLMs tend to mix information of distinct entities to form non-factual paragraphs, making their D-FActScore much lower than FActScore by over 10%.

Sergio Servantez, Joe Barrow, Kristian Hammond, Rajiv Jain

Rule-based reasoning, a fundamental type of legal reasoning, enables us to draw conclusions by accurately applying a rule to a set of facts. We explore causal language models as rule-based reasoners, specifically with respect to compositional rules - rules consisting of multiple elements which form a complex logical expression. Reasoning about compositional rules is challenging because it requires multiple reasoning steps, and attending to the logical relationships between elements. We introduce a new prompting method, Chain of Logic, which elicits rule-based reasoning through decomposition (solving elements as independent threads of logic), and recomposition (recombining these sub-answers to resolve the underlying logical expression). This method was inspired by the IRAC (Issue, Rule, Application, Conclusion) framework, a sequential reasoning approach used by lawyers. We evaluate chain of logic across eight rule-based reasoning tasks involving three distinct compositional rules from the LegalBench benchmark and demonstrate it consistently outperforms other prompting methods, including chain of thought and self-ask, using open-source and commercial language models.

Fred Xu, Song Jiang, Zijie Huang, Xiao Luo, Shichang Zhang, Yuanzhou Chen, Yizhou Sun

Taxonomy Expansion, which relies on modeling concepts and concept relations, can be formulated as a set representation learning task. The generalization of set, fuzzy set, incorporates uncertainty and measures the information within a semantic concept, making it suitable for concept modeling. Existing works usually model sets as vectors or geometric objects such as boxes, which are not closed under set operations. In this work, we propose a sound and efficient formulation of set representation learning based on its volume approximation as a fuzzy set. The resulting embedding framework, Fuzzy Set Embedding, satisfies all set operations and compactly approximates the underlying fuzzy set, hence preserving information while being efficient to learn, relying on minimum neural architecture. We empirically demonstrate the power of FUSE on the task of taxonomy expansion, where FUSE achieves remarkable improvements up to 23% compared with existing baselines. Our work marks the first attempt to understand and efficiently compute the embeddings of fuzzy sets.

Junmo Kang, Hongyin Luo, Yada Zhu, Jacob Hansen, James Glass, David Cox, Alan Ritter, Rogerio Feris, Leonid Karlinsky

Recent works have demonstrated the effectiveness of self-alignment in which a large language model is aligned to follow general instructions using instructional data generated from the model itself starting from a handful of human-written seeds. Instead of general alignment, in this work, we focus on self-alignment for expert domain specialization (e.g., biomedicine, finance). As a preliminary, we quantitively show the marginal effect that generic instruction-following training has on downstream expert domains’ performance. To remedy this, we propose self-specialization - allowing for effective model specialization while achieving cross-task generalization by leveraging only a few labeled seeds. Self-specialization offers a data- and parameter-efficient way of “carving out” an expert model out of a generalist pre-trained LLM. Exploring a variety of popular open large models as a base for specialization, our experimental results in both biomedical and financial domains show that our self-specialized models outperform their base models by a large margin, and even larger models that are generally instruction-tuned or that have been adapted to the target domain by other means.

Lingfeng Shen, Weiting Tan, Sihao Chen, Yunmo Chen, Jingyu Zhang, Haoran Xu, Boyuan Zheng, Philipp Koehn, Daniel Khashabi

As the influence of large language models (LLMs) spans across global communities, their safety challenges in multilingual settings become paramount for alignment research. This paper examines the variations in safety challenges faced by LLMs across different languages and discusses approaches to alleviating such concerns. By comparing how state-of-the-art LLMs respond to the same set of malicious prompts written in higher- vs. lower-resource languages,we observe that (1) LLMs tend to generate unsafe responses much more often when a malicious prompt is written in a lower-resource language, and (2) LLMs tend to generate more irrelevant responses to malicious prompts in lower-resource languages. To understand where the discrepancy can be attributed, we study the effect of instruction tuning with reinforcement learning from human feedback (RLHF) or supervised finetuning (SFT) on the HH-RLHF dataset. Surprisingly, while training with high-resource languages improves model alignment, training in lower-resource languages yields minimal improvement. This suggests that the bottleneck of cross-lingual alignment is rooted in the pretraining stage. Our findings highlight the challenges in cross-lingual LLM safety, and we hope they inform future research in this direction.

Herun Wan, Shangbin Feng, Zhaoxuan Tan, Heng Wang, Yulia Tsvetkov, Minnan Luo

Large language models are limited by challenges in factuality and hallucinations to be directly employed off-the-shelf for judging the veracity of news articles, where factual accuracy is paramount. In this work, we propose DELL that identifies three key stages in misinformation detection where LLMs could be incorporated as part of the pipeline: 1) LLMs could generate news reactions to represent diverse perspectives and simulate user-news interaction networks; 2) LLMs could generate explanations for proxy tasks (e.g., sentiment, stance) to enrich the contexts of news articles and produce experts specializing in various aspects of news understanding; 3) LLMs could merge task-specific experts and provide an overall prediction by incorporating the predictions and confidence scores of varying experts. Extensive experiments on seven datasets with three LLMs demonstrate that DELL outperforms state-of-the-art baselines by up to 16.8% in macro f1-score. Further analysis reveals that the generated reactions and explanations are greatly helpful in misinformation detection, while our proposed LLM-guided expert merging helps produce better-calibrated predictions.

Wenxuan Ding, Shangbin Feng, Yuhan Liu, Zhaoxuan Tan, Vidhisha Balachandran, Tianxing He, Yulia Tsvetkov

We propose Knowledge Crosswords, a geometric knowledge reasoning benchmark consisting of incomplete knowledge networks bounded by structured factual constraints, where LLMs are tasked with inferring the missing facts to meet all constraints. The novel setting of geometric knowledge reasoning necessitates new LM abilities beyond existing atomic/linear multi-hop QA, such as backtracking, verifying facts and constraints, reasoning with uncertainty, and more. Knowledge Crosswords contains 2,101 individual problems, covering diverse knowledge domains, and is further divided into three difficulty levels. We conduct extensive experiments to evaluate existing LLMs and approaches on Knowledge Crosswords. Results demonstrate that baseline approaches struggle with larger knowledge networks and semantically-equivalent entity distractors. In light of their limitations, we propose two new approaches, Staged Prompting and Verify-All, to augment LLMs’ abilities for error-aware backtracking and constraint verification. Our Verify-All significantly outperforms prior methods and is more robust towards problems in the hard subset. Further analysis shows that geometric knowledge reasoning poses new challenges to LLMs’ knowledge abilities, particularly in robustness towards varying option orders, complex structural constraints in knowledge networks, “none of the above” scenarios, and more.

Zhong-Zhi Li, Ming-Liang Zhang, Fei Yin, Cheng-Lin Liu

Geometry problem solving (GPS) is a challenging mathematical reasoning task requiring multi-modal understanding, fusion, and reasoning. Existing neural solvers take GPS as a vision-language task but are short in the representation of geometry diagrams that carry rich and complex layout information. In this paper, we propose a layout-aware neural solver named LANS, integrated with two new modules: multimodal layout-aware pre-trained language module (MLA-PLM) and layout-aware fusion attention (LA-FA). MLA-PLM adopts structural-semantic pre-training (SSP) to implement global relationship modeling, and point-match pre-training (PMP) to achieve alignment between visual points and textual points. LA-FA employs a layout-aware attention mask to realize point-guided cross-modal fusion for further boosting layout awareness of LANS. Extensive experiments on datasets Geometry3K and PGPS9K validate the effectiveness of the layout-aware modules and superior problem-solving performance of our LANS solver, over existing symbolic and neural solvers. We have made our code and data publicly available.

Antonios Dimakis, Stella Markantonatou, Antonios Anastasopoulos

Multi-word expressions (MWEs) present unique challenges in natural language processing (NLP), particularly within the context of translation systems, due to their inherent scarcity, non-compositional nature, and other distinct lexical and morphosyntactic characteristics, issues that are exacerbated in low-resource settings.In this study, we elucidate and attempt to address these challenges by leveraging a substantial corpus of human-annotated Greek MWEs. To address the complexity of translating such phrases, we propose a novel method leveraging an available out-of-context lexicon.We assess the translation capabilities of current state-of-the-art systems on this task, employing both automated metrics and human evaluators.We find that by using our method when applicable, the performance of current systems can be significantly improved, however these models are still unable to produce translations comparable to those of a human speaker.

Peiran Yao, Denilson Barbosa

Open-domain question answering (Open-QA) is a common task for evaluating large language models (LLMs). However, current Open-QA evaluations are criticized for the ambiguity in questions and the lack of semantic understanding in evaluators. Complex evaluators, powered by foundation models or LLMs and pertaining to semantic equivalence, still deviate from human judgments by a large margin. We propose to study the entailment relations of answers to identify more informative and more general system answers, offering a much closer evaluation to human judgment on both NaturalQuestions and TriviaQA while being learning-free. The entailment-based evaluation we propose allows the assignment of bonus or partial marks by quantifying the inference gap between answers, enabling a nuanced ranking of answer correctness that has higher AUC than current methods.

Krishnapriya Vishnubhotla, Adam Hammond, Graeme Hirst, Saif Mohammad

Stories are rich in the emotions they exhibit in their narratives and evoke in the readers. The emotional journeys of the various characters within a story are central to their appeal. Computational analysis of the emotions of novels, however, has rarely examined the variation in the emotional trajectories of the different characters within them, instead considering the entire novel to represent a single story arc. In this work, we use character dialogue to distinguish between the emotion arcs of the narration and the various characters. We analyze the emotion arcs of the various characters in a dataset of English literary novels using the framework of Utterance Emotion Dynamics. Our findings show that the narration and the dialogue largely express disparate emotions through the course of a novel, and that the commonalities or differences in the emotional arcs of stories are more accurately captured by those associated with individual characters.

Tuna Alikaşifoğlu, Arda Aras, Aykut Koc

The emergence of transformers has revolutionized natural language processing (NLP), as evidenced in various NLP tasks. While graph neural networks (GNNs) show recent promise in NLP, they are not standalone replacements for transformers. Rather, recent research explores combining transformers and GNNs. Existing GNN-based approaches rely on static graph construction methods requiring excessive text processing, and most of them are not scalable with the increasing document and word counts. We address these limitations by proposing a novel dynamic graph construction method for text documents based on vector visibility graphs (VVGs) generated from transformer output. Then, we introduce visibility pooler (VISPool), a scalable model architecture that seamlessly integrates VVG convolutional networks into transformer pipelines. We evaluate the proposed model on the General Language Understanding Evaluation (GLUE) benchmark datasets. VISPool outperforms the baselines with less trainable parameters, demonstrating the viability of the visibility-based graph construction method for enhancing transformers with GNNs.

Haihui Yang, Xiaojun Quan

Chinese grammatical error correction (CGEC) faces serious overcorrection challenges when employing autoregressive generative models such as sequence-to-sequence (Seq2Seq) models and decoder-only large language models (LLMs). While previous methods aim to address overcorrection in Seq2Seq models, they are difficult to adapt to decoder-only LLMs. In this paper, we propose an alignment-enhanced corrector for the overcorrection problem that applies to both Seq2Seq models and decoder-only LLMs. Our method first trains a correction model to generate an initial correction of the source sentence. Then, we combine the source sentence with the initial correction and feed it through an alignment model for another round of correction, aiming to enforce the alignment model to focus on potential overcorrection. Moreover, to enhance the model’s ability to identify nuances, we further explore the reverse alignment of the source sentence and the initial correction. Finally, we transfer the alignment knowledge from two alignment models to the correction model, instructing it on how to avoid overcorrection. Experimental results on three CGEC datasets demonstrate the effectiveness of our approach in alleviating overcorrection and improving overall performance. Our code has been made publicly available.

Nedjma Ousidhoum, Shamsuddeen Muhammad, Mohamed Abdalla, Idris Abdulmumin, Ibrahim Ahmad, Sanchit Ahuja, Alham Aji, Vladimir Araujo, Abinew Ayele, Pavan Baswani 等

Exploring and quantifying semantic relatedness is central to representing language and holds significant implications across various NLP tasks. While earlier NLP research primarily focused on semantic similarity, often within the English language context, we instead investigate the broader phenomenon of semantic relatedness. In this paper, we present SemRel, a new semantic relatedness dataset collection annotated by native speakers across 13 languages: Afrikaans, Algerian Arabic, Amharic, English, Hausa, Hindi, Indonesian, Kinyarwanda, Marathi, Moroccan Arabic, Modern Standard Arabic, Spanish, and Telugu. These languages originate from five distinct language families and are predominantly spoken in Africa and Asia – regions characterised by a relatively limited availability of NLP resources. Each instance in the SemRel datasets is a sentence pair associated with a score that represents the degree of semantic textual relatedness between the two sentences. The scores are obtained using a comparative annotation framework. We describe the data collection and annotation processes, challenges when building the datasets, baseline experiments, and their impact and utility in NLP.

Xin Miao, Yongqi Li, Shen Zhou, Tieyun Qian

Large language models (LLMs) have achieved satisfactory performance in counterfactual generation. However, confined by the stochastic generation process of LLMs, there often are misalignments between LLMs and humans which hinder LLMs from handling complex tasks like relation extraction. As a result, LLMs may generate commonsense-violated counterfactuals like ‘eggs were produced by a box’. To bridge this gap, we propose to mimick the episodic memory retrieval, the working mechanism of human hippocampus, to align LLMs’ generation process with that of humans. In this way, LLMs can derive experience from their extensive memory, which keeps in line with the way humans gain commonsense. We then implement two central functions in the hippocampus, i.e., pattern separation and pattern completion, to retrieve the episodic memory from LLMs and generate commonsense counterfactuals for relation extraction. Experimental results demonstrate the improvements of our framework over existing methods in terms of the quality of counterfactuals.

Shitao Xiao, Zheng Liu, Peitian Zhang, Xingrun Xing

The pre-trained language models are continually fine-tuned to better support downstream applications. However, this operation may result in significant performance degeneration on general tasks beyond the targeted domain. To overcome this problem, we propose LM-Cocktail which enables the fine-tuned model to stay resilient in general perspectives. Our method is conducted in the form of model merging, where the fine-tuned language model is merged with the pre-trained base model or the peer models from other domains through weighted average. Despite simplicity, LM-Cocktail is surprisingly effective: the resulted model is able to achieve a strong empirical performance in the whole scope of general tasks while preserving a superior capacity in its targeted domain.

Chengfeng Dou, Ying Zhang, Zhi Jin, Wenpin Jiao, Haiyan Zhao, Yongqiang Zhao, Zhengwei Tao

The utilization of large language models for medical dialogue generation has attracted considerable attention due to its potential to enhance response richness and coherence. While previous studies have made strides in optimizing model performance, there is a pressing need to bolster the model’s capacity for diagnostic logic to ensure patient safety. In response to this need, we propose an approach termed preference learning from process feedback (PLPF), which involves integrating the doctor’s diagnostic logic into LLMs. PLPF encompasses three key components: rule modeling, preference data generation, and preference alignment. These components collectively serve to train the model to adhere to the diagnostic process. Our experimental results, utilizing Standardized Patient Testing, demonstrate that PLPF enhances the diagnostic accuracy of the baseline model in medical conversations by 17.6%, surpassing the performance of traditional approaches. Moreover, PLPF exhibits effectiveness in both multi-round and single-round dialogue tasks, thereby highlighting its potential in improving medical dialogue generation. Our dataset is available at https://github.com/Chengfeng-Dou/SpTesting.

Amit Dhurandhar, Rahul Nair, Moninder Singh, Elizabeth Daly, Karthikeyan Natesan Ramamurthy

Evaluation and ranking of large language models (LLMs) has become an important problem with the proliferation of these models and their impact. Evaluation methods either require human responses which are expensive to acquire or use pairs of LLMs to evaluate each other which can be unreliable. In this paper, we provide a novel perspective where, given a dataset of prompts (viz. questions, instructions, etc.) and a set of LLMs, we rank them without access to any ground truth or reference responses. Inspired by real life where both an expert and a knowledgeable person can identify a novice our main idea is to consider triplets of models, where each one of them evaluates the other two, correctly identifying the worst model in the triplet with high probability. We also analyze our idea and provide sufficient conditions for it to succeed. Applying this idea repeatedly we propose two methods to rank LLMs. In experiments on different generative tasks (summarization, multiple-choice, and dialog), our methods reliably recover true rankings without reference data. This points to a viable low-resource mechanism for practical use.

Amit Dhurandhar, Tejaswini Pedapati, Ronny Luss, Soham Dan, Aurelie Lozano, Payel Das, Georgios Kollias

Transformer-based Language Models have become ubiquitous in Natural Language Processing (NLP) due to their impressive performance on various tasks. However, expensive training as well as inference remains a significant impediment to their widespread applicability. While enforcing sparsity at various levels of the model architecture has found promise in addressing scaling and efficiency issues, there remains a disconnect between how sparsity affects network topology. Inspired by brain neuronal networks, we explore sparsity approaches through the lens of network topology. Specifically, we exploit mechanisms seen in biological networks, such as preferential attachment and redundant synapse pruning, and show that principled, model-agnostic sparsity approaches are performant and efficient across diverse NLP tasks, spanning both classification (such as natural language inference) and generation (summarization, machine translation), despite our sole objective not being optimizing performance. NeuroPrune is competitive with (or sometimes superior to) baselines on performance and can be up to 10x faster in terms of training time for a given level of sparsity, simultaneously exhibiting measurable improvements in inference time in many cases.

Jaehyung Seo, Jaewook Lee, Chanjun Park, SeongTae Hong, Seungjun Lee, Heuiseok Lim

The evolution of large language models (LLMs) has culminated in a multitask model paradigm where prompts drive the generation of user-specific outputs. However, this advancement has revealed a critical challenge: LLMs frequently produce outputs against socially acceptable commonsense standards in various scenarios. To address this gap in commonsense reasoning, we present KoCommonGEN v2, a fine-grained benchmark dataset focused on Korean commonsense reasoning. This dataset, enriched with human annotations, comprises multiple-choice questions across seven error categories. These categories include commonsense memorization, numerical commonsense, toxic speech, and more, which are vulnerable to undermining the reliability of LLMs’ commonsense reasoning capabilities. The empirical results present that LLMs struggle with Korean commonsense reasoning. With human accuracy benchmarked at approximately 85%, GPT-4’s performance lags at about 74%, and other LLMs demonstrate an average accuracy of around 42%. Our findings emphasize the need for targeted improvements in Korean commonsense reasoning within LLMs, paving the way for more socially and contextually sensitive AI models.