The information stored in large language models (LLMs) falls out of date quickly, and retraining from scratch is often not an option. This has recently given rise to a range of techniques for injecting new facts through updating model weights. Current evaluation paradigms are extremely limited, mainly validating the recall of edited facts, but changing one fact should cause rippling changes to the model’s related beliefs. If we edit the UK Prime Minister to now be Rishi Sunak, then we should get a different answer to Who is married to the British Prime Minister? In this work, we present a benchmark MQuAKE (Multi-hop Question Answering for Knowledge Editing) comprising multi-hop questions that assess whether edited models correctly answer questions where the answer should change as an entailed consequence of edited facts. While we find that current knowledge-editing approaches can recall edited facts accurately, they fail catastrophically on the constructed multi-hop questions. We thus propose a simple memory-based approach, MeLLo, which stores all edited facts externally while prompting the language model iteratively to generate answers that are consistent with the edited facts. While MQuAKE remains challenging, we show that MeLLo scales well with LLMs (up to 175B) and outperforms previous model editors by a large margin.
论文检索
输入标题、作者或关键词，从 7,591 篇学术成果中精准定位
Document-level NER approaches use global information via word-based key-value memory for accurate and consistent predictions. However, such global information on word level can introduce noise when the same word appears in different token sequences and has different labels. This work proposes a two-stage document-level NER model, ScdNER, for more accurate and consistent predictions via adaptive span-level global feature fusion. In the first stage, ScdNER trains a binary classifier to predict if a token sequence is an entity with a probability. Via a span-based key-value memory, the probabilities are further used to obtain the entity’s global features with reduced impact of non-entity sequences. The second stage predicts the entity types using a gate mechanism to balance its local and global information, leading to adaptive global feature fusion. Experiments on benchmark datasets from scientific, biomedical, and general domains show the effectiveness of the proposed methods.
Learning Knowledge-Enhanced Contextual Language Representations for Domain Natural Language Understanding
PDF ↗Knowledge-Enhanced Pre-trained Language Models (KEPLMs) improve the performance of various downstream NLP tasks by injecting knowledge facts from large-scale Knowledge Graphs (KGs). However, existing methods for pre-training KEPLMs with relational triples are difficult to be adapted to close domains due to the lack of sufficient domain graph semantics. In this paper, we propose a Knowledge-enhanced language representation learning framework for various closed domains (KANGAROO) via capturing the implicit graph structure among the entities. Specifically, since the entity coverage rates of closed-domain KGs can be relatively low and may exhibit the global sparsity phenomenon for knowledge injection, we consider not only the shallow relational representations of triples but also the hyperbolic embeddings of deep hierarchical entity-class structures for effective knowledge fusion. Moreover, as two closed-domain entities under the same entity-class often havel locally dense neighbor subgraphs counted by max point biconnected component, we further propose a data augmentation strategy based on contrastive learning over subgraphs to construct hard negative samples of higher quality. It makes the underlying KELPMs better distinguish the semantics of these neighboring entities to further complement the global semantic sparsity. In the experiments, we evaluate KANGAROO over various knowledge-aware and general NLP tasks in both full and few-shot learning settings, outperforming various KEPLM training paradigms performance in closed-domains significantly.
Large language models (LLMs) have exhibited striking in-context learning (ICL) ability to adapt to target tasks with a few input-output demonstrations. For better ICL, different methods are proposed to select representative demonstrations from existing training corpora. However, such settings are not aligned with real-world practices, as end-users usually query LMs without access to demonstration pools. In this work, we introduce Self-ICL—a simple framework which bootstraps LMs’ intrinsic capabilities to perform zero-shot ICL. Given a test input, Self-ICL first prompts the model to generate pseudo-inputs. Next, the model predicts pseudo-labels for the pseudo-inputs via zero-shot prompting. Finally, we perform ICL for the test input with the pseudo-input-label pairs as demonstrations. Evaluation on 23 BIG-Bench Hard tasks shows Self-ICL outperforms zero-shot baselines on both average accuracy and head-to-head comparison. Moreover, with zero-shot chain-of-thought, Self-ICL achieves results comparable to using real demonstrations. Additionally, we conduct a range of analyses to validate Self-ICL’s effectiveness and provide insights for its behaviors under different settings.
A Training-Free Debiasing Framework with Counterfactual Reasoning for Conversational Emotion Detection
PDF ↗Unintended dataset biases typically exist in existing Emotion Recognition in Conversations (ERC) datasets, including label bias, where models favor the majority class due to imbalanced training data, as well as the speaker and neutral word bias, where models make unfair predictions because of excessive correlations between specific neutral words or speakers and classes. However, previous studies in ERC generally focus on capturing context-sensitive and speaker-sensitive dependencies, ignoring the unintended dataset biases of data, which hampers the generalization and fairness in ERC. To address this issue, we propose a Training-Free Debiasing framework (TFD) that operates during prediction without additional training. To ensure compatibility with various ERC models, it does not balance data or modify the model structure. Instead, TFD extracts biases from the model by generating counterfactual utterances and contexts and mitigates them using simple yet empirically robust element-wise subtraction operations. Extensive experiments on three public datasets demonstrate that TFD effectively improves generalization ability and fairness across different ERC models.
Language Models (LMs) have demonstrated impressive molecule understanding ability on various 1D text-related tasks. However, they inherently lack 2D graph perception — a critical ability of human professionals in comprehending molecules’ topological structures. To bridge this gap, we propose MolCA: Molecular Graph-Language Modeling with Cross-Modal Projector and Uni-Modal Adapter. MolCA enables an LM (i.e., Galactica) to understand both text- and graph-based molecular contents via the cross-modal projector. Specifically, the cross-modal projector is implemented as a Q-Former to connect a graph encoder’s representation space and an LM’s text space. Further, MolCA employs a uni-modal adapter (i.e., LoRA) for the LM’s efficient adaptation to downstream tasks. Unlike previous studies that couple an LM with a graph encoder via cross-modal contrastive learning, MolCA retains the LM’s ability of open-ended text generation and augments it with 2D graph information. To showcase its effectiveness, we extensively benchmark MolCA on tasks of molecule captioning, IUPAC name prediction, and molecule-text retrieval, on which MolCA significantly outperforms the baselines.
In this paper, we propose methods for discovering semantic differences in words appearing in two corpora. The key idea is to measure the coverage of meanings of a word in a corpus through the norm of its mean word vector, which is equivalent to examining a kind of variance of the word vector distribution. The proposed methods do not require alignments between words and/or corpora for comparison that previous methods do. All they require are to compute variance (or norms of mean word vectors) for each word type. Nevertheless, they rival the best-performing system in the SemEval-2020 Task 1. In addition, they are (i) robust for the skew in corpus sizes; (ii) capable of detecting semantic differences in infrequent words; and (iii) effective in pinpointing word instances that have a meaning missing in one of the two corpora under comparison. We show these advantages for historical corpora and also for native/non-native English corpora.
Nearest Neighbor Machine Translation (kNN-MT) has achieved great success in domain adaptation tasks by integrating pre-trained Neural Machine Translation (NMT) models with domain-specific token-level retrieval. However, the reasons underlying its success have not been thoroughly investigated. In this paper, we comprehensively analyze kNN-MT through theoretical and empirical studies. Initially, we provide new insights into the working mechanism of kNN-MT as an efficient technique to implicitly execute gradient descent on the output projection layer of NMT, indicating that it is a specific case of model fine-tuning. Subsequently, we conduct multi-domain experiments and word-level analysis to examine the differences in performance between kNN-MT and entire-model fine-tuning. Our findings suggest that: (i) Incorporating kNN-MT with adapters yields comparable translation performance to fine-tuning on in-domain test sets, while achieving better performance on out-of-domain test sets; (ii) Fine-tuning significantly outperforms kNN-MT on the recall of in-domain low-frequency words, but this gap could be bridged by optimizing the context representations with additional adapter layers.
Scaling up language models has been empirically shown to improve performance on a wide range of downstream tasks. However, if we were to observe worse performance as a function of scale (inverse scaling) on certain tasks, this would indicate that scaling can also encourage behaviors that are misaligned with human preferences. The Inverse Scaling Prize (McKenzie et al. 2023) identified eleven such inverse scaling tasks, evaluated on models of up to 280B parameters and up to 500 zettaFLOPs of training compute. This paper takes a closer look at these inverse scaling tasks. In this paper, we evaluate models of up to 540B parameters, trained on five times more compute than those evaluated in the Inverse Scaling Prize. With this increased range of model sizes and compute, only four out of the eleven tasks remain inverse scaling. Six tasks exhibit U-shaped scaling, where performance decreases up to a certain size, and then increases again up to the largest model evaluated (the one remaining task displays positive scaling). In addition, 1-shot examples and chain-of-thought can help mitigate undesirable scaling patterns even further. U-shaped scaling suggests that the inverse scaling trend observed in McKenzie et al. (2023) may not continue to hold for larger models, which we attribute to the presence of distractor tasks that only sufficiently large models can avoid.
Novels are often adapted into feature films, but the differences between the two media usually require dropping sections of the source text from the movie script. Here we study this screen adaptation process by constructing narrative alignments using the Smith-Waterman local alignment algorithm coupled with SBERT embedding distance to quantify text similarity between scenes and book units. We use these alignments to perform an automated analysis of 40 adaptations, revealing insights into the screenwriting process concerning (i) faithfulness of adaptation, (ii) importance of dialog, (iii) preservation of narrative order, and (iv) gender representation issues reflective of the Bechdel test.
From Values to Opinions: Predicting Human Behaviors and Stances Using Value-Injected Large Language Models
PDF ↗Being able to predict people’s opinions on issues and behaviors in realistic scenarios can be helpful in various domains, such as politics and marketing. However, conducting large-scale surveys like the European Social Survey to solicit people’s opinions on individual issues can incur prohibitive costs. Leveraging prior research showing influence of core human values on individual decisions and actions, we propose to use value-injected large language models (LLM) to predict opinions and behaviors. To this end, we present Value Injection Method (VIM), a collection of two methods—argument generation and question answering—designed to inject targeted value distributions into LLMs via fine-tuning. We then conduct a series of experiments on four tasks to test the effectiveness of VIM and the possibility of using value-injected LLMs to predict opinions and behaviors of people. We find that LLMs value-injected with variations of VIM substantially outperform the baselines. Also, the results suggest that opinions and behaviors can be better predicted using value-injected LLMs than the baseline approaches.
Eye movements in reading play a crucial role in psycholinguistic research studying the cognitive mechanisms underlying human language processing. More recently, the tight coupling between eye movements and cognition has also been leveraged for language-related machine learning tasks such as the interpretability, enhancement, and pre-training of language models, as well as the inference of reader- and text-specific properties. However, scarcity of eye movement data and its unavailability at application time poses a major challenge for this line of research. Initially, this problem was tackled by resorting to cognitive models for synthesizing eye movement data. However, for the sole purpose of generating human-like scanpaths, purely data-driven machine-learning-based methods have proven to be more suitable. Following recent advances in adapting diffusion processes to discrete data, we propose ScanDL, a novel discrete sequence-to-sequence diffusion model that generates synthetic scanpaths on texts. By leveraging pre-trained word representations and jointly embedding both the stimulus text and the fixation sequence, our model captures multi-modal interactions between the two inputs. We evaluate ScanDL within- and across-dataset and demonstrate that it significantly outperforms state-of-the-art scanpath generation methods. Finally, we provide an extensive psycholinguistic analysis that underlines the model’s ability to exhibit human-like reading behavior. Our implementation is made available at https://github.com/DiLi-Lab/ScanDL.
The dramatic increase in the use of social media platforms for information sharing has also fueled a steep growth in online abuse. A simple yet effective way of abusing individuals or communities is by creating memes, which often integrate an image with a short piece of text layered on top of it. Such harmful elements are in rampant use and are a threat to online safety. Hence it is necessary to develop efficient models to detect and flag abusive memes. The problem becomes more challenging in a low-resource setting (e.g., Bengali memes, i.e., images with Bengali text embedded on it) because of the absence of benchmark datasets on which AI models could be trained. In this paper we bridge this gap by building a Bengali meme dataset. To setup an effective benchmark we implement several baseline models for classifying abusive memes using this dataset. We observe that multimodal models that use both textual and visual information outperform unimodal models. Our best-performing model achieves a macro F1 score of 70.51. Finally, we perform a qualitative error analysis of the misclassified memes of the best-performing text-based, image-based and multimodal models.
Continual relation extraction (CRE) aims to solve the problem of catastrophic forgetting when learning a sequence of newly emerging relations. Recent CRE studies have found that catastrophic forgetting arises from the model’s lack of robustness against future analogous relations. To address the issue, we introduce rationale, i.e., the explanations of relation classification results generated by Large Language Models (LLM), into CRE task. Specifically, we design the multi-task rationale tuning strategy to help the model learn current relations robustly. We also conduct contrastive rationale replay to further distinguish analogous relations. Experimental results on two standard benchmarks demonstrate that our method outperforms the state-of-the-art CRE models.
All Things Considered: Detecting Partisan Events from News Media with Cross-Article Comparison
PDF ↗Public opinion is shaped by the information news media provide, and that information in turn may be shaped by the ideological preferences of media outlets. But while much attention has been devoted to media bias via overt ideological language or topic selection, a more unobtrusive way in which the media shape opinion is via the strategic inclusion or omission of \textit{partisan events} that may \textit{support} one side or the other. We develop a latent variable-based framework to predict the ideology of news articles by comparing multiple articles on the same story and identifying partisan events whose inclusion or omission reveals ideology. Our experiments first validate the existence of partisan event selection, and then show that article alignment and cross-document comparison detect partisan events and article ideology better than competitive baselines. Our results reveal the high-level form of media bias, which is present even among mainstream media with strong norms of objectivity and nonpartisanship. Our codebase and dataset are available at https://github.com/launchnlp/ATC.
In the financial domain, conducting entity-level sentiment analysis is crucial for accurately assessing the sentiment directed toward a specific financial entity. To our knowledge, no publicly available dataset currently exists for this purpose. In this work, we introduce an entity-level sentiment classification dataset, called FinEntity, that annotates financial entity spans and their sentiment (positive, neutral, and negative) in financial news. We document the dataset construction process in the paper. Additionally, we benchmark several pre-trained models (BERT, FinBERT, etc.) and ChatGPT on entity-level sentiment classification. In a case study, we demonstrate the practical utility of using FinEntity in monitoring cryptocurrency markets. The data and code of FinEntity is available at https://github.com/yixuantt/FinEntity.
Document-level relation extraction (DocRE) involves identifying relations between entities distributed in multiple sentences within a document. Existing methods focus on building a heterogeneous document graph to model the internal structure of an entity and the external interaction between entities. However, there are two drawbacks in existing methods. On one hand, anaphor plays an important role in reasoning to identify relations between entities but is ignored by these methods. On the other hand, these methods achieve cross-sentence entity interactions implicitly by utilizing a document or sentences as intermediate nodes. Such an approach has difficulties in learning fine-grained interactions between entities across different sentences, resulting in sub-optimal performance. To address these issues, we propose an Anaphor-Assisted (AA) framework for DocRE tasks. Experimental results on the widely-used datasets demonstrate that our model achieves a new state-of-the-art performance.
CorefPrompt: Prompt-based Event Coreference Resolution by Measuring Event Type and Argument Compatibilities
PDF ↗Event coreference resolution (ECR) aims to group event mentions referring to the same real-world event into clusters. Most previous studies adopt the “encoding first, then scoring” framework, making the coreference judgment rely on event encoding. Furthermore, current methods struggle to leverage human-summarized ECR rules, e.g., coreferential events should have the same event type, to guide the model. To address these two issues, we propose a prompt-based approach, CorefPrompt, to transform ECR into a cloze-style MLM (masked language model) task. This allows for simultaneous event modeling and coreference discrimination within a single template, with a fully shared context. In addition, we introduce two auxiliary prompt tasks, event-type compatibility and argument compatibility, to explicitly demonstrate the reasoning process of ECR, which helps the model make final predictions. Experimental results show that our method CorefPrompt performs well in a state-of-the-art (SOTA) benchmark.
While existing neural network-based approaches have shown promising results in Handwritten Text Recognition (HTR) for high-resource languages and standardized/machine-written text, their application to low-resource languages often presents challenges, resulting in reduced effectiveness. In this paper, we propose an innovative HTR approach that leverages the Transformer architecture for recognizing handwritten Old Occitan language. Given the limited availability of data, which comprises only word pairs of graphical variants and lemmas, we develop and rely on elaborate data augmentation techniques for both text and image data. Our model combines a custom-trained Swin image encoder with a BERT text decoder, which we pre-train using a large-scale augmented synthetic data set and fine-tune on the small human-labeled data set. Experimental results reveal that our approach surpasses the performance of current state-of-the-art models for Old Occitan HTR, including open-source Transformer-based models such as a fine-tuned TrOCR and commercial applications like Google Cloud Vision. To nurture further research and development, we make our models, data sets, and code publicly available.
Word-level auto-completion (WLAC) plays a crucial role in Computer-Assisted Translation. While previous studies have primarily focused on designing complex model architectures, this paper takes a different perspective by rethinking the fundamental question: what kind of words are good auto-completions? We introduce a measurable criterion to address this question and discover that existing WLAC models often fail to meet this criterion. Building upon this observation, we propose an effective approach to enhance WLAC performance by promoting adherence to the criterion. Notably, the proposed approach is general and can be applied to various encoder-based architectures. Through extensive experiments, we demonstrate that our approach outperforms the top-performing system submitted to the WLAC shared tasks in WMT2022, while utilizing significantly smaller model sizes.