Handwritten text recognition (HTR) yields textual output that comprises errors, which are considerably more compared to that of recognised printed (OCRed) text. Post-correcting methods can eliminate such errors but may also introduce errors. In this study, we investigate the issues arising from this reality in Byzantine Greek. We investigate the properties of the texts that lead post-correction systems to this adversarial behaviour and we experiment with text classification systems that learn to detect incorrect recognition output. A large masked language model, pre-trained in modern and fine-tuned in Byzantine Greek, achieves an Average Precision score of 95%. The score improves to 97% when using a model that is pre-trained in modern and then in ancient Greek, the two language forms Byzantine Greek combines elements from. A century-based analysis shows that the advantage of the classifier that is further-pre-trained in ancient Greek concerns texts of older centuries. The application of this classifier before a neural post-corrector on HTRed text reduced significantly the post-correction mistakes.
论文检索
输入标题、作者或关键词，从 7,876 篇学术成果中精准定位
With growing capabilities of large language models, prompting them has become the dominant way to access them. This has motivated the development of strategies for automatically selecting effective language prompts. In this paper, we introduce **pFlat** (prompt flatness), a new metric to quantify the expected utility of a language prompt. This metric is inspired by *flatness* regularization in statistical learning that quantifies the robustness of the model towards its parameter perturbations. We provide theoretical foundations for this metric and its relationship with other prompt selection metrics, providing a comprehensive understanding of existing methods. Empirically, we show that combining **pFlat** with existing metrics improves both performance and sample efficiency. Our metric outperforms the previous prompt selection metrics with an average increase of 10% in Pearson correlation across 6 classification benchmarks, and the prompt selected by our metric gains 5% higher accuracy than previous metrics across the benchmarks.
Text2Tree: Aligning Text Representation to the Label Tree Hierarchy for Imbalanced Medical Classification
PDF ↗Deep learning approaches exhibit promising performances on various text tasks. However, they are still struggling on medical text classification since samples are often extremely imbalanced and scarce. Different from existing mainstream approaches that focus on supplementary semantics with external medical information, this paper aims to rethink the data challenges in medical texts and present a novel framework-agnostic algorithm called Text2Tree that only utilizes internal label hierarchy in training deep learning models. We embed the ICD code tree structure of labels into cascade attention modules for learning hierarchy-aware label representations. Two new learning schemes, Similarity Surrogate Learning (SSL) and Dissimilarity Mixup Learning (DML), are devised to boost text classification by reusing and distinguishing samples of other labels following the label representation hierarchy, respectively. Experiments on authoritative public datasets and real-world medical records show that our approach stably achieves superior performances over classical and advanced imbalanced classification methods. Our code is available at https://github.com/jyansir/Text2Tree.
BERT-based models have had strong performance on leaderboards, yet have been demonstrably worse in real-world settings requiring generalization. Limited quantities of training data is considered a key impediment to achieving generalizability in machine learning. In this paper, we examine the impact of training data quality, not quantity, on a model’s generalizability. We consider two characteristics of training data: the portion of human-adversarial (h-adversarial), i.e. sample pairs with seemingly minor differences but different ground-truth labels, and human-affable (h-affable) training samples, i.e. sample pairs with minor differences but the same ground-truth label. We find that for a fixed size of training samples, as a rule of thumb, having 10-30% h-adversarial instances improves the precision, and therefore F1, by up to 20 points in the tasks of text classification and relation extraction. Increasing h-adversarials beyond this range can result in performance plateaus or even degradation. In contrast, h-affables may not contribute to a model’s generalizability and may even degrade generalization performance.
We propose a framework for online multimodal dialog act (DA) classification based on raw audio and ASR-generated transcriptions of current and past utterances. Existing multimodal DA classification approaches are limited by ineffective audio modeling and late-stage fusion. We showcase significant improvements in multimodal DA classification by integrating modalities at a more granular level and incorporating recent advancements in large language and audio models for audio feature extraction. We further investigate the effectiveness of self-attention and cross-attention mechanisms in modeling utterances and dialogs for DA classification. We achieve a substantial increase of 3 percentage points in the F1 score relative to current state-of-the-art models on two prominent DA classification datasets, MRDA and EMOTyDA.
Large Language Models (LLMs) have demonstrated impressive performance on a range of Natural Language Processing (NLP) tasks. Unfortunately, the immense amount of computations and memory accesses required for LLM training makes them prohibitively expensive in terms of hardware cost, and thus challenging to deploy in use cases such as on-device learning. In this paper, motivated by the observation that LLM training is memory-bound, we propose a novel dynamic quantization strategy, termed Dynamic Stashing Quantization (DSQ), that puts a special focus on reducing the memory operations, but also enjoys the other benefits of low precision training, such as the reduced arithmetic cost. We conduct a thorough study on two translation tasks (trained-from-scratch) and three classification tasks (fine-tuning). DSQ reduces the amount of arithmetic operations by 20.95\times and the number of DRAM operations by 2.55\times on IWSLT17 compared to the standard 16-bit fixed-point, which is widely used in on-device learning.
Implicit Discourse Relation Recognition (IDRR) involves identifying the sense label of an implicit connective between adjacent text spans. This has traditionally been approached as a classification task. However, some downstream tasks require more than just a sense label as well as the specific connective used. This paper presents Implicit Sense-labeled Connective Recognition (ISCR), which identifies the implicit connectives and their sense labels between adjacent text spans. ISCR can be treated as a classification task, but a large number of potential categories, sense labels, and uneven distribution of instances among them make this difficult. Instead, this paper handles the task as a text-generation task, using an encoder-decoder model to generate both connectives and their sense labels. Here, we explore a classification method and three kinds of text-generation methods. From our evaluation results on PDTB-3.0, we found that our method outperforms the conventional classification-based method.
Recent studies propose various data augmentation approaches to resolve the low-resource problem in natural language processing tasks. Data augmentation is a successful solution to this problem and recent strategies give variation on sentence structures to boost performance. However, these approaches can potentially lead to semantic errors and produce semantically noisy data due to the unregulated variation of sentence structures. In an effort to combat these semantic errors, we leverage slot information, the representation of the context of keywords from a sentence, and form a data augmentation strategy which we propose, called GDA. Our strategy employs algorithms that construct and manipulate rules of context-aware grammar, utilizing this slot information. The algorithms extract recurrent patterns by distinguishing words with slots and form the “rules of grammar”—a set of injective relations between a sentence’s semantics and its syntactical structure—to augment the dataset. The augmentation is done in an automated manner with the constructed rules and thus, GDA is explainable and reliable without any human intervention. We evaluate GDA with state-of-the-art data augmentation techniques, including those using pre-trained language models, and the result illustrates that GDA outperforms all other data augmentation methods by 19.38%. Extensive experiments show that GDA is an effective data augmentation strategy that incorporates word semantics for more accurate and diverse data.
Transformer-based pretrained models like BERT, GPT-2 and T5 have been finetuned for a large number of natural language processing (NLP) tasks, and have been shown to be very effective. However, while finetuning, what changes across layers in these models with respect to pretrained checkpoints is under-studied. Further, how robust are these models to perturbations in input text? Does the robustness vary depending on the NLP task for which the models have been finetuned? While there exists some work on studying robustness of BERT finetuned for a few NLP tasks, there is no rigorous study which compares this robustness across encoder only, decoder only and encoder-decoder models. In this paper, we characterize changes between pretrained and finetuned language model representations across layers using two metrics: CKA and STIR. Further, we study the robustness of three language models (BERT, GPT-2 and T5) with eight different text perturbations on classification tasks from General Language Understanding Evaluation (GLUE) benchmark, and generation tasks like summarization, free-form generation and question generation. GPT-2 representations are more robust than BERT and T5 across multiple types of input perturbation. Although models exhibit good robustness broadly, dropping nouns, verbs or changing characters are the most impactful. Overall, this study provides valuable insights into perturbation-specific weaknesses of popular Transformer-based models which should be kept in mind when passing inputs.
‘Person’ == Light-skinned, Western Man, and Sexualization of Women of Color: Stereotypes in Stable Diffusion
PDF ↗We study stereotypes embedded within one of the most popular text-to-image generators: Stable Diffusion. We answer the question: what stereotypes of gender and nationality/continental identity does Stable Diffusion display in the absence of such information i.e. what gender and nationality/continental identity is assigned to ‘a person,’ or to ‘a person from Asia.’ Using CLIP-cosine similarity for zero-shot classification of images generated by CLIP-based Stable Diffusion v2.1 verified by manual examination, we chronicle results from 136 prompts (50 results/prompt) of front-facing images of faces from 6 different continents, 27 countries and 3 genders. We observe how Stable Diffusion results of ‘a person’ without any additional gender/nationality information correspond closest to images of men (avg. similarity 0.64) and least with persons of nonbinary gender (avg. similarity 0.41), and to persons from Europe/North America (avg. similarities 0.71 and 0.68, respectively) over Africa/Asia (avg. similarities 0.43 and 0.41, respectively), pointing towards Stable Diffusion having a concerning representation of personhood to be a European/North American man. We also show continental stereotypes and resultant harms e.g. a person from Oceania is deemed to be Australian/New Zealander (avg. similarities 0.77 and 0.74, respectively) over Papua New Guinean (avg. similarity 0.31), pointing to the erasure of Indigenous Oceanic peoples, who form a majority over descendants of colonizers both in Papua New Guinea and in Oceania overall. Finally, we unexpectedly observe a pattern of sexualization of women, specifically Latin American, Mexican, Indian and Egyptian women, confirmed through an NSFW detector and verified by manual examination. This demonstrates how Stable Diffusion perpetuates Western fetishization of women of color through objectification in media, which if left unchecked will worsen this stereotypical representation. All code and relevant data will be made publicly available.
In recent years, Pre-trained Language Models (PLMs) have shown their superiority by pre-training on unstructured text corpus and then fine-tuning on downstream tasks. On entity-rich textual resources like Wikipedia, Knowledge-Enhanced PLMs (KEPLMs) incorporate the interactions between tokens and mentioned entities in pre-training, and are thus more effective on entity-centric tasks such as entity linking and relation classification. Although exploiting Wikipedia’s rich structures to some extent, conventional KEPLMs still neglect a unique layout of the corpus where each Wikipedia page is around a topic entity (identified by the page URL and shown in the page title). In this paper, we demonstrate that KEPLMs without incorporating the topic entities will lead to insufficient entity interaction and biased (relation) word semantics. We thus propose KEPLET, a novel Knowledge-Énhanced Pre-trained LanguagE model with Topic entity awareness. In an end-to-end manner, KEPLET identifies where to add the topic entity’s information in a Wikipedia sentence, fuses such information into token and mentioned entities representations, and supervises the network learning, through which it takes topic entities back into consideration. Experiments demonstrated the generality and superiority of KEPLET which was applied to two representative KEPLMs, achieving significant improvements on four entity-centric tasks.
Recent advancements in natural language processing have demonstrated the efficacy of pre-trained language models for various downstream tasks through prompt-based fine-tuning. In contrast to standard fine-tuning, which relies solely on labeled examples, prompt-based fine-tuning combines a few labeled examples (few shot) with guidance through prompts tailored for the specific language and task. For low-resource languages, where labeled examples are limited, prompt-based fine-tuning appears to be a promising alternative. In this paper, we compare prompt-based and standard fine-tuning for the popular task of text classification in Urdu and Roman Urdu languages. We conduct experiments using five datasets, covering different domains, and pre-trained multilingual transformers. The results reveal that significant improvement of up to 13% in accuracy is achieved by prompt-based fine-tuning over standard fine-tuning approaches. This suggests the potential of prompt-based fine-tuning as a valuable approach for low-resource languages with limited labeled data.
Retrieval-augmented methods are successful in the standard scenario where the retrieval space is sufficient; whereas in the few-shot scenario with limited retrieval space, this paper shows it is non-trivial to put them into practice. First, it is impossible to retrieve semantically similar examples by using an off-the-shelf metric and it is crucial to learn a task-specific retrieval metric; Second, our preliminary experiments demonstrate that it is difficult to optimize a plausible metric by minimizing the standard cross-entropy loss. The in-depth analyses quantitatively show minimizing cross-entropy loss suffers from the weak supervision signals and the severe gradient vanishing issue during the optimization. To address these issues, we introduce two novel training objectives, namely EM-L and R-L, which provide more task-specific guidance to the retrieval metric by the EM algorithm and a ranking-based loss, respectively. Extensive experiments on 10 datasets prove the superiority of the proposed retrieval augmented methods on the performance.
Unmasking the Hidden Meaning: Bridging Implicit and Explicit Hate Speech Embedding Representations
PDF ↗Research on automatic hate speech (HS) detection has mainly focused on identifying explicit forms of hateful expressions on user-generated content. Recently, a few works have started to investigate methods to address more implicit and subtle abusive content. However, despite these efforts, automated systems still struggle to correctly recognize implicit and more veiled forms of HS. As these systems heavily rely on proper textual representations for classification, it is crucial to investigate the differences in embedding implicit and explicit messages. Our contribution to address this challenging task is fourfold. First, we present a comparative analysis of transformer-based models, evaluating their performance across five datasets containing implicit HS messages. Second, we examine the embedding representations of implicit messages across different targets, gaining insight into how veiled cases are encoded. Third, we compare and link explicit and implicit hateful messages across these datasets through their targets, enforcing the relation between explicitness and implicitness and obtaining more meaningful embedding representations. Lastly, we show how these newer representation maintains high performance on HS labels, while improving classification in borderline cases.
Code-switching is a common phenomenon in multilingual communities and is often used on social media. However, sentiment analysis of code-switched data is a challenging yet less explored area of research. This paper aims to develop a sentiment analysis system for code-switched data. In this paper, we present a novel approach combining two transformers using logits of their output and feeding them to a neural network for classification. We show the efficacy of our approach using two benchmark datasets, viz., English-Hindi (En-Hi), and English-Spanish (En-Es) availed by Microsoft GLUECoS. Our approach results in an F1 score of 73.66% for En-Es and 61.24% for En-Hi, significantly higher than the best model reported for the GLUECoS benchmark dataset.
TK-KNN: A Balanced Distance-Based Pseudo Labeling Approach for Semi-Supervised Intent Classification
PDF ↗The ability to detect intent in dialogue systems has become increasingly important in modern technology. These systems often generate a large amount of unlabeled data, and manually labeling this data requires substantial human effort. Semi-supervised methods attempt to remedy this cost by using a model trained on a few labeled examples and then by assigning pseudo-labels to further a subset of unlabeled examples that has a model prediction confidence higher than a certain threshold. However, one particularly perilous consequence of these methods is the risk of picking an imbalanced set of examples across classes, which could lead to poor labels. In the present work, we describe Top-K K-Nearest Neighbor (TK-KNN), which uses a more robust pseudo-labeling approach based on distance in the embedding space while maintaining a balanced set of pseudo-labeled examples across classes through a ranking-based approach. Experiments on several datasets show that TK-KNN outperforms existing models, particularly when labeled data is scarce on popular datasets such as CLINC150 and Banking77.
During crisis events, people often use social media platforms such as Twitter to disseminate information about the situation, warnings, advice, and support. Emergency relief organizations leverage such information to acquire timely crisis circumstances and expedite rescue operations. While existing works utilize such information to build models for crisis event analysis, fully-supervised approaches require annotating vast amounts of data and are impractical due to limited response time. On the other hand, semi-supervised models can be biased, performing moderately well for certain classes while performing extremely poorly for others, resulting in substantially negative effects on disaster monitoring and rescue. In this paper, we first study two recent debiasing methods on semi-supervised crisis tweet classification. Then we propose a simple but effective debiasing method, DeCrisisMB, that utilizes a Memory Bank to store and perform equal sampling for generated pseudo-labels from each class at each training iteration. Extensive experiments are conducted to compare different debiasing methods’ performance and generalization ability in both in-distribution and out-of-distribution settings. The results demonstrate the superior performance of our proposed method. Our code is available at https://github.com/HenryPengZou/DeCrisisMB.
Visual Elements Mining as Prompts for Instruction Learning for Target-Oriented Multimodal Sentiment Classification
PDF ↗Target-oriented Multimodal Sentiment Classification (TMSC) aims to incorporate visual modality with text modality to identify the sentiment polarity towards a specific target within a sentence. To address this task, we propose a Visual Elements Mining as Prompts (VEMP) method, which describes the semantic information of visual elements with Text Symbols Embedded in the Image (TSEI), Target-aware Adjective-Noun Pairs (TANPs) and image scene caption, and then transform them into prompts for instruction learning of the model Tk-Instruct. In our VEMP, the text symbols embedded in the image may contain the textual descriptions of fine-grained visual elements, and are extracted as input TSEI; we extract adjective-noun pairs from the image and align them with the target to obtain TANPs, in which the adjectives provide emotional embellishments for the relevant target; finally, to effectively fuse these visual elements with text modality for sentiment prediction, we integrate them to construct instruction prompts for instruction-tuning Tk-Instruct which possesses powerful learning capabilities under instructions. Extensive experimental results show that our method achieves state-of-the-art performance on two benchmark datasets. And further analysis demonstrates the effectiveness of each component of our method.
Large, general purpose language models have demonstrated impressive performance across many different conversational domains. While multi-domain language models achieve low overall perplexity, their outputs are not guaranteed to stay within the domain of a given input prompt. This paper proposes domain privacy as a novel way to quantify how likely a conditional language model will leak across domains. We also develop policy functions based on token-level domain classification, and propose an efficient fine-tuning method to improve the trained model’s domain privacy. Experiments on membership inference attacks show that our proposed method has comparable resiliency to methods adapted from recent literature on differentially private language models.
ASSERT: Automated Safety Scenario Red Teaming for Evaluating the Robustness of Large Language Models
PDF ↗As large language models are integrated into society, robustness toward a suite of prompts is increasingly important to maintain reliability in a high-variance environment.Robustness evaluations must comprehensively encapsulate the various settings in which a user may invoke an intelligent system. This paper proposes ASSERT, Automated Safety Scenario Red Teaming, consisting of three methods – semantically aligned augmentation, target bootstrapping, and adversarial knowledge injection. For robust safety evaluation, we apply these methods in the critical domain of AI safety to algorithmically generate a test suite of prompts covering diverse robustness settings – semantic equivalence, related scenarios, and adversarial. We partition our prompts into four safety domains for a fine-grained analysis of how the domain affects model performance. Despite dedicated safeguards in existing state-of-the-art models, we find statistically significant performance differences of up to 11% in absolute classification accuracy among semantically related scenarios and error rates of up to 19% absolute error in zero-shot adversarial settings, raising concerns for users’ physical safety.