论文检索

Arijit Nag, Animesh Mukherjee, Niloy Ganguly, Soumen Chakrabarti

Large Language Models (LLMs) exhibit impressive zero/few-shot inference and generation quality for high-resource languages (HRLs). A few of them have been trained on low-resource languages (LRLs) and give decent performance. Owing to the prohibitive costs of training LLMs, they are usually used as a network service, with the client charged by the count of input and output tokens. The number of tokens strongly depends on the script and language, as well as the LLM’s subword vocabulary. We show that LRLs are at a pricing disadvantage, because the well-known LLMs produce more tokens for LRLs than HRLs. This is because most currently popular LLMs are optimized for HRL vocabularies. Our objective is to level the playing field: reduce the cost of processing LRLs in contemporary LLMs while ensuring that predictive and generative qualities are not compromised. As means to reduce the number of tokens processed by the LLM, we consider code-mixing, translation, and transliteration of LRLs to HRLs. We perform an extensive study using the IndicXTREME classification and six generative tasks dataset, covering 15 Indic and 3 other languages, while using GPT-4 (one of the costliest LLM services released so far) as a commercial LLM. We observe and analyze interesting patterns involving token count, cost, and quality across a multitude of languages and tasks. We show that choosing the best policy to interact with the LLM can reduce cost by ~90% while giving better or comparable performance, compared to communicating with the LLM in the original LRL.

Yeo Jin Kim, Halim Acosta, Wookhee Min, Jonathan Rowe, Bradford Mott, Snigdha Chaturvedi, James Lester

Dialogue act recognition is the task of classifying conversational utterances based on their communicative intent or function. To address this problem, we propose a novel two-phase processing approach called Dual-Process Masking. This approach streamlines the task by masking less important tokens in the input, identified through retrospective analysis of their estimated contribution during training. It enhances interpretability by using the masks applied during classification learning. Dual-Process Masking significantly improves performance over strong baselines for dialogue act recognition on a collaborative problem-solving dataset and three public dialogue benchmarks.

Aiqi Jiang, Nikolas Vitsakis, Tanvi Dinkar, Gavin Abercrombie, Ioannis Konstas

Gender-Based Violence (GBV) is an increasing problem online, but existing datasets fail to capture the plurality of possible annotator perspectives or ensure the representation of affected groups. We revisit two important stages in the moderation pipeline for GBV: (1) manual data labelling; and (2) automated classification. For (1), we examine two datasets to investigate the relationship between annotator identities and attitudes and the responses they give to two GBV labelling tasks. To this end, we collect demographic and attitudinal information from crowd-sourced annotators using three validated surveys from Social Psychology. We find that higher Right Wing Authoritarianism scores are associated with a higher propensity to label text as sexist, while for Social Dominance Orientation and Neosexist Attitudes, higher scores are associated with a negative tendency to do so.For (2), we conduct classification experiments using Large Language Models and five prompting strategies, including infusing prompts with annotator information. We find: (i) annotator attitudes affect the ability of classifiers to predict their labels; (ii) including attitudinal information can boost performance when we use well-structured brief annotator descriptions; and (iii) models struggle to reflect the increased complexity and imbalanced classes of the new label sets.

Marcio Monteiro, Charu Karakkaparambil James, Marius Kloft, Sophie Fellenz

Fine-tuning pretrained language models on task-specific data is a common practice in Natural Language Processing (NLP) applications. However, the number of pretrained models available to choose from can be very large, and it remains unclear how to select the optimal model without spending considerable amounts of computational resources, especially for the text domain. To address this problem, we introduce PsyMatrix, a novel framework designed to efficiently characterize text datasets. PsyMatrix evaluates multiple dimensions of text and discourse, producing interpretable, low-dimensional embeddings. Our framework has been tested using a meta-dataset repository that includes the performance of 24 pretrained large language models fine-tuned across 146 classification datasets. Using the proposed embeddings, we successfully developed a meta-learning system capable of recommending the most effective pretrained models (optimal and near-optimal) for fine-tuning on new datasets.

Hakyung Lee, Keon-Hee Park, Hoyoon Byun, Jeyoon Yeom, Jihee Kim, Gyeong-Moon Park, Kyungwoo Song

Detecting out-of-distribution (OOD) samples is crucial for ensuring the safety and robustness of models deployed in real-world scenarios. While most studies on OOD detection focus on fine-tuned models trained on in-distribution (ID) data, detecting OOD in pre-trained models is also important due to computational limitations and the widespread use of open-source pre-trained models. However, in the same domain shift setting, the OOD detection performance of pre-trained models is insufficient because both ID and OOD samples originate from the same domain, leading to a high overlap in their embeddings. To address this issue, we introduce a new method called CED, a training-free OOD detection technique designed to enhance the distinction between ID and OOD datasets. We theoretically validate that specific auxiliary and oracle samples that satisfy certain conditions improve this distinction. Motivated by our theoretical analysis, CED enhances the differentiation by utilizing these specially designed auxiliary and oracle samples. As a result, CED significantly improves the ability of pre-trained models to distinguish between ID and OOD samples in text classification and hallucination detection tasks. Furthermore, we verify that CED is a plug-and-play method compatible with various backbone networks, such as RoBERTa, Llama, and OpenAI Embedding.

Witold Sosnowski, Arkadiusz Modzelewski, Kinga Skorupska, Jahna Otterbacher, Adam Wierzbicki

As narratives shape public opinion and influence societal actions, distinguishing between truthful and misleading narratives has become a significant challenge. To address this, we introduce the EU DisinfoTest, a novel benchmark designed to evaluate the efficacy of Language Models in identifying disinformation narratives. Developed through a Human-in-the-Loop methodology and grounded in research from EU DisinfoLab, the EU DisinfoTest comprises more than 1,300 narratives. Our benchmark includes persuasive elements under Logos, Pathos, and Ethos rhetorical dimensions. We assessed state-of-the-art LLMs, including the newly released GPT-4o, on their capability to perform zero-shot classification of disinformation narratives versus credible narratives. Our findings reveal that LLMs tend to regard narratives with authoritative appeals as trustworthy, while those with emotional appeals are frequently incorrectly classified as disinformative. These findings highlight the challenges LLMs face in nuanced content interpretation and suggest the need for tailored adjustments in LLM training to better handle diverse narrative structures.

Md Fahim, Fariha Tanjim Shifat, Fabiha Haider, Deeparghya Dutta Barua, MD Sakib Ul Rahman Sourove, Md Farhan Ishmam, Md Farhad Alam Bhuiyan

Low-resource languages like Bangla are severely limited by the lack of datasets. Romanized Bangla texts are ubiquitous on the internet, offering a rich source of data for Bangla NLP tasks and extending the available data sources. However, due to the informal nature of romanized text, they often lack the structure and consistency needed to provide insights. We address these challenges by proposing: (1) BanglaTLit, the large-scale Bangla transliteration dataset consisting of 42.7k samples, (2) BanglaTLit-PT, a pre-training corpus on romanized Bangla with 245.7k samples, (3) encoders further-pretrained on BanglaTLit-PT achieving state-of-the-art performance in several romanized Bangla classification tasks, and (4) multiple back-transliteration baseline methods, including a novel encoder-decoder architecture using further pre-trained encoders. Our results show the potential of automated Bangla back-transliteration in utilizing the untapped sources of romanized Bangla to enrich this language. The code and datasets are publicly available: https://github.com/farhanishmam/BanglaTLit.

Eunji Kim, Kyuhong Shim, Simyung Chang, Sungroh Yoon

A text encoder within Vision-Language Models (VLMs) like CLIP plays a crucial role in translating textual input into an embedding space shared with images, thereby facilitating the interpretative analysis of vision tasks through natural language. Despite the varying significance of different textual elements within a sentence depending on the context, efforts to account for variation of importance in constructing text embeddings have been lacking. We propose a framework of Semantic Token Reweighting to build Interpretable text embeddings (SToRI), which incorporates controllability as well. SToRI refines the text encoding process in CLIP by differentially weighting semantic elements based on contextual importance, enabling finer control over emphasis responsive to data-driven insights and user preferences. The efficacy of SToRI is demonstrated through comprehensive experiments on few-shot image classification and image retrieval tailored to user preferences.

Zhe Yang, Yi Huang, Yaqin Chen, Xiaoting Wu, Junlan Feng, Chao Deng

In extremely low resource relation identification scenario, small language models (SLMs) incline to overfit, which significantly diminishes their accuracy. Recently, large language models (LLMs) are gradually applied to classification tasks with converting original objective into the generation task via in-context learning. However, abundance of the classifier categories poses challenges in selecting demonstrations. Moreover, the mapping between category labels and textual descriptions requires expensive expert knowledge, thereby constraining the efficacy of in-context learning for LLMs. We uphold that SLM is optimal for handling classification tasks, and its shortcomings in the low resource setting can be mitigated by leveraging LLM. Hence, we propose a co-evolution strategy on SLM & LLM for relation identification. Specifically, LLM provides essential background knowledge to assist training process of the SLM classifier, while evaluation metrics from the classifier, in turn, offer valuable insights to refine the generation prompts of the LLM. We conduct experiments on several datasets which demonstrates preponderance of the proposed model.

Sheng-Wei Chen, Chih-Jen Lin

Text classification is one of the essential topics in natural language processing, and each text is often associated with multiple labels. Recently, the number of labels has become larger and larger, especially in the applications of e-commerce, so handling text-related e-commerce problems further requires a large memory space in many existing multi-label learning methods. To address the space concern, utilizing a distributed system to share that large memory requirement is a possible solution. We propose “random label forests,” a distributed ensemble method with label subsampling, for handling extremely large-scale labels. Random label forests can reduce memory usage per computer while keeping competitive performances over real-world data sets.

Anirudh Ajith, Sameer Singh, Danish Pruthi

Watermarking involves implanting an imperceptible signal into generated text that can later be detected via statistical tests. A prominent family of watermarking strategies for LLMs embeds this signal by upsampling a (pseudorandomly-chosen) subset of tokens at every generation step. However, such signals alter the model’s output distribution and can have unintended effects on its downstream performance. In this work, we evaluate the performance of LLMs watermarked using three different strategies over a diverse suite of tasks including those cast as k-class classification (CLS), multiple choice question answering (MCQ), short-form generation (e.g., open-ended question answering) and long-form generation (e.g., translation) tasks. We find that watermarks (under realistic hyperparameters) can cause significant drops in LLMs’ effective utility across all tasks. We observe drops of 10 to 20% in CLS tasks in the average case, which shoot up to 100% in the worst case. We notice degradations of about 7% in MCQ tasks, 10-15% in short-form generation, and 5-15% in long-form generation tasks. Our findings highlight the trade-offs that users should be cognizant of when using watermarked models.

Trang Le, Daniel Lazar, Suyoun Kim, Shan Jiang, Duc Le, Adithya Sagar, Aleksandr Livshits, Ahmed A Aly, Akshat Shrivastava

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

Zheng Xin Yong, Cristina Menghini, Stephen Bach

Data scarcity in low-resource languages can be addressed with word-to-word translations from labeled task data in high-resource languages using bilingual lexicons. However, bilingual lexicons often have limited lexical overlap with task data, which results in poor translation coverage and lexicon utilization. We propose lexicon-conditioned data generation LexC-Gen, a method that generates low-resource-language classification task data at scale. Specifically, LexC-Gen first uses high-resource-language words from bilingual lexicons to generate lexicon-compatible task data, and then it translates them into low-resource languages with bilingual lexicons via word translation. Across 17 extremely low-resource languages, LexC-Gen generated data is competitive with expert-translated gold data, and yields on average 5.6 and 8.9 points improvement over existing lexicon-based word translation methods on sentiment analysis and topic classification tasks respectively. Through ablation study, we show that conditioning on bilingual lexicons is the key component of LexC-Gen. LexC-Gen serves as a potential solution to close the performance gap between open-source multilingual models, such as BLOOMZ and Aya-101, and state-of-the-art commercial models like GPT-4o on low-resource-language tasks.

Theron S. Wang, Xingyuan Li, Chunhao Zhang, Mengyue Wu, Kenny Q. Zhu

This paper attempts to discover communication patterns automatically within dog vocalizations in a data-driven approach, which breaks the barrier previous approaches that rely on human prior knowledge on limited data. We present a self-supervised approach with HuBERT, enabling the accurate classification of phones, and an adaptive grammar induction method that identifies phone sequence patterns that suggest a preliminary vocabulary within dog vocalizations. Our results show that a subset of this vocabulary has substantial causality relations with certain canine activities, suggesting signs of stable semantics associated with these “words”.

Lilian Ngweta, Mayank Agarwal, Subha Maity, Alex Gittens, Yuekai Sun, Mikhail Yurochkin

Large Language Models (LLMs) need to be aligned with human expectations to ensure their safety and utility in most applications. Alignment is challenging, costly, and needs to be repeated for every LLM and alignment criterion. We propose to decouple LLMs and alignment by training *aligner* models that can be used to align any LLM for a given criteria on an as-needed basis, thus also reducing the potential negative impacts of alignment on performance. Our recipe for training the aligner models solely relies on synthetic data generated with a (prompted) LLM and can be easily adjusted for a variety of alignment criteria. We use the same synthetic data to train *inspectors*, binary miss-alignment classification models to guide a *squad* of multiple aligners. Our empirical results demonstrate consistent improvements when applying aligner squad to various LLMs, including chat-aligned models, across several instruction-following and red-teaming datasets.

Poojitha Thota, Shirin Nilizadeh

Large Language Models (LLMs) have introduced novel opportunities for text comprehension and generation. Yet, they are vulnerable to adversarial perturbations and data poisoning attacks, particularly in tasks like text classification and translation. However, the adversarial robustness of abstractive text summarization models remains less explored. In this work, we unveil a novel approach by exploiting the inherent lead bias in summarization models, to perform adversarial perturbations. Furthermore, we introduce an innovative application of influence functions, to execute data poisoning, which compromises the model’s integrity. This approach not only shows a skew in the models’ behavior to produce desired outcomes but also shows a new behavioral change, where models under attack tend to generate extractive summaries rather than abstractive summaries.

Tannon Kew, Florian Schottmann, Rico Sennrich

The vast majority of today’s large language models (LLMs) are English-centric, having been pretrained predominantly on English text. Yet, in order to meet user expectations, models need to be able to respond appropriately in multiple languages once deployed in downstream applications. This requires strong cross-lingual transfer abilities.In this work, we investigate the minimal amount of multilinguality required during finetuning to elicit cross-lingual generalisation in English-centric LLMs. In experiments across four LLMs, we find that multilingual instruction tuning with as few as two to three languages is both necessary and sufficient to elicit effective cross-lingual generalisation, with the limiting factor being the degree to which a target language is seen during pretraining. Evaluations on five different tasks further reveal that multilingual instruction tuning is most beneficial for generative tasks that assume input/output language agreement, such as in chat settings, while being of less importance for highly structured classification-style tasks.

Junehyung Kim, Sungjae Hwang

This paper introduces LADAM, a novel method for enhancing the performance of text classification tasks. LADAM employs attention mechanisms to exchange semantically similar words between sentences. This approach generates a greater diversity of synthetic sentences compared to simpler operations like random insertions, while maintaining the context of the original sentences. Additionally, LADAM is an easy-to-use, lightweight technique that does not require external datasets or large language models. Our experimental results across five datasets demonstrate that LADAM consistently outperforms baseline methods across diverse text classification conditions.

Zhivar Sourati, Darshan Girish Deshpande, Filip Ilievski, Kiril Gashteovski, Sascha Saralajew

Downstream applications often require text classification models to be accurate and robust. While the accuracy of state-of-the-art Language Models (LMs) approximates human performance, they often exhibit a drop in performance on real-world noisy data. This lack of robustness can be concerning, as even small perturbations in text, irrelevant to the target task, can cause classifiers to incorrectly change their predictions. A potential solution can be the family of Prototype-Based Networks (PBNs) that classifies examples based on their similarity to prototypical examples of a class (prototypes) and has been shown to be robust to noise for computer vision tasks. In this paper, we study whether the robustness properties of PBNs transfer to text classification tasks under both targeted and static adversarial attack settings. Our results show that PBNs, as a mere architectural variation of vanilla LMs, offer more robustness compared to vanilla LMs under both targeted and static settings. We showcase how PBNs’ interpretability can help us understand PBNs’ robustness properties. Finally, our ablation studies reveal the sensitivity of PBNs’ robustness to the strictness of clustering and the number of prototypes in the training phase, as tighter clustering and a low number of prototypes result in less robust PBNs.

Hemank Lamba, Anton Abilov, Ke Zhang, Elizabeth M Olson, Henry Kudzanai Dambanemuya, João Cordovil Bárcia, David S. Batista, Christina Wille, Aoife Cahill, Joel R. Tetreault 等

Humanitarian organizations can enhance their effectiveness by analyzing data to discover trends, gather aggregated insights, manage their security risks, support decision-making, and inform advocacy and funding proposals. However, data about violent incidents with direct impact and relevance for humanitarian aid operations is not readily available. An automatic data collection and NLP-backed classification framework aligned with humanitarian perspectives can help bridge this gap. In this paper, we present HumVI – a dataset comprising news articles in three languages (English, French, Arabic) containing instances of different types of violent incidents categorized by the humanitarian sector they impact, e.g., aid security, education, food security, health, and protection. Reliable labels were obtained for the dataset by partnering with a data-backed humanitarian organization, Insecurity Insight. We provide multiple benchmarks for the dataset, employing various deep learning architectures and techniques, including data augmentation and mask loss, to address different task-related challenges, e.g., domain expansion. The dataset is publicly available at https://github.com/dataminr-ai/humvi-dataset.