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

*Do larger and more performant models resolve NLP’s longstanding robustness issues?* We investigate this question using over 20 models of different sizes spanning different architectural choices and pretraining objectives. We conduct evaluations using (a) out-of-domain and challenge test sets, (b) behavioral testing with CheckLists, (c) contrast sets, and (d) adversarial inputs. Our analysis reveals that not all out-of-domain tests provide insight into robustness. Evaluating with CheckLists and contrast sets shows significant gaps in model performance; merely scaling models does not make them adequately robust. Finally, we point out that current approaches for adversarial evaluations of models are themselves problematic: they can be easily thwarted, and in their current forms, do not represent a sufficiently deep probe of model robustness. We conclude that not only is the question of robustness in NLP as yet unresolved, but even some of the approaches to measure robustness need to be reassessed.

To completely understand a document, the use of textual information is not enough. Understanding visual cues, such as layouts and charts, is also required. While the current state-of-the-art approaches for document understanding (both OCR-based and OCR-free) work well, a thorough analysis of their capabilities and limitations has not yet been performed. Therefore, in this work, we addresses the limitation of current VisualQA models when applied to charts and plots. To investigate shortcomings of the state-of-the-art models, we conduct a comprehensive behavioral analysis, using ChartQA as a case study. Our findings indicate that existing models particularly underperform in answering questions related to the chart’s structural and visual context, as well as numerical information. To address these issues, we propose three simple pre-training tasks that enforce the existing model in terms of both structural-visual knowledge, as well as its understanding of numerical questions. We evaluate our pre-trained model (called MatCha-v2) on three chart datasets - both extractive and abstractive question datasets - and observe that it achieves an average improvement of 1.7 % over the baseline model.

We release Saamayik, a dataset of around 53,000 parallel English-Sanskrit sentences, written in contemporary prose. Sanskrit is a classical language still in sustenance and has a rich documented heritage. However, due to the limited availability of digitized content, it still remains a low-resource language. Existing Sanskrit corpora, whether monolingual or bilingual, have predominantly focused on poetry and offer limited coverage of contemporary written materials. Saamayik is curated from a diverse range of domains, including language instruction material, textual teaching pedagogy, and online tutorials, among others. It stands out as a unique resource that specifically caters to the contemporary usage of Sanskrit, with a primary emphasis on prose writing. Translation models trained on our dataset demonstrate statistically significant improvements when translating out-of-domain contemporary corpora, outperforming models trained on older classical-era poetry datasets. Finally, we also release benchmark models by adapting four multilingual pre-trained models, three of them have not been previously exposed to Sanskrit for translating between English and Sanskrit while one of them is multi-lingual pre-trained translation model including English and Sanskrit. The dataset and source code can be found at https://github.com/ayushbits/saamayik.

Can language models transform inputs to protect text classifiers against adversarial attacks? In this work, we present ATINTER, a model that intercepts and learns to rewrite adversarial inputs to make them non-adversarial for a downstream text classifier. Our experiments on four datasets and five attack mechanisms reveal that ATINTER is effective at providing better adversarial robustness than existing defense approaches, without compromising task accuracy. For example, on sentiment classification using the SST-2 dataset, our method improves the adversarial accuracy over the best existing defense approach by more than 4% with a smaller decrease in task accuracy (0.5 % vs 2.5%). Moreover, we show that ATINTER generalizes across multiple downstream tasks and classifiers without having to explicitly retrain it for those settings. For example, we find that when ATINTER is trained to remove adversarial perturbations for the sentiment classification task on the SST-2 dataset, it even transfers to a semantically different task of news classification (on AGNews) and improves the adversarial robustness by more than 10%.

Identifying intents from dialogue utterances forms an integral component of task-oriented dialogue systems. Intent-related tasks are typically formulated either as a classification task, where the utterances are classified into predefined categories or as a clustering task when new and previously unknown intent categories need to be discovered from these utterances. Further, the intent classification may be modeled in a multiclass (MC) or multilabel (ML) setup. While typically these tasks are modeled as separate tasks, we propose IntenDD a unified approach leveraging a shared utterance encoding backbone. IntenDD uses an entirely unsupervised contrastive learning strategy for representation learning, where pseudo-labels for the unlabeled utterances are generated based on their lexical features. Additionally, we introduce a two-step post-processing setup for the classification tasks using modified adsorption. Here, first, the residuals in the training data are propagated followed by smoothing the labels both modeled in a transductive setting. Through extensive evaluations on various benchmark datasets, we find that our approach consistently outperforms competitive baselines across all three tasks. On average, IntenDD reports percentage improvements of 2.32 %, 1.26 %, and 1.52 % in their respective metrics for few-shot MC, few-shot ML, and the intent discovery tasks respectively.

Clean-label (CL) attack is a form of data poisoning attack where an adversary modifies only the textual input of the training data, without requiring access to the labeling function. CL attacks are relatively unexplored in NLP, as compared to label flipping (LF) attacks, where the latter additionally requires access to the labeling function as well. While CL attacks are more resilient to data sanitization and manual relabeling methods than LF attacks, they often demand as high as ten times the poisoning budget than LF attacks. In this work, we first introduce an Adversarial Clean Label attack which can adversarially perturb in-class training examples for poisoning the training set. We then show that an adversary can significantly bring down the data requirements for a CL attack, using the aforementioned approach, to as low as 20 % of the data otherwise required. We then systematically benchmark and analyze a number of defense methods, for both LF and CL attacks, some previously employed solely for LF attacks in the textual domain and others adapted from computer vision. We find that text-specific defenses greatly vary in their effectiveness depending on their properties.

Few-shot Question Generation (QG) is an important and challenging problem in the Natural Language Generation (NLG) domain. Multilingual BERT (mBERT) has been successfully used in various Natural Language Understanding (NLU) applications. However, the question of how to utilize mBERT for few-shot QG, possibly with cross-lingual transfer, remains. In this paper, we try to explore how mBERT performs in few-shot QG (cross-lingual transfer) and also whether applying meta-learning on mBERT further improves the results. In our setting, we consider mBERT as the base model and fine-tune it using a seq-to-seq language modeling framework in a cross-lingual setting. Further, we apply the model agnostic meta-learning approach to our base model. We evaluate our model for two low-resource Indian languages, Bengali and Telugu, using the TyDi QA dataset. The proposed approach consistently improves the performance of the base model in few-shot settings and even works better than some heavily parameterized models. Human evaluation also confirms the effectiveness of our approach.

In this work, we introduce : the largest publicly available multilingual dataset for factual verification of naturally existing real-world claims. The dataset contains short statements in 25 languages and is labeled for veracity by expert fact-checkers. The dataset includes a multilingual evaluation benchmark that measures both out-of-domain generalization, and zero-shot capabilities of the multilingual models. Using state-of-the-art multilingual transformer-based models, we develop several automated fact-checking models that, along with textual claims, make use of additional metadata and evidence from news stories retrieved using a search engine. Empirically, our best model attains an F-score of around 40%, suggesting that our dataset is a challenging benchmark for the evaluation of multilingual fact-checking models.

Neural dependency parsing has achieved remarkable performance for many domains and languages. The bottleneck of massive labelled data limits the effectiveness of these approaches for low resource languages. In this work, we focus on dependency parsing for morphological rich languages (MRLs) in a low-resource setting. Although morphological information is essential for the dependency parsing task, the morphological disambiguation and lack of powerful analyzers pose challenges to get this information for MRLs. To address these challenges, we propose simple auxiliary tasks for pretraining. We perform experiments on 10 MRLs in low-resource settings to measure the efficacy of our proposed pretraining method and observe an average absolute gain of 2 points (UAS) and 3.6 points (LAS).

Neural sequence labelling approaches have achieved state of the art results in morphological tagging. We evaluate the efficacy of four standard sequence labelling models on Sanskrit, a morphologically rich, fusional Indian language. As its label space can theoretically contain more than 40,000 labels, systems that explicitly model the internal structure of a label are more suited for the task, because of their ability to generalise to labels not seen during training. We find that although some neural models perform better than others, one of the common causes for error for all of these models is mispredictions due to syncretism.

We propose a framework using energy-based models for multiple structured prediction tasks in Sanskrit. Ours is an arc-factored model, similar to the graph-based parsing approaches, and we consider the tasks of word segmentation, morphological parsing, dependency parsing, syntactic linearization, and prosodification, a “prosody-level” task we introduce in this work. Ours is a search-based structured prediction framework, which expects a graph as input, where relevant linguistic information is encoded in the nodes, and the edges are then used to indicate the association between these nodes. Typically, the state-of-the-art models for morphosyntactic tasks in morphologically rich languages still rely on hand-crafted features for their performance. But here, we automate the learning of the feature function. The feature function so learned, along with the search space we construct, encode relevant linguistic information for the tasks we consider. This enables us to substantially reduce the training data requirements to as low as 10%, as compared to the data requirements for the neural state-of-the-art models. Our experiments in Czech and Sanskrit show the language-agnostic nature of the framework, where we train highly competitive models for both the languages. Moreover, our framework enables us to incorporate language-specific constraints to prune the search space and to filter the candidates during inference. We obtain significant improvements in morphosyntactic tasks for Sanskrit by incorporating language-specific constraints into the model. In all the tasks we discuss for Sanskrit, we either achieve state-of-the-art results or ours is the only data-driven solution for those tasks.

Morphologically rich languages seem to benefit from joint processing of morphology and syntax, as compared to pipeline architectures. We propose a graph-based model for joint morphological parsing and dependency parsing in Sanskrit. Here, we extend the Energy based model framework (Krishna et al., 2020), proposed for several structured prediction tasks in Sanskrit, in 2 simple yet significant ways. First, the framework’s default input graph generation method is modified to generate a multigraph, which enables the use of an exact search inference. Second, we prune the input search space using a linguistically motivated approach, rooted in the traditional grammatical analysis of Sanskrit. Our experiments show that the morphological parsing from our joint model outperforms standalone morphological parsers. We report state of the art results in morphological parsing, and in dependency parsing, both in standalone (with gold morphological tags) and joint morphosyntactic parsing setting.

Automatic processing of figurative languages is gaining popularity in NLP community for their ubiquitous nature and increasing volume. In this era of web 2.0, automatic analysis of sarcasm and metaphors is important for their extensive usage. Metaphors are a part of figurative language that compares different concepts, often on a cognitive level. Many approaches have been proposed for automatic detection of metaphors, even using sequential models or neural networks. In this paper, we propose a method for detection of metaphors at the token level using a hybrid model of Bidirectional-LSTM and CRF. We used fewer features, as compared to the previous state-of-the-art sequential model. On experimentation with VUAMC, our method obtained an F-score of 0.674.