Current benchmark tasks for natural language processing contain text that is qualitatively different from the text used in informal day to day digital communication. This discrepancy has led to severe performance degradation of state-of-the-art NLP m
odels when fine-tuned on real-world data. One way to resolve this issue is through lexical normalization, which is the process of transforming non-standard text, usually from social media, into a more standardized form. In this work, we propose a sentence-level sequence-to-sequence model based on mBART, which frames the problem as a machine translation problem. As the noisy text is a pervasive problem across languages, not just English, we leverage the multi-lingual pre-training of mBART to fine-tune it to our data. While current approaches mainly operate at the word or subword level, we argue that this approach is straightforward from a technical standpoint and builds upon existing pre-trained transformer networks. Our results show that while word-level, intrinsic, performance evaluation is behind other methods, our model improves performance on extrinsic, downstream tasks through normalization compared to models operating on raw, unprocessed, social media text.
We propose the first general-purpose gradient-based adversarial attack against transformer models. Instead of searching for a single adversarial example, we search for a distribution of adversarial examples parameterized by a continuous-valued matrix
, hence enabling gradient-based optimization. We empirically demonstrate that our white-box attack attains state-of-the-art attack performance on a variety of natural language tasks, outperforming prior work in terms of adversarial success rate with matching imperceptibility as per automated and human evaluation. Furthermore, we show that a powerful black-box transfer attack, enabled by sampling from the adversarial distribution, matches or exceeds existing methods, while only requiring hard-label outputs.
Following the success of dot-product attention in Transformers, numerous approximations have been recently proposed to address its quadratic complexity with respect to the input length. While these variants are memory and compute efficient, it is not
possible to directly use them with popular pre-trained language models trained using vanilla attention, without an expensive corrective pre-training stage. In this work, we propose a simple yet highly accurate approximation for vanilla attention. We process the queries in chunks, and for each query, compute the top-*k* scores with respect to the keys. Our approach offers several advantages: (a) its memory usage is linear in the input size, similar to linear attention variants, such as Performer and RFA (b) it is a drop-in replacement for vanilla attention that does not require any corrective pre-training, and (c) it can also lead to significant memory savings in the feed-forward layers after casting them into the familiar query-key-value framework. We evaluate the quality of top-*k* approximation for multi-head attention layers on the Long Range Arena Benchmark, and for feed-forward layers of T5 and UnifiedQA on multiple QA datasets. We show our approach leads to accuracy that is nearly-identical to vanilla attention in multiple setups including training from scratch, fine-tuning, and zero-shot inference.
Modern Natural Language Processing (NLP) makes intensive use of deep learning methods because of the accuracy they offer for a variety of applications. Due to the significant environmental impact of deep learning, cost-benefit analysis including carb
on footprint as well as accuracy measures has been suggested to better document the use of NLP methods for research or deployment. In this paper, we review the tools that are available to measure energy use and CO2 emissions of NLP methods. We describe the scope of the measures provided and compare the use of six tools (carbon tracker, experiment impact tracker, green algorithms, ML CO2 impact, energy usage and cumulator) on named entity recognition experiments performed on different computational set-ups (local server vs. computing facility). Based on these findings, we propose actionable recommendations to accurately measure the environmental impact of NLP experiments.
GPT-3 shows remarkable in-context learning ability of large-scale language models (LMs) trained on hundreds of billion scale data. Here we address some remaining issues less reported by the GPT-3 paper, such as a non-English LM, the performances of d
ifferent sized models, and the effect of recently introduced prompt optimization on in-context learning. To achieve this, we introduce HyperCLOVA, a Korean variant of 82B GPT-3 trained on a Korean-centric corpus of 560B tokens. Enhanced by our Korean-specific tokenization, HyperCLOVA with our training configuration shows state-of-the-art in-context zero-shot and few-shot learning performances on various downstream tasks in Korean. Also, we show the performance benefits of prompt-based learning and demonstrate how it can be integrated into the prompt engineering pipeline. Then we discuss the possibility of materializing the No Code AI paradigm by providing AI prototyping capabilities to non-experts of ML by introducing HyperCLOVA studio, an interactive prompt engineering interface. Lastly, we demonstrate the potential of our methods with three successful in-house applications.
We develop a novel approach for confidently accelerating inference in the large and expensive multilayer Transformers that are now ubiquitous in natural language processing (NLP). Amortized or approximate computational methods increase efficiency, bu
t can come with unpredictable performance costs. In this work, we present CATs -- Confident Adaptive Transformers -- in which we simultaneously increase computational efficiency, while guaranteeing a specifiable degree of consistency with the original model with high confidence. Our method trains additional prediction heads on top of intermediate layers, and dynamically decides when to stop allocating computational effort to each input using a meta consistency classifier. To calibrate our early prediction stopping rule, we formulate a unique extension of conformal prediction. We demonstrate the effectiveness of this approach on four classification and regression tasks.
Being able to accurately perform Question Difficulty Estimation (QDE) can improve the accuracy of students' assessment and better their learning experience. Traditional approaches to QDE are either subjective or introduce a long delay before new ques
tions can be used to assess students. Thus, recent work proposed machine learning-based approaches to overcome these limitations. They use questions of known difficulty to train models capable of inferring the difficulty of questions from their text. Once trained, they can be used to perform QDE of newly created questions. Existing approaches employ supervised models which are domain-dependent and require a large dataset of questions of known difficulty for training. Therefore, they cannot be used if such a dataset is not available ( for new courses on an e-learning platform). In this work, we experiment with the possibility of performing QDE from text in an unsupervised manner. Specifically, we use the uncertainty of calibrated question answering models as a proxy of human-perceived difficulty. Our experiments show promising results, suggesting that model uncertainty could be successfully leveraged to perform QDE from text, reducing both costs and elapsed time.
The massive spread of false information on social media has become a global risk especially in a global pandemic situation like COVID-19. False information detection has thus become a surging research topic in recent months. In recent years, supervis
ed machine learning models have been used to automatically identify false information in social media. However, most of these machine learning models focus only on the language they were trained on. Given the fact that social media platforms are being used in different languages, managing machine learning models for each and every language separately would be chaotic. In this research, we experiment with multilingual models to identify false information in social media by using two recently released multilingual false information detection datasets. We show that multilingual models perform on par with the monolingual models and sometimes even better than the monolingual models to detect false information in social media making them more useful in real-world scenarios.
Recent transformer-based approaches to NLG like GPT-2 can generate syntactically coherent original texts. However, these generated texts have serious flaws: global discourse incoherence and meaninglessness of sentences in terms of entity values. We a
ddress both of these flaws: they are independent but can be combined to generate original texts that will be both consistent and truthful. This paper presents an approach to estimate the quality of discourse structure. Empirical results confirm that the discourse structure of currently generated texts is inaccurate. We propose the research directions to correct it using discourse features during the fine-tuning procedure. The suggested approach is universal and can be applied to different languages. Apart from that, we suggest a method to correct wrong entity values based on Web Mining and text alignment.
We developed a system for task 6 sub-task 1 for detecting propaganda in memes. An external dataset and augmentation data-set were used to extend the official competition data-set. Data augmentation techniques were applied on the external data-set and
competition data-set to come up with the augmented data-set. We trained 5 transformers (DeBERTa, and 4 RoBERTa) and ensembled them to make the prediction. We trained 1 RoBERTa model initially on the augmented data-set for a few epochs and then fine-tuned it on the competition data-set which improved the f1-micro up to 0.1 scores. After that, another initial RoBERTa model was trained on the external data-set merged with the augmented data-set for few epochs and fine-tuned it on the competition data-set. Furthermore, we ensembled the initial models with the models after fine-tuning. For the final model in the ensemble, we trained a DeBERTa model on the augmented data-set without fine-tuning it on the competition data-set. Finally, we averaged the output of each model in the ensemble to make the prediction.
