The Impact of Quantization on the Robustness of Transformer-based Text Classifiers

Transformer-based models have made remarkable advancements in various NLP areas. Nevertheless, these models often exhibit vulnerabilities when confronted with adversarial attacks. In this paper, we explore the effect of quantization on the robustness of Transformer-based models. Quantization usually involves mapping a high-precision real number to a lower-precision value, aiming at reducing the size of the model at hand. To the best of our knowledge, this work is the first application of quantization on the robustness of NLP models. In our experiments, we evaluate the impact of quantization on BERT and DistilBERT models in text classification using SST-2, Emotion, and MR datasets. We also evaluate the performance of these models against TextFooler, PWWS, and PSO adversarial attacks. Our findings show that quantization significantly improves (by an average of 18.68%) the adversarial accuracy of the models. Furthermore, we compare the effect of quantization versus that of the adversarial training approach on robustness. Our experiments indicate that quantization increases the robustness of the model by 18.80% on average compared to adversarial training without imposing any extra computational overhead during training. Therefore, our results highlight the effectiveness of quantization in improving the robustness of NLP models.

Views Are My Own, But Also Yours: Benchmarking Theory of Mind using Common Ground

Evaluating the theory of mind (ToM) capabilities of language models (LMs) has recently received much attention. However, many existing benchmarks rely on synthetic data which risks misaligning the resulting experiments with human behavior. We introduce the first ToM dataset based on naturally occurring spoken dialogs, Common-ToM, and show that LMs struggle to demonstrate ToM. We then show that integrating a simple, explicit representation of beliefs improves LM performance on Common-ToM.

Finding Common Ground: Annotating and Predicting Common Ground in Spoken Conversations

When we communicate with other humans, we do not simply generate a sequence of words. Rather, we use our cognitive state (beliefs, desires, intentions) and our model of the audience's cognitive state to create utterances that affect the audience's cognitive state in the intended manner. An important part of cognitive state is the common ground, which is the content the speaker believes, and the speaker believes the audience believes, and so on. While much attention has been paid to common ground in cognitive science, there has not been much work in natural language processing. In this paper, we introduce a new annotation and corpus to capture common ground. We then describe some initial experiments extracting propositions from dialog and tracking their status in the common ground from the perspective of each speaker.

Prose2Poem: The Blessing of Transformers in Translating Prose to Persian Poetry

Persian Poetry has consistently expressed its philosophy, wisdom, speech, and rationale on the basis of its couplets, making it an enigmatic language on its own to both native and non-native speakers. Nevertheless, the notice able gap between Persian prose and poem has left the two pieces of literature medium-less. Having curated a parallel corpus of prose and their equivalent poems, we introduce a novel Neural Machine Translation (NMT) approach to translate prose to ancient Persian poetry using transformer-based Language Models in an extremely low-resource setting. More specifically, we trained a Transformer model from scratch to obtain initial translations and pretrained different variations of BERT to obtain final translations. To address the challenge of using masked language modelling under poeticness criteria, we heuristically joined the two models and generated valid poems in terms of automatic and human assessments. Final results demonstrate the eligibility and creativity of our novel heuristically aided approach among Literature professionals and non-professionals in generating novel Persian poems.

Improving Question Answering Performance Using Knowledge Distillation and Active Learning

Contemporary question answering (QA) systems, including transformer-based architectures, suffer from increasing computational and model complexity which render them inefficient for real-world applications with limited resources. Further, training or even fine-tuning such models requires a vast amount of labeled data which is often not available for the task at hand. In this manuscript, we conduct a comprehensive analysis of the mentioned challenges and introduce suitable countermeasures. We propose a novel knowledge distillation (KD) approach to reduce the parameter and model complexity of a pre-trained BERT system and utilize multiple active learning (AL) strategies for immense reduction in annotation efforts. In particular, we demonstrate that our model achieves the performance of a 6-layer TinyBERT and DistilBERT, whilst using only 2% of their total parameters. Finally, by the integration of our AL approaches into the BERT framework, we show that state-of-the-art results on the SQuAD dataset can be achieved when we only use 20% of the training data.

Defeating Author Gender Identification with Text Style Transfer

Text Style Transfer can be named as one of the most important Natural Language Processing tasks. Up until now, there have been several approaches and methods experimented for this purpose. In this work, we introduce PGST, a novel polyglot text style transfer approach in gender domain composed of different building blocks. If they become fulfilled with required elements, our method can be applied in multiple languages. We have proceeded with a pre-trained word embedding for token replacement purposes, a character-based token classifier for gender exchange purposes, and the beam search algorithm for extracting the most fluent combination among all suggestions. Since different approaches are introduced in our research, we determine a trade-off value for evaluating different models' success in faking our gender identification model with transferred text. To demonstrate our method's multilingual applicability, we applied our method on both English and Persian corpora and finally ended up defeating our proposed gender identification model by 45.6% and 39.2%, respectively, and obtained highly competitive evaluation results in an analogy among English state of the art methods.

DeepSentiPers: Novel Deep Learning Models Trained Over Proposed Augmented Persian Sentiment Corpus

This paper focuses on how to extract opinions over each Persian sentence-level text. Deep learning models provided a new way to boost the quality of the output. However, these architectures need to feed on big annotated data as well as an accurate design. To best of our knowledge, we do not merely suffer from lack of well-annotated Persian sentiment corpus, but also a novel model to classify the Persian opinions in terms of both multiple and binary classification. So in this work, first we propose two novel deep learning architectures comprises of bidirectional LSTM and CNN. They are a part of a deep hierarchy designed precisely and also able to classify sentences in both cases. Second, we suggested three data augmentation techniques for the low-resources Persian sentiment corpus. Our comprehensive experiments on three baselines and two different neural word embedding methods show that our data augmentation methods and intended models successfully address the aims of the research.

LexiPers: An ontology based sentiment lexicon for Persian

Sentiment analysis refers to the use of natural language processing to identify and extract subjective information from textual resources. One approach for sentiment extraction is using a sentiment lexicon. A sentiment lexicon is a set of words associated with the sentiment orientation that they express. In this paper, we describe the process of generating a general purpose sentiment lexicon for Persian. A new graph-based method is introduced for seed selection and expansion based on an ontology. Sentiment lexicon generation is then mapped to a document classification problem. We used the K-nearest neighbors and nearest centroid methods for classification. These classifiers have been evaluated based on a set of hand labeled synsets. The final sentiment lexicon has been generated by the best classifier. The results show an acceptable performance in terms of accuracy and F-measure in the generated sentiment lexicon.

Genetic Neural Architecture Search for automatic assessment of human sperm images

Male infertility is a disease which affects approximately 7% of men. Sperm morphology analysis (SMA) is one of the main diagnosis methods for this problem. Manual SMA is an inexact, subjective, non-reproducible, and hard to teach process. As a result, in this paper, we introduce a novel automatic SMA based on a neural architecture search algorithm termed Genetic Neural Architecture Search (GeNAS). For this purpose, we used a collection of images called MHSMA dataset contains 1,540 sperm images which have been collected from 235 patients with infertility problems. GeNAS is a genetic algorithm that acts as a meta-controller which explores the constrained search space of plain convolutional neural network architectures. Every individual of the genetic algorithm is a convolutional neural network trained to predict morphological deformities in different segments of human sperm (head, vacuole, and acrosome), and its fitness is calculated by a novel proposed method named GeNAS-WF especially designed for noisy, low resolution, and imbalanced datasets. Also, a hashing method is used to save each trained neural architecture fitness, so we could reuse them during fitness evaluation and speed up the algorithm. Besides, in terms of running time and computation power, our proposed architecture search method is far more efficient than most of the other existing neural architecture search algorithms. Additionally, other proposed methods have been evaluated on balanced datasets, whereas GeNAS is built specifically for noisy, low quality, and imbalanced datasets which are common in the field of medical imaging. In our experiments, the best neural architecture found by GeNAS has reached an accuracy of 92.66%, 77.33%, and 77.66% in the vacuole, head, and acrosome abnormality detection, respectively. In comparison to other proposed algorithms for MHSMA dataset, GeNAS achieved state-of-the-art results.

SentiPers: A Sentiment Analysis Corpus for Persian

Sentiment Analysis (SA) is a major field of study in natural language processing, computational linguistics and information retrieval. Interest in SA has been constantly growing in both academia and industry over the recent years. Moreover, there is an increasing need for generating appropriate resources and datasets in particular for low resource languages including Persian. These datasets play an important role in designing and developing appropriate opinion mining platforms using supervised, semi-supervised or unsupervised methods. In this paper, we outline the entire process of developing a manually annotated sentiment corpus, SentiPers, which covers formal and informal written contemporary Persian. To the best of our knowledge, SentiPers is a unique sentiment corpus with such a rich annotation in three different levels including document-level, sentence-level, and entity/aspect-level for Persian. The corpus contains more than 26000 sentences of users opinions from digital product domain and benefits from special characteristics such as quantifying the positiveness or negativity of an opinion through assigning a number within a specific range to any given sentence. Furthermore, we present statistics on various components of our corpus as well as studying the inter-annotator agreement among the annotators. Finally, some of the challenges that we faced during the annotation process will be discussed as well.